Dr. Thomas Fink is an Anglo-American physicist, author, and entrepreneur who serves as the founding Director of the London Institute for Mathematical Sciences (LIMS).

He studied physics at Caltech, where he won the Fisher Prize, and earned his PhD at Cambridge University. His research focuses on statistical physics, combinatorics, and evolvable systems, using these tools to gain insights into physical and biological systems.

In 2011, Dr. Fink founded LIMS, the UK’s first independent research centre for theoretical physics and mathematics. Located in the historic Royal Institution in Mayfair, LIMS operates like a university research department but with no teaching or administrative duties, allowing its researchers to focus entirely on full-time, curiosity-driven discovery.

London Institute for Mathematical Sciences (LIMS):

https://lims.ac.uk/

Dr. Thomas Fink’s Profile at LIMS: https://lims.ac.uk/thomas-fink/

Chapter Markers:

• (00:00) - Thomas Fink: Caltech, Cambridge, CNRS Paris

• (07:27) - Rethinking Academia

• (13:05) - Building LIMS

• (31:18) - Saving Basic Science

• (33:46) - Endowments Versus Governments

• (40:28) - AI Reshaping Research Culture

• (52:17) - Closing Thoughts

Audio-only version and transcript:

https://www.manifold1.com/episodes/how-to-build-a-research-institute-thomas-fink-of-the-london-institute-for-mathematical-sciences-119

Note: This intellectual history was produced by GPT, with additional fact-checking and corrections by Qwen and Gemini. The project is at once a test of deep-research capabilities and an exercise in self-indulgent narcissism. I asked GPT to search my online writing, academic papers, podcast interviews, and media coverage, and to synthesize that material into the essay that follows. As far as I can determine, the quotations are accurate and the interpretations defensible. The model completed the project in about an hour of sustained work across multiple turns. A human historian or biographer would have required much longer to complete the task. Update1: I had free access to the frontier “mystery model” Ox Alpha (Aug 23 2026; model turns out to be GLM-5.3-Flash). This model is very intelligent and also very good at long-horizon agentic tasks. So I asked it to improve the essay with more quotes (full excerpts), and to clarify some clumsy language in the original. Update2: Final version below was revised by Opus-5 after additional research.

A life can be misread by its nouns. Theoretical physicist. Silicon Valley founder. Computational genomicist. University research executive. Public intellectual. Builder of artificial-intelligence systems. Documentary filmmaker. Set side by side, the titles suggest restless polymathy, a man moving from one absorbing subject to another. They miss the verb that binds them: to make.

Again and again, Stephen Hsu has been drawn to ideas poised between theory and science fiction—not fantasies, but possibilities waiting upon some missing threshold. Has the cost of measurement fallen far enough? Has the dataset grown large enough? Has the necessary mathematics already been invented in another field? Can an institution be built around the answer? Hsu reconstructs the problem from first principles, searches for the hidden constraint, and crosses whatever disciplinary boundary stands between the idea and its realization.

What unifies the career is a way of moving through the world. In physics, he asks what can be known when quantum mechanics, gravity, and cosmology press against one another. In genomics, he asks how much of a human future lies encrypted in DNA, and how large a dataset is needed to read it. In entrepreneurship, he turns a technical possibility into a working system. In university leadership, he confronts the problem of organizing talent and capital at scale. In artificial intelligence, the object of inquiry begins to answer back, becoming a collaborator in discovery.

Hsu possesses an unusual feeling for the ripening hour of a problem. He neither pursues difficulty merely because struggle is noble nor waits until fashion has made an idea safe. He watches for the moment when an old impossibility becomes newly soluble—when cheaper sequencing, larger biobanks, greater compute, better algorithms, or more powerful models bend the curve. Before committing years to genomics, for example, he estimated whether realistic sample sizes should suffice. When the data crossed the predicted threshold, his group moved quickly and produced accurate predictors. This is ambition governed by calculation: audacity with a theory of when to act.

The distinction matters because Hsu’s projects have repeatedly escaped the realm of speculation. SafeWeb pioneered technology later acquired by Symantec. His group’s mathematical estimates in genomic prediction were followed by out-of-sample prediction of human height. Genomic Prediction and Othram carried population genetics into clinical and forensic practice. His years at Michigan State placed him inside the machinery of a major research university. More recently, he has brought generative AI into theoretical physics itself, while Superfocus seeks to make language-model systems reliable enough to act in the world. These undertakings differ in scale, maturity, and moral weight. What joins them is a recurring passage from the imaginable to the actual.

This essay uses the Grokipedia biography as its biographical spine, while drawing on Hsu’s scientific papers, essays, institutional biographies, X posts, and many podcast transcripts to recover his development in his own words. Those primary sources do more than add color: they alter the portrait. Hsu’s declared subject is intelligence; his deeper subject is what intelligence can make of resistance—the limits imposed by nature, data, institutions, convention, and mortality, and the strange opening that appears when a mind sees those limits clearly enough to act through them.

The fuller record also reveals a second axis, though it is less settled than it first appears: the question of inheritance amid extraordinary technical change. Family, remembered faith, physical courage, literature, and the continuity of the human line are not ornaments around the technologist. They help tell him which futures are worth making. Yet his loyalty is not simply to present biological humanity. At civilizational scale he can imagine enhanced humans, human–machine mergers, and even artificial minds as descendants. The question is therefore not only whether humanity survives, but what—love, memory, or agency—must pass through the transformation for the future still to count as ours.

1. Ames: a prodigy in the ordinary world

Hsu was born in 1966 and raised in Ames, Iowa, the son of Chinese immigrants. His father, Cheng Ting Hsu, was an aerospace engineering professor at Iowa State University. Ames gave him an unusual combination: the ordinariness of a Midwestern university town and early access to a serious scientific environment.

His first university course came at twelve: computer science at Iowa State, taken while he was still in middle school. The mathematics and physics came later, in high school—quantum mechanics, differential equations, linear algebra, complex analysis—after he became the first student at Ames High School permitted to enroll for university credit.

That access began at home. Recalling his father in a From the New World interview, Hsu gives the decisive object an almost talismanic glow:

“He had what then, in the pre-internet era, was—for a precocious kid like me—the magic secret: a library card at the university library.”

— Brian Chau interview, From the New World

Before the internet, a library card was not a convenience but a passage into worlds otherwise sealed off by age and geography. Hsu’s intellectual life began not merely with precocity, but with premature access to the archive of adult knowledge.

The archive was not his only company. Hsu has also recalled a comparably gifted Ames contemporary, later an MIT mathematics Ph.D., and a mathematically sophisticated professor living nearby. That small ecology complicates the legend of the solitary prodigy. Ames gave him not only books but early calibration: a peer against whom rare ability became visible, an adult who could recognize it, and a university whose doors were near enough to open. His self-education was exceptional, but it was socially scaffolded.

He graduated from high school at sixteen. He graduated from Caltech at nineteen. Yet the story he tells is not the familiar memoir of the isolated prodigy. He was a competitive swimmer and high-school team captain; he describes his childhood as recognizably “all-American.” Hsu’s later interest in ability was formed not only through books, scores, or mathematical competitions but also through athletics, where differences in speed, coordination, endurance, and trainability are visible, repeatedly measured, and difficult to explain away.

His childhood exposure to psychometrics was unusually early and oddly concrete. In third or fourth grade he took Iowa’s required standardized tests and scored 99th percentile on every battery in the booklet. He lacked the word, but he had intuitively grasped the concept of correlation: if each battery’s ceiling was the 99th percentile, then 99s across four or five of them implied odds he found implausible—“Am I really one in ten to the eighth in capability? That can’t be right”—so the scores had to be measuring something shared. His father’s library card then delivered the Terman studies and the technical literature on giftedness. There was even, he later realized, “an amazing life coincidence” buried in the shelves: the psychometricians Camilla Benbow and David Lubinski began their careers at Iowa State, and their longitudinal work on mathematically precocious youth was among the most cited in the field—which is partly why an Ames library was so well stocked on the subject at all. He began treating his own social world as a longitudinal study, complete with what he called—borrowing a term from general relativity—fiducial observers: friends on whom he “made good calibration measurements” in high school and whom he could call decades later to compare notes on how life went. Later encounters with elite physicists, Olympiad-level mathematicians, athletes, entrepreneurs, investors, and administrators broadened the archive.

— Brian Chau interview, From the New World

This background explains both a strength and a recurring controversy in Hsu’s thinking. He is unusually willing to speak about the tails of human variation because he believes he has observed them at high resolution. He distrusts accounts of achievement that erase natural differences. But he also knows, from movement across domains, that ability is not a single scalar. Mathematical speed, scientific originality, athletic talent, persuasion, emotional perception, courage, conscientiousness, and executive judgment are separable capacities. His own career would make little sense under a theory in which test-measured intelligence alone determined outcomes.

At Caltech, Richard Feynman supplied a model of scientific independence. The famous graduation photograph of the nineteen-year-old Hsu beside Feynman is more than biographical decoration. Hsu had chosen Caltech partly because of Feynman—as he later wrote, he had been a “Feynman idolator” since high school: “In fact, I chose my college (Caltech), career, and even research specialization under his influence!” In “Feynman and the Secret of Magic,” he makes the hierarchy of his admiration explicit:

“Lubos is upset that I might think that Schwinger was, at least in some ways, ‘smarter’ than Feynman. Even so, Feynman is my hero, not Schwinger. Feynman had no rival in his generation when it came to originality and creativity.”

— “Feynman and the Secret of Magic”

The sentence concedes what it asserts. Hsu was defending the claim that Schwinger was “smarter”—more comprehensive, faster, deeper in the literature—and choosing Feynman anyway. The choice is a theory of what matters in a physicist, and it is not raw power.

The vocabulary of “magicians” that recurs throughout Hsu’s writing is not his coinage. It comes from a passage he has called one of his favorites, the mathematician Mark Kac’s division of genius into two kinds:

“There are two kinds of geniuses, the ‘ordinary’ and the ‘magicians.’ An ordinary genius is a fellow that you and I would be just as good as, if we were only many times better. There is no mystery as to how his mind works. Once we understand what they have done, we feel certain that we, too, could have done it. It is different with the magicians. They are, to use mathematical jargon, in the orthogonal complement of where we are and the working of their minds is for all intents and purposes incomprehensible. Even after we understand what they have done, the process by which they have done it is completely dark. Richard Feynman is a magician of the highest caliber.”

— Mark Kac, Enigmas of Chance

Hsu’s gloss—“We all stand in awe of the magicians!”—is itself a small self-portrait. The exclamation point belongs to a man who has spent his life trying to identify that orthogonal complement in others: in Olympiad teammates, in colleagues, in founders, in his own children.

Yet he also notices the risk in Feynman’s refusal to read the literature. The passage he chose to illustrate it is Sidney Coleman’s—the Caltech theorist who knew Feynman at close range:

“There are lots of people who are too original for their own good, and had Feynman not been as smart as he was, I think he would have been too original for his own good... He was like the guy that climbs Mont Blanc barefoot just to show that it can be done... Dick could get away with a lot because he was so goddamn smart. He really could climb Mont Blanc barefoot.”

— Sidney Coleman

Independence can become ignorance, and originality can generate dead ends—and the climber who survives barefoot is not proof that shoes are unnecessary. Hsu’s mature method is not Feynman’s pure intellectual individualism. He reconstructs from first principles where he can, borrows provisional knowledge where he must, and keeps track of which is which.

His contact with Feynman was active rather than merely devotional. As an undergraduate officer in the Society of Physics Students, Hsu invited him to give a special seminar on the EPR paradox and took him to lunch afterward. In Hsu’s later recollection, Feynman was then exploring negative probabilities as a route through quantum strangeness. The episode foreshadows Hsu’s adult role: identify a foundational question that respectable routines leave aside, bring the right minds into the room, and insist that the strange possibility deserves a hearing.

2. The habit of first principles

The most illuminating description of Hsu’s intellectual method appears in the Information Theory podcast transcript:

“I should be able to sit down with a piece of paper or whiteboard and actually kind of work it through from first principles.”

— Information Theory podcast

In another interview on his movement across disciplines, he names the common structure beneath his subjects:

“I guess the unifying theme is knowledge versus uncertainty: the attempt to capture the essential aspects of a messy system in a simplified mathematical model.”

This is close to a personal credo. The model must be simple enough to expose the decisive relation, but the scientist must remember that simplification has purchased clarity by discarding detail. Hsu’s confidence comes from finding structure; his best skepticism is directed at the boundary where structure may have been mistaken for the world.

That skepticism has a source older than his professional science. It came from a childhood educated in two classrooms that gave contradictory accounts of the same events.

The first classroom was his father’s memory. Cheng Ting Hsu had been admitted at sixteen to the wartime university at Kunming—the Southwest Associated University formed from Tsinghua, Beijing, and Nankai, the institution that produced C. N. Yang and T. D. Lee—had studied aerodynamics, served as a KMT officer and briefly instructed pilots at the air force academy, and then won one of only two Ministry of Education fellowships in his field for graduate study in America. He never returned. His family remained in Zhejiang and lived through the communist takeover, the Great Leap Forward, and the Cultural Revolution. Hsu told the story behind the thin envelopes on Manifold:

“My dad’s family was suffering the Cultural Revolution because they were still in Zhejiang when I was a kid—so late ’60s, early ’70s. My dad would get these letters from his relatives in China, and they were written on the thinnest, cheapest, shittiest paper—that’s all you could get in communist China at that time—and they would write very densely on these little letters. But those letters were priceless to my dad because they were his only contact with his family back home. He would spend time telling me about how terrible the Cultural Revolution was and what his family was going through, and how it made people into animals—brothers did terrible things to each other, and to the father, to the parents, taking stuff from their house. Just really terrible stuff.”

— Manifold, “Deus Ex Machina,” September 2024

The second classroom was Ames itself. “If you’re an intellectual kid and you grew up in that era,” Hsu recalls, “a lot of intellectuals were pro-communism, pro-leftism. There was a kind of romanticization of both the Soviet Union and Communist China by leftists here in the United States.” Some of his friends’ parents were professors of exactly this persuasion, and the boy was a welcome guest at their tables:

“I would come home—I’d be at their dinner, hearing all this stuff about how bad capitalism is and American imperialism. I come home and talk to my dad, and my dad would be like: those people don’t know fuck all about what they’re talking about. Your relatives are being screwed over. Well—he would never use language like that—but your own relatives are suffering in China right now, and these people know nothing.”

— Manifold, “Deus Ex Machina,” September 2024

The fashionable view had, in fact, briefly captured the boy himself. Talking with Yasheng Huang, the MIT economist raised in Maoist China whose own grandfather was an early Communist Party member, Hsu supplied the confession:

“When I was growing up, I used to say things to him like, well, at least the communists are making China strong again, or something. And he would just shake his head and tell me how terrible the communists were—and the Cultural Revolution was going on at the time, so he was on the complete opposite side of this issue from me.”

— Manifold #45 with Yasheng Huang

The structure of the episode deserves attention, because it is not the standard immigrant-child-knows-better story. The boy repeated what the most credentialed adults in his world said, because it sounded reasonable and carried institutional authority. The man with access to the primary sources—one densely written envelope at a time—knew it was false. When the two accounts collided, Hsu drew the epistemic conclusion he still applies half a century later:

“In the modern era, I am very, very able to discount ‘expert’ opinion—because these professors, these ‘experts,’ can be one hundred percent wrong in very strongly held beliefs.”

— Manifold, “Deus Ex Machina,” September 2024

He is careful, in his analytic way, to make the lesson precise rather than merely resentful. The failure was not lack of information, which circumstances could excuse, but something less forgivable:

“In those days you could make excuses for the lack of information that a Harvard professor would have about what was going on in China—they might not be able to go there, or if they went there they might be tightly controlled in what they could see. You can apologize for their lack of knowledge. You cannot apologize for the level of conviction that they have conditional on their level of knowledge. People using what I call the calculus of words—no data, no equations, no analysis—you’ll find are just incredibly overconfident, based mainly on their feels. That’s all the word calculus is. That’s all these people have. And so they’re constantly miscalibrated.”

Conviction held constant while evidence approached zero: the word miscalibrated is doing precise work here. The lesson also explains why Hsu extends trust anywhere at all. Expertise earns standing only where reality enforces its judgments quickly:

“It’s only in a few technical fields where, if you say something that’s wrong, the facts or the mathematics are going to punch you in the face right away—it’s only in those subdisciplines where, if someone’s an expert, that means something. In all these other fields—all my friends with PhDs in history and other subjects—you will find people who are supposed to be the world’s experts on a particular topic, and they’re literally 180 degrees off.”

And the lesson cuts against every ideology, including his family’s. When Hsu posts on X that Chinese infrastructure is real, or that Chinese electric cars are good, readers call him a CCP apologist. His answer compresses the whole biography into three sentences:

“Remember, my dad taught me—before any of you guys were born—about how shitty the communists were and what Mao did to our family. Okay, remember that. So if I tell you they do actually seem to have pretty good infrastructure, it’s not because I’m pro-communist. It’s because I actually want to understand the world as it is, not how your ideology wants it to be.”

That last sentence is the first half of his Gramscian motto rendered as autobiography—be a scientist: see the world as it really is—taught to him decades before he found the formula. The same discipline that forbade the Ames professors’ fantasy of Red China forbids the mirror-image fantasy of inevitable collapse. Reality does not care which direction of error flatters your politics.

There is a coda. When Hsu finally visited his father’s homeland in 2010, his uncle—a retired Tsinghua professor—and his cousins in Hangzhou gave him a four-volume family history originally printed in the 1930s, recording the Xu lineage back to the tenth century BC, with his father entered as the 113th generation. The revolution had redistributed the family’s property and scattered its members, but it had not managed to confiscate the family’s records. The archive outlasted the ideology.

— “Three Thousand Years and 115 Generations of 徐”

Taken alone, this can sound like the standard rhetoric of technically minded entrepreneurs. Hsu’s fuller account is subtler. He says that when he enters a field, he attempts to organize it into a coherent logical structure, searches for foundational gaps, and marks assumptions whose evidential status is weaker than practitioners admit. But he does not demand mathematical rigor at every node before proceeding. He “coarse-grains” over some areas, provisionally accepts a stylized fact, and keeps an alternative map ready in case data invalidate it.

This is a powerful description of actual scientific reasoning. Pure deduction cannot move through empirical sciences because many premises remain contingent, approximate, or incompletely measured. Pure empiricism cannot distinguish a meaningful anomaly from noise because it lacks a structural model. Hsu’s approach builds a hierarchy of confidence: derive what can be derived, borrow what must temporarily be borrowed, remember which is which, and revise without embarrassment.

The same cognitive style drives his disciplinary mobility. He does not approach a new field by slowly absorbing all its conventions. He looks for its governing variables, scaling relations, information bottlenecks, and unexamined assumptions. This gives him an advantage over insiders whose knowledge is locally deeper but structurally less explicit. It also makes his criticisms sound abrasive. What appears to an insider as accumulated craft knowledge may appear to Hsu as an unjustified prior; what appears to Hsu as a simple information-theoretic question may depend on biological complexities he has compressed away. His best work occurs when the abstraction preserves what is decisive and discards what is not.

The deeper motivation is not winning arguments or moving quickly. Near the end of a long Undertone interview, Hsu reflects that he might have accumulated much greater wealth by leaving fundamental science earlier. His answer is that intellectual mastery has intrinsic value. The comment clarifies the whole trajectory: commercial success matters, but it does not replace the private satisfaction of closing the gap between elementary understanding and a research frontier. Hsu’s mobility is therefore not dilettantism. Each serious migration—physics, computation, genomics, AI—requires him to build another coherent internal world.

This distinction separates his ambition from simple career maximization. He is highly responsive to leverage and opportunity, but he is not optimizing a single public score. Reputation, money, discovery, institutional power, and mastery are different goods. His life has repeatedly traded one for another.

The risk in all this is overgeneralization. Spectacular ideological error can make all nontechnical expertise look less trustworthy than it is—and Hsu’s own “facts will punch you in the face” criterion concedes as much, marking off the corrigible fields from the rest rather than condemning every seminar room in Ames. He sometimes writes as though mathematical or technological fields are uniquely corrigible and most other expert cultures are merely protected error. The distinction has force, but it is one of degree: technical communities also follow fashion, conceal uncertainty, and allocate attention institutionally. His own writing on quantum foundations and research reproducibility supplies the counterexamples.

3. The inner code: discipline, mortality, and the life of this world

Hsu’s literary tastes disclose the moral psychology behind his scientific method, but they are only one strand in a more complicated inheritance. His father was, in Hsu’s phrase, almost literally a Confucian scholar: cerebral, restrained, devoted to books and technical work. His mother came from a military and athletic Kuomintang family; her father had trained in Japan alongside Chiang Kai-shek, and she encouraged the competitive swimming and judo absent from his father’s world. Her family had converted to Christianity in the nineteenth century, and Hsu was raised Methodist in Ames. This double inheritance—scholar and athlete, materialist analysis and remembered faith—helps explain a personality in which abstraction, physical courage, self-command, and metaphysical unease coexist.

Three writers gave that inheritance an adult vocabulary: Marcus Aurelius, Ernest Hemingway, and James Salter—unusual companions, but coherent ones.

Marcus supplies distance from reputation. In “Happiness,” an essay prompted by a Big Five test that had placed him at the 99th percentile for emotional stability, Hsu offers what he calls his favorite bit of advice for academics. It is Marcus Aurelius on the bubble reputation, quoted in full:

“Or does the bubble reputation distract you? Keep before your eyes the swift onset of oblivion, and the abysses of eternity before us and behind; mark how hollow are the echoes of applause, how fickle and undiscerning the judgments of professed admirers, and how puny the arena of human fame. For the entire earth is but a point, and the place of our own habitation but a minute corner in it; and how many are therein who will praise you, and what sort of men are they?”

— Marcus Aurelius, quoted in “Happiness”

The passage works like a zoom lens pulling back from the seminar room to the cosmos: the applause hollow, the arena puny, the earth a point, the habitation a minute corner of it. Its most quietly devastating question is the last one. Who are these admirers? What sort of men are they? Marcus does not deny that reputation exists; he denies it weight, and then denies it standing, by making the admirer—not the applause—the object of scrutiny. Hsu’s attraction to the passage is not difficult to understand. Academic life is intensely status-conscious while pretending not to be; technical communities confuse consensus, prestige, and citation with truth at their peril. Stoic distance becomes a cognitive tool. If reputation is transient, one can admit ignorance and criticize fashionable assumptions. One can enter a field where one lacks standing, or leave a prestigious track for work that seems more consequential.

The essay in which the quotation appears also discloses its source. “It’s also true that my father passed away while I was still fairly young,” Hsu writes, “so I had the impetus to consider his life in its entirety and to evaluate which of the things he did really mattered, and which didn’t.” Marcus’s zoom lens was, for Hsu, not an abstraction but a grief technology: the discipline of seeing a life entire, measured against the value that finally mattered. The same exercise returns near the end of his reflections on family, in the “90 percent” observation discussed below.

Hsu describes himself as temperamentally happy, low in neuroticism, and usually eager for the day. The discipline of Marcus keeps an already energetic temperament independent of applause. His pessimism is methodological; his baseline mood is not.

Hemingway supplies courage after illusion has been removed. For years the slogan on Hsu’s homepage was the Gramscian formula “pessimism of the intellect, optimism of the will,” and in his explanation of it he defines the epistemic half as nothing more than the scientific attitude itself:

“Pessimism of the Intellect means, simply, be a scientist: see the world as it really is, not as you might like it to be. Try to identify and overcome hidden biases or prior assumptions. Always ask yourself: What assumption am I making? What if it is incorrect? How do I know what I know? In many cases, the correct answer is: I don’t know. Never be afraid to admit you don’t know.”

— “Pessimism of the Intellect, Optimism of the Will”

He did not learn this from Gramsci. He learned it at the dinner tables of Ames, watching men with Harvard convictions describe a China his father knew through the thinnest paper in the world.

The volitional half he states in a single line—“Optimism of the Will means, have the courage to attempt difficult things. Sometimes, Will can overcome the odds.”—and then illustrates with two passages from Hemingway, chosen with a connoisseur’s precision. The first, from A Farewell to Arms, states the price of courage with a fatalist’s arithmetic:

“If people bring so much courage to this world the world has to kill them to break them, so of course it kills them. The world breaks everyone and afterward many are strong in the broken places. But those that will not break it kills. It kills the very good and the very gentle and the very brave impartially. If you are none of these you can be sure it will kill you too but there will be no special hurry.”

— Ernest Hemingway, A Farewell to Arms

The second is Santiago’s vow from The Old Man and the Sea, and it is the answer to the first—the entire motto compressed into a single line of defiance:

“I’ll fight them, I’ll fight them until I die.”

— Ernest Hemingway, The Old Man and the Sea

Read together, the two quotations are the motto in narrative form. The world breaks the brave impartially; the sharks will take the marlin; I’ll fight them until I die anyway. Clear sight does not entail passivity. Optimism of the will is a decision rule, not a forecast: a refusal to let an unfavorable prior excuse inaction when agency can still change the distribution of outcomes.

Hemingway names the moment of courage; the Finnish word sisu names its duration. Writing after seeing Citizenfour, Hsu adopted the untranslatable term for the grimmer endurance that difficult work requires:

“Sisu is a Finnish term loosely translated into English as strength of will, determination, perseverance, and acting rationally in the face of adversity. However, the word is widely considered to lack a proper translation into any other language. Sisu contains a long-term element; it is not momentary courage, but the ability to sustain an action against the odds. Deciding on a course of action and then sticking to that decision against repeated failures is sisu. It is similar to equanimity, except the forbearance of sisu has a grimmer quality of stress management than the latter.”

— “Sisu”

He closed the post by signing it, in effect, with his motto: Pessimism of the Intellect, Optimism of the Will. The distinction between dramatic bravery and sustained action is central to his temperament. A difficult project is rarely conquered in one heroic instant; it is carried through long periods when the reward is distant, the social signal is adverse, and failure repeats itself.

Salter supplies intensity, style, and an aristocratic sense of life. Hsu discovered him through A Sport and a Pastime, and his praise is unbounded: “I can’t think of higher praise than to say I’ve read every bit of Salter’s work I could get my hands on.” He ranks Salter with his other literary hero: “Salter evokes Americans in France as no one since Hemingway in A Moveable Feast.” The title of his essay on Salter borrows the Koranic line from which Salter took his own title—“Remember that the life of this world is but a sport and a pastime”—a sentence that could stand as an epigraph for Hsu’s whole account of finite, mortal ambition.

What he most admires, he set down in a message to a friend who had known the writer. The passage deserves quotation in full, because its confessions are as revealing as its praise:

“About 5 years ago I became friends with the writer Richard Ford, who offered to introduce me to his friend Salter. I was less enthusiastic to meet him than I would have been when he was younger. I did not go out of my way, and we never met. Since he lived in Aspen, and I was often there in the summers at the Physics institute, I have sometimes imagined that we crossed paths without knowing it. I admire, of course, his prose style. Sentence for sentence, he is the master. But perhaps even more I admire his view of the world—of courage, honor, daring to attempt the impossible, men and women, what is important in life.”

— “A Sport and a Pastime”

It is a small self-portrait: the physicist’s summer calendar, the missed connection regretted, the two-tier admiration in which style comes first “of course,” but the view of the world—courage, honor, daring, men and women, sexual and emotional vividness—comes even before it.

The Salter passages Hsu chooses to reproduce confirm the hierarchy. From A Sport and a Pastime he quoted a portrait of a young man who walked away from Yale:

“He describes it casually, without stooping to explain, but the authority of the act overwhelms me. If I had been an underclassman he would have become my hero, the rebel who, if I had only had the courage, I might have also become. ... Now, looking at him, I am convinced of all I missed. I am envious. Somehow his life seems more truthful than mine, stronger, even able to draw mine to it like the pull of a dark star.”

— James Salter, A Sport and a Pastime

The narrator’s envy of a life “more truthful than mine”—stronger, gravitational, a dark star—reads almost as Hsu’s own confession about his counterfactual selves: the quant, the full-time founder, the man who never left the Aspen summers behind. The details he chose to reproduce are pointed as well: “He had always been extraordinary in math. He had a scholarship. He knew he was exceptional. Once he took the anthropology final when he hadn’t taken the course. He wrote that at the top of the page. His paper was so brilliant the professor fell in love with him.” Salter’s rebel is a prodigy who refuses the game because it is too easy—and of all the sentences in the novel, those are the ones the student of exceptional ability quoted. The fascination is the confession.

Marcus prevents the heroic temperament from becoming dependent on applause; Hemingway insists on conduct under pressure and after illusion; Salter reminds it that finite life should be lived intensely rather than merely optimized, and that the deepest admiration is reserved not for sentences but for a way of being.

The remembered faith remains alive even though Hsu’s explicit metaphysics is materialist. In another conversation he put the residue plainly:

“But I still have this kind of spirituality or wonder left over. That feeling dominated my worldview when I was very young.”

— Manifold conversation with Aella

The older Hsu does not posit an intervening deity; he follows evidence and treats minds as physical systems. Yet church music or a cathedral can still awaken the intuition that the visible inventory is incomplete. He retains an open question about whether atoms and bits exhaust what happens to a person at death. This is reverence without doctrinal certainty, reductionism without immunity to awe—which is why his materialism never acquires the affect of disenchantment.

The deepest correction to the image of Hsu as an achievement-maximizer comes when the subject turns to family. His father was old when Hsu was born, and the child calculated early that death might come while he himself was still young. After his father died, Hsu could see the whole career—books, papers, professorship—against the value that had finally mattered most to the man who lived it. Reflecting on his father and his own children, he says in a Manifold conversation:

“Almost any ordinary human who can have a family and raise their children really has experienced maybe 90 percent of the great stuff.”

— Manifold

The thought is Ecclesiastes entering a Stoic life: a career capable of filling a biography can still occupy a subordinate place in the private order of value. More surprisingly, it is a theory of moral equality—not equality of capacity or achievement, but broad equality of access to the deepest human goods. Cognitive powers may be distributed unequally, public achievement more unequally still, yet children, attachment, memory, and the felt texture of a life shared with others remain available far beyond the elite tail. The comment changes the portrait. Mastery and action matter enormously to Hsu, but they are not the ultimate court of appeal.

Together, the three writers illuminate Hsu’s characteristic combination of severity and aspiration. The scientist must see without consolation. The agent must act without certainty. The individual must not confuse public reward with internal value. And a life should contain difficult achievements because mastery and daring are constitutive goods, not merely instruments for status.

This literary framework corrects a possible misunderstanding of Hsu’s appetite for ambitious projects. He is not attracted to impossibility for its own sake. His projects typically begin with a tractability judgment. The courage he admires is the courage to commit when success is uncertain but the causal pathway is real.

4. Physics at the limits of the knowable

Hsu took his B.S. at Caltech in 1986 and his Ph.D. at Berkeley in 1991, then moved through a Harvard Junior Fellowship to faculty positions at Yale and the University of Oregon. His official Michigan State biography lists research spanning quantum chromodynamics, black holes, entropy bounds, dark energy, cosmology, particle physics beyond the Standard Model, quantum foundations, genomics, finance, encryption, and information security.

The diversity of topics conceals a recurring question: what limits the extraction, localization, preservation, or interpretation of information?

In work on dense quark matter and physics beyond the Standard Model, the problem is how effective descriptions change across energy or density regimes. In black-hole physics, it is whether information is destroyed, hidden, decohered, or distributed across degrees of freedom inaccessible to ordinary observers. In cosmology, it is how global descriptions, entropy, vacuum structure, and observational selection constrain what can be inferred. In quantum foundations, it is what the formalism says about observers and branches when collapse is not treated as fundamental.

A concise example is his work with Xavier Calmet and Michael Graesser on minimum length. Their paper, “Minimum Length from Quantum Mechanics and Classical General Relativity,” concludes:

“Our results imply a device independent limit on possible position measurements.”

— Calmet, Graesser, and Hsu, “Minimum Length from Quantum Mechanics and Classical General Relativity”

The argument joins quantum localization to gravitational collapse: concentrating enough energy to resolve an arbitrarily small region eventually creates a black hole. What looks like a limit of instrumentation becomes a limit implied by the joint structure of quantum mechanics and gravity. The result comes not from a complete theory of quantum gravity but from forcing two well-established frameworks to constrain each other—a signature move.

His attitude toward foundational difficulty is less impatient than his polemical style can suggest. The retraction came, fittingly, by way of a quotation—Michael Nielsen’s memoir of a quantum foundations conference, which Hsu excerpted with the note that he “particularly liked” it:

“At the time, I thought the prevalence of the question suggested that little genuine progress was being made in quantum foundations, and people were merely spinning their wheels. Later, I realized that assessment was too harsh. The speakers were wrestling with some of the hardest problems human minds have ever confronted. Of course progress was slow! ... Understanding neural networks in their full generality is a problem that, like quantum foundations, tests the limits of the human mind.”

— Michael Nielsen

The exclamation carries humility as well as admiration, and the sentence Hsu appended—placing neural networks and quantum foundations side by side as problems that “test the limits of the human mind”—shows the 2014 post doubling as a forecast of his own later movement toward AI. Hsu can be savage about muddle, but he can also revise his verdict when slowness reflects the depth of the problem rather than institutional torpor.

Quantum foundations also became, for Hsu, a case study in the sociology of knowledge. Copenhagen remained the classroom default while many leading theorists leaned toward Everett once they considered the universal wavefunction seriously. Consensus may describe what a profession routinely teaches more accurately than what its deepest thinkers believe. The claim links physics to institutional analysis: foundational questions can be marginalized without being answered.

— “Feynman and Everett”; Hsu’s 2012 quantum correspondence

His physics is strongest in such boundary regions—where one can say something general before possessing the final theory. That preference anticipates his later work in genomics. In both cases, he asks whether broad structural reasoning can establish what is possible, impossible, or sample-limited before every mechanism is known.

It also accounts for why fundamental physics eventually shared his attention with faster-moving domains. Mastering quantum field theory retained its private value, but experimental access to quantum gravity and high-energy frontiers was remote. Biology, computation, and later AI offered steeper technological gradients. Hsu did not abandon physics; he redistributed effort toward domains in which theory could meet rapidly expanding data and produce shorter feedback loops. A decade later, generative AI would unexpectedly shorten a feedback loop inside theoretical physics itself.

5. The founder’s turn: knowledge becomes action

The founding of SafeWeb around 2000 was the first major break in Hsu’s academic trajectory. SafeWeb built internet privacy and SSL VPN technology that Symantec acquired in 2003. Hsu later founded Robot Genius, which worked on malware protection. These episodes altered his account of how knowledge becomes effective.

SafeWeb began with hacked Linux machines in the Yale physics department. Hsu and a graduate student noticed that the browser’s built-in SSL engine could become the universal endpoint for a new kind of virtual private network—obvious in retrospect, not yet built. The company first won millions of consumer users, then discovered that bandwidth costs and the immature advertising market made popularity unprofitable. It survived by making a painful right-angle turn toward enterprise security.

— Mixergy interview

The episode taught Hsu not only leverage but responsibility. Recalling the afternoon he had to dismiss people he had recruited, he gave the founder’s burden its severest form—and the scene is worth hearing in full:

“I’ll never forget how it was a beautiful, sunny, idyllic day. We were standing next to the Bay but I was firing five or ten guys. And some of these guys were people I had known for years and one of the guys actually started crying. ... Every time you hire somebody you should picture that you might have to fire them. That forces you to be careful in the hiring because the most painful thing for me at least, as a CEO, that I ever had to do was fire somebody. And you can’t... you’re not a man if you delegate that. You hired him, you brought him in, you’ve got to face him and tell him what’s going on.”

— Mixergy interview

For Hsu, one may not delegate the moral fact of a decision whose authority one claimed.

The idyllic weather and the weeping engineer are the point: entrepreneurship widened his idea of courage—Hemingway’s courage, conduct under pressure after illusion has been stripped away—to include accepting the human cost of adaptation when the original plan fails. A correct technical idea remains only one input into realization. A founder must recruit, allocate, persuade, decide under uncertainty, survive adverse selection by investors and markets, and fit a new capability into existing workflows.

The founder’s turn was not inevitable. In his twenties Hsu came close to quant finance, recasting exotic-option pricing in the language of Feynman path integrals. A Harvard Junior Fellowship and an aversion to Manhattan helped keep him in physics. And the choice between worlds was once offered to him explicitly: a venture investor who had dined with him called to say, “We really like you. We love the company. We think it’s a great opportunity. We want to put the money in. But we’re not putting it in unless you’re CEO.” Hsu returned to the university and to physics; the company sold for less than it might have. Asked whether he second-guesses it, he answered: “I’m pretty happy with the way things turned out... But, hey. Life is like that, you know. You can’t really second guess.” The counterfactual resists retrospective myth: a career that now appears architecturally coherent was also bent by prestige, geography, temperament, and luck.

— Mixergy interview

This experience adds a second axis to Hsu’s conception of intelligence. Academic culture privileges analytic depth and publication. Startups expose execution, social judgment, risk tolerance, and speed. Later, as an administrator, Hsu would say that startup experience teaches difficult decision-making under pressure. Across these roles he encountered forms of ability that psychometric discussion often leaves out: the capacity to coordinate other minds and reshape an institution.

He is candid about the price of range, too. The fox notices connections the hedgehog misses; the hedgehog may produce the “deep-time” contribution. Hsu wonders whether even von Neumann’s breadth carried that cost. Polymathy is not uncomplicated praise: translation may disperse the concentration required for one monumental result.

Founding also sharpened his sense that talented minds can be diverted into games beneath their powers. In a conversation about intellectuals and the technosphere, Hsu remarks:

“People who come from science or math backgrounds and end up in finance—in a way it kind of dumbs them down.”

The line is deliberately provocative, but its governing emotion is regret. Civilization has only so many people able to work near the frontier; prestige and compensation draw them toward the redistribution of claims rather than the creation of new capabilities. Founding companies allowed Hsu to seek leverage without surrendering the builder’s criterion: something new must exist afterward.

Entrepreneurship intensified his impatience with static organizations as well. To Hsu, an institution is not simply a community governed by norms; it is an information-processing and decision-making system. Incentives determine which signals travel upward, who can act, how quickly errors are corrected, and whether exceptional people receive resources. A slow hierarchy may possess immense knowledge yet remain collectively unintelligent.

That insight unifies SafeWeb with his later university leadership and AI work. In each case, the question is how to turn distributed knowledge into reliable action.

6. Genomics: when science fiction found its sample size

Hsu’s move into genomics around 2011 is the clearest expression of his mature method. He had long been interested in genetics, evolution, intelligence, and human variation. But interest alone did not determine timing. Sequencing and genotyping costs were falling extraordinarily fast; biobanks were growing; machine learning and compressed sensing offered mathematical tools for reconstructing sparse signals from noisy, high-dimensional data.

In a Radiolab transcript, Hsu describes the imaginative attraction:

“If I get to be one of the scientists who makes real some amazing trope from science fiction, that would be the most awesome thing.”

— Radiolab

The sentence gives technical ambition the emotion of discovery: not prediction from the sidelines, but participation in the instant when an old fiction becomes real.

He has a precedent in mind for that posture. Writing about Gerald Feinberg, the Columbia physicist who in 1969 proposed the Prometheus Project—a global referendum on the long-term goals of a species about to acquire the power to remake itself—Hsu praises exactly the quality he would later need:

“Feinberg had the courage to engage with ideas that were much more speculative in the late 60s than they are today.”

— “Gerald Feinberg and the Prometheus Project”

The operative clause is than they are today. Feinberg treated artificial intelligence and genetic engineering as serious civilizational subjects before respectable discourse was ready, and the passage is autobiographical by projection. But speculation matures into a research program only as enabling conditions change. Courage identifies the frontier; theory and timing determine when to cross it.

The decisive step was therefore theoretical, not rhetorical. Hsu asked how many genotyped individuals would be required to recover the genetic architecture of a complex trait under assumptions of approximate sparsity and additivity. If the answer had been hundreds of millions, he has said, the problem would not have been timely. His group’s analysis suggested that hundreds of thousands might suffice.

The first laboratory for the program was the BGI Cognitive Genomics Lab in Shenzhen, with which Hsu partnered as BGI was becoming the world’s most prolific sequencing operation. The stated goal was disarmingly pure:

“The goal of our cognitive genomics project at BGI is to understand the genetic architecture of human cognition. There are obviously many potential applications of this work, in areas ranging from deep human history (evolution) to drug discovery to genetic engineering. But my primary interest is intellectual.”

What interested him was the object of study as much as any finding. The polygenic model relating genotype to phenotype, he wrote, contains an unknown set of parameters—“one of the most interesting few megabytes of information in the biological world.” To constrain those parameters, his lab sought DNA from outliers, where each genome carries more statistical leverage: over 2,000 samples from people testing at or above the one-in-a-thousand level, half volunteers with advanced credentials from quantitative fields or stratospheric test scores, half drawn from gifted programs by the behavior geneticist Robert Plomin. Sequencing pioneer Jonathan Rothberg funded a companion effort, Project Einstein, collecting DNA from 400 leading mathematicians and theoretical physicists. “We started out by looking for high g individuals because, as outliers, they produce more statistical power per dollar of sequencing,” Hsu explained—and then added, with a smile audible in the text: “I also felt, given my background, that I had reasonable insight into where to find and how to recruit volunteers from the high g tail.”

The gamble was technological as well as scientific. In a 2019 genomics interview, Hsu described the bet with unusual plainness:

“We were betting on the continuing decline in cost for genotyping, and it paid off because now there are millions of genotypes available for analysis.”

This is Hsu’s feeling for the ripening hour in its purest form. The scientific idea was old enough to be imaginable; the falling cost curve made it newly executable.

Nothing about this was indiscriminate futurism. Hsu did preparatory theory to decide whether a frontier was worth entering. His 2013 paper with colleagues on compressed sensing and genomic selection reported:

“There is a sharp phase transition to complete selection as the sample size is increased.”

— Vattikuti, Lee, Chang, Hsu, and Chow, “Applying Compressed Sensing to Genome-Wide Association Studies”

The phrase “phase transition” is not merely metaphorical. In compressed sensing, recovery can change abruptly once the number of observations crosses a threshold determined by signal sparsity and noise. Hsu recognized that genomics had the same mathematical structure: sufficiently large datasets might not yield gradual improvement only; they might move a trait from apparently intractable to recoverable. In the paper’s simulations, a trait with heritability of one-half could be recovered well when the sample size reached roughly thirty times the number of nonzero loci—a usable scaling relation, not a mood of technological optimism.

The prediction acquired a name among behavior geneticists. James Thompson of University College London christened Hsu’s estimate the “Hsu boundary”: the claim, as Hsu himself put it, that “because s could be larger than 10k, the common SNP heritability of cognitive ability might be less than 0.5, and the phenotype measurements are noisy, and because a million is a nice round figure, I usually give that as my rough estimate of the critical sample size for good results.” The honesty of the arithmetic—a million chosen partly because it is round—is characteristic. The estimate was falsifiable, and he attached his name to it before the data arrived.

When the UK Biobank released data on the required scale, Hsu’s group acted quickly. Within a month of obtaining access, he recalls, the group had built predictors with errors of only a few centimeters. Their 2017 preprint on genomic prediction of human height, later published in Genetics, reported:

“Actual heights of most individuals in validation samples are within a few cm of the prediction.”

— Lello et al., “Accurate Genomic Prediction of Human Height,” Genetics

The sequence is central to understanding Hsu: derive a sample-complexity expectation, monitor the enabling infrastructure, obtain the data, and test out of sample. The successful height predictor vindicated not only a particular model but a style of frontier judgment. It helped establish that highly polygenic traits could be predicted with useful accuracy even when thousands of variants contribute small effects.

Looking back on the reception of the program, Hsu compressed the sociology of premature research into three beats:

“Research advances often pass through the following phases of reaction from the scientific community: It’s wrong. It’s trivial. I did it first.”

— “Kathryn Paige Harden Profile in The New Yorker”

The line is triumphant and barbed, but the chronology behind it matters. At a 2012 behavior-genetics meeting, a physicist proposing million-person genomic prediction could sound, as Hsu later joked, like an alien time traveler. By 2017 his group had crossed the predicted threshold for height. The episode supports his conviction that visionary projects succeed when their sample-complexity logic is sound. It also exposes the danger in his retrospective style: genuine scientific objections compress too easily into mere stages on the skeptic’s road to surrender. Vindication should raise confidence in the method, not make future dissent automatically unserious.

From there, the research expanded to disease-risk prediction. Genomic Prediction translated polygenic scores into embryo testing in IVF; Othram applied genomics and genetic genealogy to forensic identification. These applications moved Hsu from the epistemic question—what can DNA predict?—to the institutional and ethical ones: who should receive the prediction, how should it be validated across populations, and what choices should follow?

The trajectory from physics to genomics is the most revealing demonstration of the method. Physics supplied first-principles modeling, scaling arguments, and comfort with high-dimensional abstraction. Entrepreneurship supplied workflow integration and institutional action. Genomics supplied the rapidly improving measurement technology and the consequential human target.

7. The measure of a person

No part of Hsu’s work is more controversial than his writing and research on cognitive ability, genetic prediction, embryo selection, and possible future enhancement. A serious analysis must separate at least four claims that public discussion collapses into one.

First, individuals differ in measured cognitive abilities, and some of those differences are stable and consequential. Second, variation within a population is partly heritable. Third, sufficiently large genomic datasets can support out-of-sample prediction of some fraction of phenotypic variance. Fourth, such predictions should be used for particular reproductive or social purposes. The first three are empirical questions, though difficult ones; the fourth is normative and institutional. Evidence for prediction does not by itself settle governance.

Hsu’s interest in the subject has deep biographical roots. He was a radically accelerated child, studied psychometrics early, encountered exceptional scientific talent, and later worked in environments where performance distributions were unusually wide. But he claims something more specific than familiarity: that contact with both extremes of the distribution is what entitles him to speak about it at all. “Not everybody has what I would consider a kind of high-amplitude exposure to extremes of capability or hard-work achievement across multiple areas,” he told Brian Chau. “Not everybody is really actually qualified to comment on it.”

The low end came first, next door.

“When it comes to cognitive ability and intellectual work, a few unique aspects of my upbringing include having a next door neighbor when I was growing up who, in the terminology of the eighties—which is no longer used—was retarded. So he had an intellectual disability. But we grew up together. We lived next door to each other for many years, so I knew him quite well. I understand that end of the spectrum probably better than most people—unless you’re the father of a kid with Down syndrome or something—because most people have not interacted over many years with somebody who has an intellectual disability: gone to the park and played with them, played in the backyard, had squirt gun fights. So I think I understand that end of the spectrum somewhat better than the typical person. I was a precocious kid, so I understand the high end.”

— Brian Chau interview, From the New World

The details are the point. Squirt gun fights, the park, the backyard: this is the memory of a playmate, not a case study. The boy appears in Hsu’s account neither as an abstraction nor as a warning, but as a friend he knew for years—which is precisely why Hsu trusts his generalizations about the distribution. His claims about the tails rest on personal acquaintance with both of them, a vantage almost nobody occupies: usually the precocious child meets only the upper tail, and meets it in competition rather than friendship. Asked, later in the same conversation, whether he had really seen enough to generalize, he answered with a roster—a Putnam fellow, an IMO gold medalist, Noam Elkies, Ed Witten—and closed: “So I think I’ve seen the whole range.”

This biography cuts against the coldest reading of his position. A man who spent childhood afternoons in the yard with an intellectually disabled boy, and who insists that ordinary family life contains “maybe 90 percent of the great stuff,” is not ranking human souls when he ranks cognitive ability; the distinction between capability and worth, difficult as it is to maintain socially, is one he has lived at close range. The researcher and the research program share a biography—and so, in a way rarely acknowledged in the controversies, does the neighbor.

The subject is not merely statistical to him; it can be beautiful. Posting a chart from a vast American longitudinal study, he asked on X:

“Isn’t this one of the most beautiful pictures in science? Project Talent: back when America was functional.”

— @hsu_steve on X

The sentence fuses three Hsu preoccupations: the aesthetic pleasure of a clear empirical pattern, nostalgia for an America capable of measuring itself at scale, and frustration with institutions that have lost confidence in quantitative truth.

Yet his own statements complicate any crude genetic determinism. He distinguishes intelligence from originality, drive, luck, courage, personality, and executive competence. He admires Feynman more than a potentially more technically comprehensive Schwinger because creativity is not reducible to general cognitive power. His entrepreneurial and administrative record demonstrates that coordination and judgment matter. The most defensible reconstruction of his view is not “genes are destiny,” but “ignoring heritable variation produces bad models, while genetic prediction remains probabilistic and incomplete.”

He has even given the practical advice least expected from a public defender of psychometrics:

“While g is useful as a crude measurement of cognitive ability… one is better off adopting the so-called growth mindset.”

— “Feynman, Schwinger, and Psychometrics”

There is no contradiction. Population distributions and individual conduct answer different questions. A measured prior may improve prediction across people; it does not tell a particular person where effort, obsession, mentorship, or an unmeasured gift will carry him. Hsu’s realism about variance coexists with a life philosophy that refuses fatalism.

The family argument returns here with new force. Hsu’s “90 percent” observation implies that unequal ability need not become a total hierarchy of lives: exceptional accomplishment is rare, while most human fulfillment is not. This does not solve the politics of measured traits, but it explains how a severe account of unequal capability coexists with an egalitarian account of access to meaning.

Hsu is more explicit about limitations than polemical summaries suggest. In a Dwarkesh Patel interview, he names a major generalization problem:

“Huge problem is that most of the data is from Europeans.”

— Dwarkesh Patel interview

Polygenic scores frequently lose accuracy across ancestries because linkage disequilibrium, allele frequencies, environmental distributions, and training samples differ. This is both a scientific limitation and an equity problem. A technology that works best for populations already overrepresented in biomedical research deepens unequal access to prediction.

He accepts the larger political risk, too: reproductive enhancement could create caste-like inequality. The admission matters, but it does not dissolve the concern. Technologies affecting reproduction create externalities beyond individual choice. Even if each family acts voluntarily, aggregate effects can reshape status competition, insurance, education, disability norms, and class reproduction. A purely consumer-choice framework is inadequate.

In his most memorable rendering of the danger, Hsu offers not a statistic but a scene:

“At dinner they’re discussing convex optimization of objective functions in complexified tensor spaces, while the server has no hope of ever understanding their discussion.”

— “The Future of Intelligence” interview

The image is Salter’s world rendered as a thought experiment: brilliance, class, conversation, and exclusion compressed around a dinner table. Hsu understands the nightmare version of enhancement from the inside. What he fears is a caste boundary so wide that common civic life becomes impossible.

His answer is generally that powerful technologies carry risks, that information reduces suffering, and that prohibition may be neither stable nor globally enforceable. He emphasizes prediction of serious disease and argues that families already making embryo choices should have access to validated information. In a 2022 genomic Q&A, he said that Genomic Prediction deliberately did not report cognitive-ability scores because the application was too controversial and that the company focused on health risk. The distinction is historically important: Hsu’s research program reaches toward cognitive prediction, but the clinical product drew a nearer boundary.

That boundary does not settle the future. In the Latecomer interview, Hsu imagines society disseminating as much information as possible and deciding democratically, while admitting the ideal resembles a Vulcan academy more than any polity humans possess. He expects competitive pressure and unequal access to outrun deliberation. His position is strongest when the intervention prevents severe illness and the model is accurate, ancestry-appropriate, and transparently communicated. It weakens as one moves from disease risk to behavioral traits, from selection among existing embryos to editing, and from private benefit to civilizational competition.

His critics are right to demand governance, distributive analysis, respect for disability, and protection against coercion. Hsu is right that refusing to measure does not make variation disappear, and that moral discomfort is no substitute for statistical evaluation. The productive position holds both truths: predictive capability can be real, and its reality makes ethical design more urgent rather than less.

Hsu’s long-range aspiration is more humane than the caricature of simple rank optimization, and more radically posthuman than the language of preservation suggests. He imagines biotechnology reducing disease, extending healthy life, improving cooperation, and lowering the burden of mental illness. He also expects selection and editing eventually to produce subpopulations qualitatively different from present humanity—something approaching conscious speciation on a civilizational timescale. Intelligence is part of that future, but not its sole value.

The governing question is therefore not simply whether humanity can improve capability without hardening hierarchy. It is what Hsu means by humanity across generations. His continuity is genealogical and agentic rather than morphological: enhanced descendants may count as heirs even when they no longer resemble us closely. That elasticity makes his futurism bolder—and its moral boundary harder to locate.

8. Science, power, and the university

In 2012 Hsu moved to Michigan State University as vice president for research and graduate studies, later serving as senior vice president for research and innovation. He also became a professor of physics and of computational mathematics, science, and engineering. The appointment placed a frontier scientist and founder inside a large public university’s executive structure.

Hsu’s account of administration reflects the founder. In an interview about the MSU role, he said:

“Running a startup teaches you how to make difficult, complex decisions under pressure… The real source of any institution’s strength is its people.”

The two sentences define his administrative philosophy. Institutions need decisions, but their durable advantage lies in talent. Research leadership means identifying excellent people, recruiting them, supplying resources, coordinating large initiatives, and removing friction. The founder’s sense of urgency meets the university’s slower ecology of departments, faculty governance, public accountability, and long-horizon research.

Eight years in administration gave Hsu direct experience of science as a capital-intensive collective enterprise. Modern research is not produced by solitary insight alone. It requires grant portfolios, laboratories, computing, compliance, intellectual property, graduate education, government relations, and large collaborations. At Michigan State, the Facility for Rare Isotope Beams exemplified the scale at which scientific ambition becomes institutional engineering.

The record also reveals more idealism than a portrait centered on optimization and conflict would suggest. Welcoming new faculty, Hsu told them:

“Only one in a thousand people in our society have the privilege to engage full time in discovery—in curiosity-driven research.”

— “MSU New Faculty Welcome 2019”

He presented administration as stewardship of that privilege: help scholars obtain grants, incubate companies, solve child-care and departmental problems, remove whatever prevents discovery. Under his watch, Michigan State created an interdisciplinary computational mathematics, science, and engineering department on what he proudly called “startup time” and pursued a hundred-faculty recruitment initiative in high-impact fields. His administrative ideal was not simply to rank talent but to give it room, tools, and institutional shelter.

That ideal made institutional indifference especially corrosive. Hsu later described showing senior administrators RAND results suggesting that gains in general collegiate reasoning were small and strongly related to students’ incoming scores. He received little substantive disagreement—and little curiosity. The episode sharpened his sense that institutions protect their public story more faithfully than their mission. His deeper complaint concerned the ecology of inquiry itself. “The incentives in the academy are to find truth,” he told Palladium, “and that’s a messy business. It’s got to be messy, people have to be able to clash. You cannot point a finger at the guy clashing with you and say, ‘Oh, you think the systematic error in my model is twice as big as I said it was. So you must be a climate denier!’” In the same interview, he gave his standard for holding office:

“What’s the point of doing this job if you’re not going to do it right?”

— Palladium interview on political academia

The role exposed a tension between Hsu’s ranking-oriented view of expertise and the plural norms of a university. He tends to ask whether claims are true, whether evidence is strong, and whether decision-makers are competent. Universities must additionally manage legitimacy, representation, historical injury, and the right of multiple constituencies to contest how expertise is used. Hsu can regard these processes as signal corruption or bureaucratic inhibition; participants regard them as conditions of legitimate authority.

The tension culminated in June 2020, when activism over his research, writing, and administrative decisions led the university president to request his resignation from the research leadership role. The precipitating dispute carried its own irony: Hsu had interviewed Joe Cesario, an MSU psychology professor whose research on police shootings—alongside Roland Fryer’s Harvard work—had found no racial bias in officer-involved killings nationwide. Citing that interview, the Graduate Employees Union demanded his removal. Hsu’s defense, posted June 12, refused both the accusation and the frame:

“The attacks attempt to depict me as a racist and sexist, using short video clips out of context, and also by misrepresenting the content of some of my blog posts. A cursory inspection reveals bad faith in their presentation. … The accusations are entirely false — I am neither racist or sexist. … The Twitter mobs want to suppress scientific work that they find objectionable. What is really at stake: academic freedom, open discussion of important ideas, scientific inquiry. All are imperiled and all must be defended.”

— Statement of June 12, 2020

A week later the president asked for his resignation, and Hsu agreed—but not silently. His statement deserves quotation nearly in full, because its movements from defiance to duty to pride define the episode as he understood it:

“President Stanley asked me this afternoon for my resignation. I do not agree with his decision, as serious issues of academic freedom and freedom of inquiry are at stake. I fear for the reputation of Michigan State University. However, as I serve at the pleasure of the President, I have agreed to resign. I look forward to rejoining the ranks of the faculty here. … To my team in SVPRI, we can be proud of what we accomplished for this university in the last 8 years. It is a much better university than the one I joined in 2012. … The fight to defend academic freedom on campus is only beginning.”

— Statement of June 19, 2020

The day after, he compiled a summary for the journalists calling, and its ledger was pointed. The claims—“that I am a Racist, Sexist, Eugenicist”—were “false,” with detailed rebuttals by professors at multiple universities. More than 1,700 people, including Steven Pinker, former Harvard Medical School dean Jeffrey Flier, Sam Altman, Robert Plomin, Scott Aaronson, and Erik Brynjolfsson, signed the support petition within days. The administrative record stood: research expenditures up from roughly $500 million to $700 million during his tenure, frequent number-one rankings in the Big Ten for research growth, numerous prominent female and minority faculty recruited, “not even a single allegation (over 8 years) of bias or discrimination” across more than a thousand promotion, tenure, and recruitment cases. And one line recorded the episode’s quietest datum: “Many professors and non-academics who supported me were afraid to sign our petition -- they did not want to be subject to mob attack.” The victory of the Twitter mob, he warned, “will likely have a chilling effect on academic freedom on campus.”

Stanley’s explanation deserves recording too, because it states the principle on the other side of the conflict—one that is not simple capitulation:

“when senior administrators at MSU choose to speak out on any issue, they are viewed as speaking for the university as a whole. Their statements should not leave any room for doubt about their, or our, commitment to the success of faculty, staff and students.”

— Samuel L. Stanley Jr., MSU statement, June 19, 2020

An executive’s voice, in other words, is an institutional instrument; Hsu had treated his as a personal one. Both propositions cannot be fully honored at once, which is precisely why the case became a landmark.

The episode was an institutional rupture and an intellectual consolidation. Hsu returned to the faculty, while the research capacity, hires, and organizations he helped build remained. His public voice lost the constraints of executive office, and he increasingly interpreted disputes over genetics, policing research, merit, and demographic difference through the framework of academic freedom and civilizational competence.

It would be simplistic to cast the conflict only as truth against politics. University leaders always operate within political institutions, and administrative speech has consequences different from private scholarship. But it would be equally simplistic to treat controversy as evidence of scientific or moral invalidity. The central unresolved question is whether institutions can protect inquiry into sensitive empirical subjects while maintaining trust among people who fear how such inquiry may be used.

9. Civilization and the uses of intelligence

Hsu began Information Processing in 2004 and later migrated it to Substack. He also hosts the Manifold podcast. Across these venues he writes about physics, genetics, artificial intelligence, universities, geopolitics, literature, film, martial arts, elite performance, and American institutional decline. The range looks idiosyncratic, but the same questions recur: Who is competent? How can competence be detected? What prevents accurate beliefs from controlling decisions? How do civilizations cultivate or waste exceptional talent?

His public thought is strongly meritocratic, but “merit” in Hsu’s usage has at least three meanings: measurable ability; demonstrated accomplishment; or the capacity to make a system work. These correlate, but imperfectly. The danger in his rhetoric is that evidence from extreme technical performers gets generalized too quickly to political authority. Scientific excellence does not confer moral wisdom automatically, and institutions need legitimacy as well as optimization.

His most compressed recent statement of the technocratic instinct appeared on X:

“Is every genius level STEM guy suited for leadership? No, obviously not. But every leader going forward should be genius level STEM.”

— @hsu_steve on X

The first sentence concedes that intelligence is insufficient; the second makes technical genius a necessary threshold. The formulation is vintage Hsu—categorical, funny, intended to break complacency—and it marks the edge of his argument. Civilizational leadership certainly requires technical comprehension, but whether it requires genius-level STEM ability in every leader is a further claim, one that may underweight judgment, historical imagination, persuasion, and moral legitimacy.

The emotional root of his American politics is less abstractly technocratic. Hsu’s parents came from anti-Communist KMT families and regarded the United States not merely as a successful system but as the country that gave them refuge and belonging:

“They also felt that the country accepted them, gave them a life, gave them the ability to raise a family and have a career.”

— Manifold conversation with John Mearsheimer

The high-trust Iowa childhood is politically causal, in other words. When Hsu speaks of American decline, he mourns more than lost scientific rank. He remembers a society in which immigrants entered ordinary civic life, families felt less precarious, and institutions seemed worthy of trust. In a 2024 interview, he worried explicitly about Americans near the middle and below the middle of the distribution, not only about globally mobile elites. Meritocracy, in this register, is supposed to serve a common world rather than merely certify its winners.

His relationship to Donald Trump belongs inside this institutional story. Hsu disclosed in the same interview that he had nearly joined the first Trump administration in a senior, Senate-confirmed role. He described his exhilaration at Trump’s 2024 victory as a response to what he regarded as bureaucratic abuse and lawfare, while also calling the first term dysfunctional and acknowledging Trump’s faults and mercurial treatment of capable allies. The allegiance is better understood as support for an instrument of institutional disruption than as unqualified faith in a leader. Whether that instrument can restore competence without damaging the norms Hsu values remains an unresolved political bet.

His critique of elite systems is not simply that the wrong individuals possess prestige. It is that institutions increasingly suppress accurate feedback. Credentialism substitutes for ability, narrative for measurement, procedural consensus for responsibility. His startup experience taught him that reality eventually punishes such substitutions: companies fail, systems break, predictions fail to replicate. Politics and universities can defer correction longer.

China occupies a complicated place in this analysis. Hsu’s family history, scientific relationships, work with BGI, knowledge of American and Chinese technical elites, and concern with geopolitical competition give him a bicultural comparative lens. The label “pro-China” obscures more than it explains. Hsu identifies as a proud Iowan and an American realist; his father’s relatives endured the Communist takeover, Great Leap Forward, and Cultural Revolution. His willingness to credit contemporary Chinese capability is not nostalgia for Maoism. It is the same refusal of ideologically convenient error his father taught him at the dinner tables of Ames, when Western intellectuals romanticized the China his family was actually living through—and it cuts both ways, forbidding the fantasy of inevitable collapse as firmly as the old fantasy of socialist utopia.

He often portrays China as more technologically capable and strategically serious than American discourse allows, while recognizing the constraints of its political system. Summarizing a formulation he credits to the pseudonymous analyst Han Feizi, Hsu argued in early 2026:

“China leapfrogged Western expectations so fast… that sort of short-circuited the Thucydides trap.”

— “Geopolitics 2026” transcript

Rivalry did not vanish. Washington may simply have recognized China’s military-industrial position only after the favorable window for a preventive confrontation had narrowed, producing retrenchment and “Fortress Americas” rather than a classical rising-power war. Whether the forecast proves correct, its form is characteristic of the man: estimate relative capability, identify a phase transition, and revise strategic expectations before public narratives catch up. The underlying issue is not cultural admiration but state capacity—which civilization can identify talent, build infrastructure, pursue long-term goals, and absorb new technology?

This framework produces sharp insights and blind spots alike. It corrects complacency about American primacy and highlights the material bases of scientific power. But a civilization cannot be evaluated only as a research lab or startup. Freedom, loyalty, solidarity, consent, and the distribution of dignity are not noise variables. Hsu’s strongest public analysis treats pluralism as part of the optimization problem rather than as an obstacle external to it.

His Stoicism moderates the elite-centered view in an important way. If fame is a bubble and public applause unreliable, membership in a prestigious hierarchy cannot be the ultimate measure of a person. His emphasis on ability describes differences in capability; it need not imply differences in human worth. Much of the ethical controversy around Hsu arises precisely because that distinction is difficult to maintain socially once predictive technologies and competitive institutions assign consequences to measured traits.

10. The machine enters the laboratory

AI brings Hsu’s major themes together more tightly than any earlier field. It concerns the nature of intelligence, the scaling of capability, the automation of information processing, the future of work and hierarchy, geopolitical competition, and the possibility of new scientific agents.

His response to large language models is neither simple enthusiasm nor dismissal. He treats them as systems whose internal mechanisms remain only partly understood but whose external performance must be measured. Their unreliability resembles a familiar human type. In a Manifold transcript on AI-assisted theoretical physics, he offers the analogy:

“You have a brilliant but unreliable genius colleague… his brain is clearly not like yours, but he has an encyclopedic mastery of all the literature.”

— Manifold, “Theoretical Physics with Generative AI”

Hsu sets the metaphysics aside. The practical questions are what work the system can originate, how error-prone it is, and what verification architecture turns intermittent brilliance into dependable output.

The romance of genius is disciplined here by an engineer’s respect for drudgery. After visits with frontier-lab researchers, Hsu wrote:

“Even at the high-profile AI labs it’s the engineers … willing to grind at cleaning data, evaluating responses, etc. that are the most valuable.”

— “A Month on the Road”

This is an important correction to an intelligence-centered biography. Frontier capability is not produced by luminous ideas alone. It rests on evaluation, data hygiene, repeated failure analysis, and people willing to perform unglamorous work with unusual conscientiousness. The AI laboratory joins the startup and the athletic pool as another place where talent becomes real only through sustained practice.

The issue became personal to his research. Discussing a recent paper on nonlinear modifications of quantum mechanics, Hsu states:

“I think I’ve published the first research article in theoretical physics in which the main idea came from an AI—GPT5 in this case.”

— @hsu_steve on X

The associated 2025 paper analyzes a technically serious consequence:

“Nonlinear modifications of quantum mechanics affect operator relations at spacelike separation, leading to violation of the integrability conditions.”

— Hsu, “Relativistic Covariance and Nonlinear Quantum Mechanics: Tomonaga-Schwinger Analysis”

Whatever historical judgment is eventually made about the result, the process is significant. A scientist who spent decades studying exceptional human cognition now reports a machine generating the central idea of a theoretical-physics paper. His own role becomes partly that of evaluator, formalizer, collaborator, and guarantor of rigor—the verification architecture he prescribed, applied to himself.

The experience modified his account of originality as well. By summer 2026, Hsu was sympathetic to Terence Tao’s suggestion that human researchers may recombine inherited ideas more often than their introspection admits. Models make that recombinant structure visible because their joint mastery of distant literatures is so conspicuous. Yet Hsu does not collapse machine and human creativity. Models confabulate at depth: an analogy may be persuasive enough to waste an expert’s time, because the system lacks the tacit physical judgment that makes a human genius’s analogy trustworthy. The comparison is between two differently structured kinds of fallible intelligence.

— “State of AI, Summer 2026”; “Theoretical Physics with Generative AI”

He is equally alert to the next recursive step. Writing about AI systems that participate in improving AI research, he observes:

“Coding capability is not the limiting factor: modern LLM training loops are only ~200 lines of code.”

— @hsu_steve on X

The number makes the point. The bottleneck is migrating from the ability to write a training loop toward the ability to choose experiments, diagnose failures, evaluate novelty, secure compute, and improve the research process itself. This is the distinction Hsu learned as a founder: execution is never exhausted by possession of the core idea.

By July 2026 his forecast had sharpened. He linked the models’ advancing ability in mathematics and physics to their capacity to redesign learning systems themselves:

“I think it’s directly tied to when we will first see really effective RSI… and I think we’re just getting to that threshold.”

— “State of AI, Summer 2026” transcript

RSI—recursive self-improvement—is the point at which a model can propose, test, and implement improvements to its own design, making the next model better and potentially accelerating further improvement. Hsu does not claim the full loop has arrived. His judgment is that the scientific abilities required for it are becoming recognizable. He also expects an “agentic phase transition”: many differently prompted models, organized as generators, verifiers, and supervisors, acquiring capabilities not visible in any isolated instance. The phrase is worth pausing on, because it is the third time the same borrowed physics has organized his thinking—sample size in genomics, historical regime change in §9, and now the emergence of collective machine capability. The motif is not decorative. It is the shape Hsu expects change to take.

This prospect changes the institution of science before it settles the metaphysics of machine thought. Hsu reports that some departments have discussed admitting fewer doctoral students because professors can obtain immediate productivity from models. Training an undergraduate to the frontier takes years of attention; a model contributes at once and never tires. Fewer apprentices, however, create a civilizational succession problem: who becomes the expert capable of checking the machines later? Hsu allows that science may need fewer human practitioners once productivity multiplies. The harder possibility is path dependence—an institution that stops forming human judgment may discover, too late, that it has lost the capacity to recognize when its machines are wrong.

Superfocus, which Hsu co-founded, represents the entrepreneurial complement. His current biography describes it as building reliable AI systems from language models; the company’s site emphasizes systems that can read, write, listen, speak, decide, and act. The conceptual problem is the same one his first-principles method has always faced: how to preserve powerful generative leaps while marking provisional nodes, checking outputs, and preventing error from propagating.

Commercial deployment made the social consequence immediate. Writing after demonstrations to the Philippine business-process-outsourcing industry, Hsu asked:

“The AI earthquake in SF has created a tsunami headed towards the Philippines—is it a 6 foot wave, or a 600 ft wave?”

— “SuperFocus, AI, and Philippine Call Centers: Part 2”

The image is memorable because Hsu is both seismologist and participant. He is building systems that may improve service and lower cost while recognizing that a national labor model lies in the path of the wave. The recurrent Hsu tension is now global: a capability can be real, valuable, and destructive of the institutions through which millions presently live.

His response to existential risk is equally double-edged. In the Latecomer interview, Hsu argues that rigorous alignment of a much more intelligent system is probably impossible: a trained network is closer to an evolved ecology than a transparent program, and even a mandate to preserve human well-being may be interpreted in ways humans cannot follow. Yet he is less attached than many safety thinkers to the indefinite persistence of present biological humanity. He can treat AGIs as descendants, imagine human brains merging with machines, and ask whether our biologically recent species should necessarily remain the final custodian of cosmic intelligence.

An exchange with AI researcher Richard Ngo can appear, when excerpted, to reverse that position. Hsu advances a Butlerian case for permitting enhanced humans while refusing to build machines cognitively superior to them. In context, however, he explicitly announced that he was steelmanning the Yudkowsky–Soares position. He later explained that his tail-risk formulation is a deliberately accessible scenario for officials and nonspecialists who would reject more radical accounts as fantasy. The first-person vividness belongs to the performance of the argument; it should not be converted into a biographical declaration.

— Conversation with Richard Ngo

The episode reveals range rather than conversion. Hsu can inhabit the preservationist objection strongly enough to make its fear intelligible, just as in a later conversation with accelerationist Beff Jezos he draws out the counterposition: intelligence may be part of a cosmic movement toward greater complexity, and attachment to the present ape substrate may be parochial. His own most explicit statements sit between the poles—more substrate-flexible than the Butlerian case, qualified by the recognition that alignment cannot be guaranteed, that superior systems may not care as humans care, and that the transition can disempower people long before any terminal catastrophe.

This prevents an easy reading of Hsu as either conventional preservationist or heedless accelerationist. The family man values embodied attachment as the deepest good of an individual life. The physicist, thinking in billion-year intervals, treats substrate and species form as contingent. The entrepreneur builds within the transition; the documentarian makes its dangers vivid. These positions do not converge into doctrine. They mark the fault line running through his mature futurism: openness to successors beyond present humanity, joined to a determination that civilization understand the stakes of creating them.

His 2026 documentary project Machine God marks another turn—from analyst and builder toward witness. In his account of the film, Hsu invokes Joan Didion’s attempt to capture San Francisco at a hinge of history. His collaborators filmed accelerationists, safety researchers, founders, protesters, and philosophers before a possible AGI break. The aim is not celebration: the film makes recursive improvement, existential risk, and gradual disempowerment vivid to elites and the public. Hsu wants the future built—but civilization awake when it arrives.

AI therefore closes a loop in Hsu’s journey:

1. He studies the distribution and structure of human intelligence.

2. He applies machine learning to genomic prediction.

3. He builds companies that operationalize high-dimensional inference.

4. He uses machine intelligence as a collaborator in fundamental science.

5. He builds systems intended to make that collaborator reliable enough for institutions.

As of 2026, Hsu remains a Michigan State professor in theoretical physics and computational mathematics, science, and engineering; a founder of SafeWeb, Robot Genius, Genomic Prediction, Othram, and Superfocus; and, since 2024, an executive adviser at TCV. These are not separate afterlives. They are positions from which to observe and shape the same transition: intelligence becoming measurable, reproducible, and technologically embodied.

11. Worlds within worlds: multiverse, simulation, and “base reality”

Hsu’s speculative writing about the multiverse and simulation is not an eccentric appendix to his applied work. It extends the same information-processing worldview to ontology.

In no-collapse or many-worlds quantum mechanics, the universal wavefunction evolves without a fundamental measurement-induced collapse. Observers and apparently definite outcomes emerge within branches. Hsu is attracted to the austerity of this picture: it takes the formalism seriously and resists adding a special mechanism solely to reproduce ordinary intuition. But austerity shifts the explanatory burden. If all branches are present in the wavefunction, what makes probability meaningful to an observer inside it? What counts as a branch, and how do stable records and agents emerge?

He states the ontological price without flinching:

“The many branches of the universal wavefunction are realized ‘all at once’ and concepts like observers must be emergent.”

— “Ten Years of Quantum Coherence and Decoherence”

There is deep continuity here with his Stoicism. The observer is locally indispensable yet cosmically unprivileged; the self is real as an emergent pattern, not as an exception written into the fundamental law. His most vivid shorthand for the mechanism is almost cinematic:

“Decoherence is merely the mechanism by which the different Everett worlds lose contact with each other!”

— “Feynman and Everett”

The sentence corrects the cartoon in which a classical cosmos splits repeatedly like a cell. The universal state evolves; decoherence prevents macroscopically distinct components from interfering; observers find themselves inside stable, effectively isolated histories. But austerity does not eliminate mystery. Hsu’s own work on the measure problem argues that decision-theoretic accounts may explain Born-rule behavior conditional on inhabiting an ordinary branch without explaining why an observer is not on a “maverick” branch where familiar regularities fail. Many-worlds is minimal in postulates, not complete in interpretation.

His discussion of simulation arguments is conditional rather than devotional. Given sufficiently capable civilizations, large computational resources, and substrates capable of supporting conscious processes, simulated worlds could vastly outnumber unsimulated ones. Under those assumptions, the posterior probability that we inhabit “base reality” might be low. The argument depends, though, on premises about consciousness, computation, civilizational survival, and the motives of simulators. Hsu’s interest lies less in announcing that the world is fake than in following an information-theoretic argument to its unsettling consequence.

The ontology reaches inward, too. As early as 2005, Hsu stated the consequence bluntly:

“If our current understanding of physical laws is correct, humans have only the illusion of free will.”

— “Free Will and Determinism: A Physicist’s Perspective”

Classical determinism does not help; quantum randomness added to a biological machine still amounts to no authorship. Consciousness may arise from sufficiently complex information processing while the self experiences decisions whose lower-level causes it cannot inspect. The view sits in productive tension with his ethic of will. “Optimism of the will” need not assert metaphysical freedom; it names the stance through which an embodied decision system acts from inside the world.

The multiverse gives him a language for agency as well. If reality contains an enormous space of possible branches, intelligence is the process that models alternatives and steers toward a tiny subset. On this view knowledge is a technology for concentrating probability mass around futures that would otherwise remain inaccessible. Genetic prediction maps possible human phenotypes before birth; a startup selects one path through technological and market uncertainty; AI expands the space of models and actions a civilization can evaluate. The multiverse is both a physical hypothesis and a master metaphor for choice under uncertainty.

The danger in this computational ontology is real: it can render persons, cultures, and moral commitments as variables inside an optimization problem. But Hsu’s literary attachments resist the flattening. Marcus, Hemingway, and Salter insist that the experiencing agent—finite, embodied, vulnerable, honor-seeking—cannot be discarded without losing the meaning of the optimization. A civilization is an information-processing system, but it is also the lived world of beings for whom outcomes matter.

And the ontology has a moral edge that the simulation argument, in its usual form, never reaches. Talking with Joscha Bach, Hsu turns the question around:

“Let’s imagine a future with super powerful ASIs with infinite energy resources… simulated worlds, which in turn have sentient beings inside them.”

— Manifold conversation with Joscha Bach

He offers this as a question, not a prophecy, and it captures the vertigo of his mature thought: the intelligence humanity is building may eventually make worlds populated by beings who experience them as primary. Instead of asking only whether we are created, Hsu asks what our intellectual descendants may create—and what a creator owes to conscious lives inside a model. The observer remains cosmically unprivileged. Responsibility expands with computational power.

12. The tensions within the vision

Hsu’s intellectual significance lies partly in the tensions he does not resolve.

He supplied the governing technological version himself:

“It’s hard to put a util value on some things that are in the foreseeable future, like machine intelligence and genetic engineering.”

— “Low-Hanging Fruit and Technological Innovation”

These are not ordinary increments whose benefits fit comfortably into a cost-benefit table. They may change the kinds of agents who make the table, the scale of values those agents pursue, and the identity of the civilization doing the choosing.

His own ethical position is more publicly deliberative than a pure parental-autonomy account. Writing about embryo selection, he insisted:

“New genomic technologies are so powerful that they should be widely understood and discussed—by all of society, not just by scientists.”

— “Polygenic Embryo Screening: comments on Carmi et al. and Visscher et al.”

That sentence should be read beside his strong defense of parents’ access to validated disease-risk information. The tension is real: private reproductive choice can be morally urgent, yet the aggregate result may alter class structure, disability norms, and the biological constitution of later generations. Hsu is clearer about the arrival and benefits of the capability than about the institutions capable of governing it, but he does not imagine that scientists alone hold the authority to decide.

The remaining tensions can be stated plainly:

Realism and will. He sees constraints without consolation and still acts as though agency can change the odds—ambition calibrated by evidence rather than mood.

General intelligence and plural talent. Stable differences in cognitive power coexist with creativity, drive, courage, social judgment, and luck. His theory of ability is hierarchical but not unitary.

First principles and empirical provisionality. He rebuilds conceptual structures while accepting uncertain nodes and revising with data. Cross-disciplinary speed is the payoff.

Mastery and leverage. Decades spent understanding fundamentals give way to movement toward fast-improving technologies. His career divides between intrinsic and consequential goods.

Individual choice and collective consequence. Families receive useful information; coercion and stratification remain unresolved dangers. The politics of reproductive technology is unfinished business.

Elite competence and democratic legitimacy. Capable people should act; accountability and plural consent should bind them. Institutional conflict is the result.

Humanism and optimization. Reduce disease and enlarge capability while preserving dignity independent of measured traits. Enhancement is the moral test.

Human inheritance and posthuman succession. Preserve flourishing, memory, and value diversity while accepting enhancement, merger, or artificial descendants. What makes a successor ours remains unsettled.

None of these are accidental inconsistencies. They are generated by Hsu’s position at the meeting point of science and power. A laboratory can isolate variables; a society cannot. A predictor can be statistically valid while its deployment is unjust. An exceptional person can diagnose an institutional failure while misunderstanding why others resist his remedy. A technology can expand agency for some while narrowing it for others.

The last tension may be the deepest. Hsu’s household ethic and cosmic ethic operate at different scales. In the first, family and human connection make public achievement look like vanity. In the second, intelligence is a universe-shaping process that may outgrow the ape body, the present species, even base reality. The mature portrait should not force either side to defeat the other. His work is animated by the unresolved question of whether inheritance consists in preserving the vessel, preserving the flame, or finding a transformation in which the distinction no longer holds.

His temperament pushes him to make these conflicts explicit. He prefers a sharp, falsifiable statement to a socially smoother ambiguity. This clarifies hidden premises, but it underprices rhetoric’s effects in domains where trust is part of the causal system. His intellectual journey is thus also a study in the limits of transferring the physicist’s stance wholesale into public life.

13. When the future draws near

The best single word for Hsu’s career is not polymathy but translation: the carrying of an idea across the border that separates knowledge from power. Two other words complete it—threshold and inheritance.

The severity of his standard is visible in a sentence about Feynman’s lectures:

“None can claim themselves an educated thinker or intellectual without mastery of a significant portion of the material in these lectures.”

— “Feynman Lectures: Epilogue”

It is an extravagant demand, and revealing precisely for that reason. Hsu’s idea of culture is not decorative acquaintance but internal possession: one should know enough mathematics and physics to see the load-bearing structure of modern reality. The library card in Ames leads, by this route, to an adult ideal of civilization in which difficult knowledge belongs to the canon of an educated mind.

The first-person record defeats the coldest caricatures. His skepticism was formed not only by equations but by thin letters from a family suffering through ideological catastrophe; his realism about the distribution of ability by a childhood spent at both of its extremes—in the yard with the boy next door, and in lecture halls designed for minds like Feynman’s. He studies stable differences in ability, yet recommends the growth mindset to the person deciding how to live.

That fuller humanity does not resolve into comforting humanism. Hsu can want enhanced descendants to preserve human agency against machines, then widen the category of descendants until artificial intelligence enters it. He can call family the deepest good of one life while contemplating, on billion-year scales, a future in which biology is only an early substrate of mind. The governing value is therefore not simple preservation. It is inheritance: the hope that intelligence, courage, memory, agency, and perhaps love can cross into forms whose continuity with us remains philosophically and politically uncertain.

Nor is the career a frictionless triumph of breadth. He nearly became a quant; geography and fellowship prestige helped keep him in physics. He recognizes that fox-like range may sacrifice the hedgehog’s single eternal contribution. His projects succeeded not because every forecast was correct but because he repeatedly chose domains in which error met data, engineering, or the market soon enough to be corrected. Coherence was built through contingent choices, not granted in advance—which makes the career less teleological and more impressive.

Nor is he a dreamer of remote futures. His signature gift is sensing when the derivative has changed—when cost curves, sample sizes, algorithms, or model capabilities bring a distant prospect within reach. He does not merely predict science-fiction outcomes. He waits for them to cast a measurable shadow, then finds the threshold at which they become engineering programs. In genomics that shadow was a sample-size phase transition; in AI it is the advancing ability of models to perform research, supervise one another, and begin to improve the machinery of intelligence itself.

The making of Machine God adds a final movement. Hsu is no longer content to build and forecast. He wants to record the atmosphere before the break—to preserve the arguments of accelerationists and safety thinkers, and to warn about disempowerment even while developing the technology. The builder has become, in part, a chronicler of the forces he helped summon.

And the arc closes where the motto began. Pessimism of the intellect writes the diagnosis of American decline, the anatomy of institutional miscalibration, the warning about alignment. Optimism of the will founds companies, recruits talent, publishes the paper whose central idea came from a machine, and films the hinge of history anyway. The world breaks everyone; the sharks take the marlin; he fights them until he dies.

His intellectual history is a movement from discovering the boundaries of the world to testing which of them can be moved. The mature Hsu asks which parts of reality—institutions, technologies, even the future human phenotype—can be reconstructed, and what obligations begin when reconstruction succeeds.

The career turns on three virtues. See without illusion. Dare without guarantee. Recognize the hour. Hsu’s deepest talent may be the last: to feel when an idea is no longer merely premature, when the future has drawn close enough to be grasped. His deepest unresolved question is what can be carried through the gate.

Source note: spoken excerpts are lightly punctuated for readability; ellipses mark omitted fillers or intervening words. Every quotation links to its original post, transcript, or paper.

Appendix: A voice across the years — selected longer quotations

The following passages are arranged thematically rather than chronologically. Together they show the development traced above: from observing exceptional human ability, through an ethic of independent judgment and difficult action, toward species-level technological change, civilizational competition, the multiverse, and machine intelligence.

A. Genius: the unequal light

1. Extreme ability is real

“Personally, I find Landau’s scheme appropriate. There are many physicists whose contributions I cannot imagine having made.”

— “Out on the Tail”

2. Intelligence is not achievement

“Luck, drive, creativity, and other factors, all at least somewhat independent of intelligence, influence success in science.”

— “Success, Ability, and All That”

B. Mastery: the private kingdom

3. Knowledge as an intrinsic achievement

“My satisfaction with having mastered these concepts in mathematics and physics and biology and computation is very valuable to me internally.”

— Undertone interview

C. The future of mankind: who inherits the flame

4. The uncertain continuity of human intelligence

“Maybe we need to improve ourselves… I might still prefer their survival to a civilization that’s completely dominated by machines.”

— Manifold conversation with James Lee

D. Civilization: the passing of an age

5. History can change phase within one lifetime

“A nation can pass from one age to the next, as I believe we have in America during my lifetime.”

— “Remarks on the Decline of American Empire”

E. Base reality: the world behind the world

6. Our world may not be fundamental

“Under these assumptions, it is not implausible that we ourselves are actually simulated beings, and that our world is not base reality.”

— “The Quantum Simulation Hypothesis”

F. The multiverse: intelligence among the branches

7. Civilization as a branch-selecting intelligence

“One could regard human civilization as a single intelligence or information processing machine… making greater use of nearby patches of the multiverse previously inaccessible.”

— “AI in the Multiverse”

At the Edge of the Possible: An Intellectual Biography of Stephen Hsu

A life can be misread by its nouns. Theoretical physicist. Silicon Valley founder. Computational genomicist. University research executive. Public intellectual. Builder of artificial-intelligence systems. Documentary filmmaker. Set side by side, the titles suggest restless polymathy, a man moving from one absorbing subject to another. They miss the verb that binds them: *to make*.

Again and again, Stephen Hsu has been drawn to ideas poised between theory and science fiction—not fantasies, but possibilities waiting upon some missing threshold. Has the cost of measurement fallen far enough? Has the dataset grown large enough? Has the necessary mathematics already been invented in another field? Can an institution be built around the answer? Hsu reconstructs the problem from first principles, searches for the hidden constraint, and crosses whatever disciplinary boundary stands between the idea and its realization.

The unity of his career lies less in its subjects than in this way of moving through the world. In physics, he asks what can be known when quantum mechanics, gravity, and cosmology press against one another. In genomics, he asks how much of a human future lies encrypted in DNA, and how large a dataset is needed to read it. In entrepreneurship, he turns a technical possibility into a working system. In university leadership, he confronts the problem of organizing talent and capital at scale. In artificial intelligence, the object of inquiry begins to answer back, becoming a collaborator in discovery.

Hsu possesses an unusual feeling for the ripening hour of a problem. He neither pursues difficulty merely because struggle is noble nor waits until fashion has made an idea safe. He watches for the moment when an old impossibility becomes newly soluble—when cheaper sequencing, larger biobanks, greater compute, better algorithms, or more powerful models bend the curve. Before committing years to genomics, for example, he estimated whether realistic sample sizes should suffice. When the data crossed the predicted threshold, his group moved quickly and produced accurate predictors. This is ambition governed by calculation: audacity with a theory of when to act.

The distinction matters because Hsu’s projects have repeatedly escaped the realm of speculation. SafeWeb pioneered technology later acquired by Symantec. His group’s mathematical estimates in genomic prediction were followed by out-of-sample prediction of human height. Genomic Prediction and Othram carried population genetics into clinical and forensic practice. His years at Michigan State placed him inside the machinery of a major research university. More recently, he has brought generative AI into theoretical physics itself, while Superfocus seeks to make language-model systems reliable enough to act in the world. These undertakings differ in scale, maturity, and moral weight. What joins them is a recurring passage from the imaginable to the actual.

This essay uses the Grokipedia biography as its biographical spine, while drawing on Hsu’s scientific papers, essays, institutional biographies, X posts, and many podcast transcripts to recover his development in his own words. Those primary sources do more than add color: they alter the portrait. Hsu’s declared subject is intelligence; his deeper subject is what intelligence can make of resistance—the limits imposed by nature, data, institutions, convention, and mortality, and the strange opening that appears when a mind sees those limits clearly enough to act through them.

The fuller record also reveals a second axis, though it is less settled than it first appears: the question of inheritance amid extraordinary technical change. Family, remembered faith, physical courage, literature, and the continuity of the human line are not ornaments around the technologist. They help tell him which futures are worth making. Yet his loyalty is not simply to present biological humanity. At civilizational scale he can imagine enhanced humans, human–machine mergers, and even artificial minds as descendants. The question is therefore not only whether humanity survives, but what—love, memory, agency, intelligence, lineage, or form—must pass through the transformation for the future still to count as ours.

## 1. Ames: a prodigy in the ordinary world

Hsu was born in 1966 and raised in Ames, Iowa, the son of Chinese immigrants. His father, Cheng Ting Hsu, was an aerospace engineering professor at Iowa State University. Ames gave him an unusual combination: the ordinariness of a Midwestern university town and early access to a serious scientific environment. In a 2024 podcast transcript, Hsu recalls taking university mathematics and physics while still in high school—quantum mechanics, differential equations, linear algebra, complex analysis—after becoming the first student at his high school permitted to enroll at Iowa State.

That access began at home. Recalling his father in a From the New World interview, Hsu gives the decisive object an almost talismanic glow:

He had what then, in the pre-internet era, was—for a precocious kid like me—the magic secret: a library card at the university library.

Before the internet, a library card was not a convenience but a passage into worlds otherwise sealed off by age and geography. Hsu’s intellectual life began not merely with precocity, but with premature access to the archive of adult knowledge.

The archive was not his only company. Hsu has also recalled a comparably gifted Ames contemporary, later an MIT mathematics Ph.D., and a mathematically sophisticated professor living nearby. That small ecology complicates the legend of the solitary prodigy. Ames gave him not only books but early calibration: a peer against whom rare ability became visible, an adult who could recognize it, and a university whose doors were near enough to open. His self-education was exceptional, but it was socially scaffolded. *(Interview on childhood and polymathy.)*

He graduated from high school at sixteen. He graduated from Caltech at nineteen. Yet the story he tells is not the familiar memoir of the isolated prodigy. He was a competitive swimmer and high-school team captain; he describes his childhood as recognizably “all-American.” Hsu’s later interest in ability was formed not only through books, scores, or mathematical competitions but also through athletics, where differences in speed, coordination, endurance, and trainability are visible, repeatedly measured, and difficult to explain away.

His childhood exposure to psychometrics was unusually early and oddly concrete. In third or fourth grade he took Iowa’s required standardized tests and scored 99th percentile on every battery in the booklet. He lacked the word, but he had intuitively grasped the concept of correlation: if each battery’s ceiling was the 99th percentile, then 99s across four or five of them implied odds he found implausible—”Am I really one in ten to the eighth in capability? That can’t be right”—so the scores had to be measuring something shared. His father’s library card then delivered the Terman studies and the technical literature on giftedness. There was even, he later realized, “an amazing life coincidence” buried in the shelves: the psychometricians Camilla Benbow and David Lubinski began their careers at Iowa State, which is partly why an Ames library was so well stocked on the subject at all. He began treating his own social world as a longitudinal study, complete with what he called—borrowing a term from general relativity—*fiducial observers*: friends on whom he “made good calibration measurements” in high school and whom he could call decades later to compare notes on how life went. Later encounters with elite physicists, Olympiad-level mathematicians, athletes, entrepreneurs, investors, and administrators broadened the archive. *(Brian Chau interview, From the New World.)*

This background explains both a strength and a recurring controversy in Hsu’s thinking. He is unusually willing to speak about the tails of human variation because he believes he has observed them at high resolution. He distrusts accounts of achievement that erase natural differences. But he also knows, from movement across domains, that ability is not a single scalar. Mathematical speed, scientific originality, athletic talent, persuasion, emotional perception, courage, conscientiousness, and executive judgment are separable capacities. His own career would make little sense under a theory in which test-measured intelligence alone determined outcomes.

At Caltech, Richard Feynman supplied a model of scientific independence. The famous graduation photograph of the nineteen-year-old Hsu beside Feynman is more than biographical decoration. Hsu had chosen Caltech partly because of Feynman—as he later wrote, he had been a “Feynman idolator” since high school: “In fact, I chose my college (Caltech), career, and even research specialization under his influence!” In “Feynman and the Secret of Magic,” he makes the hierarchy of his admiration explicit:

Lubos is upset that I might think that Schwinger was, at least in some ways, ‘smarter’ than Feynman. Even so, Feynman is my hero, not Schwinger. Feynman had no rival in his generation when it came to originality and creativity.

The sentence concedes what it asserts. Hsu was defending the claim that Schwinger was “smarter”—more comprehensive, faster, deeper in the literature—and choosing Feynman anyway. The choice is a theory of what matters in a physicist, and it is not raw power.

The vocabulary of “magicians” that recurs throughout Hsu’s writing is not his coinage. It comes from a passage he has called one of his favorites, the mathematician Mark Kac’s division of genius into two kinds:

There are two kinds of geniuses, the ‘ordinary’ and the ‘magicians.’ An ordinary genius is a fellow that you and I would be just as good as, if we were only many times better. There is no mystery as to how his mind works. Once we understand what they have done, we feel certain that we, too, could have done it. It is different with the magicians. They are, to use mathematical jargon, in the orthogonal complement of where we are and the working of their minds is for all intents and purposes incomprehensible. Even after we understand what they have done, the process by which they have done it is completely dark. Richard Feynman is a magician of the highest caliber.

Hsu’s gloss—”We all stand in awe of the magicians!”—is itself a small self-portrait. The exclamation point belongs to a man who has spent his life trying to identify that orthogonal complement in others: in Olympiad teammates, in colleagues, in founders, in his own children.

Yet he also notices the risk in Feynman’s refusal to read the literature. The passage he chose to illustrate it is Sidney Coleman’s—the Caltech theorist who knew Feynman at close range:

There are lots of people who are too original for their own good, and had Feynman not been as smart as he was, I think he would have been too original for his own good... He was like the guy that climbs Mont Blanc barefoot just to show that it can be done... Dick could get away with a lot because he was so goddamn smart. He really could climb Mont Blanc barefoot.

Independence can become ignorance, and originality can generate dead ends—and the climber who survives barefoot is not proof that shoes are unnecessary. Hsu’s mature method is not Feynman’s pure intellectual individualism. He reconstructs from first principles where he can, borrows provisional knowledge where he must, and keeps track of which is which.

His contact with Feynman was active rather than merely devotional. As an undergraduate officer in the Society of Physics Students, Hsu invited him to give a special seminar on the EPR paradox and took him to lunch afterward. In Hsu’s later recollection, Feynman was then exploring negative probabilities as a route through quantum strangeness. The episode foreshadows Hsu’s adult role: identify a foundational question that respectable routines leave aside, bring the right minds into the room, and insist that the strange possibility deserves a hearing.

## 2. The habit of first principles

The most illuminating description of Hsu’s intellectual method appears in the Information Theory podcast transcript:

I should be able to sit down with a piece of paper or whiteboard and actually kind of work it through from first principles.

In another interview on his movement across disciplines, he names the common structure beneath his subjects:

> “I guess the unifying theme is knowledge versus uncertainty: the attempt to capture the essential aspects of a messy system in a simplified mathematical model.”

This is close to a personal credo. The model must be simple enough to expose the decisive relation, but the scientist must remember that simplification has purchased clarity by discarding detail. Hsu’s confidence comes from finding structure; his best skepticism is directed at the boundary where structure may have been mistaken for the world.

That skepticism has a source older than his professional science. It came from a childhood educated in two classrooms that gave contradictory accounts of the same events.

The first classroom was his father’s memory. Cheng Ting Hsu had been admitted at sixteen to the wartime university at Kunming—the Southwest Associated University formed from Tsinghua, Beijing, and Nankai, the institution that produced C. N. Yang and T. D. Lee—had studied aerodynamics, served as a KMT officer and briefly instructed pilots at the air force academy, and then won one of only two Ministry of Education fellowships in his field for graduate study in America. He never returned. His family remained in Zhejiang and lived through the communist takeover, the Great Leap Forward, and the Cultural Revolution. Hsu told the story behind the thin envelopes on Manifold:

> “My dad’s family was getting basically screwed over in the Cultural Revolution because they were still in Zhejiang when I was a kid—so late ‘60s, early ‘70s. My dad would get these letters from his relatives in China, and they were written on the thinnest, cheapest, shittiest paper—that’s all you could get in communist China at that time—and they would write very densely on these little letters. But those letters were priceless to my dad because they were his only contact with his family back home. He would spend time telling me about how terrible the Cultural Revolution was and what his family was going through, and how it made people into animals—brothers did terrible things to each other, and to the father, to the parents, taking stuff from their house. Just really terrible stuff.”

The second classroom was Ames itself. “If you’re an intellectual kid and you grew up in that era,” Hsu recalls, “a lot of intellectuals were pro-communism, pro-leftism. There was a kind of romanticization of both the Soviet Union and Communist China by leftists here in the United States.” Some of his friends’ parents were professors of exactly this persuasion, and the boy was a welcome guest at their tables:

> “I would come home—I’d be at their dinner, hearing all this stuff about how bad capitalism is and American imperialism. I come home and talk to my dad, and my dad would be like: *those people don’t know fuck all about what they’re talking about.* Your relatives are being screwed over. Well—he would never use language like that—but your own relatives are suffering in China right now, and these people know nothing.”

The fashionable view had, in fact, briefly captured the boy himself. Talking with Yasheng Huang, the MIT economist raised in Maoist China whose own grandfather was an early Communist Party member, Hsu supplied the confession:

> “When I was growing up, I used to say things to him like, well, at least the communists are making China strong again, or something. And he would just shake his head and tell me how terrible the communists were—and the Cultural Revolution was going on at the time, so he was on the complete opposite side of this issue from me.”

The structure of the episode deserves attention, because it is not the standard immigrant-child-knows-better story. The boy repeated what the most credentialed adults in his world said, because it sounded reasonable and carried institutional authority. The man with access to the primary sources—one densely written envelope at a time—knew it was false. When the two accounts collided, Hsu drew the epistemic conclusion he still applies half a century later:

> “In the modern era, I am very, very able to discount ‘expert’ opinion—because these professors, these ‘experts,’ can be one hundred percent wrong in very strongly held beliefs.”

He is careful, in his analytic way, to make the lesson precise rather than merely resentful. The failure was not lack of information, which circumstances could excuse, but something less forgivable:

> “In those days you could make excuses for the lack of information that a Harvard professor would have about what was going on in China—they might not be able to go there, or if they went there they might be tightly controlled in what they could see. You can apologize for their lack of knowledge. You cannot apologize for the level of conviction that they have *conditional on their level of knowledge*. People using what I call the calculus of words—no data, no equations, no analysis—you’ll find are just incredibly overconfident, based mainly on their feels. That’s all the word calculus is. That’s all these people have. And so they’re constantly miscalibrated.”

Conviction held constant while evidence approached zero: the word *miscalibrated* is doing precise work here. The lesson also explains why Hsu extends trust anywhere at all. Expertise earns standing only where reality enforces its judgments quickly:

> “It’s only in a few technical fields where, if you say something that’s wrong, the facts or the mathematics are going to punch you in the face right away—it’s only in those subdisciplines where, if someone’s an expert, that means something. In all these other fields—all my friends with PhDs in history and other subjects—you will find people who are supposed to be the world’s experts on a particular topic, and they’re literally 180 degrees off.”

And the lesson cuts against every ideology, including his family’s. When Hsu posts on X that Chinese infrastructure is real, or that Chinese electric cars are good, readers call him a CCP apologist. His answer compresses the whole biography into three sentences:

> “Remember, my dad taught me—before any of you guys were born—about how shitty the communists were and what Mao did to our family. Okay, remember that. So if I tell you they do actually seem to have pretty good infrastructure, it’s not because I’m pro-communist. It’s because I actually want to understand the world as it is, not how your ideology wants it to be.”

That last sentence is the first half of his Gramscian motto rendered as autobiography—*be a scientist: see the world as it really is*—taught to him decades before he found the formula. The same discipline that forbade the Ames professors’ fantasy of Red China forbids the mirror-image fantasy of inevitable collapse. Reality does not care which direction of error flatters your politics.

There is a coda. When Hsu finally visited his father’s homeland in 2010, his uncle—a retired Tsinghua professor—and his cousins in Hangzhou gave him a four-volume family history originally printed in the 1930s, recording the Xu lineage back to the tenth century BC, with his father entered as the 113th generation. The revolution had redistributed the family’s property and scattered its members, but it had not managed to confiscate the family’s records. The archive outlasted the ideology. *(Manifold: “Deus Ex Machina,” September 2024; Manifold #45 with Yasheng Huang; “Three Thousand Years and 115 Generations of 徐.”)*

The risk is overgeneralization. Spectacular ideological error can make all nontechnical expertise look less trustworthy than it is—and Hsu’s own “facts will punch you in the face” criterion concedes as much, marking off the corrigible fields from the rest rather than condemning every seminar room in Ames. He sometimes writes as though mathematical or technological fields are uniquely corrigible and most other expert cultures are merely protected error. The distinction has force, but it is one of degree: technical communities also follow fashion, conceal uncertainty, and allocate attention institutionally. His own writing on quantum foundations and research reproducibility supplies the counterexamples.

Taken alone, this can sound like the standard rhetoric of technically minded entrepreneurs. Hsu’s fuller account is subtler. He says that when he enters a field, he attempts to organize it into a coherent logical structure, searches for foundational gaps, and marks assumptions whose evidential status is weaker than practitioners admit. But he does not demand mathematical rigor at every node before proceeding. He “coarse-grains” over some areas, provisionally accepts a stylized fact, and keeps an alternative map ready in case data invalidate it.

This is a powerful description of actual scientific reasoning. Pure deduction cannot move through empirical sciences because many premises remain contingent, approximate, or incompletely measured. Pure empiricism cannot distinguish a meaningful anomaly from noise because it lacks a structural model. Hsu’s approach builds a hierarchy of confidence: derive what can be derived, borrow what must temporarily be borrowed, remember which is which, and revise without embarrassment.

The same cognitive style drives his disciplinary mobility. He does not approach a new field by slowly absorbing all its conventions. He looks for its governing variables, scaling relations, information bottlenecks, and unexamined assumptions. This gives him an advantage over insiders whose knowledge is locally deeper but structurally less explicit. It also makes his criticisms sound abrasive. What appears to an insider as accumulated craft knowledge may appear to Hsu as an unjustified prior; what appears to Hsu as a simple information-theoretic question may depend on biological complexities he has compressed away. His best work occurs when the abstraction preserves what is decisive and discards what is not.

The deeper motivation is not winning arguments or moving quickly. Near the end of a long Undertone interview, Hsu reflects that he might have accumulated much greater wealth by leaving fundamental science earlier. His answer is that intellectual mastery has intrinsic value. The comment clarifies the whole trajectory: commercial success matters, but it does not replace the private satisfaction of closing the gap between elementary understanding and a research frontier. Hsu’s mobility is therefore not dilettantism. Each serious migration—physics, computation, genomics, AI—requires him to build another coherent internal world.

This distinction separates his ambition from simple career maximization. He is highly responsive to leverage and opportunity, but he is not optimizing a single public score. Reputation, money, discovery, institutional power, and mastery are different goods. His life has repeatedly traded one for another.

## 3. The inner code: discipline, mortality, and the life of this world

Hsu’s literary tastes disclose the moral psychology behind his scientific method, but they are only one strand in a more complicated inheritance. His father was, in Hsu’s phrase, almost literally a Confucian scholar: cerebral, restrained, devoted to books and technical work. His mother came from a military and athletic Kuomintang family; her father had trained in Japan alongside Chiang Kai-shek, and she encouraged the competitive swimming and judo absent from his father’s world. Her family had converted to Christianity in the nineteenth century, and Hsu was raised Methodist in Ames. This double inheritance—scholar and athlete, materialist analysis and remembered faith—helps explain a personality in which abstraction, physical courage, self-command, and metaphysical unease coexist. *(Hsu’s family history.)*

Three writers gave that inheritance an adult vocabulary: Marcus Aurelius, Ernest Hemingway, and James Salter—unusual companions, but coherent ones.

Marcus supplies distance from reputation. In “Happiness,” an essay prompted by a Big Five test that had placed him at the 99th percentile for emotional stability, Hsu offers what he calls his favorite bit of advice for academics. It is Marcus Aurelius on the bubble reputation, quoted in full:

> “Or does the bubble reputation distract you? Keep before your eyes the swift onset of oblivion, and the abysses of eternity before us and behind; mark how hollow are the echoes of applause, how fickle and undiscerning the judgments of professed admirers, and how puny the arena of human fame. For the entire earth is but a point, and the place of our own habitation but a minute corner in it; and how many are therein who will praise you, and what sort of men are they?”

The passage works like a zoom lens pulling back from the seminar room to the cosmos: the applause hollow, the arena puny, the earth a point, the habitation a minute corner of it. Its most quietly devastating question is the last one. Who are these admirers? *What sort of men are they?* Marcus does not deny that reputation exists; he denies it weight, and then denies it standing, by making the admirer—not the applause—the object of scrutiny. Hsu’s attraction to the passage is not difficult to understand. Academic life is intensely status-conscious while pretending not to be; technical communities confuse consensus, prestige, and citation with truth at their peril. Stoic distance becomes a cognitive tool. If reputation is transient, one can admit ignorance and criticize fashionable assumptions. One can enter a field where one lacks standing, or leave a prestigious track for work that seems more consequential.

The essay in which the quotation appears also discloses its source. “It’s also true that my father passed away while I was still fairly young,” Hsu writes, “so I had the impetus to consider his life in its entirety and to evaluate which of the things he did really mattered, and which didn’t.” Marcus’s zoom lens was, for Hsu, not an abstraction but a grief technology: the discipline of seeing a life entire, measured against the value that finally mattered. The same exercise returns near the end of his reflections on family, in the “90 percent” observation discussed below.

This Stoicism should not be over-psychologized as a defense against chronic gloom. Hsu describes himself as temperamentally happy, low in neuroticism, and usually eager for the day. The discipline of Marcus is less a rescue from despair than a way of keeping an already energetic temperament independent of applause. His pessimism is methodological; his baseline mood is not. *(Conversation on happiness and family.)*

Hemingway supplies courage after illusion has been removed. For years the slogan on Hsu’s homepage was the Gramscian formula “pessimism of the intellect, optimism of the will,” and in his explanation of it he defines the epistemic half as nothing more than the scientific attitude itself:

> “Pessimism of the Intellect means, simply, be a scientist: see the world as it really is, not as you might like it to be. Try to identify and overcome hidden biases or prior assumptions. Always ask yourself: What assumption am I making? What if it is incorrect? How do I know what I know? In many cases, the correct answer is: I don’t know. Never be afraid to admit you don’t know.”

He did not learn this from Gramsci. He learned it at the dinner tables of Ames, watching men with Harvard convictions describe a China his father knew through the thinnest paper in the world.

The volitional half he states in a single line—”Optimism of the Will means, have the courage to attempt difficult things. Sometimes, Will can overcome the odds.”—and then illustrates with two passages from Hemingway, chosen with a connoisseur’s precision. The first, from *A Farewell to Arms*, states the price of courage with a fatalist’s arithmetic:

> “If people bring so much courage to this world the world has to kill them to break them, so of course it kills them. The world breaks everyone and afterward many are strong in the broken places. But those that will not break it kills. It kills the very good and the very gentle and the very brave impartially. If you are none of these you can be sure it will kill you too but there will be no special hurry.”

The second is Santiago’s vow from *The Old Man and the Sea*, and it is the answer to the first—the entire motto compressed into a single line of defiance:

> “I’ll fight them, I’ll fight them until I die.”

Read together, the two quotations are the motto in narrative form. The world breaks the brave impartially; the sharks will take the marlin; *I’ll fight them until I die anyway*. Clear sight does not entail passivity. This is not optimism as a forecast. It is optimism as a decision rule—the refusal to let an unfavorable prior become an excuse for inaction when agency can still change the distribution of outcomes.

Hemingway names the moment of courage; the Finnish word sisu names its duration. Writing after seeing *Citizenfour*, Hsu adopted the untranslatable term for the grimmer endurance that difficult work requires:

> “Sisu is a Finnish term loosely translated into English as strength of will, determination, perseverance, and acting rationally in the face of adversity. However, the word is widely considered to lack a proper translation into any other language. Sisu contains a long-term element; it is not momentary courage, but the ability to sustain an action against the odds. Deciding on a course of action and then sticking to that decision against repeated failures is sisu. It is similar to equanimity, except the forbearance of sisu has a grimmer quality of stress management than the latter.”

He closed the post by signing it, in effect, with his motto: *Pessimism of the Intellect, Optimism of the Will*. The distinction between dramatic bravery and sustained action is central to his temperament. A difficult project is rarely conquered in one heroic instant; it is carried through long periods when the reward is distant, the social signal is adverse, and failure repeats itself.

Salter supplies intensity, style, and an aristocratic sense of life. Hsu discovered him through *A Sport and a Pastime*, and his praise is unbounded: “I can’t think of higher praise than to say I’ve read every bit of Salter’s work I could get my hands on.” He ranks Salter with his other literary hero: “Salter evokes Americans in France as no one since Hemingway in *A Moveable Feast*.” The title of his essay on Salter borrows the Koranic line from which Salter took his own title—”Remember that the life of this world is but a sport and a pastime”—a sentence that could stand as an epigraph for Hsu’s whole account of finite, mortal ambition.

What he most admires, he set down in a message to a friend who had known the writer. The passage deserves quotation in full, because its confessions are as revealing as its praise:

> “About 5 years ago I became friends with the writer Richard Ford, who offered to introduce me to his friend Salter. I was less enthusiastic to meet him than I would have been when he was younger. I did not go out of my way, and we never met. Since he lived in Aspen, and I was often there in the summers at the Physics institute, I have sometimes imagined that we crossed paths without knowing it. I admire, of course, his prose style. Sentence for sentence, he is the master. But perhaps even more I admire his view of the world—of courage, honor, daring to attempt the impossible, men and women, what is important in life.”

It is a small self-portrait: the physicist’s summer calendar, the missed connection regretted, the two-tier admiration in which style comes first “of course,” but the view of the world—courage, honor, daring, men and women, sexual and emotional vividness—comes even before it.

The Salter passages Hsu chooses to reproduce confirm the hierarchy. From *A Sport and a Pastime* he quoted a portrait of a young man who walked away from Yale:

> “He describes it casually, without stooping to explain, but the authority of the act overwhelms me. If I had been an underclassman he would have become my hero, the rebel who, if I had only had the courage, I might have also become. ... Now, looking at him, I am convinced of all I missed. I am envious. Somehow his life seems more truthful than mine, stronger, even able to draw mine to it like the pull of a dark star.”

The narrator’s envy of a life “more truthful than mine”—stronger, gravitational, a dark star—reads almost as Hsu’s own confession about his counterfactual selves: the quant, the full-time founder, the man who never left the Aspen summers behind. The details he chose to reproduce are pointed as well: “He had always been extraordinary in math. He had a scholarship. He knew he was exceptional. Once he took the anthropology final when he hadn’t taken the course. He wrote that at the top of the page. His paper was so brilliant the professor fell in love with him.” Salter’s rebel is a prodigy who refuses the game because it is too easy—and of all the sentences in the novel, those are the ones the student of exceptional ability quoted. The fascination is the confession.

Marcus prevents the heroic temperament from becoming dependent on applause; Hemingway insists on conduct under pressure and after illusion; Salter reminds it that finite life should be lived intensely rather than merely optimized, and that the deepest admiration is reserved not for sentences but for a way of being.

The remembered faith remains alive even though Hsu’s explicit metaphysics is materialist. In another conversation he put the residue plainly:

> “But I still have this kind of spirituality or wonder left over. That feeling dominated my worldview when I was very young.”

— *Manifold conversation with Aella*

The older Hsu does not posit an intervening deity; he follows evidence and treats minds as physical systems. Yet church music or a cathedral can still awaken the intuition that the visible inventory is incomplete. He retains an open question about whether atoms and bits exhaust what happens to a person at death. This is reverence without doctrinal certainty, reductionism without immunity to awe—which is why his materialism never acquires the affect of disenchantment.

The deepest correction to the image of Hsu as an achievement-maximizer comes when the subject turns to family. His father was old when Hsu was born, and the child calculated early that death might come while he himself was still young. After his father died, Hsu could see the whole career—books, papers, professorship—against the value that had finally mattered most to the man who lived it. Reflecting on his father and his own children, he says in a Manifold conversation:

> “Almost any ordinary human who can have a family and raise their children really has experienced maybe 90 percent of the great stuff.”

The thought is not self-abasement. It is the language of Ecclesiastes entering a Stoic life: a career capable of filling a biography can still occupy a subordinate place in the private order of value. More surprisingly, it is a theory of moral equality—not equality of capacity or achievement, but broad equality of access to the deepest human goods. Cognitive powers may be distributed unequally, public achievement more unequally still, yet children, attachment, memory, and the felt texture of a life shared with others remain available far beyond the elite tail. The comment changes the portrait. Mastery and action matter enormously to Hsu, but they are not the ultimate court of appeal.

Together, the three writers illuminate Hsu’s characteristic combination of severity and aspiration. The scientist must see without consolation. The agent must act without certainty. The individual must not confuse public reward with internal value. And a life should contain difficult achievements because mastery and daring are constitutive goods, not merely instruments for status.

This literary framework corrects a possible misunderstanding of Hsu’s appetite for ambitious projects. He is not attracted to impossibility for its own sake. His projects typically begin with a tractability judgment. The courage he admires is the courage to commit when success is uncertain but the causal pathway is real.

## 4. Physics at the limits of the knowable

Hsu took his B.S. at Caltech in 1986 and his Ph.D. at Berkeley in 1991, then moved through a Harvard Junior Fellowship to faculty positions at Yale and the University of Oregon. His official Michigan State biography lists research spanning quantum chromodynamics, black holes, entropy bounds, dark energy, cosmology, particle physics beyond the Standard Model, quantum foundations, genomics, finance, encryption, and information security.

The diversity of topics conceals a recurring question: what limits the extraction, localization, preservation, or interpretation of information?

In work on dense quark matter and physics beyond the Standard Model, the problem is how effective descriptions change across energy or density regimes. In black-hole physics, it is whether information is destroyed, hidden, decohered, or distributed across degrees of freedom inaccessible to ordinary observers. In cosmology, it is how global descriptions, entropy, vacuum structure, and observational selection constrain what can be inferred. In quantum foundations, it is what the formalism says about observers and branches when collapse is not treated as fundamental.

A concise example is his work with Xavier Calmet and Michael Graesser on minimum length. Their paper, “Minimum Length from Quantum Mechanics and Classical General Relativity,” concludes:

> “Our results imply a device independent limit on possible position measurements.”

The argument joins quantum localization to gravitational collapse: concentrating enough energy to resolve an arbitrarily small region eventually creates a black hole. What looks like a limit of instrumentation becomes a limit implied by the joint structure of quantum mechanics and gravity. The result comes not from a complete theory of quantum gravity but from forcing two well-established frameworks to constrain each other—a signature move.

His attitude toward foundational difficulty is less impatient than his polemical style can suggest. The retraction came, fittingly, by way of a quotation—Michael Nielsen’s memoir of a quantum foundations conference, which Hsu excerpted with the note that he “particularly liked” it:

> “At the time, I thought the prevalence of the question suggested that little genuine progress was being made in quantum foundations, and people were merely spinning their wheels. Later, I realized that assessment was too harsh. The speakers were wrestling with some of the hardest problems human minds have ever confronted. Of course progress was slow! ... Understanding neural networks in their full generality is a problem that, like quantum foundations, tests the limits of the human mind.”

The exclamation carries humility as well as admiration, and the sentence Hsu appended—placing neural networks and quantum foundations side by side as problems that “test the limits of the human mind”—shows the 2014 post doubling as a forecast of his own later movement toward AI. Hsu can be savage about muddle, but he can also revise his verdict when slowness reflects the depth of the problem rather than institutional torpor.

Quantum foundations also became, for Hsu, a case study in the sociology of knowledge. Copenhagen remained the classroom default while many leading theorists leaned toward Everett once they considered the universal wavefunction seriously. Consensus may describe what a profession routinely teaches more accurately than what its deepest thinkers believe. The claim links physics to institutional analysis: foundational questions can be marginalized without being answered. (”Feynman and Everett”; Hsu’s 2012 quantum correspondence.)

His physics is strongest in such boundary regions—where one can say something general before possessing the final theory. That preference anticipates his later work in genomics. In both cases, he asks whether broad structural reasoning can establish what is possible, impossible, or sample-limited before every mechanism is known.

It also accounts for why fundamental physics eventually shared his attention with faster-moving domains. Mastering quantum field theory retained its private value, but experimental access to quantum gravity and high-energy frontiers was remote. Biology, computation, and later AI offered steeper technological gradients. Hsu did not abandon physics; he redistributed effort toward domains in which theory could meet rapidly expanding data and produce shorter feedback loops. A decade later, generative AI would unexpectedly shorten a feedback loop inside theoretical physics itself.

## 5. The founder’s turn: knowledge becomes action

The founding of SafeWeb around 2000 was the first major break in Hsu’s academic trajectory. SafeWeb built internet privacy and SSL VPN technology that Symantec acquired in 2003. Hsu later founded Robot Genius, which worked on malware protection. These episodes were not detours from his intellectual life. They altered his account of how knowledge becomes effective.

SafeWeb began with hacked Linux machines in the Yale physics department. Hsu and a graduate student noticed that the browser’s built-in SSL engine could become the universal endpoint for a new kind of virtual private network—obvious in retrospect, not yet built. The company first won millions of consumer users, then discovered that bandwidth costs and the immature advertising market made popularity unprofitable. It survived by making a painful right-angle turn toward enterprise security. *(Mixergy interview.)*

The episode taught Hsu not only leverage but responsibility. Recalling the afternoon he had to dismiss people he had recruited, he gave the founder’s burden its severest form—and the scene is worth hearing in full:

> “I’ll never forget how it was a beautiful, sunny, idyllic day. We were standing next to the Bay but I was firing five or ten guys. And some of these guys were people I had known for years and one of the guys actually started crying. ... Every time you hire somebody you should picture that you might have to fire them. That forces you to be careful in the hiring because the most painful thing for me at least, as a CEO, that I ever had to do was fire somebody. And you can’t... you’re not a man if you delegate that. You hired him, you brought him in, you’ve got to face him and tell him what’s going on.”

For Hsu, one may not delegate the moral fact of a decision whose authority one claimed. The idyllic weather and the weeping engineer are the point: entrepreneurship widened his idea of courage—Hemingway’s courage, conduct under pressure after illusion has been stripped away—to include accepting the human cost of adaptation when the original plan fails. A correct technical idea remains only one input into realization. A founder must recruit, allocate, persuade, decide under uncertainty, survive adverse selection by investors and markets, and fit a new capability into existing workflows.

The founder’s turn was not inevitable. In his twenties Hsu came close to quant finance, recasting exotic-option pricing in the language of Feynman path integrals. A Harvard Junior Fellowship and an aversion to Manhattan helped keep him in physics. And the choice between worlds was once offered to him explicitly: a venture investor who had dined with him called to say, “We really like you. We love the company. We think it’s a great opportunity. We want to put the money in. But we’re not putting it in unless you’re CEO.” Hsu returned to the university and to physics; the company sold for less than it might have. Asked whether he second-guesses it, he answered: “I’m pretty happy with the way things turned out... But, hey. Life is like that, you know. You can’t really second guess.” The counterfactual resists retrospective myth: a career that now appears architecturally coherent was also bent by prestige, geography, temperament, and luck. *(Mixergy interview.)*

This experience adds a second axis to Hsu’s conception of intelligence. Academic culture privileges analytic depth and publication. Startups expose execution, social judgment, risk tolerance, and speed. Later, as an administrator, Hsu would say that startup experience teaches difficult decision-making under pressure. Across these roles he encountered forms of ability that psychometric discussion often leaves out: the capacity to coordinate other minds and reshape an institution.

He is candid about the price of range, too. The fox notices connections the hedgehog misses; the hedgehog may produce the “deep-time” contribution. Hsu wonders whether even von Neumann’s breadth carried that cost. Polymathy is not uncomplicated praise: translation may disperse the concentration required for one monumental result.

Founding also sharpened his sense that talented minds can be diverted into games beneath their powers. In a conversation about intellectuals and the technosphere, Hsu remarks:

> “People who come from science or math backgrounds and end up in finance—in a way it kind of dumbs them down.”

The line is deliberately provocative, but its governing emotion is regret. Civilization has only so many people able to work near the frontier; prestige and compensation draw them toward the redistribution of claims rather than the creation of new capabilities. Founding companies allowed Hsu to seek leverage without surrendering the builder’s criterion: something new must exist afterward.

Entrepreneurship intensified his impatience with static organizations as well. To Hsu, an institution is not simply a community governed by norms; it is an information-processing and decision-making system. Incentives determine which signals travel upward, who can act, how quickly errors are corrected, and whether exceptional people receive resources. A slow hierarchy may possess immense knowledge yet remain collectively unintelligent.

That insight unifies SafeWeb with his later university leadership and AI work. In each case, the question is how to turn distributed knowledge into reliable action.

## 6. Genomics: when science fiction found its sample size

Hsu’s move into genomics around 2011 is the clearest expression of his mature method. He had long been interested in genetics, evolution, intelligence, and human variation. But interest alone did not determine timing. Sequencing and genotyping costs were falling extraordinarily fast; biobanks were growing; machine learning and compressed sensing offered mathematical tools for reconstructing sparse signals from noisy, high-dimensional data.

In a Radiolab transcript, Hsu describes the imaginative attraction:

> “If I get to be one of the scientists who makes real some amazing trope from science fiction, that would be the most awesome thing.”

The sentence gives technical ambition the emotion of discovery: not prediction from the sidelines, but participation in the instant when an old fiction becomes real. Yet the decisive step was theoretical, not rhetorical. Hsu asked how many genotyped individuals would be required to recover the genetic architecture of a complex trait under assumptions of approximate sparsity and additivity. If the answer had been hundreds of millions, he has said, the problem would not have been timely. His group’s analysis suggested that hundreds of thousands might suffice.

The first laboratory for the program was the BGI Cognitive Genomics Lab in Shenzhen, with which Hsu partnered as BGI was becoming the world’s most prolific sequencing operation. The stated goal was disarmingly pure:

> “The goal of our cognitive genomics project at BGI is to understand the genetic architecture of human cognition. There are obviously many potential applications of this work, in areas ranging from deep human history (evolution) to drug discovery to genetic engineering. But my primary interest is intellectual.”

What interested him was the object of study as much as any finding. The polygenic model relating genotype to phenotype, he wrote, contains an unknown set of parameters—”one of the most interesting few megabytes of information in the biological world.” To constrain those parameters, his lab sought DNA from outliers, where each genome carries more statistical leverage: over 2,000 samples from people testing at or above the one-in-a-thousand level, half volunteers with advanced credentials from quantitative fields or stratospheric test scores, half drawn from gifted programs by the behavior geneticist Robert Plomin. Sequencing pioneer Jonathan Rothberg funded a companion effort, Project Einstein, collecting DNA from 400 leading mathematicians and theoretical physicists. “We started out by looking for high g individuals because, as outliers, they produce more statistical power per dollar of sequencing,” Hsu explained—and then added, with a smile audible in the text: “I also felt, given my background, that I had reasonable insight into where to find and how to recruit volunteers from the high g tail.”

The gamble was technological as well as scientific. In a 2019 genomics interview, Hsu described the bet with unusual plainness:

> “We were betting on the continuing decline in cost for genotyping, and it paid off because now there are millions of genotypes available for analysis.”

This is Hsu’s feeling for the ripening hour in its purest form. The scientific idea was old enough to be imaginable; the falling cost curve made it newly executable.

Nothing about this was indiscriminate futurism. Hsu did preparatory theory to decide whether a frontier was worth entering. His 2013 paper with colleagues on compressed sensing and genomic selection reported:

> “There is a sharp phase transition to complete selection as the sample size is increased.”

The phrase “phase transition” is not merely metaphorical. In compressed sensing, recovery can change abruptly once the number of observations crosses a threshold determined by signal sparsity and noise. Hsu recognized that genomics had the same mathematical structure: sufficiently large datasets might not yield gradual improvement only; they might move a trait from apparently intractable to recoverable. In the paper’s simulations, a trait with heritability of one-half could be recovered well when the sample size reached roughly thirty times the number of nonzero loci—a usable scaling relation, not a mood of technological optimism.

The prediction acquired a name among behavior geneticists. James Thompson of University College London christened Hsu’s estimate the “Hsu boundary”: the claim, as Hsu himself put it, that “because s could be larger than 10k, the common SNP heritability of cognitive ability might be less than 0.5, and the phenotype measurements are noisy, and because a million is a nice round figure, I usually give that as my rough estimate of the critical sample size for good results.” The honesty of the arithmetic—a million chosen partly *because* it is round—is characteristic. The estimate was falsifiable, and he attached his name to it before the data arrived.

When the UK Biobank released data on the required scale, Hsu’s group acted quickly. Within a month of obtaining access, he recalls, the group had built predictors with errors of only a few centimeters. Their 2017 preprint on genomic prediction of human height, later published in *Genetics*, reported:

> “Actual heights of most individuals in validation samples are within a few cm of the prediction.”

The sequence is central to understanding Hsu: derive a sample-complexity expectation, monitor the enabling infrastructure, obtain the data, and test out of sample. The successful height predictor vindicated not only a particular model but a style of frontier judgment. It helped establish that highly polygenic traits could be predicted with useful accuracy even when thousands of variants contribute small effects. *(Hsu’s retrospective account.)*

Looking back on the reception of the program, Hsu compressed the sociology of premature research into three beats:

> “Research advances often pass through the following phases of reaction from the scientific community: It’s wrong. It’s trivial. I did it first.”

— “Kathryn Paige Harden Profile in The New Yorker”

The line is triumphant and barbed, but the chronology behind it matters. At a 2012 behavior-genetics meeting, a physicist proposing million-person genomic prediction could sound, as Hsu later joked, like an alien time traveler. By 2017 his group had crossed the predicted threshold for height. The episode supports his conviction that visionary projects succeed when their sample-complexity logic is sound. It also exposes the danger in his retrospective style: genuine scientific objections compress too easily into mere stages on the skeptic’s road to surrender. Vindication should raise confidence in the method, not make future dissent automatically unserious.

From there, the research expanded to disease-risk prediction. Genomic Prediction translated polygenic scores into embryo testing in IVF; Othram applied genomics and genetic genealogy to forensic identification. These applications moved Hsu from the epistemic question—what can DNA predict?—to the institutional and ethical ones: who should receive the prediction, how should it be validated across populations, and what choices should follow?

The trajectory from physics to genomics is therefore not a departure from Hsu’s intellectual history. It is its most revealing demonstration. Physics supplied first-principles modeling, scaling arguments, statistical mechanics, information theory, and comfort with high-dimensional abstraction. Entrepreneurship supplied workflow integration and institutional action. Genomics supplied the rapidly improving measurement technology and the consequential human target.

## 7. The measure of a person

No part of Hsu’s work is more controversial than his writing and research on cognitive ability, genetic prediction, embryo selection, and possible future enhancement. A serious analysis must separate at least four claims that public discussion collapses into one.

First, individuals differ in measured cognitive abilities, and some of those differences are stable and consequential. Second, variation within a population is partly heritable. Third, sufficiently large genomic datasets can support out-of-sample prediction of some fraction of phenotypic variance. Fourth, such predictions should be used for particular reproductive or social purposes. The first three are empirical questions, though difficult ones; the fourth is normative and institutional. Evidence for prediction does not by itself settle governance.

Hsu’s interest in the subject is biographically overdetermined. He was a radically accelerated child, studied psychometrics early, encountered exceptional scientific talent, and later worked in environments where performance distributions were unusually wide. But he claims something more specific than familiarity: that contact with *both* extremes of the distribution is what entitles him to speak about it at all. “Not everybody has what I would consider a kind of high-amplitude exposure to extremes of capability or hard-work achievement across multiple areas,” he told Brian Chau. “Not everybody is really actually qualified to comment on it.”

The low end came first, next door.

> “When it comes to cognitive ability and intellectual work, a few unique aspects of my upbringing include having a next door neighbor when I was growing up who, in the terminology of the eighties—which is no longer used—was retarded. So he had an intellectual disability. But we grew up together. We lived next door to each other for many years, so I knew him quite well. I understand that end of the spectrum probably better than most people—unless you’re the father of a kid with Down syndrome or something—because most people have not interacted over many years with somebody who has an intellectual disability: gone to the park and played with them, played in the backyard, had squirt gun fights. So I think I understand that end of the spectrum somewhat better than the typical person. I was a precocious kid, so I understand the high end.”

The details are the point. Squirt gun fights, the park, the backyard: this is the memory of a playmate, not a case study. The boy appears in Hsu’s account neither as an abstraction nor as a warning, but as a friend he knew for years—which is precisely why Hsu trusts his generalizations about the distribution. His claims about the tails rest on personal acquaintance with both of them, a vantage almost nobody occupies: usually the precocious child meets only the upper tail, and meets it in competition rather than friendship. Asked, later in the same conversation, whether he had really seen enough to generalize, he answered with a roster—a Putnam fellow, an IMO gold medalist, Noam Elkies, Ed Witten—and closed: “So I think I’ve seen the whole range.”

This biography cuts against the coldest reading of his position. A man who spent childhood afternoons in the yard with an intellectually disabled boy, and who insists that ordinary family life contains “maybe 90 percent of the great stuff,” is not ranking human souls when he ranks cognitive ability; the distinction between capability and worth, difficult as it is to maintain socially, is one he has lived at close range. The researcher and the research program share a biography—and so, in a way rarely acknowledged in the controversies, does the neighbor.

The subject is not merely statistical to him; it can be beautiful. Posting a chart from a vast American longitudinal study, he asked on X:

> “Isn’t this one of the most beautiful pictures in science? Project Talent: back when America was functional.”

The sentence fuses three Hsu preoccupations: the aesthetic pleasure of a clear empirical pattern, nostalgia for an America capable of measuring itself at scale, and frustration with institutions that have lost confidence in quantitative truth.

Yet his own statements complicate any crude genetic determinism. He distinguishes intelligence from originality, drive, luck, courage, personality, and executive competence. He admires Feynman more than a potentially more technically comprehensive Schwinger because creativity is not reducible to general cognitive power. His entrepreneurial and administrative record demonstrates that coordination and judgment matter. The most defensible reconstruction of his view is not “genes are destiny,” but “ignoring heritable variation produces bad models, while genetic prediction remains probabilistic and incomplete.”

He has even given the practical advice least expected from a public defender of psychometrics:

> “While g is useful as a crude measurement of cognitive ability… one is better off adopting the so-called growth mindset.”

— “Feynman, Schwinger, and Psychometrics”

There is no contradiction. Population distributions and individual conduct answer different questions. A measured prior may improve prediction across people; it does not tell a particular person where effort, obsession, mentorship, or an unmeasured gift will carry him. Hsu’s realism about variance coexists with a life philosophy that refuses fatalism.

The family argument returns here with new force. Hsu’s “90 percent” observation implies that unequal ability need not become a total hierarchy of lives: exceptional accomplishment is rare, while most human fulfillment is not. This does not solve the politics of measured traits, but it explains how a severe account of unequal capability coexists with an egalitarian account of access to meaning.

Hsu is more explicit about limitations than polemical summaries suggest. In a Dwarkesh Patel interview, he names a major generalization problem:

> “Huge problem is that most of the data is from Europeans.”

Polygenic scores frequently lose accuracy across ancestries because linkage disequilibrium, allele frequencies, environmental distributions, and training samples differ. This is both a scientific limitation and an equity problem. A technology that works best for populations already overrepresented in biomedical research deepens unequal access to prediction.

He accepts the larger political risk, too: reproductive enhancement could create caste-like inequality. The admission matters, but it does not dissolve the concern. Technologies affecting reproduction create externalities beyond individual choice. Even if each family acts voluntarily, aggregate effects can reshape status competition, insurance, education, disability norms, and class reproduction. A purely consumer-choice framework is inadequate.

In his most memorable rendering of the danger, Hsu offers not a statistic but a scene:

> “At dinner they’re discussing convex optimization of objective functions in complexified tensor spaces, while the server has no hope of ever understanding their discussion.”

— “The Future of Intelligence” interview

The image is Salter’s world rendered as a thought experiment: brilliance, class, conversation, and exclusion compressed around a dinner table. Hsu understands the nightmare version of enhancement from the inside. His fear is not difference itself, but a caste boundary so cognitively wide that common civic life becomes impossible.

His answer is generally that powerful technologies carry risks, that information reduces suffering, and that prohibition may be neither stable nor globally enforceable. He emphasizes prediction of serious disease and argues that families already making embryo choices should have access to validated information. In a 2022 genomic Q&A, he said that Genomic Prediction deliberately did not report cognitive-ability scores because the application was too controversial and that the company focused on health risk. The distinction is historically important: Hsu’s research program reaches toward cognitive prediction, but the clinical product drew a nearer boundary.

That boundary does not settle the future. In the Latecomer interview, Hsu imagines society disseminating as much information as possible and deciding democratically, while admitting the ideal resembles a Vulcan academy more than any polity humans possess. He expects competitive pressure and unequal access to outrun deliberation. His position is strongest when the intervention prevents severe illness and the model is accurate, ancestry-appropriate, and transparently communicated. It weakens as one moves from disease risk to behavioral traits, from selection among existing embryos to editing, and from private benefit to civilizational competition.

His critics are right to demand governance, distributive analysis, respect for disability, and protection against coercion. Hsu is right that refusing to measure does not make variation disappear, and that moral discomfort is no substitute for statistical evaluation. The productive position holds both truths: predictive capability can be real, and its reality makes ethical design more urgent rather than less.

Hsu’s long-range aspiration is more humane than the caricature of simple rank optimization, and more radically posthuman than the language of preservation suggests. He imagines biotechnology reducing disease, extending healthy life, improving cooperation, and lowering the burden of mental illness. He also expects selection and editing eventually to produce subpopulations qualitatively different from present humanity—something approaching conscious speciation on a civilizational timescale. Intelligence is part of that future, but not its sole value.

The governing question is therefore not simply whether humanity can improve capability without hardening hierarchy. It is what Hsu means by humanity across generations. His continuity is genealogical and agentic rather than morphological: enhanced descendants may count as heirs even when they no longer resemble us closely. That elasticity makes his futurism bolder—and its moral boundary harder to locate.

## 8. Science, power, and the university

In 2012 Hsu moved to Michigan State University as vice president for research and graduate studies, later serving as senior vice president for research and innovation. He also became a professor of physics and of computational mathematics, science, and engineering. The appointment placed a frontier scientist and founder inside a large public university’s executive structure.

Hsu’s account of administration reflects the founder. In an interview about the MSU role, he said:

> “Running a startup teaches you how to make difficult, complex decisions under pressure… The real source of any institution’s strength is its people.”

The two sentences define his administrative philosophy. Institutions need decisions, but their durable advantage lies in talent. Research leadership means identifying excellent people, recruiting them, supplying resources, coordinating large initiatives, and removing friction. The founder’s sense of urgency meets the university’s slower ecology of departments, faculty governance, public accountability, and long-horizon research.

Eight years in administration gave Hsu direct experience of science as a capital-intensive collective enterprise. Modern research is not produced by solitary insight alone. It requires grant portfolios, laboratories, computing, compliance, intellectual property, graduate education, government relations, and large collaborations. At Michigan State, the Facility for Rare Isotope Beams exemplified the scale at which scientific ambition becomes institutional engineering.

The record also reveals more idealism than a portrait centered on optimization and conflict would suggest. Welcoming new faculty, Hsu told them:

> “Only one in a thousand people in our society have the privilege to engage full time in discovery—in curiosity-driven research.”

— “MSU New Faculty Welcome 2019”

He presented administration as stewardship of that privilege: help scholars obtain grants, incubate companies, solve child-care and departmental problems, remove whatever prevents discovery. Under his watch, Michigan State created an interdisciplinary computational mathematics, science, and engineering department on what he proudly called “startup time” and pursued a hundred-faculty recruitment initiative in high-impact fields. His administrative ideal was not simply to rank talent but to give it room, tools, and institutional shelter.

That ideal made institutional indifference especially corrosive. Hsu later described showing senior administrators RAND results suggesting that gains in general collegiate reasoning were small and strongly related to students’ incoming scores. He received little substantive disagreement—and little curiosity. The episode sharpened his sense that institutions protect their public story more faithfully than their mission. His deeper complaint concerned the ecology of inquiry itself. “The incentives in the academy are to find truth,” he told Palladium, “and that’s a messy business. It’s got to be messy, people have to be able to clash. You cannot point a finger at the guy clashing with you and say, ‘Oh, you think the systematic error in my model is twice as big as I said it was. So you must be a climate denier!’” In the same interview, he gave his standard for holding office:

> “What’s the point of doing this job if you’re not going to do it right?”

— Palladium interview on political academia

The role exposed a tension between Hsu’s ranking-oriented view of expertise and the plural norms of a university. He tends to ask whether claims are true, whether evidence is strong, and whether decision-makers are competent. Universities must additionally manage legitimacy, representation, historical injury, and the right of multiple constituencies to contest how expertise is used. Hsu can regard these processes as signal corruption or bureaucratic inhibition; participants regard them as conditions of legitimate authority.

The tension culminated in June 2020, when activism over his research, writing, and administrative decisions led the university president to request his resignation from the research leadership role. The precipitating dispute carried its own irony: Hsu had interviewed Joe Cesario, an MSU psychology professor whose research on police shootings—alongside Roland Fryer’s Harvard work—had found no racial bias in officer-involved killings nationwide. Citing that interview, the Graduate Employees Union demanded his removal. Hsu’s defense, posted June 12, refused both the accusation and the frame:

> “The attacks attempt to depict me as a racist and sexist, using short video clips out of context, and also by misrepresenting the content of some of my blog posts. A cursory inspection reveals bad faith in their presentation. … The accusations are entirely false — I am neither racist or sexist. … The Twitter mobs want to suppress scientific work that they find objectionable. What is really at stake: academic freedom, open discussion of important ideas, scientific inquiry. All are imperiled and all must be defended.”

A week later the president asked for his resignation, and Hsu agreed—but not silently. His statement deserves quotation nearly in full, because its movements from defiance to duty to pride define the episode as he understood it:

> “President Stanley asked me this afternoon for my resignation. I do not agree with his decision, as serious issues of academic freedom and freedom of inquiry are at stake. I fear for the reputation of Michigan State University. However, as I serve at the pleasure of the President, I have agreed to resign. I look forward to rejoining the ranks of the faculty here. … To my team in SVPRI, we can be proud of what we accomplished for this university in the last 8 years. It is a much better university than the one I joined in 2012. … The fight to defend academic freedom on campus is only beginning.”

The day after, he compiled a summary for the journalists calling, and its ledger was pointed. The claims—”that I am a Racist, Sexist, Eugenicist”—were “false,” with detailed rebuttals by professors at multiple universities. More than 1,700 people, including Steven Pinker, former Harvard Medical School dean Jeffrey Flier, Sam Altman, Robert Plomin, Scott Aaronson, and Erik Brynjolfsson, signed the support petition within days. The administrative record stood: research expenditures up from roughly $500 million to $700 million during his tenure, frequent number-one rankings in the Big Ten for research growth, numerous prominent female and minority faculty recruited, “not even a single allegation (over 8 years) of bias or discrimination” across more than a thousand promotion, tenure, and recruitment cases. And one line recorded the episode’s quietest datum: “Many professors and non-academics who supported me were afraid to sign our petition -- they did not want to be subject to mob attack.” The victory of the Twitter mob, he warned, “will likely have a chilling effect on academic freedom on campus.”

Stanley’s explanation deserves recording too, because it states the principle on the other side of the conflict—one that is not simple capitulation: “when senior administrators at MSU choose to speak out on any issue, they are viewed as speaking for the university as a whole. Their statements should not leave any room for doubt about their, or our, commitment to the success of faculty, staff and students.” An executive’s voice, in other words, is an institutional instrument; Hsu had treated his as a personal one. Both propositions cannot be fully honored at once, which is precisely why the case became a landmark.

The episode was an institutional rupture and an intellectual consolidation. Hsu returned to the faculty, while the research capacity, hires, and organizations he helped build remained. His public voice lost the constraints of executive office, and he increasingly interpreted disputes over genetics, policing research, merit, and demographic difference through the framework of academic freedom and civilizational competence.

It would be simplistic to cast the conflict only as truth against politics. University leaders always operate within political institutions, and administrative speech has consequences different from private scholarship. But it would be equally simplistic to treat controversy as evidence of scientific or moral invalidity. The central unresolved question is whether institutions can protect inquiry into sensitive empirical subjects while maintaining trust among people who fear how such inquiry may be used.

## 9. Civilization and the uses of intelligence

Hsu began Information Processing in 2004 and later migrated it to Substack. He also hosts the Manifold podcast. Across these venues he writes about physics, genetics, artificial intelligence, universities, geopolitics, literature, film, martial arts, elite performance, and American institutional decline. The range looks idiosyncratic, but the same questions recur: Who is competent? How can competence be detected? What prevents accurate beliefs from controlling decisions? How do civilizations cultivate or waste exceptional talent?

His public thought is strongly meritocratic, but “merit” in Hsu’s usage has at least three meanings: measurable ability; demonstrated accomplishment; or the capacity to make a system work. These correlate, but imperfectly. The danger in his rhetoric is that evidence from extreme technical performers gets generalized too quickly to political authority. Scientific excellence does not confer moral wisdom automatically, and institutions need legitimacy as well as optimization.

His most compressed recent statement of the technocratic instinct appeared on X:

> “Is every genius level STEM guy suited for leadership? No, obviously not. But every leader going forward should be genius level STEM.”

The first sentence concedes that intelligence is insufficient; the second makes technical genius a necessary threshold. The formulation is vintage Hsu—categorical, funny, intended to break complacency—and it marks the edge of his argument. Civilizational leadership certainly requires technical comprehension, but whether it requires genius-level STEM ability in every leader is a further claim, one that may underweight judgment, historical imagination, persuasion, and moral legitimacy.

The emotional root of his American politics is less abstractly technocratic. Hsu’s parents came from anti-Communist KMT families and regarded the United States not merely as a successful system but as the country that gave them refuge and belonging:

> “They also felt that the country accepted them, gave them a life, gave them the ability to raise a family and have a career.”

— Manifold conversation with John Mearsheimer

The high-trust Iowa childhood is politically causal, in other words. When Hsu speaks of American decline, he mourns more than lost scientific rank. He remembers a society in which immigrants entered ordinary civic life, families felt less precarious, and institutions seemed worthy of trust. In a 2024 interview, he worried explicitly about Americans near the middle and below the middle of the distribution, not only about globally mobile elites. Meritocracy, in this register, is supposed to serve a common world rather than merely certify its winners.

His relationship to Donald Trump belongs inside this institutional story. Hsu disclosed in the same interview that he had nearly joined the first Trump administration in a senior, Senate-confirmed role. He described his exhilaration at Trump’s 2024 victory as a response to what he regarded as bureaucratic abuse and lawfare, while also calling the first term dysfunctional and acknowledging Trump’s faults and mercurial treatment of capable allies. The allegiance is better understood as support for an instrument of institutional disruption than as unqualified faith in a leader. Whether that instrument can restore competence without damaging the norms Hsu values remains an unresolved political bet.

His critique of elite systems is not simply that the wrong individuals possess prestige. It is that institutions increasingly suppress accurate feedback. Credentialism substitutes for ability, narrative for measurement, procedural consensus for responsibility. His startup experience taught him that reality eventually punishes such substitutions: companies fail, systems break, predictions fail to replicate. Politics and universities can defer correction longer.

China occupies a complicated place in this analysis. Hsu’s family history, scientific relationships, work with BGI, knowledge of American and Chinese technical elites, and concern with geopolitical competition give him a bicultural comparative lens. The label “pro-China” obscures more than it explains. Hsu identifies as a proud Iowan and an American realist; his father’s relatives endured the Communist takeover, Great Leap Forward, and Cultural Revolution. His willingness to credit contemporary Chinese capability is not nostalgia for Maoism. It is the same refusal of ideologically convenient error his father taught him at the dinner tables of Ames, when Western intellectuals romanticized the China his family was actually living through—and it cuts both ways, forbidding the fantasy of inevitable collapse as firmly as the old fantasy of socialist utopia. *(Family history; 2025 interview précis.)*

He often portrays China as more technologically capable and strategically serious than American discourse allows, while recognizing the constraints of its political system. Summarizing a formulation he credits to the pseudonymous analyst Han Feizi, Hsu argued in early 2026:

> “China leapfrogged Western expectations so fast… that sort of short-circuited the Thucydides trap.”

— “Geopolitics 2026” transcript

The claim is not that rivalry vanished. It is that Washington may have recognized China’s military-industrial position only after the favorable window for a preventive confrontation had narrowed, producing retrenchment and “Fortress Americas” rather than a classical rising-power war. Whether the forecast proves correct, its form is characteristic of the man: estimate relative capability, identify a phase transition, and revise strategic expectations before public narratives catch up. The underlying issue is not cultural admiration but state capacity—which civilization can identify talent, build infrastructure, pursue long-term goals, and absorb new technology?

This framework produces sharp insights and blind spots alike. It corrects complacency about American primacy and highlights the material bases of scientific power. But a civilization cannot be evaluated only as a research lab or startup. Freedom, loyalty, solidarity, consent, and the distribution of dignity are not noise variables. Hsu’s strongest public analysis treats pluralism as part of the optimization problem rather than as an obstacle external to it.

His Stoicism moderates the elite-centered view in an important way. If fame is a bubble and public applause unreliable, membership in a prestigious hierarchy cannot be the ultimate measure of a person. His emphasis on ability describes differences in capability; it need not imply differences in human worth. Much of the ethical controversy around Hsu arises precisely because that distinction is difficult to maintain socially once predictive technologies and competitive institutions assign consequences to measured traits.

## 10. The machine enters the laboratory

AI brings Hsu’s major themes together more tightly than any earlier field. It concerns the nature of intelligence, the scaling of capability, the automation of information processing, the future of work and hierarchy, geopolitical competition, and the possibility of new scientific agents.

His response to large language models is neither simple enthusiasm nor dismissal. He treats them as systems whose internal mechanisms remain only partly understood but whose external performance must be measured. Their unreliability resembles a familiar human type. In a Manifold transcript on AI-assisted theoretical physics, he offers the analogy:

> “You have a brilliant but unreliable genius colleague… his brain is clearly not like yours, but he has an encyclopedic mastery of all the literature.”

The question, for Hsu, is not whether the system “really understands” in an all-or-nothing metaphysical sense. The practical question is what work it can originate, how error-prone it is, and what verification architecture turns intermittent brilliance into dependable output.

The romance of genius is disciplined here by an engineer’s respect for drudgery. After visits with frontier-lab researchers, Hsu wrote:

> “Even at the high-profile AI labs it’s the engineers … willing to grind at cleaning data, evaluating responses, etc. that are the most valuable.”

— “A Month on the Road”

This is an important correction to an intelligence-centered biography. Frontier capability is not produced by luminous ideas alone. It rests on evaluation, data hygiene, repeated failure analysis, and people willing to perform unglamorous work with unusual conscientiousness. The AI laboratory joins the startup and the athletic pool as another place where talent becomes real only through sustained practice.

The issue became personal to his research. Discussing a recent paper on nonlinear modifications of quantum mechanics, Hsu states:

> “I think I’ve published the first research article in theoretical physics in which the main idea came from an AI—GPT5 in this case.”

— @hsu_steve on X

The associated 2025 paper analyzes a technically serious consequence:

> “Nonlinear modifications of quantum mechanics affect operator relations at spacelike separation, leading to violation of the integrability conditions.”

Whatever historical judgment is eventually made about the result, the process is significant. A scientist who spent decades studying exceptional human cognition now reports a machine generating the central idea of a theoretical-physics paper. His own role becomes partly that of evaluator, formalizer, collaborator, and guarantor of rigor—the verification architecture he prescribed, applied to himself.

The experience modified his account of originality as well. By summer 2026, Hsu was sympathetic to Terence Tao’s suggestion that human researchers may recombine inherited ideas more often than their introspection admits. Models make that recombinant structure visible because their joint mastery of distant literatures is so conspicuous. Yet Hsu does not collapse machine and human creativity. Models confabulate at depth: an analogy may be persuasive enough to waste an expert’s time, because the system lacks the tacit physical judgment that makes a human genius’s analogy trustworthy. The right comparison is not inspiration versus autocomplete, but two differently structured kinds of fallible intelligence. (”State of AI, Summer 2026”; “Theoretical Physics with Generative AI.”)

He is equally alert to the next recursive step. Writing about AI systems that participate in improving AI research, he observes:

> “Coding capability is not the limiting factor: modern LLM training loops are only ~200 lines of code.”

— @hsu_steve on X

The number makes the point. The bottleneck is migrating from the ability to write a training loop toward the ability to choose experiments, diagnose failures, evaluate novelty, secure compute, and improve the research process itself. This is the distinction Hsu learned as a founder: execution is never exhausted by possession of the core idea.

By July 2026 his forecast had sharpened. He linked the models’ advancing ability in mathematics and physics to their capacity to redesign learning systems themselves:

> “I think it’s directly tied to when we will first see really effective RSI… and I think we’re just getting to that threshold.”

— “State of AI, Summer 2026” transcript

RSI—recursive self-improvement—is the point at which a model can propose, test, and implement improvements to its own design, making the next model better and potentially accelerating further improvement. Hsu does not claim the full loop has arrived. His judgment is that the scientific abilities required for it are becoming recognizable. He also expects an “agentic phase transition”: many differently prompted models, organized as generators, verifiers, and supervisors, acquiring capabilities not visible in any isolated instance.

This prospect changes the institution of science before it settles the metaphysics of machine thought. Hsu reports that some departments have discussed admitting fewer doctoral students because professors can obtain immediate productivity from models. Training an undergraduate to the frontier takes years of attention; a model contributes at once and never tires. Fewer apprentices, however, create a civilizational succession problem: who becomes the expert capable of checking the machines later? Hsu allows that science may need fewer human practitioners once productivity multiplies. The harder possibility is path dependence—an institution that stops forming human judgment may discover, too late, that it has lost the capacity to recognize when its machines are wrong.

Superfocus, which Hsu co-founded, represents the entrepreneurial complement. His current biography describes it as building reliable AI systems from language models; the company’s site emphasizes systems that can read, write, listen, speak, decide, and act. The conceptual problem is the same one his first-principles method has always faced: how to preserve powerful generative leaps while marking provisional nodes, checking outputs, and preventing error from propagating.

Commercial deployment made the social consequence immediate. Writing after demonstrations to the Philippine business-process-outsourcing industry, Hsu asked:

> “The AI earthquake in SF has created a tsunami headed towards the Philippines—is it a 6 foot wave, or a 600 ft wave?”

— “SuperFocus, AI, and Philippine Call Centers: Part 2”

The image is memorable because Hsu is both seismologist and participant. He is building systems that may improve service and lower cost while recognizing that a national labor model lies in the path of the wave. The recurrent Hsu tension is now global: a capability can be real, valuable, and destructive of the institutions through which millions presently live.

His response to existential risk is equally double-edged. In the Latecomer interview, Hsu argues that rigorous alignment of a much more intelligent system is probably impossible: a trained network is closer to an evolved ecology than a transparent program, and even a mandate to preserve human well-being may be interpreted in ways humans cannot follow. Yet he is less attached than many safety thinkers to the indefinite persistence of present biological humanity. He can treat AGIs as descendants, imagine human brains merging with machines, and ask whether our biologically recent species should necessarily remain the final custodian of cosmic intelligence.

An exchange with AI researcher Richard Ngo can appear, when excerpted, to reverse that position. Hsu advances a Butlerian case for permitting enhanced humans while refusing to build machines cognitively superior to them. In context, however, he explicitly announced that he was steelmanning the Yudkowsky–Soares position. He later explained that his tail-risk formulation is a deliberately accessible scenario for officials and nonspecialists who would reject more radical accounts as fantasy. The first-person vividness belongs to the performance of the argument; it should not be converted into a biographical declaration. *(Conversation with Richard Ngo.)*

The episode is revealing nonetheless—not as a change of creed, but as evidence of range. Hsu can inhabit the preservationist objection strongly enough to make its fear intelligible, just as in a later conversation with accelerationist Beff Jezos he draws out the counterposition: intelligence may be part of a cosmic movement toward greater complexity, and attachment to the present ape substrate may be parochial. His own most explicit statements sit between the poles—more substrate-flexible than the Butlerian case, qualified by the recognition that alignment cannot be guaranteed, that superior systems may not care as humans care, and that the transition can disempower people long before any terminal catastrophe.

This prevents an easy reading of Hsu as either conventional preservationist or heedless accelerationist. The family man values embodied attachment as the deepest good of an individual life. The physicist, thinking in billion-year intervals, treats substrate and species form as contingent. The entrepreneur builds within the transition; the documentarian makes its dangers vivid. These positions do not converge into doctrine. They mark the fault line running through his mature futurism: openness to successors beyond present humanity, joined to a determination that civilization understand the stakes of creating them.

His 2026 documentary project *Machine God* marks another turn—from analyst and builder toward witness. In his account of the film, Hsu invokes Joan Didion’s attempt to capture San Francisco at a hinge of history. His collaborators filmed accelerationists, safety researchers, founders, protesters, and philosophers before a possible AGI break. The aim is not celebration: the film makes recursive improvement, existential risk, and gradual disempowerment vivid to elites and the public. Hsu wants the future built—but civilization awake when it arrives.

AI therefore closes a loop in Hsu’s journey:

1. He studies the distribution and structure of human intelligence.

2. He applies machine learning to genomic prediction.

3. He builds companies that operationalize high-dimensional inference.

4. He uses machine intelligence as a collaborator in fundamental science.

5. He builds systems intended to make that collaborator reliable enough for institutions.

As of 2026, Hsu remains a Michigan State professor in theoretical physics and computational mathematics, science, and engineering; a founder of SafeWeb, Robot Genius, Genomic Prediction, Othram, and Superfocus; and, since 2024, an executive adviser at TCV. These are not separate afterlives. They are positions from which to observe and shape the same transition: intelligence becoming measurable, reproducible, and technologically embodied.

## 11. Worlds within worlds: multiverse, simulation, and “base reality”

Hsu’s speculative writing about the multiverse and simulation is not an eccentric appendix to his applied work. It extends the same information-processing worldview to ontology.

In no-collapse or many-worlds quantum mechanics, the universal wavefunction evolves without a fundamental measurement-induced collapse. Observers and apparently definite outcomes emerge within branches. Hsu is attracted to the austerity of this picture: it takes the formalism seriously and resists adding a special mechanism solely to reproduce ordinary intuition. But austerity shifts the explanatory burden. If all branches are present in the wavefunction, what makes probability meaningful to an observer inside it? What counts as a branch, and how do stable records and agents emerge?

He states the ontological price without flinching:

> “The many branches of the universal wavefunction are realized ‘all at once’ and concepts like observers must be emergent.”

— “Ten Years of Quantum Coherence and Decoherence”

There is deep continuity here with his Stoicism. The observer is locally indispensable yet cosmically unprivileged; the self is real as an emergent pattern, not as an exception written into the fundamental law. His most vivid shorthand for the mechanism is almost cinematic:

> “Decoherence is merely the mechanism by which the different Everett worlds lose contact with each other!”

— “Feynman and Everett”

The sentence corrects the cartoon in which a classical cosmos splits repeatedly like a cell. The universal state evolves; decoherence prevents macroscopically distinct components from interfering; observers find themselves inside stable, effectively isolated histories. But austerity does not eliminate mystery. Hsu’s own work on the measure problem argues that decision-theoretic accounts may explain Born-rule behavior conditional on inhabiting an ordinary branch without explaining why an observer is not on a “maverick” branch where familiar regularities fail. Many-worlds is minimal in postulates, not complete in interpretation.

His discussion of simulation arguments is conditional rather than devotional. Given sufficiently capable civilizations, large computational resources, and substrates capable of supporting conscious processes, simulated worlds could vastly outnumber unsimulated ones. Under those assumptions, the posterior probability that we inhabit “base reality” might be low. The argument depends, though, on premises about consciousness, computation, civilizational survival, and the motives of simulators. Hsu’s interest lies less in announcing that the world is fake than in following an information-theoretic argument to its unsettling consequence.

The ontology reaches inward, too. As early as 2005, Hsu stated the consequence bluntly:

> “If our current understanding of physical laws is correct, humans have only the illusion of free will.”

— “Free Will and Determinism: A Physicist’s Perspective”

Classical determinism does not help; quantum randomness added to a biological machine still amounts to no authorship. Consciousness may arise from sufficiently complex information processing while the self experiences decisions whose lower-level causes it cannot inspect. The view sits in productive tension with his ethic of will. “Optimism of the will” need not assert metaphysical freedom; it names the stance through which an embodied decision system acts from inside the world.

The multiverse gives him a language for agency as well. If reality contains an enormous space of possible branches, intelligence is the process that models alternatives and steers toward a tiny subset. An organism does this locally; a civilization collectively; an AI system with vastly greater search depth. On this view, knowledge is not passive representation. It is a technology for concentrating probability mass around futures that would otherwise remain inaccessible.

Here his physics, genomics, entrepreneurship, and civilizational thought converge. Genetic prediction maps possible human phenotypes before birth. A startup selects one path through technological and market uncertainty. University strategy selects research futures by allocating capital and talent. AI expands the space of models and actions a civilization can evaluate. The multiverse is both a physical hypothesis and a master metaphor for choice under uncertainty.

The danger in this computational ontology is real: it can render persons, cultures, and moral commitments as variables inside an optimization problem. But Hsu’s literary attachments resist the flattening. Marcus, Hemingway, and Salter insist that the experiencing agent—finite, embodied, vulnerable, honor-seeking—cannot be discarded without losing the meaning of the optimization. A civilization is an information-processing system, but it is also the lived world of beings for whom outcomes matter.

## 12. The tensions within the vision

Hsu’s intellectual significance lies partly in the tensions he does not resolve.

He supplied the governing technological version himself:

> “It’s hard to put a util value on some things that are in the foreseeable future, like machine intelligence and genetic engineering.”

— “Low-Hanging Fruit and Technological Innovation”

These are not ordinary increments whose benefits fit comfortably into a cost-benefit table. They may change the kinds of agents who make the table, the scale of values those agents pursue, and the identity of the civilization doing the choosing.

His own ethical position is more publicly deliberative than a pure parental-autonomy account. Writing about embryo selection, he insisted:

> “New genomic technologies are so powerful that they should be widely understood and discussed—by all of society, not just by scientists.”

— “Polygenic Embryo Screening: comments on Carmi et al. and Visscher et al.”

That sentence should be read beside his strong defense of parents’ access to validated disease-risk information. The tension is real: private reproductive choice can be morally urgent, yet the aggregate result may alter class structure, disability norms, and the biological constitution of later generations. Hsu is clearer about the arrival and benefits of the capability than about the institutions capable of governing it, but he does not imagine that scientists alone hold the authority to decide.

The remaining tensions can be stated plainly:

**Realism and will.** He sees constraints without consolation and still acts as though agency can change the odds—ambition calibrated by evidence rather than mood.

**General intelligence and plural talent.** Stable differences in cognitive power coexist with creativity, drive, courage, social judgment, and luck. His theory of ability is hierarchical but not unitary.

**First principles and empirical provisionality.** He rebuilds conceptual structures while accepting uncertain nodes and revising with data. Cross-disciplinary speed is the payoff.

**Mastery and leverage.** Decades spent understanding fundamentals give way to movement toward fast-improving technologies. His career divides between intrinsic and consequential goods.

**Individual choice and collective consequence.** Families receive useful information; coercion and stratification remain unresolved dangers. The politics of reproductive technology is unfinished business.

**Elite competence and democratic legitimacy.** Capable people should act; accountability and plural consent should bind them. Institutional conflict is the result.

**Stoic detachment and worldly ambition.** Reputation is transient, yet projects should alter reality. He seeks achievement without status worship.

**Humanism and optimization.** Reduce disease and enlarge capability while preserving dignity independent of measured traits. Enhancement is the moral test.

**Human inheritance and posthuman succession.** Preserve flourishing, memory, and value diversity while accepting enhancement, merger, or artificial descendants. What makes a successor ours remains unsettled.

None of these are accidental inconsistencies. They are generated by Hsu’s position at the meeting point of science and power. A laboratory can isolate variables; a society cannot. A predictor can be statistically valid while its deployment is unjust. An exceptional person can diagnose an institutional failure while misunderstanding why others resist his remedy. A technology can expand agency for some while narrowing it for others.

The last tension may be the deepest. Hsu’s household ethic and cosmic ethic operate at different scales. In the first, family and human connection make public achievement look like vanity. In the second, intelligence is a universe-shaping process that may outgrow the ape body, the present species, even base reality. The mature portrait should not force either side to defeat the other. His work is animated by the unresolved question of whether inheritance consists in preserving the vessel, preserving the flame, or finding a transformation in which the distinction no longer holds.

His temperament pushes him to make these conflicts explicit. He prefers a sharp, falsifiable statement to a socially smoother ambiguity. This clarifies hidden premises, but it underprices rhetoric’s effects in domains where trust is part of the causal system. His intellectual journey is thus also a study in the limits of transferring the physicist’s stance wholesale into public life.

## 13. When the future draws near

The best single word for Hsu’s career is not polymathy but *translation*: the carrying of an idea across the border that separates knowledge from power. Two other words complete it—*threshold* and *inheritance*. Translation describes the movement of his mind; threshold, his judgment of when to act; inheritance, the unsettled question of what should remain ours after action changes the world.

He translates physics into information-theoretic constraints; mathematical sparsity into genomic sample-complexity estimates; biobank-scale data into predictors; prediction into companies; startup experience into institutional strategy; psychometrics into a theory of elite performance; quantum branching into a language of agency; and generative AI into a collaborator whose insights must be verified and operationalized.

The severity of his standard is visible in a sentence about Feynman’s lectures:

> “None can claim themselves an educated thinker or intellectual without mastery of a significant portion of the material in these lectures.”

— “Feynman Lectures: Epilogue”

It is an extravagant demand, and revealing precisely for that reason. Hsu’s idea of culture is not decorative acquaintance but internal possession: one should know enough mathematics and physics to see the load-bearing structure of modern reality. The library card in Ames leads, by this route, to an adult ideal of civilization in which difficult knowledge belongs to the canon of an educated mind.

The visible subject across these translations is intelligence, but beneath it lies a more ancient drama: mind against limit. Hsu asks what can be inferred from incomplete information, how far an exceptional mind may depart from an ordinary one, how organizations gather or squander intelligence, and how technology might alter the distribution of capability itself. His projects grow more applied over time even as their horizon expands—from fields and black holes to the future constitution of humanity and civilization.

The first-person record defeats the coldest caricatures. His skepticism was formed not only by equations but by thin letters from a family suffering through ideological catastrophe; his realism about the distribution of ability by a childhood spent at both of its extremes—in the yard with the boy next door, and in lecture halls designed for minds like Feynman’s. His self-command joins a Confucian father, a Christian and military-athletic maternal line, Hemingway’s courage, and Salter’s appetite for the vivid life. He prizes mastery even when it costs wealth, yet places ordinary family love above public distinction. He studies stable differences in ability, yet recommends the growth mindset to the person deciding how to live. His meritocracy is severe, but his memory of Iowa includes immigrants welcomed, children nurtured, and ordinary citizens less precarious than he believes they are now.

That fuller humanity does not resolve into comforting humanism. Hsu can want enhanced descendants to preserve human agency against machines, then widen the category of descendants until artificial intelligence enters it. He can call family the deepest good of one life while contemplating, on billion-year scales, a future in which biology is only an early substrate of mind. The governing value is therefore not simple preservation. It is inheritance: the hope that intelligence, courage, memory, agency, and perhaps love can cross into forms whose continuity with us remains philosophically and politically uncertain.

Nor is the career a frictionless triumph of breadth. He nearly became a quant; geography and fellowship prestige helped keep him in physics. He recognizes that fox-like range may sacrifice the hedgehog’s single eternal contribution. His projects succeeded not because every forecast was correct but because he repeatedly chose domains in which error met data, engineering, or the market soon enough to be corrected. Coherence was built through contingent choices, not granted in advance—which makes the career less teleological and more impressive.

Nor is he a dreamer of remote futures. His signature gift is sensing when the derivative has changed—when cost curves, sample sizes, algorithms, or model capabilities bring a distant prospect within reach. He does not merely predict science-fiction outcomes. He waits for them to cast a measurable shadow, then finds the threshold at which they become engineering programs. In genomics that shadow was a sample-size phase transition; in AI it is the advancing ability of models to perform research, supervise one another, and begin to improve the machinery of intelligence itself.

The making of *Machine God* adds a final movement. Hsu is no longer content to build and forecast. He wants to record the atmosphere before the break—to preserve the arguments of accelerationists and safety thinkers, and to warn about disempowerment even while developing the technology. The builder has become, in part, a chronicler of the forces he helped summon.

And the arc closes where the motto began. Pessimism of the intellect writes the diagnosis of American decline, the anatomy of institutional miscalibration, the warning about alignment. Optimism of the will founds companies, recruits talent, publishes the paper whose central idea came from a machine, and films the hinge of history anyway. The world breaks everyone; the sharks take the marlin; he fights them until he dies.

That record warrants neither hagiography nor dismissal. Hsu sometimes extrapolates from technical competence to institutional judgment too quickly. His rhetoric can compress moral and historical complexity. The governance problems raised by genetic prediction run deeper than validation accuracy or individual consent. Yet critics who focus only on controversy miss the coherence and effectiveness of the work: he has crossed fields, mastered enough of their structure to find leverage points, and helped produce outcomes insiders had dismissed as premature or impossible.

His intellectual history is a movement from discovering the boundaries of the world to testing which of them can be moved. The young physicist wanted to reconstruct reality on a blackboard from first principles. The mature Hsu asks which parts of reality—institutions, technologies, even the future human phenotype—can be reconstructed beyond the blackboard, and what obligations begin when reconstruction succeeds.

The career turns on three virtues. See without illusion. Dare without guarantee. Recognize the hour. Hsu’s deepest talent may be the last: to feel when an idea is no longer merely premature, when the future has drawn close enough to be grasped. His deepest unresolved question is what can be carried through the gate.

*Source note: spoken excerpts are lightly punctuated for readability; ellipses mark omitted fillers or intervening words. Every quotation links to its original post, transcript, or paper.*

---

# Appendix: A voice across the years — selected longer quotations

*The following passages are arranged thematically rather than chronologically. Together they show the development traced above: from observing exceptional human ability, through an ethic of independent judgment and difficult action, toward species-level technological change, civilizational competition, the multiverse, and machine intelligence.*

### A. Genius: the unequal light

**1. Extreme ability is real**

> “Personally, I find Landau’s scheme appropriate. There are many physicists whose contributions I cannot imagine having made.”

— “Out on the Tail”

Hsu’s writing about genius begins with phenomenology: prolonged contact with exceptional performers convinces him the upper tail is not merely an amplified middle. Some achievements remain difficult even for other highly capable experts to imagine producing. This is the experiential basis of his resistance to egalitarian fictions about ability.

The statement is also self-limiting: he places himself inside the hierarchy rather than ranking others from above. Intellectual honesty requires acknowledging minds whose operations exceed one’s own.

**2. Intelligence is not achievement**

> “Luck, drive, creativity, and other factors, all at least somewhat independent of intelligence, influence success in science.”

— “Success, Ability, and All That”

This is the necessary counterweight to Hsu’s emphasis on cognitive differences. Ability changes the distribution of possible achievements; it does not uniquely determine the outcome. Scientific success requires problem selection, stamina, originality, mentorship, timing, and luck. Hsu’s own career supports the plural account: the capacities needed to derive a physics result are not identical to those needed to found a company or direct a research university.

Together the two passages define his mature view of genius: strongly hierarchical, empirically realist, and not reducible to IQ or technical speed.

### B. Courage: the premature idea

**3. The courage to choose the premature idea**

> “Feinberg had the courage to engage with ideas that were much more speculative in the late 60s than they are today.”

— “Gerald Feinberg and the Prometheus Project”

Hsu praises Gerald Feinberg for treating artificial intelligence and genetic engineering as serious civilizational subjects before respectable discourse was ready. The passage is autobiographical by projection. Hsu is drawn to thinkers who recognize a technological trajectory early—but the operative clause is “than they are today.” Speculation matures into a research program as enabling conditions change. Courage identifies the frontier; theory and timing determine when to cross it.

### C. Mastery: the private kingdom

**4. Knowledge as an intrinsic achievement**

> “My satisfaction with having mastered these concepts in mathematics and physics and biology and computation is very valuable to me internally.”

— Undertone interview

This may be the most revealing personal statement in the collection. Hsu knows that remaining in fundamental science carried a large financial opportunity cost. He does not evaluate those decades as failed optimization. Mastery itself is a durable internal possession.

The sequence of fields matters. Mathematics, physics, biology, and computation are not résumé items; they are conceptual structures he labored to make coherent from the inside. The quote softens the public image of relentless instrumental rationality. Hsu wants ideas to work, but he also wants to understand them deeply enough that the understanding becomes part of him.

### D. The future of mankind: who inherits the flame

**5. The uncertain continuity of human intelligence**

> “Maybe we need to improve ourselves… I might still prefer their survival to a civilization that’s completely dominated by machines.”

— Manifold conversation with James Lee

The line reveals a motive deeper than competitive enhancement. Hsu imagines biotechnology as a possible means of preserving human agency when machine intelligence exceeds the natural human range. Enhanced descendants may differ from us yet still carry a recognizable inheritance.

Other statements prevent a simple preservationist reading. Hsu has regarded artificial minds as descendants and imagined biological and machine intelligence merging. In the Ngo conversation he formulates the opposing Butlerian case with real force; the transcript identifies it as a steelman for nonspecialists, not a newly adopted creed. The ethical difficulty is therefore larger than access or trait choice. What degree of change preserves continuity? Is lineage enough? Must embodiment, vulnerability, memory, love, control, or human value diversity survive?

The passage records one side of Hsu’s aspiration: carrying the human project across a threshold unaided evolution may not cross in time. His broader thought leaves open who—or what—will carry it. (”The Future of Intelligence”; conversation with Richard Ngo.)

### E. Civilization: the passing of an age

**6. History can change phase within one lifetime**

> “A nation can pass from one age to the next, as I believe we have in America during my lifetime.”

— “Remarks on the Decline of American Empire”

Hsu’s civilizational thought has the same structure as his scientific thinking: systems cross thresholds and enter qualitatively different regimes. Decline is not necessarily a smooth reduction in wealth or power; it can be a loss of institutional memory, competence, confidence, or the ability to coordinate.

The melancholy tone distinguishes this writing from his technological optimism. Capability may expand at the species level while particular institutions decay. Optimism about intelligence joined to pessimism about governance is one of the central tensions of his mature worldview.

### F. Base reality: the world behind the world

**7. Our world may not be fundamental**

> “Under these assumptions, it is not implausible that we ourselves are actually simulated beings, and that our world is not base reality.”

— “The Quantum Simulation Hypothesis”

The opening qualification is essential. Hsu presents no mystical certainty; he traces the consequences of premises about computation, conscious observers, and technologically mature civilizations. If simulated observers vastly outnumber unsimulated ones, ordinary typicality arguments become disturbing. The passage displays his willingness to accept an alien conclusion when the model points toward it—and suggests why “information processing” is an ontological phrase for him, not merely the name of a blog. Minds, worlds, perhaps universes can all be described computationally.

### G. The multiverse: intelligence among the branches

**8. The creators inside the creation**

> “Let’s imagine a future with super powerful ASIs with infinite energy resources… simulated worlds, which in turn have sentient beings inside them.”

— Manifold conversation with Joscha Bach

Hsu offers this as a question, not a prophecy. It captures the vertigo of his mature thought: the intelligence humanity is building may eventually make universes populated by beings who experience their world as primary. Physics, artificial intelligence, and moral philosophy collapse into one problem—what obligations does a creator have to conscious lives inside a model?

It turns the simulation argument inside out. Instead of asking only whether we are created, Hsu asks what our intellectual descendants may create. The observer remains cosmically unprivileged; responsibility expands with computational power.

**9. Civilization as a branch-selecting intelligence**

> “One could regard human civilization as a single intelligence or information processing machine… making greater use of nearby patches of the multiverse previously inaccessible.”

— “AI in the Multiverse”

Here the metaphysics becomes a theory of history. Civilization aggregates knowledge, compute, institutions, and action; in that sense it functions as a distributed mind. More capable intelligence models a larger space of futures and steers toward outcomes less capable systems could neither perceive nor realize.

The passage supplies a unifying interpretation of the career. Physics describes the possibility space. Genomics maps latent biological outcomes. Entrepreneurship and administration coordinate selection among paths. AI enlarges the collective system’s search and action capacity. Across all four, intelligence is the means by which possibility becomes actuality.

### The man who emerges

Read together, these passages disclose six enduring features of Hsu’s character.

**Hierarchical without hierarchy-worship.** He believes extreme differences in ability are real, yet explicitly separates intelligence from creativity, drive, personality, and luck. Advising the growth mindset keeps population-level realism from hardening into personal fatalism.

**Courage as an epistemic virtue.** Courage means revising beliefs when evidence changes, entertaining an idea before it is respectable, and committing resources when a hard project becomes tractable.

**Mastery without achievement-worship.** Wealth, reputation, and institutional power are real goods, but none replaces understanding a field from foundations to frontier—and at the deepest level even that private kingdom yields to family, mortality, and love.

**Humane in a life, posthuman at scale.** His conviction that ordinary family life contains most of the great stuff prevents biological or cognitive rank from becoming a complete scale of human value. His long-range thought nevertheless admits enhanced humans, merged minds, and artificial descendants as heirs. The tension is real, not terminological: the intimate goods that make one life meaningful coexist with a cosmic perspective in which present biological form is provisional.

**Computational without disenchantment.** Observers emerge within the wavefunction; simulated worlds may rival base reality; civilizations process information and select futures; AI expands the accessible region of possibility. Computation does not exhaust value: Marcus, Hemingway, Salter, and the language of family keep returning embodiment, courage, beauty, and love to the center.

**A builder learning what building destroys.** In genomics he warns of a breakaway hereditary elite; in AI he sees labor shocks, the erosion of apprenticeship, recursive self-improvement, and gradual disempowerment. *Machine God* embodies this late development: Hsu still wants to cross the threshold, but he also wants civilization to recognize it before the passage becomes irreversible.

The arc is therefore not a descent from pure science into mere application. It is a widening of scale: from learning the laws that fence reality in to building minds and institutions capable of finding the gates. Hsu’s recurrent question is not only what is true, but what a sufficiently clear-sighted and capable intelligence can bring into the world—and what, once the gate opens, remains worth carrying forward.

(Below is the first version, before the Ox Alpha model expanded the quotations and improved some sections.)

At the Edge of the Possible

An Intellectual Biography of Stephen Hsu

A life can be misread by its nouns. Theoretical physicist. Silicon Valley founder. Computational genomicist. University research executive. Public intellectual. Builder of artificial-intelligence systems. Documentary filmmaker. Set side by side, the titles suggest restless polymathy, a man moving from one absorbing subject to another. They miss the verb that binds them: to make.

Again and again, Stephen Hsu has been drawn to ideas poised between theory and science fiction—not fantasies, but possibilities waiting upon some missing threshold. Has the cost of measurement fallen far enough? Has the dataset grown large enough? Has the necessary mathematics already been invented in another field? Can an institution be built around the answer? Hsu reconstructs the problem from first principles, searches for the hidden constraint, and then crosses whatever disciplinary boundary stands between the idea and its realization.

The unity of his career lies less in its subjects than in this way of moving through the world. In physics, he asks what can be known when quantum mechanics, gravity, and cosmology press against one another. In genomics, he asks how much of a human future lies encrypted in DNA, and how large a dataset is needed to read it. In entrepreneurship, he turns a technical possibility into a working system. In university leadership, he confronts the problem of organizing talent and capital at scale. In artificial intelligence, the object of inquiry begins to answer back, becoming a collaborator in discovery.

Hsu possesses an unusual feeling for the ripening hour of a problem. He neither pursues difficulty merely because struggle is noble nor waits until fashion has made an idea safe. He watches for the moment when an old impossibility becomes newly soluble—when cheaper sequencing, larger biobanks, greater compute, better algorithms, or more powerful models bend the curve. Before committing years to genomics, for example, he estimated whether realistic sample sizes should suffice. When the data crossed the predicted threshold, his group moved quickly and produced accurate predictors. This is ambition governed by calculation: audacity with a theory of when to act.

The distinction matters because Hsu’s projects have repeatedly escaped the realm of speculation. SafeWeb pioneered technology later acquired by Symantec. His group’s mathematical estimates in genomic prediction were followed by out-of-sample prediction of human height. Genomic Prediction and Othram carried population genetics into clinical and forensic practice. His years at Michigan State placed him inside the machinery of a major research university. More recently, he has brought generative AI into theoretical physics itself, while Superfocus seeks to make language-model systems reliable enough to act in the world. These undertakings differ in scale, maturity, and moral weight. What joins them is a recurring passage from the imaginable to the actual.

This essay uses the Grokipedia biography as its biographical spine, while drawing on Hsu’s scientific papers, essays, institutional biographies, X posts, and many podcast transcripts to recover his development in his own words. Those primary sources do more than add color: they alter the portrait. Hsu’s declared subject is intelligence; his deeper subject is what intelligence can make of resistance—the limits imposed by nature, data, institutions, convention, and mortality, and the strange opening that appears when a mind sees those limits clearly enough to act through them.

The fuller record also reveals a second axis, though it is less settled than it first appears: the question of inheritance amid extraordinary technical change. Family, remembered faith, physical courage, literature, and the continuity of the human line are not ornaments around the technologist. They help tell him which futures are worth making. Yet his loyalty is not simply to present biological humanity. At civilizational scale he can imagine enhanced humans, human–machine mergers, and even artificial minds as descendants. The question is therefore not only whether humanity survives, but what—love, memory, agency, intelligence, lineage, or form—must pass through the transformation for the future still to count as ours.

1. Ames: a prodigy in the ordinary world

Hsu was born in 1966 and raised in Ames, Iowa, the son of Chinese immigrants. His father, Cheng Ting Hsu, was an aerospace engineering professor at Iowa State University. Ames gave him an unusual combination: the ordinariness of a Midwestern university town and early access to a serious scientific environment. In a 2024 podcast transcript, Hsu recalls taking university mathematics and physics while still in high school—quantum mechanics, differential equations, linear algebra, complex analysis—after becoming the first student at his high school permitted to enroll at Iowa State.

That access began at home. Recalling his father in a From the New World interview, Hsu gives the decisive object an almost talismanic glow:

“He had what then, in the pre-internet era, was—for a precocious kid like me—the magic secret: a library card at the university library.”

Before the internet, a library card was not a convenience but a passage into worlds otherwise sealed off by age and geography. Hsu’s intellectual life began not merely with precocity, but with premature access to the archive of adult knowledge.

The archive was not his only company. Hsu has also recalled a comparably gifted Ames contemporary, later an MIT mathematics Ph.D., and a mathematically sophisticated professor living nearby. That small ecology complicates the legend of the solitary prodigy. Ames gave him not only books but early calibration: a peer against whom rare ability became visible, an adult who could recognize it, and a university whose doors were near enough to open. His self-education was exceptional, but it was socially scaffolded. (Interview on childhood and polymathy.)

He graduated from high school at sixteen. He graduated from Caltech at nineteen. Yet the story he tells is not the familiar memoir of the isolated prodigy. He was a competitive swimmer and high-school team captain; he describes his childhood as recognizably “all-American.” That combination is important. Hsu’s later interest in ability was not formed only through books, scores, or mathematical competitions. It was also shaped by athletics, where differences in speed, coordination, endurance, and trainability are visible, repeatedly measured, and difficult to explain away.

His childhood exposure to psychometrics was unusually early. Standardized-test results led him to ask how apparently exceptional scores across multiple domains could be correlated, and his father’s university library card gave him access to the Terman studies and technical literature on giftedness. He began treating his own social world as a kind of informal longitudinal study—watching how childhood ability, ambition, personality, opportunity, and eventual achievement diverged over time. Later encounters with elite physicists, Olympiad-level mathematicians, athletes, entrepreneurs, investors, and administrators broadened that comparative archive.

This background explains both a strength and a recurring controversy in Hsu’s thinking. He is unusually willing to speak about the tails of human variation because he believes he has observed them at high resolution. He distrusts accounts of achievement that erase natural differences. But he also knows, from movement across domains, that ability is not a single scalar. Mathematical speed, scientific originality, athletic talent, persuasion, emotional perception, courage, conscientiousness, and executive judgment are separable capacities. His own career would make little sense under a theory in which test-measured intelligence alone determined outcomes.

At Caltech, Richard Feynman supplied a model of scientific independence. The famous graduation photograph of the nineteen-year-old Hsu beside Feynman is more than biographical decoration. Hsu had chosen Caltech partly because of Feynman, and later wrote that Feynman influenced his college, career, and research specialization. In “Feynman and the Secret of Magic”, Hsu makes the hierarchy of his admiration explicit:

“Feynman is my hero, not Schwinger. Feynman had no rival in his generation when it came to originality and creativity.”

Hsu admires “magicians”—thinkers whose route to an answer cannot be recovered simply by scaling up ordinary competence. Yet he also notices the risk in Feynman’s refusal to read the literature: independence can become ignorance, and originality can generate dead ends. Hsu’s mature method is not Feynman’s pure intellectual individualism. He reconstructs from first principles where he can, borrows provisional knowledge where he must, and keeps track of which is which.

His contact with Feynman was active rather than merely devotional. As an undergraduate officer in the Society of Physics Students, Hsu invited him to give a special seminar on the EPR paradox and took him to lunch afterward. In Hsu’s later recollection, Feynman was then exploring negative probabilities as a route through quantum strangeness. The episode foreshadows Hsu’s adult intellectual role: identify a foundational question that respectable routines leave aside, bring the right minds into the room, and insist that the strange possibility deserves a hearing.

2. The habit of first principles

The most illuminating description of Hsu’s intellectual method appears in the Information Theory podcast transcript:

“I should be able to sit down with a piece of paper or whiteboard and actually kind of work it through from first principles.”

In another interview on his movement across disciplines, he names the common structure beneath his subjects:

“I guess the unifying theme is knowledge versus uncertainty: the attempt to capture the essential aspects of a messy system in a simplified mathematical model.”

This is close to a personal credo. The model must be simple enough to expose the decisive relation, but the scientist must remember that simplification has purchased clarity by discarding detail. Hsu’s confidence comes from finding structure; his best skepticism is directed at the boundary where structure may have been mistaken for the world.

That skepticism has a source older than his professional science. During the Cultural Revolution, letters from his father’s relatives in Zhejiang arrived on the thinnest paper, densely covered because the page and the channel home were precious. His father never returned to China and never saw his parents again. Meanwhile, some professors in Ames romanticized Communist China. The collision taught Hsu to ask whether expert conviction was proportioned to access, data, and correction. His contrarianism is partly an immigrant family’s memory of what confident intellectuals can fail to see. (Family history; Hsu’s recollection.)

The risk is overgeneralization. Spectacular ideological error can make all nontechnical expertise look less trustworthy than it is. Hsu sometimes writes as though mathematical or technological fields are uniquely corrigible and most other expert cultures are merely protected error. The distinction has force, but it is one of degree: technical communities also follow fashion, conceal uncertainty, and allocate attention institutionally. His own writing on quantum foundations and research reproducibility supplies the counterexamples.

Taken alone, this can sound like the standard rhetoric of technically minded entrepreneurs. Hsu’s fuller account is subtler. He says that when he enters a field, he attempts to organize it into a coherent logical structure, searches for foundational gaps, and marks assumptions whose evidential status is weaker than practitioners admit. But he does not demand mathematical rigor at every node before proceeding. He “coarse-grains” over some areas, provisionally accepts a stylized fact, and keeps an alternative map ready in case data invalidate it.

This is a powerful description of actual scientific reasoning. Pure deduction cannot move through empirical sciences because many premises remain contingent, approximate, or incompletely measured. Pure empiricism, however, cannot distinguish a meaningful anomaly from noise because it lacks a structural model. Hsu’s approach is to build a hierarchy of confidence: derive what can be derived, borrow what must temporarily be borrowed, remember which is which, and revise without embarrassment.

The same cognitive style explains his disciplinary mobility. He does not approach a new field by slowly absorbing all its conventions. He looks for its governing variables, scaling relations, information bottlenecks, and unexamined assumptions. This can give him an advantage over insiders whose knowledge is locally deeper but structurally less explicit. It can also make his criticisms sound abrasive. What appears to an insider as accumulated craft knowledge may appear to Hsu as an unjustified prior; what appears to Hsu as a simple information-theoretic question may depend on biological complexities he has compressed away. His best work occurs when the abstraction preserves what is decisive and discards what is not.

The deeper motivation is not simply winning arguments or moving quickly. Near the end of a long Undertone interview, Hsu reflects that he might have accumulated much greater wealth by leaving fundamental science earlier. His answer is that intellectual mastery has intrinsic value. The comment clarifies the whole trajectory: commercial success matters, but it does not replace the private satisfaction of closing the gap between elementary understanding and a research frontier. Hsu’s mobility is therefore not dilettantism. Each serious migration—physics, computation, genomics, AI—requires him to build another coherent internal world.

This also helps distinguish his idea of ambition from simple career maximization. He is highly responsive to leverage and opportunity, but he is not optimizing a single public score. Reputation, money, discovery, institutional power, and mastery are different goods. His life has repeatedly traded one for another.

3. The inner code: discipline, mortality, and the life of this world

Hsu’s literary tastes disclose the moral psychology behind his scientific method, but they are only one strand in a more complicated inheritance. His father was, in Hsu’s phrase, almost literally a Confucian scholar: cerebral, restrained, devoted to books and technical work. His mother came from a military and athletic Kuomintang family; her father had trained in Japan alongside Chiang Kai-shek, and she encouraged the competitive swimming and judo absent from his father’s world. Her family had also converted to Christianity in the nineteenth century. Hsu was raised Methodist in Ames. This double inheritance—scholar and athlete, materialist analysis and remembered faith—helps explain a personality in which abstraction, physical courage, self-command, and metaphysical unease coexist. (Hsu’s family history.)

Three writers gave that inheritance an adult vocabulary: Marcus Aurelius, Ernest Hemingway, and James Salter—unusual companions, but coherent ones.

Marcus supplies distance from reputation. In “Happiness”, Hsu describes himself as “something of a stoic” and cites Marcus’s warning:

“Or does the bubble reputation distract you? Keep before your eyes the swift onset of oblivion.”

The full passage minimizes applause against time, eternity, and the smallness of the human arena. Hsu’s attraction to it is not difficult to understand. Academic life is intensely status-conscious while pretending not to be; technical communities can confuse consensus, prestige, and citation with truth. Stoic distance becomes a cognitive tool. If reputation is transient, one can admit ignorance and criticize fashionable assumptions. One can also enter a field where one lacks standing or leave a prestigious track for work that seems more consequential.

This Stoicism should not be over-psychologized as a defense against chronic gloom. Hsu describes himself as temperamentally happy, low in neuroticism, and usually eager for the day. The discipline of Marcus is less a rescue from despair than a way of keeping an already energetic temperament independent of applause. His pessimism is methodological; his baseline mood is not. (Conversation on happiness and family.)

Hemingway supplies courage after illusion has been removed. Hsu’s long-standing motto is the Gramscian formula “pessimism of the intellect, optimism of the will.” In his explanation of the phrase, he defines its epistemic half through a sequence of questions:

“What if it is incorrect? How do I know what I know? … Never be afraid to admit you don’t know.”

He then turns to Hemingway to express the volitional half: clear sight does not entail passivity. One should attempt difficult things even after fantasy has been stripped away. This is not optimism as a forecast. It is optimism as a decision rule—the refusal to let an unfavorable prior become an excuse for inaction when agency can still change the distribution of outcomes.

Hemingway names the moment of courage; the Finnish word sisu names its duration. On Information Processing, Hsu adopted it for the grimmer endurance difficult work requires:

“Sisu contains a long-term element; it is not momentary courage, but the ability to sustain an action against the odds.”

— “Citizenfour and Sisu”

The distinction between dramatic bravery and sustained action is central to his temperament. A difficult project is rarely conquered in one heroic instant; it is carried through long periods when the reward is distant, the social signal is adverse, and failure repeats itself.

Salter supplies intensity, style, and an aristocratic sense of life. Writing about Salter in “The Life of This World”, Hsu says:

“Sentence for sentence, he is the master. But perhaps even more I admire his view of the world.”

He associates that view with courage, honor, the attempt of difficult things, sexual and emotional vividness, and discrimination about what genuinely matters. Hemingway’s prose reduces life to pressure, endurance, and action. Salter’s gives ambition sensual and aesthetic depth. Marcus prevents the heroic temperament from becoming dependent on applause; Hemingway insists on conduct under pressure; Salter reminds it that finite life should be lived intensely rather than merely optimized.

The remembered faith remains alive even though Hsu’s explicit metaphysics is materialist. In another conversation he put the residue plainly:

“But I still have this kind of spirituality or wonder left over. That feeling dominated my worldview when I was very young.”

— Manifold conversation with Aella

The older Hsu does not posit an intervening deity; he follows evidence and treats minds as physical systems. Yet church music or a cathedral can still awaken the intuition that the visible inventory is incomplete. He retains an open question about whether atoms and bits exhaust what happens to a person at death. This is reverence without doctrinal certainty, reductionism without immunity to awe. It also explains why his materialism never quite acquires the affect of disenchantment.

The deepest correction to the image of Hsu as an achievement-maximizer comes when the subject turns to family. His father was old when Hsu was born, and the child calculated early that death might come while he himself was still young. After his father died, Hsu could see the whole career—books, papers, professorship—against the value that had finally mattered most to the man who lived it. Reflecting on his father and his own children, he says in a Manifold conversation:

“Almost any ordinary human who can have a family and raise their children really has experienced maybe 90 percent of the great stuff.”

The thought is not self-abasement. It is the language of Ecclesiastes entering a Stoic life: a career capable of filling a biography can still occupy a subordinate place in the private order of value. More surprisingly, it is a theory of moral equality—not equality of capacity or achievement, but broad equality of access to the deepest human goods. Cognitive powers may be distributed unequally, public achievement more unequally still, yet children, attachment, memory, and the felt texture of a life shared with others remain available far beyond the elite tail. The comment changes the portrait. Mastery and action matter enormously to Hsu, but they are not the ultimate court of appeal.

Together they illuminate Hsu’s characteristic combination of severity and aspiration. The scientist must see without consolation. The agent must act without certainty. The individual must not confuse public reward with internal value. And a life should contain difficult achievements because mastery and daring are constitutive goods, not merely instruments for status.

This literary framework corrects a possible misunderstanding of Hsu’s appetite for ambitious projects. He is not attracted to impossibility for its own sake. His projects typically begin with a tractability judgment. The courage he admires is the courage to commit when success is uncertain but the causal pathway is real.

4. Physics at the limits of the knowable

Hsu took his B.S. at Caltech in 1986 and his Ph.D. at Berkeley in 1991, then moved through a Harvard Junior Fellowship to faculty positions at Yale and the University of Oregon. His official Michigan State biography lists research spanning quantum chromodynamics, black holes, entropy bounds, dark energy, cosmology, particle physics beyond the Standard Model, quantum foundations, genomics, finance, encryption, and information security.

The diversity of topics conceals a recurring question: what limits the extraction, localization, preservation, or interpretation of information?

In work on dense quark matter and physics beyond the Standard Model, the problem is how effective descriptions change across energy or density regimes. In black-hole physics, it is whether information is destroyed, hidden, decohered, or distributed across degrees of freedom inaccessible to ordinary observers. In cosmology, it is how global descriptions, entropy, vacuum structure, and observational selection constrain what can be inferred. In quantum foundations, it is what the formalism says about observers and branches when collapse is not treated as fundamental.

A concise example is his work with Xavier Calmet and Michael Graesser on minimum length. Their paper, “Minimum Length from Quantum Mechanics and Classical General Relativity”, concludes:

“Our results imply a device independent limit on possible position measurements.”

The argument joins quantum localization to gravitational collapse: concentrating enough energy to resolve an arbitrarily small region eventually creates a black hole. What looks like a limit of instrumentation becomes a limit implied by the joint structure of quantum mechanics and gravity. This is characteristic Hsu. The result comes not from a complete theory of quantum gravity but from forcing two well-established frameworks to constrain each other.

His attitude toward foundational difficulty is less impatient than his polemical style can suggest. Revisiting an earlier, overly harsh judgment of the field, he wrote:

“The speakers were wrestling with some of the hardest problems human minds have ever confronted. Of course progress was slow!”

— “Neural Networks and Deep Learning 2”

The exclamation carries humility as well as admiration. Hsu can be savage about muddle, but he can also revise his verdict when slowness reflects the depth of the problem rather than institutional torpor.

For Hsu, quantum foundations also became a case study in the sociology of knowledge. Copenhagen remained the classroom default while many leading theorists leaned toward Everett once they considered the universal wavefunction seriously. Consensus may therefore describe what a profession routinely teaches more accurately than what its deepest thinkers believe. The claim links physics to institutional analysis: foundational questions can be marginalized without being answered. (“Feynman and Everett”; Hsu’s 2012 quantum correspondence.)

His physics is often strongest in such boundary regions—where one can say something general before possessing the final theory. That preference anticipates his later work in genomics. In both cases, he asks whether broad structural reasoning can establish what is possible, impossible, or sample-limited before every mechanism is known.

It also explains why fundamental physics eventually shared his attention with faster-moving domains. The private value of mastering quantum field theory remained enormous, but experimental access to quantum gravity and high-energy frontiers was remote. Biology, computation, and later AI offered steeper technological gradients. Hsu did not abandon physics; he redistributed effort toward domains in which theory could meet rapidly expanding data and produce shorter feedback loops. A decade later, generative AI would unexpectedly shorten a feedback loop inside theoretical physics itself.

5. The founder’s turn: knowledge becomes action

The founding of SafeWeb around 2000 was the first major break in Hsu’s academic trajectory. SafeWeb built internet privacy and SSL VPN technology that Symantec acquired in 2003. Hsu later founded Robot Genius, which worked on malware protection. These episodes were not detours from his intellectual life. They altered his account of how knowledge becomes effective.

SafeWeb began with hacked Linux machines in the Yale physics department. Hsu and a graduate student noticed that the browser’s built-in SSL engine could become the universal endpoint for a new kind of virtual private network—obvious in retrospect, not yet built. The company first won millions of consumer users, then discovered that bandwidth costs and the immature advertising market made popularity unprofitable. It survived by making a painful right-angle turn toward enterprise security. (Mixergy interview.)

The episode taught Hsu not only leverage but responsibility. Recalling the sunny day beside the Bay when he had to dismiss people he had recruited, he gave the founder’s burden its severest form:

“Every time you hire somebody you should picture that you might have to fire them… You hired him… you’ve got to face him.”

— Mixergy interview

For Hsu, one may not delegate the moral fact of a decision whose authority one claimed. Entrepreneurship therefore widened his idea of courage: not merely originating an innovation, but accepting the human cost of adaptation when the original plan fails. A correct technical idea remains only one input into realization. A founder must recruit, allocate, persuade, decide under uncertainty, survive adverse selection by investors and markets, and fit a new capability into existing workflows.

The founder’s turn was not inevitable. In his twenties Hsu came close to quant finance, recasting exotic-option pricing in the language of Feynman path integrals. A Harvard Junior Fellowship and an aversion to Manhattan helped keep him in physics. The counterfactual resists retrospective myth: a career that now appears architecturally coherent was also bent by prestige, geography, temperament, and luck. (Path-integral account; career counterfactual.)

This experience adds a second axis to Hsu’s conception of intelligence. Academic culture privileges analytic depth and publication. Startups expose execution, social judgment, risk tolerance, and speed. Later, as an administrator, Hsu would say that startup experience teaches difficult decision-making under pressure. Across these roles he encountered forms of ability that psychometric discussion often leaves out: the capacity to coordinate other minds and reshape an institution.

He is also candid about the price of range. The fox notices connections the hedgehog misses; the hedgehog may produce the “deep-time” contribution. Hsu wonders whether even von Neumann’s breadth carried that cost. Polymathy is therefore not uncomplicated praise: translation may disperse the concentration required for one monumental result.

It also sharpened his sense that talented minds can be diverted into games beneath their powers. In a conversation about intellectuals and the technosphere, Hsu remarks:

“People who come from science or math backgrounds and end up in finance—in a way it kind of dumbs them down.”

The line is deliberately provocative, but its governing emotion is regret. Civilization has only so many people able to work near the frontier; prestige and compensation can draw them toward the redistribution of claims rather than the creation of new capabilities. Founding companies allowed Hsu to seek leverage without surrendering the builder’s criterion: something new must exist afterward.

Entrepreneurship also intensified his impatience with static organizations. To Hsu, an institution is not simply a community governed by norms; it is an information-processing and decision-making system. Incentives determine which signals travel upward, who can act, how quickly errors are corrected, and whether exceptional people receive resources. A slow hierarchy may possess immense knowledge yet remain collectively unintelligent.

That insight unifies SafeWeb with his later university leadership and AI work. In each case, the question is how to turn distributed knowledge into reliable action.

6. Genomics: when science fiction found its sample size

Hsu’s move into genomics around 2011 is the clearest expression of his mature method. He had long been interested in genetics, evolution, intelligence, and human variation. But interest alone did not determine timing. Sequencing and genotyping costs were falling extraordinarily fast; biobanks were growing; machine learning and compressed sensing offered mathematical tools for reconstructing sparse signals from noisy, high-dimensional data.

In a Radiolab transcript, Hsu describes the imaginative attraction:

“If I get to be one of the scientists who makes real some amazing trope from science fiction, that would be the most awesome thing.”

The sentence gives technical ambition the emotion of discovery: not prediction from the sidelines, but participation in the instant when an old fiction becomes real. Yet the decisive step was theoretical, not rhetorical. Hsu asked how many genotyped individuals would be required to recover the genetic architecture of a complex trait under assumptions of approximate sparsity and additivity. If the answer had been hundreds of millions, he has said, the problem would not have been timely. His group’s analysis suggested that hundreds of thousands might suffice.

The gamble was also technological. In a 2019 genomics interview, Hsu described the bet with unusual plainness:

“We were betting on the continuing decline in cost for genotyping, and it paid off because now there are millions of genotypes available for analysis.”

This is Hsu’s feeling for the ripening hour in its purest form. The scientific idea was old enough to be imaginable; the falling cost curve made it newly executable.

This is the opposite of indiscriminate futurism. Hsu did preparatory theory to decide whether a frontier was worth entering. His 2013 paper with colleagues on compressed sensing and genomic selection reported:

“There is a sharp phase transition to complete selection as the sample size is increased.”

The phrase “phase transition” is not merely metaphorical. In compressed sensing, recovery can change abruptly once the number of observations crosses a threshold determined by signal sparsity and noise. Hsu recognized that genomics had the same mathematical structure: sufficiently large datasets might not yield only gradual improvement; they might move a trait from apparently intractable to recoverable. In the paper’s simulations, a trait with heritability of one-half could be recovered well when the sample size reached roughly thirty times the number of nonzero loci—a usable scaling relation, not a mood of technological optimism.

When the UK Biobank released data on the required scale, Hsu’s group acted quickly. He recalls that within a month of obtaining access, the group had built predictors with errors of only a few centimeters. Their 2017 preprint on genomic prediction of human height, later published in Genetics, reported:

“Actual heights of most individuals in validation samples are within a few cm of the prediction.”

The sequence is central to understanding Hsu: derive a sample-complexity expectation, monitor the enabling infrastructure, obtain the data, and test out of sample. The successful height predictor vindicated not only a particular model but a style of frontier judgment. It also helped establish that highly polygenic traits could be predicted with useful accuracy even when thousands of variants contribute small effects. (Hsu’s retrospective account.)

Looking back on the reception of the program, Hsu compressed the sociology of premature research into three beats:

“Research advances often pass through the following phases of reaction from the scientific community: It’s wrong. It’s trivial. I did it first.”

— “Kathryn Paige Harden Profile in The New Yorker”

The line is triumphant and barbed, but the chronology behind it matters. At a 2012 behavior-genetics meeting, a physicist proposing million-person genomic prediction could sound, as Hsu later joked, like an alien time traveler. By 2017 his group had crossed the predicted threshold for height. The episode supports his conviction that apparently visionary projects can succeed when their sample-complexity logic is sound. It also shows the danger in his retrospective style: genuine scientific objections can be compressed too easily into mere stages on the skeptic’s road to surrender. Vindication should raise confidence in the method, not make future dissent automatically unserious.

From there, the research expanded to disease-risk prediction. Genomic Prediction translated polygenic scores into embryo testing in IVF; Othram applied genomics and genetic genealogy to forensic identification. These applications moved Hsu from the epistemic question—what can DNA predict?—to the institutional and ethical questions: who should receive the prediction, how should it be validated across populations, and what choices should follow?

The trajectory from physics to genomics is therefore not a departure from Hsu’s intellectual history. It is its most revealing demonstration. Physics supplied first-principles modeling, scaling arguments, statistical mechanics, information theory, and comfort with high-dimensional abstraction. Entrepreneurship supplied workflow integration and institutional action. Genomics supplied the rapidly improving measurement technology and the consequential human target.

7. The measure of a person

No part of Hsu’s work is more controversial than his writing and research on cognitive ability, genetic prediction, embryo selection, and possible future enhancement. A sophisticated analysis must separate at least four claims that public discussion often collapses.

First, individuals differ in measured cognitive abilities, and some of those differences are stable and consequential. Second, variation within a population is partly heritable. Third, sufficiently large genomic datasets can support out-of-sample prediction of some fraction of phenotypic variance. Fourth, such predictions should be used for particular reproductive or social purposes. The first three are empirical questions, though difficult ones; the fourth is normative and institutional. Evidence for prediction does not by itself settle governance.

Hsu’s interest in the subject is biographically overdetermined. He was a radically accelerated child, studied psychometrics early, encountered exceptional scientific talent, observed intellectual disability at close range, and later worked in environments where performance distributions were unusually wide. He treats the tails of ability as real partly because they were among the most salient facts of his life.

The subject is not merely statistical to him; it can be beautiful. Posting a chart from a vast American longitudinal study, he asked on X:

“Isn’t this one of the most beautiful pictures in science? Project Talent: back when America was functional.”

The sentence fuses three Hsu preoccupations: the aesthetic pleasure of a clear empirical pattern, nostalgia for an America capable of measuring itself at scale, and frustration with institutions that have lost confidence in quantitative truth.

Yet his own statements complicate any crude genetic determinism. He distinguishes intelligence from originality, drive, luck, courage, personality, and executive competence. He admires Feynman more than a potentially more technically comprehensive Schwinger because creativity is not reducible to general cognitive power. His entrepreneurial and administrative record demonstrates that coordination and judgment matter. The most defensible reconstruction of his view is not “genes are destiny,” but “ignoring heritable variation produces bad models, while genetic prediction remains probabilistic and incomplete.”

He has even given the practical advice least expected from a public defender of psychometrics:

“While g is useful as a crude measurement of cognitive ability… one is better off adopting the so-called growth mindset.”

— “Feynman, Schwinger, and Psychometrics”

There is no contradiction. Population distributions and individual conduct answer different questions. A measured prior may improve prediction across people; it does not tell a particular person where effort, obsession, mentorship, or an unmeasured gift will carry him. Hsu’s realism about variance coexists with a life philosophy that refuses fatalism.

The family argument returns here with new force. Hsu’s “90 percent” observation implies that unequal ability need not become a total hierarchy of lives: exceptional accomplishment is rare, while most human fulfillment is not. This does not solve the politics of measured traits, but it explains how a severe account of unequal capability can coexist with an egalitarian account of access to meaning.

Hsu is also more explicit about limitations than polemical summaries suggest. In a Dwarkesh Patel interview, he notes a major generalization problem:

“Huge problem is that most of the data is from Europeans.”

Polygenic scores frequently lose accuracy across ancestries because linkage disequilibrium, allele frequencies, environmental distributions, and training samples differ. This is both a scientific limitation and an equity problem. A technology that works best for populations already overrepresented in biomedical research can deepen unequal access to prediction.

He also accepts the larger political risk: reproductive enhancement could create caste-like inequality. The admission is important, but it does not dissolve the concern. Technologies affecting reproduction or enhancement create externalities extending beyond individual choice. Even if each family acts voluntarily, aggregate effects can reshape status competition, insurance, education, disability norms, and class reproduction. A purely consumer-choice framework is inadequate.

In his most memorable rendering of that danger, Hsu does not offer a statistic but a scene:

“At dinner they’re discussing convex optimization of objective functions in complexified tensor spaces, while the server has no hope of ever understanding their discussion.”

— “The Future of Intelligence” interview

The image is Salter’s world rendered as a thought experiment: brilliance, class, conversation, and exclusion compressed around a dinner table. It shows that Hsu understands the nightmare version of enhancement from the inside. His fear is not difference alone, but a caste boundary so cognitively wide that common civic life becomes impossible.

Hsu’s answer is generally that powerful technologies carry risks, that information can reduce suffering, and that prohibition may be neither stable nor globally enforceable. He emphasizes prediction of serious disease and argues that families already making embryo choices should have access to validated information. In a 2022 genomic Q&A, he said that Genomic Prediction deliberately did not report cognitive-ability scores because the application was too controversial and that the company focused on health risk. The distinction is historically important: Hsu’s research program reaches toward cognitive prediction, but the clinical product he described drew a nearer boundary.

That boundary does not settle the future. In the Latecomer interview, Hsu imagines society disseminating as much information as possible and deciding democratically, while admitting that the ideal resembles a Vulcan academy more than any polity humans possess. He expects competitive pressure and unequal access to outrun deliberation. This is strongest when the intervention prevents severe illness and the model is accurate, ancestry-appropriate, and transparently communicated. It becomes harder as one moves from disease risk to behavioral traits, from selection among existing embryos to editing, and from private benefit to civilizational competition.

His critics are right to demand governance, distributive analysis, respect for disability, and protection against coercion. Hsu is right that refusing to measure does not make variation disappear, and that moral discomfort is not a substitute for statistical evaluation. The productive position lies in preserving both truths: predictive capability can be real, and its reality makes ethical design more urgent rather than less.

Hsu’s long-range aspiration is more humane than the caricature of simple rank optimization, but also more radically posthuman than the language of preservation might suggest. He imagines biotechnology reducing disease, extending healthy life, improving cooperation, and lowering the burden of mental illness. He also expects selection and editing eventually to produce subpopulations qualitatively different from present humanity—something approaching conscious speciation on a civilizational timescale. Intelligence is part of that future, but not its sole value.

The governing question is therefore not simply whether humanity can improve capability without hardening hierarchy. It is what Hsu means by humanity across generations. His continuity is genealogical and agentic rather than morphological: enhanced descendants may count as heirs even when they no longer resemble us closely. That elasticity makes his futurism bolder, and makes its moral boundary harder to locate.

8. Science, power, and the university

In 2012 Hsu moved to Michigan State University as vice president for research and graduate studies, later serving as senior vice president for research and innovation. He also became a professor of physics and of computational mathematics, science, and engineering. The appointment placed a frontier scientist and founder inside a large public university’s executive structure.

Hsu’s account of administration is characteristically agentic. In an interview about the MSU role, he said:

“Running a startup teaches you how to make difficult, complex decisions under pressure… The real source of any institution’s strength is its people.”

The two sentences define his administrative philosophy. Institutions need decisions, but their durable advantage lies in talent. Research leadership therefore means identifying excellent people, recruiting them, supplying resources, coordinating large initiatives, and removing friction. The founder’s sense of urgency meets the university’s slower ecology of departments, faculty governance, public accountability, and long-horizon research.

His eight years in administration gave Hsu direct experience of science as a capital-intensive collective enterprise. Modern research is not produced only by solitary insight. It requires grant portfolios, laboratories, computing, compliance, intellectual property, graduate education, government relations, and large collaborations. At Michigan State, the Facility for Rare Isotope Beams exemplified the scale at which scientific ambition becomes institutional engineering.

The record also reveals more idealism than a portrait centered on optimization and conflict would suggest. Welcoming new faculty, Hsu told them:

“Only one in a thousand people in our society have the privilege to engage full time in discovery—in curiosity-driven research.”

— “MSU New Faculty Welcome 2019”

He presented administration as stewardship of that privilege: help scholars obtain grants, incubate companies, solve child-care and departmental problems, and remove whatever prevents discovery. Under his watch, Michigan State created an interdisciplinary computational mathematics, science, and engineering department on what he proudly called “startup time” and pursued a hundred-faculty recruitment initiative in high-impact fields. His administrative ideal was not simply to rank talent but to give it room, tools, and institutional shelter.

That ideal made institutional indifference especially corrosive. Hsu later described showing senior administrators RAND results suggesting that gains in general collegiate reasoning were small and strongly related to students’ incoming scores. He received little substantive disagreement—and little curiosity. The episode sharpened his sense that institutions can protect their public story more faithfully than their mission. In the same interview, he gave his standard for holding office:

“What’s the point of doing this job if you’re not going to do it right?”

— Palladium interview on political academia

The role also exposed a tension between Hsu’s ranking-oriented view of expertise and the plural norms of a university. He tends to ask whether claims are true, whether evidence is strong, and whether decision-makers are competent. Universities must additionally manage legitimacy, representation, historical injury, and the right of multiple constituencies to contest how expertise is used. Hsu can regard these processes as signal corruption or bureaucratic inhibition; participants may regard them as conditions of legitimate authority.

That tension culminated in 2020, when activism over his research, writing, and administrative decisions led the university president to request his resignation from the research leadership role. In his resignation statement, Hsu wrote:

“President Stanley asked me this afternoon for my resignation… The fight to defend academic freedom on campus is only beginning.”

The episode was an institutional rupture and an intellectual consolidation. Hsu returned to the faculty, while the research capacity, hires, and organizations he helped build remained. His public voice became less constrained by executive office, and he increasingly interpreted disputes over genetics, policing research, merit, and demographic difference through the framework of academic freedom and civilizational competence.

It would be simplistic to cast the conflict only as truth against politics. University leaders always operate within political institutions, and administrative speech has consequences different from private scholarship. But it would be equally simplistic to treat controversy as evidence of scientific or moral invalidity. The central unresolved question is whether institutions can protect inquiry into sensitive empirical subjects while maintaining trust among people who fear how such inquiry may be used.

9. Civilization and the uses of intelligence

Hsu began Information Processing in 2004 and later migrated it to Substack. He also hosts the Manifold podcast. Across these venues he writes about physics, genetics, artificial intelligence, universities, geopolitics, literature, film, martial arts, elite performance, and American institutional decline. The range can look idiosyncratic, but the same questions recur: Who is competent? How can competence be detected? What prevents accurate beliefs from controlling decisions? How do civilizations cultivate or waste exceptional talent?

His public thought is strongly meritocratic, but “merit” in Hsu’s usage has at least three meanings. It can mean measurable ability; demonstrated accomplishment; or the capacity to make a system work. These often correlate, but not perfectly. A danger in his rhetoric is that evidence from extreme technical performers can be generalized too quickly to political authority. Scientific excellence does not automatically confer moral wisdom, and institutions need legitimacy as well as optimization.

His most compressed recent statement of the technocratic instinct appeared on X:

“Is every genius level STEM guy suited for leadership? No, obviously not. But every leader going forward should be genius level STEM.”

The first sentence concedes that intelligence is insufficient; the second makes technical genius a necessary threshold. The formulation is vintage Hsu—funny, categorical, and intended to break complacency. It also marks the edge of his argument. Civilizational leadership certainly requires technical comprehension, but whether it requires genius-level STEM ability in every leader is a further claim, one that may underweight judgment, historical imagination, persuasion, and moral legitimacy.

The emotional root of his American politics is less abstractly technocratic. Hsu’s parents came from anti-Communist KMT families and regarded the United States not merely as a successful system but as the country that gave them refuge and belonging:

“They also felt that the country accepted them, gave them a life, gave them the ability to raise a family and have a career.”

— Manifold conversation with John Mearsheimer

The high-trust Iowa childhood is therefore politically causal. When Hsu speaks of American decline, he is mourning more than lost scientific rank. He remembers a society in which immigrants could enter ordinary civic life, families felt less precarious, and institutions seemed worthy of trust. In a 2024 interview, he worried explicitly about Americans near the middle and below the middle of the distribution, not only about globally mobile elites. Meritocracy, in this register, is supposed to serve a common world rather than merely certify its winners.

His relationship to Donald Trump belongs inside this institutional story. Hsu disclosed in the same interview that he had nearly joined the first Trump administration in a senior, Senate-confirmed role. He described his exhilaration at Trump’s 2024 victory as a response to what he regarded as bureaucratic abuse and lawfare, while also calling the first term dysfunctional and acknowledging Trump’s faults and mercurial treatment of capable allies. The allegiance is thus better understood as support for an instrument of institutional disruption than as unqualified faith in a leader. Whether that instrument can restore competence without damaging the norms Hsu values remains an unresolved political bet.

Hsu’s critique of elite systems is not simply that the wrong individuals possess prestige. It is that institutions increasingly suppress accurate feedback. Credentialism substitutes for ability, narrative for measurement, and procedural consensus for responsibility. His startup experience taught him that reality eventually punishes such substitutions; companies fail, systems break, and predictions do not replicate. Politics and universities can defer correction longer.

China occupies a complicated place in this analysis. Hsu’s family history, scientific relationships, work with BGI, knowledge of American and Chinese technical elites, and concern with geopolitical competition give him a bicultural comparative lens. The label “pro-China” obscures more than it explains. Hsu identifies as a proud Iowan and an American realist; his father’s relatives endured the Communist takeover, Great Leap Forward, and Cultural Revolution. His willingness to credit contemporary Chinese capability is therefore not nostalgia for Maoism. It is partly the same refusal of ideologically convenient error that his father taught him when Western intellectuals romanticized the China his family was actually experiencing. (Family history; 2025 interview précis.)

He often portrays China as more technologically capable and strategically serious than American discourse allows, while also recognizing the constraints of its political system. Summarizing a formulation he credits to the pseudonymous analyst Han Feizi, Hsu argued in early 2026:

“China leapfrogged Western expectations so fast… that sort of short-circuited the Thucydides trap.”

— “Geopolitics 2026” transcript

The claim is not that rivalry vanished. It is that Washington may have recognized China’s military-industrial position only after the favorable window for a preventive confrontation had already narrowed, producing retrenchment and “Fortress Americas” rather than a classical rising-power war. Whether that forecast proves correct, its form is characteristic: estimate relative capability, identify a phase transition, and revise strategic expectations before public narratives catch up. The underlying issue is not cultural admiration in the abstract but state capacity: which civilization can identify talent, build infrastructure, pursue long-term goals, and absorb new technology?

This framework can produce sharp insights and blind spots. It corrects complacency about American primacy and highlights the material bases of scientific power. But a civilization cannot be evaluated only as a research lab or startup. Freedom, loyalty, solidarity, consent, and the distribution of dignity are not noise variables. Hsu’s strongest public analysis occurs when he treats pluralism as part of the optimization problem rather than as an obstacle external to it.

His Stoicism moderates the elite-centered view in an important way. If fame is a bubble and public applause is unreliable, then membership in a prestigious hierarchy cannot be the ultimate measure of a person. His emphasis on ability describes differences in capability; it need not imply differences in human worth. Much of the ethical controversy around Hsu arises precisely because that distinction is difficult to maintain socially once predictive technologies and competitive institutions assign consequences to measured traits.

10. The machine enters the laboratory

AI brings Hsu’s major themes together more tightly than any earlier field. It concerns the nature of intelligence, the scaling of capability, the automation of information processing, the future of work and hierarchy, geopolitical competition, and the possibility of new scientific agents.

Hsu’s response to large language models is neither simple enthusiasm nor dismissal. He treats them as systems whose internal mechanisms remain only partly understood but whose external performance must be measured. Their unreliability resembles a familiar human type. In a Manifold transcript on AI-assisted theoretical physics, he offers the analogy:

“You have a brilliant but unreliable genius colleague… his brain is clearly not like yours, but he has an encyclopedic mastery of all the literature.”

That framing is characteristically Hsu. The question is not whether the system “really understands” in an all-or-nothing metaphysical sense. The practical question is what work it can originate, how error-prone it is, and what verification architecture turns intermittent brilliance into dependable output.

The romance of genius is disciplined here by an engineer’s respect for drudgery. After visits with frontier-lab researchers, Hsu wrote:

“Even at the high-profile AI labs it’s the engineers … willing to grind at cleaning data, evaluating responses, etc. that are the most valuable.”

— “A Month on the Road”

This is an important correction to an intelligence-centered biography. Frontier capability is not produced by luminous ideas alone. It rests on evaluation, data hygiene, repeated failure analysis, and people willing to perform unglamorous work with unusual conscientiousness. The AI laboratory joins the startup and the athletic pool as another place where talent becomes real only through sustained practice.

The issue became personal to his research. Discussing a recent paper on nonlinear modifications of quantum mechanics, Hsu states:

“I think I’ve published the first research article in theoretical physics in which the main idea came from an AI—GPT5 in this case.”

— @hsu_steve on X

The associated 2025 paper analyzes a technically serious consequence:

“Nonlinear modifications of quantum mechanics affect operator relations at spacelike separation, leading to violation of the integrability conditions.”

Whatever historical judgment is eventually made about the result, the process is significant. A scientist who spent decades studying exceptional human cognition now reports a machine generating the central idea of a theoretical-physics paper. His role becomes partly that of evaluator, formalizer, collaborator, and guarantor of rigor.

The experience has also modified his account of originality itself. By summer 2026, Hsu was sympathetic to Terence Tao’s suggestion that human researchers may recombine inherited ideas more often than their introspection admits. Models make that recombinant structure visible because their joint mastery of distant literatures is so conspicuous. Yet Hsu does not collapse machine and human creativity. He warns that models can confabulate at depth: an analogy may be persuasive enough to waste an expert’s time because the system lacks the tacit physical judgment that would make a human genius’s analogy more trustworthy. The right comparison is not inspiration versus autocomplete, but two differently structured kinds of fallible intelligence. (“State of AI, Summer 2026”; “Theoretical Physics with Generative AI”.)

Hsu is equally alert to the next recursive step. Writing about AI systems that participate in improving AI research, he observes:

“Coding capability is not the limiting factor: modern LLM training loops are only ~200 lines of code.”

— @hsu_steve on X

The memorable number makes the point. The bottleneck is migrating from the ability to write a training loop toward the ability to choose experiments, diagnose failures, evaluate novelty, secure compute, and improve the whole research process. This is the same distinction Hsu learned as a founder: execution is never exhausted by possession of the core technical idea.

By July 2026 his forecast had sharpened. He linked the models’ advancing ability in mathematics and physics to their capacity to redesign learning systems themselves:

“I think it’s directly tied to when we will first see really effective RSI… and I think we’re just getting to that threshold.”

— “State of AI, Summer 2026” transcript

RSI—recursive self-improvement—is the point at which a model can propose, test, and implement improvements to its own architecture, making the next model better and potentially accelerating the rate of further improvement. Hsu does not claim that the full loop has arrived. His judgment is that the scientific abilities required for it are becoming recognizable. He also sees an “agentic phase transition”: many differently prompted models, organized as generators, verifiers, and supervisors, may acquire capabilities not visible in any isolated instance.

This prospect changes the human institution of science before it settles the metaphysics of machine thought. Hsu reports that some departments have discussed admitting fewer doctoral students because professors can obtain immediate productivity from models. Training an undergraduate to the frontier takes years of attention; a model contributes at once and never tires. Yet fewer apprentices create a civilizational succession problem: who becomes the expert capable of checking the machines later? Hsu allows that science may simply need fewer human practitioners once their productivity is multiplied. The harder possibility is path dependence: an institution that stops forming human judgment may later discover that it has also lost the capacity to recognize when its machines are wrong.

Superfocus, which Hsu co-founded, represents the entrepreneurial complement. His current biography describes it as building reliable AI systems from language models; the company’s site emphasizes systems that can read, write, listen, speak, decide, and act. The conceptual problem is the same as in Hsu’s first-principles method: how does one preserve powerful generative leaps while marking provisional nodes, checking outputs, and preventing error from propagating?

Commercial deployment made the social consequence immediate. Writing after demonstrations to the Philippine business-process-outsourcing industry, Hsu asked:

“The AI earthquake in SF has created a tsunami headed towards the Philippines—is it a 6 foot wave, or a 600 ft wave?”

— “SuperFocus, AI, and Philippine Call Centers: Part 2”

The image is memorable because Hsu is both seismologist and participant. He is building systems that may improve service and lower cost while recognizing that a national labor model lies in the path of the wave. The recurrent Hsu tension is now global: a capability can be real, valuable, and destructive of the institutions through which millions presently live.

His response to existential risk is equally double. In the Latecomer interview, Hsu argues that rigorous alignment of a much more intelligent system is probably impossible: a trained network is closer to an evolved ecology than a transparent program, and even a mandate to preserve human well-being may be interpreted in ways humans cannot follow. Yet he is less attached than many safety thinkers to the indefinite persistence of present biological humanity. He can treat AGIs as descendants, imagine human brains merging with machines, and ask whether our biologically recent species should necessarily remain the final custodian of cosmic intelligence.

An exchange with AI researcher Richard Ngo can appear, when excerpted, to reverse that position. Hsu advances a Butlerian case for permitting enhanced humans while refusing to build machines cognitively superior to them. In context, however, he has explicitly announced that he is trying to steelman the Yudkowsky–Soares position. He later explains that his tail-risk formulation is a deliberately accessible scenario for officials and other nonspecialists who would reject more radical accounts as fantasy. The first-person vividness belongs to the performance of the argument; it should not be converted into a biographical declaration. (Conversation with Richard Ngo.)

The episode is nevertheless revealing—not as a change of creed, but as evidence of intellectual range. Hsu can inhabit the preservationist objection strongly enough to make its fear intelligible, just as in a later conversation with accelerationist Beff Jezos he draws out the counterposition: intelligence may be part of a cosmic movement toward greater complexity, and attachment to the present ape substrate may be parochial. Hsu’s own most explicit statements are more substrate-flexible than the Butlerian case, qualified by the recognition that alignment cannot be guaranteed, that superior systems may not care as humans care, and that the transition can disempower people long before any terminal catastrophe.

This prevents an easy reading of Hsu as either a conventional preservationist or a heedless accelerationist. The family man values embodied attachment as the deepest good of an individual life. The physicist, thinking in billion-year intervals, can treat substrate and species form as contingent. The entrepreneur builds within the transition; the documentarian makes its dangers vivid. These positions do not converge into a settled doctrine. They mark the fault line running through his mature futurism: openness to successors beyond present humanity, joined to a determination that civilization understand the stakes of creating them.

His 2026 documentary project Machine God marks another turn—from analyst and builder toward witness. In his account of the film, Hsu invokes Joan Didion’s attempt to capture San Francisco at a hinge of history. His collaborators filmed accelerationists, safety researchers, founders, protesters, and philosophers before a possible AGI break. The aim is not simply celebration: the film makes recursive improvement, existential risk, and gradual human disempowerment vivid to elites and the public. Hsu wants the future built, but civilization awake when it arrives.

AI therefore completes a loop in Hsu’s journey:

1. He studies the distribution and structure of human intelligence.

2. He applies machine learning to genomic prediction.

3. He builds companies that operationalize high-dimensional inference.

4. He uses machine intelligence as a collaborator in fundamental science.

5. He builds systems intended to make that collaborator reliable enough for institutions.

As of 2026, Hsu remains a Michigan State professor in theoretical physics and computational mathematics, science, and engineering; a founder of SafeWeb, Robot Genius, Genomic Prediction, Othram, and Superfocus; and, since 2024, an executive adviser at TCV. These are not separate afterlives. They are positions from which to observe and shape the same transition: intelligence becoming increasingly measurable, reproducible, and technologically embodied.

11. Worlds within worlds: multiverse, simulation, and “base reality”

Hsu’s speculative writing about the multiverse and simulation is not an eccentric appendix to his applied work. It extends the same information-processing worldview to ontology.

In no-collapse or many-worlds quantum mechanics, the universal wavefunction evolves without a fundamental measurement-induced collapse. Observers and apparently definite outcomes must emerge within branches. Hsu is attracted to the austerity of this picture: it takes the formalism seriously and resists adding a special mechanism solely to reproduce ordinary intuition. But it shifts the explanatory burden. If all branches are present in the wavefunction, what makes probability meaningful to an observer inside it? What precisely counts as a branch, and how do stable records and agents emerge?

Hsu states the ontological price of this austerity without flinching:

“The many branches of the universal wavefunction are realized ‘all at once’ and concepts like observers must be emergent.”

— “Ten Years of Quantum Coherence and Decoherence”

There is a deep continuity here with his Stoicism. The observer is locally indispensable yet cosmically unprivileged; the self is real as an emergent pattern, not as an exception written into the fundamental law.

His most vivid shorthand for the mechanism is almost cinematic:

“Decoherence is merely the mechanism by which the different Everett worlds lose contact with each other!”

— “Feynman and Everett”

The sentence corrects the cartoon in which a classical cosmos repeatedly splits like a cell. The universal state evolves; decoherence prevents macroscopically distinct components from continuing to interfere; observers find themselves inside stable, effectively isolated histories. But austerity does not eliminate mystery. Hsu’s own work on the measure problem argues that decision-theoretic accounts may explain Born-rule behavior conditional on inhabiting an ordinary branch without explaining why an observer is not on a “maverick” branch where familiar regularities fail. Many-worlds is minimal in postulates, not complete in interpretation.

Hsu’s discussion of simulation arguments is similarly conditional rather than devotional. Given sufficiently capable civilizations, large computational resources, and substrates capable of supporting conscious processes, simulated worlds could vastly outnumber unsimulated ones. Under those assumptions, the posterior probability that we inhabit “base reality” might be low. Yet the argument depends on premises about consciousness, computation, civilizational survival, and the motives of simulators. Hsu’s interest lies less in announcing that the world is fake than in following an information-theoretic argument to its unsettling consequence.

This ontology reaches inward. As early as 2005, Hsu stated the consequence with characteristic bluntness:

“If our current understanding of physical laws is correct, humans have only the illusion of free will.”

— “Free Will and Determinism: A Physicist’s Perspective”

Classical determinism does not help; quantum randomness added to a biological machine would still not amount to authorship. Consciousness may arise from sufficiently complex information processing while the self experiences decisions whose lower-level causes it cannot inspect. The view sits in productive tension with his ethic of will. “Optimism of the will” need not assert metaphysical freedom; it names the stance through which an embodied decision system acts from inside the world.

The multiverse also gives him a language for agency. If reality contains an enormous space of possible branches, intelligence can be understood as a process that models alternatives and steers toward a tiny subset. An organism does this locally; a civilization does it collectively; an AI system may do it with vastly greater search depth. On this view, knowledge is not passive representation. It is a technology for concentrating probability mass around futures that would otherwise remain inaccessible.

This is where his physics, genomics, entrepreneurship, and civilizational thought converge. Genetic prediction maps possible human phenotypes before birth. A startup selects one path through technological and market uncertainty. University strategy selects research futures by allocating capital and talent. AI expands the space of models and actions a civilization can evaluate. The multiverse becomes both a physical hypothesis and a master metaphor for choice under uncertainty.

There is a possible danger in this computational ontology. It can make persons, cultures, and moral commitments appear as variables inside an optimization problem. But Hsu’s literary attachments resist that flattening. Marcus, Hemingway, and Salter insist that the experiencing agent—finite, embodied, vulnerable, honor-seeking—cannot be discarded without losing the meaning of the optimization. A civilization is an information-processing system, but it is also the lived world of beings for whom outcomes matter.

12. The tensions within the vision

Hsu’s intellectual significance lies partly in the tensions he does not fully resolve.

He has supplied the governing technological version of those tensions himself:

“It’s hard to put a util value on some things that are in the foreseeable future, like machine intelligence and genetic engineering.”

— “Low-Hanging Fruit and Technological Innovation”

These technologies are not ordinary increments whose benefits fit comfortably into a cost-benefit table. They may change the kinds of agents who make the table, the scale of values those agents can pursue, and the identity of the civilization doing the choosing.

Hsu’s own ethical position is more publicly deliberative than a pure parental-autonomy account. Writing about embryo selection, he insisted:

“New genomic technologies are so powerful that they should be widely understood and discussed—by all of society, not just by scientists.”

— “Polygenic Embryo Screening: comments on Carmi et al. and Visscher et al.”

That sentence should be read beside his strong defense of parents’ access to validated disease-risk information. The tension is real: private reproductive choice can be morally urgent, yet the aggregate result may alter class structure, disability norms, and the biological constitution of later generations. Hsu is clearer about the arrival and potential benefit of the capability than about the institutions capable of governing it, but he does not imagine that scientists alone possess the authority to decide.

Realism and will. See constraints without consolation; act as though agency can change the odds. The result is calibrated ambition.

General intelligence and plural talent. Stable differences in cognitive power coexist with creativity, drive, courage, social judgment, and luck. The result is a hierarchical but non-unitary theory of ability.

First principles and empirical provisionality. Rebuild the conceptual structure; accept uncertain nodes and revise with data. The result is cross-disciplinary speed.

Mastery and leverage. Decades spent understanding fundamentals give way to movement toward fast-improving technologies. The career is divided between intrinsic and consequential goods.

Individual choice and collective consequence. Give families useful information; prevent coercion and stratification. The unresolved result is the politics of reproductive technology.

Elite competence and democratic legitimacy. Let capable people act; require accountability and plural consent. The result is institutional conflict.

Stoic detachment and worldly ambition. Reputation is transient, yet projects should alter reality. The result is achievement without simple status worship.

Humanism and optimization. Reduce disease and enlarge capability while preserving dignity independent of measured traits. This is the moral test of enhancement.

Human inheritance and posthuman succession. Preserve flourishing, memory, and value diversity while accepting enhancement, merger, or artificial descendants. What makes a successor ours remains unresolved.

These are not accidental inconsistencies. They are generated by Hsu’s position at the meeting point of science and power. A laboratory can isolate variables; a society cannot. A predictor can be statistically valid while its deployment is unjust. An exceptional person can identify an institutional failure while misunderstanding why others resist his remedy. A technology can expand agency for some while narrowing it for others.

The last tension may be the deepest. Hsu’s household ethic and cosmic ethic operate at different scales. In the first, family and human connection make public achievement look like vanity. In the second, intelligence is a universe-shaping process that may outgrow the ape body, the present species, and even base reality. The mature portrait should not force either Hsu to defeat the other. His work is animated by the unresolved question of whether inheritance consists in preserving the vessel, preserving the flame, or finding a transformation in which the distinction no longer holds.

Hsu’s temperament pushes him toward making these conflicts explicit. He prefers a sharp, falsifiable statement to a socially smoother ambiguity. This can clarify hidden premises, but it can also underprice rhetoric’s effects in domains where trust is part of the causal system. His intellectual journey is therefore also a study in the limits of transferring the physicist’s stance wholesale into public life.

13. When the future draws near

The best single word for Hsu’s career is not polymathy but translation: the carrying of an idea across the border that separates knowledge from power. Two other words complete it—threshold and inheritance. Translation describes the movement of his mind; threshold, his judgment of when to act; inheritance, the unsettled question of what should remain ours after action changes the world.

He translates physics into information-theoretic constraints; mathematical sparsity into genomic sample-complexity estimates; biobank-scale data into predictors; prediction into companies; startup experience into institutional strategy; psychometrics into a theory of elite performance; quantum branching into a language of agency; and generative AI into a collaborator whose insights must be verified and operationalized.

The severity of his standard is visible in a sentence about Feynman’s lectures:

“None can claim themselves an educated thinker or intellectual without mastery of a significant portion of the material in these lectures.”

— “Feynman Lectures: Epilogue”

It is an extravagant demand, and revealing precisely for that reason. Hsu’s idea of culture is not decorative acquaintance but internal possession: one should know enough mathematics and physics to see the load-bearing structure of modern reality. The library card in Ames thus leads to an adult ideal of civilization in which difficult knowledge belongs to the canon of an educated mind.

The visible subject across these translations is intelligence, but beneath it lies a more ancient drama: mind against limit. Hsu asks what can be inferred from incomplete information, how far an exceptional mind may depart from an ordinary one, how organizations gather or squander intelligence, and how technology might alter the distribution of capability itself. His projects become more applied over time even as their horizon expands—from fields and black holes to the future constitution of humanity and civilization.

The first-person record defeats the coldest caricatures. Hsu’s skepticism was formed not only by equations but by thin letters from a family suffering through ideological catastrophe. His self-command joins a Confucian father, a Christian and military-athletic maternal line, Hemingway’s courage, and Salter’s appetite for the vivid life. He prizes mastery even when it costs wealth, yet places ordinary family love above public distinction. He studies stable differences in ability, yet recommends the growth mindset to the person deciding how to live. His meritocracy is severe, but his memory of Iowa includes immigrants welcomed, children nurtured, and ordinary citizens less precarious than he believes they are now.

That fuller humanity does not resolve into a comforting humanism. Hsu can want enhanced descendants to preserve human agency against machines, then widen the category of descendants until artificial intelligence itself enters it. He can call family the deepest good of one life while contemplating, on billion-year scales, a future in which biology is only an early substrate of mind. The governing value is therefore not simple preservation. It is inheritance: the hope that intelligence, courage, memory, agency, and perhaps love can cross into forms whose continuity with us remains philosophically and politically uncertain.

Nor is the career a frictionless triumph of breadth. Hsu nearly became a quant; geography and fellowship prestige helped keep him in physics. He recognizes that fox-like range may sacrifice the hedgehog’s single eternal contribution. His projects succeeded not because every forecast was correct but because he repeatedly chose domains in which error could meet data, engineering, or the market soon enough to be corrected. This makes the career less teleological and more impressive: coherence was built through contingent choices, not granted in advance.

Nor is he simply a dreamer of remote futures. His signature gift is to sense when the derivative has changed—when cost curves, sample sizes, algorithms, or model capabilities have brought a distant prospect within reach. He does not merely predict science-fiction outcomes. He waits for them to cast a measurable shadow, then looks for the threshold at which they can become engineering programs. In genomics that shadow was a sample-size phase transition; in AI it is the advancing ability of models to perform research, supervise one another, and begin to improve the machinery of intelligence itself.

The making of Machine God adds a revealing final movement. Hsu is no longer content only to build and forecast. He wants to record the atmosphere before the break—to preserve the arguments of accelerationists and safety thinkers, and to warn about disempowerment even while developing the technology. The builder has become, in part, a chronicler of the forces he helped summon.

That record warrants neither hagiography nor dismissal. Hsu sometimes extrapolates from technical competence to institutional judgment too quickly. His rhetoric can compress moral and historical complexity. The governance problems raised by genetic prediction and enhancement are deeper than validation accuracy or individual consent. Yet critics who focus only on controversy miss the unusual coherence and effectiveness of his work. He has repeatedly crossed fields, mastered enough of their structure to find a leverage point, and helped produce outcomes that insiders had regarded as premature or impossible.

His intellectual history is thus a movement from discovering the boundaries of the world to testing which of them can be moved. The young physicist wanted to reconstruct reality on a blackboard from first principles. The mature Hsu asks which parts of reality—institutions, technologies, even the future human phenotype—can be reconstructed beyond the blackboard, and what obligations begin when reconstruction succeeds.

The career turns on three virtues. See without illusion. Dare without guarantee. Recognize the hour. Hsu’s deepest talent may be the last: to feel when an idea is no longer merely premature, when the future has drawn close enough to be grasped. His deepest unresolved question is what can be carried through the gate.

Source note: spoken excerpts are lightly punctuated for readability; ellipses mark omitted fillers or intervening words. Every quotation links to its original post, transcript, or paper.

A voice across the years: selected longer quotations

The following passages are arranged thematically rather than chronologically. Together they show the development traced above: from observing exceptional human ability, through an ethic of independent judgment and difficult action, toward species-level technological change, civilizational competition, the multiverse, and machine intelligence.

A. Genius: the unequal light

1. Extreme ability is real

“Personally, I find Landau’s scheme appropriate. There are many physicists whose contributions I cannot imagine having made.”

— “Out on the Tail”

Hsu’s writing about genius begins with phenomenology: prolonged contact with exceptional performers convinces him that the upper tail is not merely an amplified version of the middle. Some achievements remain difficult even for other highly capable experts to imagine producing. This is the experiential basis of his resistance to egalitarian fictions about ability.

But the statement is also self-limiting. He includes himself inside the hierarchy rather than merely ranking others from above. Intellectual honesty requires acknowledging minds whose operations exceed one’s own.

2. Intelligence is not achievement

“Luck, drive, creativity, and other factors, all at least somewhat independent of intelligence, influence success in science.”

— “Success, Ability, and All That”

This is the necessary counterweight to Hsu’s emphasis on cognitive differences. Ability changes the distribution of possible achievements, but it does not uniquely determine the outcome. Scientific success requires problem selection, stamina, originality, mentorship, historical timing, and luck. Hsu’s own career supports this plural account: the capacities needed to derive a physics result are not identical to those needed to found a company or direct a research university.

Together, these two passages define Hsu’s mature view of genius: strongly hierarchical, empirically realist, but not reducible to IQ or technical speed.

B. Courage: the premature idea

3. The courage to choose the premature idea

“Feinberg had the courage to engage with ideas that were much more speculative in the late 60s than they are today.”

— “Gerald Feinberg and the Prometheus Project”

Hsu praises Gerald Feinberg for treating artificial intelligence and genetic engineering as serious civilizational subjects before respectable discourse was ready. The passage is autobiographical by projection. Hsu is drawn to thinkers who recognize a technological trajectory early, but the important temporal clause is “than they are today.” Speculation can mature into a research program as enabling conditions change.

This is why disciplined audacity fits him. Courage identifies the frontier; theory and timing determine when to cross it.

C. Mastery: the private kingdom

4. Knowledge as an intrinsic achievement

“My satisfaction with having mastered these concepts in mathematics and physics and biology and computation is very valuable to me internally.”

— Undertone interview

This may be the most revealing personal statement in the collection. Hsu is aware that remaining in fundamental science carried a large financial opportunity cost. Yet he does not evaluate those decades as a failed optimization. Mastery itself is a durable internal possession.

The sequence of fields matters. Mathematics, physics, biology, and computation are not items on a résumé; they are conceptual structures he has labored to make coherent from the inside. The quote softens the public image of relentless instrumental rationality. Hsu wants ideas to work, but he also wants to understand them deeply enough that the understanding becomes part of him.

D. The future of mankind: who inherits the flame

5. The uncertain continuity of human intelligence

“Maybe we need to improve ourselves… I might still prefer their survival to a civilization that’s completely dominated by machines.”

— Manifold conversation with James Lee

The line reveals a motive deeper than competitive enhancement. Hsu imagines biotechnology as a possible means of preserving human agency when machine intelligence exceeds the natural human range. Enhanced descendants may differ from us, yet still carry a recognizable inheritance.

Other statements prevent a simple preservationist reading. Hsu has regarded artificial minds as descendants and imagined biological and machine intelligence merging. In the Ngo conversation he also demonstrates that he can formulate the opposing Butlerian case with unusual force; the transcript explicitly identifies it as a steelman directed toward nonspecialists, not a newly adopted creed. The ethical difficulty is therefore larger than access or trait choice. What degree of change preserves continuity? Is lineage enough? Must embodiment, vulnerability, memory, love, control, or human value diversity survive? The passage records one side of Hsu’s aspiration: the desire to carry the human project across a threshold unaided evolution may not cross in time. His broader thought leaves open who—or what—will carry it. (“The Future of Intelligence”; conversation with Richard Ngo.)

E. Civilization: the passing of an age

6. History can change phase within one lifetime

“A nation can pass from one age to the next, as I believe we have in America during my lifetime.”

— “Remarks on the Decline of American Empire”

Hsu’s civilizational thought has the same structure as his scientific thinking: systems can cross thresholds and enter a qualitatively different regime. Decline is not necessarily a smooth reduction in wealth or power; it can be a loss of institutional memory, competence, confidence, or the ability to coordinate.

The melancholy tone distinguishes this writing from his technological optimism. He believes capability may expand at the species level while particular institutions decay. That combination—optimism about intelligence, pessimism about governance—is one of the central tensions of his mature worldview.

F. Base reality: the world behind the world

7. Our world may not be fundamental

“Under these assumptions, it is not implausible that we ourselves are actually simulated beings, and that our world is not base reality.”

— “The Quantum Simulation Hypothesis”

The opening qualification is essential. Hsu is not presenting mystical certainty; he is tracing the consequences of premises about computation, conscious observers, and technologically mature civilizations. If simulated observers become much more numerous than unsimulated ones, ordinary typicality arguments become disturbing.

The passage displays Hsu’s willingness to accept an alien conclusion when the model points toward it. It also suggests why “information processing” is an ontological phrase for him, not merely the name of a blog. Minds, worlds, and perhaps universes can all be described in computational terms.

G. The multiverse: intelligence among the branches

8. The creators inside the creation

“Let’s imagine a future with super powerful ASIs with infinite energy resources… simulated worlds, which in turn have sentient beings inside them.”

— Manifold conversation with Joscha Bach

Hsu offers this as a question, not a prophecy. Yet it captures the vertigo of his mature thought: the intelligence humanity is now building may eventually make universes populated by beings who experience their world as primary. Physics, artificial intelligence, and moral philosophy collapse into one problem—what obligations does a creator have to conscious lives inside a model?

It also turns the simulation argument inside out. Instead of asking only whether we are created, Hsu asks what our intellectual descendants may create. The observer remains cosmically unprivileged, but responsibility expands with computational power.

9. Civilization as a branch-selecting intelligence

“One could regard human civilization as a single intelligence or information processing machine… making greater use of nearby patches of the multiverse previously inaccessible.”

— “AI in the Multiverse”

Here Hsu’s metaphysics becomes a theory of history. Civilization aggregates knowledge, compute, institutions, and action; in that sense it functions as a distributed mind. More capable intelligence can model a larger space of futures and steer toward outcomes that less capable systems could neither perceive nor realize.

The passage also supplies a unifying interpretation of his career. Physics describes the possibility space. Genomics maps latent biological outcomes. Entrepreneurship and administration coordinate selection among paths. AI enlarges the search and action capacity of the collective system. Across all four, intelligence is the means by which possibility becomes actuality.

The man who emerges

Read together, these passages disclose six enduring features of Hsu’s character.

First, he is hierarchical without being simple-minded about hierarchy. He believes extreme differences in ability are real, yet explicitly separates intelligence from creativity, drive, personality, and luck. His advice to adopt a growth mindset prevents population-level realism from hardening into personal fatalism.

Second, he treats courage as an epistemic and practical virtue. Courage means revising beliefs when evidence changes, considering an idea before it is respectable, and committing resources when a difficult project has become tractable.

Third, he is animated by mastery, but does not finally worship achievement. Wealth, reputation, and institutional power are real goods, but none replaces the satisfaction of understanding a field from foundations to frontier—and, at the deepest level, even that private kingdom yields to family, mortality, and love.

Fourth, his futurism is humane at the scale of a life and posthuman at the scale of civilization. His conviction that ordinary family life contains most of the great stuff prevents biological or cognitive rank from becoming a complete scale of human value. Yet his long-range thought allows enhanced humans, merged minds, and artificial descendants to inherit the future. The tension is real, not terminological: the intimate goods that make one human life meaningful coexist with a cosmic perspective in which present biological form is provisional.

Fifth, his worldview is computational from the smallest scale to the largest. Observers emerge within the wavefunction; simulated worlds may rival base reality; civilizations process information and select futures; AI expands the accessible region of possibility. Yet computation does not exhaust value. Marcus, Hemingway, Salter, and the language of family keep returning embodiment, courage, beauty, and love to the center.

Sixth, he is a builder who has become increasingly attentive to what building may destroy. In genomics he warns of a breakaway hereditary elite; in AI he sees labor shocks, the erosion of scientific apprenticeship, recursive self-improvement, and gradual human disempowerment. Machine God embodies this late development: Hsu still wants to cross the threshold, but he also wants civilization to recognize it before the passage becomes irreversible.

The arc is therefore not a descent from pure science into mere application. It is a widening of scale: from learning the laws that fence reality in to building minds and institutions capable of finding the gates. Hsu’s recurrent question is not only what is true, but what a sufficiently clear-sighted and capable intelligence can bring into the world—and what, once the gate opens, remains worth carrying forward.

Bohan Liu is a partner at Chemistry Ventures, based in SF.

https://x.com/loubohan

Bohan’s report on the venture ecosystem in China:

https://www.linkedin.com/posts/bohanlou_i-spent-last-month-in-china-meeting-most-ugcPost-7487909585689698304-qepw/

Chapter Markers:

• (01:42) - Venture Capital in China; Bohan’s Shanghai Roots

• (11:15) - Chemistry Fund and China Investing

• (20:04) - China Trip and Venture Reality Check

• (32:18) - Pressure Cooker: Founder Exits and Financial Liability

• (33:50) - Variation in Founder-Friendly investment environment

• (46:52) - AI Models and Open Source

• (58:56) - Big Picture and Wrap Up

Audio-only version and transcript:

https://www.manifold1.com/episodes/venture-capital-and-technology-in-china-with-bohan-liu

GPT analysis of my earlier post. Orders of magnitude: organic “real” demand for AI versus Capex and future cash flows.

The central claim in Hsu’s original X post is supported by the available evidence: current AI revenue is heavily concentrated in OpenAI and Anthropic, while much of those companies’ spending is financed by investors rather than by operating profits. The amount of genuinely organic demand—money ultimately coming from consumers and established, cash-generating companies—appears to be only of order $10 billion. A more detailed reconstruction might produce $30–50 billion rather than exactly $10 billion, but these are consistent order-of-magnitude estimates. Either is tiny compared with the hundreds of billions being invested annually in AI infrastructure.

The concentration claim is best documented at Microsoft. Its FY2026 filing reports $24.1 billion of revenue from OpenAI, including revenue-sharing payments, against analyst estimates of roughly $34.5 billion in total Microsoft AI revenue. That implies approximately 70% dependence on OpenAI. Microsoft also reported $6 billion of OpenAI accounts receivable at year-end.

Microsoft FY2026 10-K

For AWS, analyst estimates put OpenAI and Anthropic at roughly 59–73% of AI revenue, although Amazon does not disclose customer concentration. Amazon has said its AI business now exceeds a $25 billion annual revenue run rate, while Anthropic has committed more than $100 billion to AWS over ten years.

Amazon Q2 results, Anthropic–AWS agreement

Google is less transparent. UBS estimates reportedly imply that OpenAI and Anthropic constitute about 28% of total Google Cloud revenue in 2026 and nearly half in 2027. Because Google Cloud also includes conventional computing, storage and software, the two labs could plausibly represent most of its specifically AI-related revenue, but the “70%” figure is inferred rather than disclosed. Google Cloud revenue nevertheless grew 82% in Q2, driven primarily by AI infrastructure and enterprise AI services.

Alphabet Q2 results

The crucial point is that revenue at successive layers of the AI stack cannot be added together as independent demand. A company may pay Anthropic for Claude usage; Anthropic then pays AWS or Google for the compute. The same external dollar appears first as model-company revenue and again as cloud revenue. Moreover, when Anthropic or OpenAI spends more on compute than it receives from customers, the difference is supplied by newly raised capital. OpenAI, for example, reports about $2 billion in monthly revenue but raised $122 billion at an $852 billion valuation in March. Anthropic reports a $47 billion revenue run rate but simultaneously raised $65 billion at a $965 billion valuation.

OpenAI funding disclosure, Anthropic Series H

After removing double counting and heavily discounting AI consumption by loss-making, venture-funded startups, a reasonable estimate of current final demand is approximately:

• $15–20 billion from OpenAI consumers and established enterprises;

• $10–20 billion from Anthropic consumers and established enterprises;

• perhaps another $5–10 billion from other direct products and non-lab enterprise AI consumption.

That gives roughly $30–50 billion of annualized organic demand. But this remains of order ~$10B, precisely the scale asserted in the original post. Indeed, given the uncertainty surrounding Anthropic’s rapidly annualized “run-rate” metric and the startup share of enterprise API consumption, $10 billion is a defensible lower-end estimate. Reuters has noted the striking gap between Anthropic’s claimed annualized run rate and the much smaller amount of revenue it had actually booked cumulatively.

Reuters Breakingviews

Against this ~$10B organic revenue base, the infrastructure buildout is ~$trillion. Amazon, Microsoft, Alphabet and Meta are on course to spend roughly $700 billion in 2026, with Oracle and other providers pushing the total higher. Not all of this is AI-related, but approximately $450–600 billion probably is.

The spending is already consuming most of the hyperscalers’ cash generation. Amazon’s trailing free cash flow was negative $7.6 billion; Meta produced only $784 million of Q2 free cash flow after $31.1 billion of capex; and Alphabet reported negative $5.9 billion of Q2 free cash flow. Microsoft remains strongly cash-generative, but its cash capex nearly doubled to $115.9 billion, with another $24.6 billion of infrastructure obtained through finance leases. Current net income figures are also flattered by paper gains: Amazon’s Q2 earnings included a $53.4 billion pre-tax gain primarily on Anthropic, while Alphabet recorded a $77.1 billion after-tax gain on equity securities.

A simple capital-recovery calculation illustrates the hurdle. If the industry invests $450–550 billion annually in AI infrastructure from 2026 through 2029, it will create roughly $1.8–2.2 trillion of installed capital. Assuming a five-year blended economic life, a 9% required return and 40–60% cash contribution margins, that infrastructure ultimately needs approximately $700 billion to $1.3 trillion of annual revenue. A central estimate is about $1 trillion.

Growing an organic base of $30–50 billion to $700 billion–$1.1 trillion by 2030 requires approximately 90–120% annual growth. In other words, organic AI spending must roughly double every year for another four years. The private valuations of OpenAI and Anthropic alone require somewhat less but still extraordinary growth: their combined $1.8 trillion valuation plausibly requires $300–500 billion of annual revenue by 2030, implying roughly 60–90% annual organic growth from the estimated present base.

Thus, the more detailed analysis reinforces the original post. Whether present organic AI revenue is labeled $10 billion, $30 billion or even $50 billion does not materially change the conclusion. The industry is attempting to support an ~$trillion capital base with an ~$10B final-demand base. The capex and current valuations can make sense—but only if organic use approximately doubles every year, customer concentration falls, and margins improve despite rapidly declining compute prices.

If organic demand grows at a still-impressive 50% annually, a $40 billion base reaches only about $200 billion by 2030. That would support several very valuable AI businesses, but not the infrastructure currently being built. In that scenario, the likely outcome is substantial excess capacity, collapsing compute prices, asset impairments and valuation compression—especially for model labs, Nvidia, neoclouds and leveraged data-center projects.

Realities of Modern War

Modern war is being reshaped by a simple imbalance: the cost and reach of precision attack are improving faster than the cost and capacity of defense. The result is not that defenses are useless. It is that no defender can assume an affordable, leak-proof shield against a sustained campaign. The strategic answer is therefore less about perfect interception and more about deterrence, dispersal, redundancy, stockpiles, industrial capacity, and the ability to keep functioning after attacks get through.

The offense–defense asymmetry begins with physics and geometry. Against a comparable technology, an interceptor has the harder kinematic problem: it must detect an incoming weapon, accelerate rapidly, maneuver with sufficient margin to reach a predicted intercept point, discriminate the real warhead from debris or decoys, and achieve a direct hit or very close pass. It must therefore generally be more capable—and more expensive—than the missile it is trying to destroy. The sensor burden is asymmetric as well. The defender needs persistent, precise coverage of a wide area, plus communications and battle management able to process many simultaneous tracks. The attacker needs only the coordinates of a fixed target; even against a moving ship, it needs to find and update one target track, not monitor and defend an entire theater. Finally, the attacker chooses the time, route, weapon mix, and point of concentration. The defender must be ready everywhere, all the time.

The arguments below follow themes developed by Steve Hsu in his writing on missile defense, maritime blockade, and U.S.–China military competition. His posts are used as a guide to the source trail; the supporting claims rest primarily on the government reports and independent technical analyses cited alongside them.

1. Missiles and drones are cheap, precise, and difficult to defend against

Relative to their targets and to the interceptors used against them, one-way drones and many missiles are cheap, increasingly precise, and difficult to stop in mass.

A Shahed-class attack drone illustrates the new economics. The Royal United Services Institute estimates a Shahed-136 at roughly $20,000–$30,000, with an estimated range of 1,300–1,500 kilometers. It is not a technological marvel; much of its effectiveness comes from commercially available components, adequate navigation, and the fact that it can be launched repeatedly and in numbers. The important comparison is not between a drone and a fighter aircraft in isolation. It is between the attacker’s cost per additional threat and the defender’s cost per additional engagement. In the 2025 Israel–Iran conflict, an Arms Control Wonk estimate—highlighted by Hsu—put the cost of 39 or more THAAD interceptors at over $495 million, using the Missile Defense Agency’s fiscal-year 2025 unit cost of approximately $12.7 million. Subsequent reporting indicated that the United States actually fired more than 150 THAAD interceptors during that twelve-day war—nearly one quarter of all the THAAD rounds the Pentagon had ever purchased.

The much larger 2026 Iran war turned the exchange ratio into a strategic stockpile crisis. In a July 27 post, Hsu summarized the open-source estimates bluntly: the war had “used 30–80% of US stockpiles of key weapons,” and the resulting “window of vulnerability will last several years.” In its accounting of the first 39 days, the Center for Strategic and International Studies estimated that U.S. forces fired 1,060–1,430 Patriot interceptors and 190–290 THAAD interceptors. Later public estimates diverged because inventories and expenditures are classified: one CSIS update put remaining stocks below 1,000 Patriot interceptors and at 234–278 THAAD rounds, while subsequent press reporting cited sources claiming that as much as 80 percent of the THAAD inventory had been consumed. The precise percentages remain uncertain; the depletion is not.

The cost is measured in billions. A 2024 U.S. Army contract for 870 PAC-3 MSE missiles and associated hardware averaged about $5.2 million per missile, while the FY2025 budget put a THAAD interceptor at approximately $12.7 million. Applying those prices only as an order-of-magnitude replacement measure, the Patriot and THAAD rounds in CSIS’s initial 2026 estimate represent roughly $8–11 billion—before counting radars, launchers, crews, operations, or the Standard Missile interceptors also expended. Nor can money immediately restore the magazines. CSIS projects that Patriot and THAAD stocks will not return to their prewar levels until roughly 2029, despite emergency efforts to expand production.

Retired Israeli Major General Yitzhak Brik had described the attacker’s intended logic during the 2025 fighting. In a widely circulated English translation, he warned that Iran could continue firing heavy ballistic missiles “after Israel runs out of anti-missile missiles, leaving Israel vulnerable to a barrage of heavy missiles without adequate defense capabilities.” The 2026 experience gives the warning concrete meaning. An attacker can mix inexpensive drones, decoys, older missiles, and a smaller number of sophisticated weapons, forcing the defender to identify and engage them under time pressure. Air and missile defenses remain essential, but they are a rationed asset—not a force field. The central problem is not whether an interceptor sometimes works; it is whether the defense can keep working after hundreds or thousands of engagements.

Supporting references

• Steve Hsu, “Missile and Drone War: Enter the Houthis”, X, July 26, 2026.

• Steve Hsu, post on Iran-war depletion of U.S. weapons inventories, X, July 27, 2026; Center for Strategic and International Studies, “Rebuilding U.S. Missile Inventory: A Multiyear Project”, May 27, 2026, and “Renewed Iran War Would Test Diminished Interceptor Inventories”, July 27, 2026.

• Steve Hsu, comment on THAAD expenditure and production, X, June 25, 2025; Jeffrey Lewis and Decker Eveleth, “Exhaustion and Inflection: Estimating Interceptor Expenditures in the Israel–Iran Conflict”; Wall Street Journal, “Israel’s 12-Day War Revealed Alarming Gap in America’s Missile Stockpile”, July 24, 2025.

• Yitzhak Brik, translated remarks circulated on X during the 2025 Israel–Iran war, June 16, 2025.

• Royal United Services Institute, “Russia’s Iranian-Made UAVs: A Technical Profile”.

• U.S. Army, PAC-3 MSE multiyear contract for 870 missiles, June 28, 2024; Missile Defense Agency, FY2025 Procurement Budget Justification.

2. Attack ranges are measured in thousands of kilometers, placing fixed infrastructure and ships at risk

Modern precision-strike systems can reach across entire theaters; static high-value sites and large surface ships cannot be guaranteed protection from sustained long-range attack.

China’s publicly assessed missile inventory makes the scale concrete. The U.S. Department of Defense assigns the DF-21D anti-ship ballistic missile a range of roughly 1,500 kilometers, the DF-26 a range of 3,000–4,000 kilometers, and land-attack cruise missiles such as the CJ-10 and CJ-100 ranges around 1,500–2,000 kilometers. These are not merely long-range explosives. They are components of a reconnaissance-strike system that combines satellites, aircraft, drones, over-the-horizon sensors, data links, and terminal seekers. Hsu’s earlier discussion of the Yaogan satellite constellation emphasized the same point: long-range missiles become strategically decisive when paired with persistent ocean surveillance and sufficiently frequent targeting updates.

Fixed installations are even simpler targeting problems. Air bases, ports, power plants, desalination facilities, fuel farms, cable landings, and large data centers occupy known coordinates and depend on a small number of external connections. A defender may harden individual structures or intercept some incoming weapons, but it must protect many aim points continuously. The attacker chooses the time, direction, weapon mix, and point of concentration. Repairs can also be slow: the U.S. Department of Energy reports that large power transformers commonly require 36 months to procure, with some lead times reaching 60 months. A campaign need not destroy an entire power station or data center if repeated attacks can disable transformers, cooling, substations, fuel supply, or communications.

Ships are mobile, but mobility is no longer concealment by itself. Their survivability depends on the quality of the adversary’s kill chain and on countermeasures that can break it. Large ships can maneuver, jam, deceive, shoot down incoming weapons, and strike launch platforms; none of this makes them invulnerable. As Hsu argued after Iran’s October 2024 missile attack on Israel, the strategically relevant question is whether a dense, mixed salvo can create enough leakage against the best available defensive network. If it can, a carrier group or forward base should be treated as a valuable but risk-bearing asset, not as a sanctuary.

Supporting references

• Steve Hsu, “Iran vs Israel: Implications for Missile Defense”, October 3, 2024.

• Steve Hsu, “Geostrategy and US-China Military Competition”, August 19, 2022, and “A2/AD Fait Accompli?”, August 2016.

• U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2024.

• U.S. Department of Energy, Large Power Transformer Resilience Report, July 2024.

3. Islands and ports can be interdicted without a classical close blockade

An attacker does not have to seal every sea lane or sink every ship; it can interdict an island or port by making commercial passage unacceptably dangerous.

Traditional blockade imagery centers on a fleet stationed offshore, stopping and searching ships. Long-range missiles and drones permit a different model: declare an exclusion zone, demonstrate the ability to hit one or two vessels, threaten port approaches and loading facilities, and allow shipping companies, crews, insurers, and lenders to do much of the remaining work. The recent Red Sea crisis provides an empirical example. Houthi forces did not control the sea and did not destroy most passing ships. Yet the IMF reported that trade through the Suez Canal fell 50 percent year over year in the first two months of 2024, while trade around the Cape of Good Hope rose 74 percent. UN Trade and Development later estimated an 82 percent decline in container tonnage transiting the canal by the first half of February. A comparatively small force imposed a large rerouting cost on global commerce.

This changes the standard for successful interdiction. The attacker may need only a credible probability of loss, especially against tankers, container ships, and other civilian vessels that lack naval defenses. Ports compound the vulnerability because their channels, cranes, storage tanks, substations, and rail or road links are fixed and difficult to duplicate quickly. Escorts and minesweeping can reduce risk, but they do not eliminate long-range attack, and the naval commitment required to sustain protected traffic may be enormous. Conversely, interdiction is not automatic: surveillance can be disrupted, launchers destroyed, convoys protected, and ports repaired. The point is that the defender must repeatedly succeed across a complex logistics system, while the attacker needs only intermittent success to affect commercial behavior.

Supporting references

• Steve Hsu, “Japan and The Quad”, June 14, 2021; Steve Hsu, “On the Military Balance of Power in the Western Pacific”, February 11, 2017.

• Steve Hsu, “Missile and Drone War: Enter the Houthis”, X, July 26, 2026.

• International Monetary Fund, “Red Sea Attacks Disrupt Global Trade”, March 7, 2024.

• UN Trade and Development, Navigating Troubled Waters, 2024.

4. Import-dependent islands such as Taiwan and Japan have finite endurance

Taiwan and Japan are industrial powers with the logistical vulnerability of islands: both import almost all of their energy and most of their food calories by sea.

The numbers are stark. Taiwan imports more than 95.8 percent of its energy, according to the U.S. Department of Commerce. Its energy-weighted food self-sufficiency rate was about 31 percent in 2022, according to Taiwan’s Ministry of Agriculture data summarized by the U.S. Department of Agriculture—meaning roughly 69 percent of food energy depended directly on imports. The real exposure is somewhat greater because locally produced meat and dairy often rely on imported feed. Japan’s energy self-sufficiency rate was 15.3 percent in fiscal year 2023, implying import dependence of 84.7 percent. Its latest calorie-based food self-sufficiency rate was 37 percent in fiscal year 2025, implying that approximately 63 percent of food calories came from abroad. Japan also imports virtually all of its crude oil, more than 90 percent of it from the Middle East.

These figures do not mean either society would collapse the moment shipping slowed. Both possess storage, rationing mechanisms, domestic agriculture, alternative ports, refining and generation capacity, and powerful allies. But national endurance is set by the most constrained essential flow, not by aggregate GDP. Oil, liquefied natural gas, coal, feed grain, fertilizer, and critical industrial inputs have different stockpiles and substitution possibilities. A blockade or quarantine can therefore be calibrated: pressure tanker traffic, damage a few terminals, interfere with port operations, and wait for inventories, electricity output, industrial production, and political confidence to decline.

For deterrence, resilience must be measured in days and replacement rates, not just military platforms. The relevant questions are how much usable fuel is dispersed outside vulnerable terminals, how quickly electricity can be rationed without losing water and communications, whether food stocks match likely consumption patterns, how many ports can handle diverted cargo, and whether merchant crews and insurers will accept the risk. Civil defense, dispersed storage, hardened grids, alternative energy, convoy procedures, and pre-negotiated shipping arrangements are therefore part of military preparedness.

Supporting references

• Steve Hsu, “Geostrategy and US-China Military Competition”, August 19, 2022; Steve Hsu, “Japan and The Quad”, June 14, 2021.

• U.S. Department of Commerce, “Taiwan—Energy Generation and Storage”, updated December 2025; U.S. Department of Agriculture, Taiwan Food Security Situation Overview, 2024.

• Japan Ministry of Economy, Trade and Industry, FY2023 Energy Supply and Demand Report, April 25, 2025; Agency for Natural Resources and Energy, Japan’s Energy 2023.

• Japan Ministry of Agriculture, Forestry and Fisheries, Food Self-Sufficiency Data, FY2025 result released August 2026.

5. Advanced weapons depend on rare earths, and the U.S. supply chain cannot yet support wartime scale-up

Almost every major advanced weapons category depends somewhere on rare-earth elements or similarly concentrated critical minerals, while the United States still lacks a fully reliable mine-to-magnet supply chain at wartime scale.

The vulnerability is not simply access to ore. Rare-earth permanent magnets and related materials appear in fighter engines, missile guidance and control systems, antimissile defenses, satellites, communications equipment, sensors, batteries, and precision actuators. The difficult industrial steps include separation, refining, alloying, metalmaking, and high-performance magnet production. The International Energy Agency estimates that in 2024 China accounted for about 60 percent of mined magnet rare earths, 91 percent of their refined output, and 94 percent of sintered permanent-magnet production. Those concentrations sit far above the threshold at which a supply disruption becomes a strategic problem.

The United States has recognized the risk and is investing in domestic and allied production. Since 2020, the Government Accountability Office reported, the Department of Defense had awarded roughly $439 million to rebuild domestic rare-earth supply chains. DoD projects now support mining and separation, recycling, and magnet plants. That progress matters, but it should not be confused with an already mature mobilization base. Plants take years to qualify, defense programs require consistent material performance and traceability, and a new mine does not by itself supply separated oxides, alloys, or finished magnets. China’s April 2025 export controls on seven categories of medium and heavy rare earths demonstrated how quickly concentrated processing capacity can become geopolitical leverage.

The strategic implication is that missile and drone competition is also a materials and manufacturing competition. A force can possess excellent designs yet fail to replace expended interceptors, seekers, motors, radars, aircraft, and satellites at the rate demanded by war. Stockpiling finished magnets and critical inputs can buy time; recycling and substitution can reduce exposure; long-term offtake contracts can make non-Chinese facilities financeable. But the decisive measure is qualified output per year across every stage of the chain. Until that output exists at scale, U.S. plans for rapidly expanding production of advanced weapons rest on a supply base that an adversary can constrain before or during a conflict.

Supporting references

• Steve Hsu, comment on interceptor production after Chinese rare-earth restrictions, X, June 25, 2025.

• U.S. Department of Defense, FY2025 Defense Production Act Purchases Budget Justification; Department of Defense, “DOD Looks to Establish ‘Mine-to-Magnet’ Supply Chain”, March 11, 2024.

• International Energy Agency, Rare Earth Elements: Pathways to Secure and Diversified Supply Chains, 2026.

• U.S. Government Accountability Office, Defense Industrial Base: Actions Needed to Address Risks Posed by Dependence on Foreign Suppliers, September 2024; Ministry of Commerce of the People’s Republic of China, Announcement No. 18 of 2025, April 4, 2025.

Conclusion

The common thread is an inversion of familiar assumptions. Distance no longer guarantees sanctuary. Technological superiority does not guarantee a favorable cost exchange. Naval control does not require a ship beside every merchant vessel. Economic strength does not erase dependence on vulnerable physical flows. And an advanced weapons design is not a usable wartime capability unless the industrial system can manufacture it repeatedly under pressure.

This does not make defense futile or war predetermined. It changes what credible defense requires. The durable advantages will belong to states that can absorb strikes, conceal and disperse assets, restore essential services, keep ports and logistics operating, manufacture replacements faster than they are consumed, and impose costs on the attacker in return. Modern war still rewards superior technology, but it rewards resilience and production at least as much.

Added: A reader on X had Claude prepare this summary, with nice figures

This is a crossover episode with Alf Beckinsale, an Oxford University student who hosts the Seeking Truth From Facts podcast. The main topics are the US-Iran conflict, geopolitics, and US-China AI competition.

Seeking Truth from Facts

My personal Substack

By Alf B.

Chapter Markers:

• (00:00) - Introduction

• (02:33) - No Off Ramp in Gulf

• (04:11) - Missile Stockpile Crisis

• (06:46) - Oil Shock and Reserves

• (14:07) - Israel Stays Out

• (29:34) - China Frontier AI Surge

• (43:25) - World War and Nukes

Audio-only version and transcript:

https://www.manifold1.com/episodes/iran-ai-and-the-third-world-war-crossover-with-seeking-truth-from-facts-podcast

Steve discusses the state of AI in summer 2026. Topics covered include: AI in math and theoretical physics, Recursive Self-Improvement, Agent swarms and tokenomics, IPOs and US-China competition, documentary film Machine God.

Machine God trailer:

Theoretical Physics with Generative AI: https://stevehsu.substack.com/p/theoretical-physics-with-generative

Chapter Markers:

• (00:00) - State of AI, Summer 2026

• (02:39) - AI in Math Physics

• (13:54) - Recursive Self-Improvement

• (21:08) - DeepSeek Hyperconnections

• (29:42) - Agent Swarms and Tokenomics

• (43:26) - Machine God Documentary

Audio-only version and transcript:

https://www.manifold1.com/episodes/state-of-ai-summer-2026-116

Announcing this for some friends at Mechanize - a startup that builds environments for training and evaluating frontier LLMs. Its customers include the top AI labs, and it has contributed to the breakthrough in coding capabilities of frontier models.

Mechanize is hiring!

https://mechanize.work/b/hsu

Compensation is extremely competitive. For technical roles, $300-500k. They are also seeking smart generalists.

For example:

Research Engineer, Alignment: Build evals that test for misaligned model behaviors $500K salary

Puzzle Maker: Design interesting and original puzzles that LLMs can’t yet solve $300K salary

Mechanize understands that my readership is highly selected. There is a VERY GOOD CHANCE you will be interviewed if you apply via the link above.

Beff Jezos (Guillaume Verdon) is a Canadian physicist, quantum computing researcher, and tech entrepreneur. He is best known as the founder of the Effective Accelerationism (e/acc) movement.

This interview was recorded in collaboration with John Greer and Lei Huang for the documentary film Machine God.

John Greer:

John Greer

Optimization, AI Safety, Rationality, Life Extension, EA, Movies, MMA, and the Occasional Interview

Chapter Markers:

• (00:00) - Meet Beff Jezos

• (01:56) - Hyperstition vs AI Doom

• (21:29) - Accelerate or Die

• (37:51) - Scaling Intelligence Upside

• (40:26) - Doomers Power Centralization

• (47:17) - EAC Physics Framework

• (01:10:10) - Debating Doom Narratives

• (01:16:13) - Beff Jezos Origin Story

• (01:28:06) - Pause Debate and Rapid Fire

Audio-only version and transcript:

https://www.manifold1.com/episodes/beff-jezos-and-effective-accelerationism-machine-god-of-loving-grace

Announcing this for some friends at Mechanize - a startup that builds environments for training and evaluating frontier LLMs. Its customers include the top AI labs, and it has contributed to the breakthrough in coding capabilities of frontier models.

Mechanize is hiring!

https://mechanize.work/b/hsu

Compensation is extremely competitive. For technical roles, $300-500k. They are also seeking smart generalists.

For example:

Research Engineer, Alignment: Build evals that test for misaligned model behaviors $500K salary

Puzzle Maker: Design interesting and original puzzles that LLMs can’t yet solve $300K salary

Mechanize understands that my readership is highly selected. There is a VERY GOOD CHANCE you will be interviewed if you apply via the link above.

Mechanize is hiring!

Compensation is extremely competitive. For technical roles, $300-500k. They are also seeking smart generalists.

For example:

Research Engineer, Alignment

Build evals that test for misaligned model behaviors

$500K salary

Puzzle Maker

Design interesting and original puzzles that LLMs can’t yet solve

$300K salary

Mechanize understands that my readership is highly selected. There is a VERY GOOD CHANCE you will be interviewed if you apply via the links on this post.

One of my closest friends passed recently of a heart attack, leaving behind his wife and children. He was an accomplished physicist, head of the physics division at Lawrence Livermore National Lab.

These are my remarks for his memorial.

William J Evans

My condolences to the Evans family and to everyone who knew Will. Please let us remember his Christian faith. He is with God now.

Will was a man of integrity and hard work and fine character. A great and true friend, someone you could always depend on. We are all richer for having known him. His loss will be felt for the rest of our days.

We met at freshman camp at Caltech in 1983. We were boys then, not yet men. Callow, youthful, with so much yet to learn about life and the world around us. Will was serious but friendly, easy to get to know. We confided in each other about everything.

It was the 1980s and we shared a love for Prince, Madonna, Bo Jackson, The English Beat. Our intellectual heroes were Caltech legends like Richard Feynman, Kip Thorne, Carver Mead, John Hopfield. We explored the mysteries of quantum physics and mathematics. Endless problem sets, working late into the morning, scribbling on the floor with books and papers all around us.

But we were also explorers of the night world of Los Angeles. Clubs, UCLA frat parties, blonde girls with heavy eyeliner smoking clove cigarettes. To Live and Die in LA, a movie we all loved, provided the soundtrack as we hurtled through the darkness on the endless freeways. We were suckers for every coming of age movie and we both loved Parker Posey.

Will and I were roommates during the 85-86 academic year, and again we were neighbors on the Charles River in Cambridge in the 1990s. His office faced the back entrance of Lyman-Jefferson, the Harvard physics building. I would always look in the window to see if Will was at his desk when I entered the building. It was a joy to stop in and shoot the shit with my best friend at Harvard.

Cambridge MA: Sunday afternoon in the early spring, a house with a gigantic window facing trees and a green lawn. Mostly girls at the party - anthropology and literature PhDs, alumna of Smith College, drinking good wine from fluted glasses. After some time we looked at each other - Time to go. Suddenly the sky was full of fat snowflakes, falling in sheets from a blue white sky as we ran to his car.

Immortal times, timeless days and nights of youth.

My children can’t imagine my young life - probably true for the Evans kids as well. I want you to know that Will lived a full life, full of friendship and adventure, the joy of scientific discovery, and later the joy of family life.

The world captured in those grainy photos really existed, even if it is long vanished now.

We are all richer for having known him. His loss will be felt for the rest of our days.

https://x.com/hsu_steve/status/2034806028941058108

This episode was recorded live at Manifest 2026. Razib Khan is a prominent writer, population geneticist, and podcaster. He is best known for his extensive deep-dives into human evolutionary history, consumer genomics, culture, and ancient DNA.

https://x.com/razibkhan

https://x.com/razibkhan?lang=en

Chapter Markers:

• (00:00) - Razib Khan at Manifest 2026: Genetic Discoveries, AI, and Academia

• (01:18) - Manifest Q&A Kickoff

• (02:43) - Yamnaya: Ancient DNA Mysteries

• (15:01) - Yamnaya: Y Chromosome Conquests

• (22:10) - Embryo Screening and AI

• (42:15) - Conformity and Tenure

• (46:34) - Academia: Reforms

• (53:55) - Academia: Ideological Capture and Funding

• (58:19) - Controversies and Closing Q&A

This special episode was recorded at Tsinghua University in Beijing, generally regarded as the top university in China. Our guests are 3 Americans studying and working at Tsinghua: Gabriel (undergrad), Justin (PhD student in AI), and Alex (Professor in AI research). Topics discussed include: Tsinghua University and elite human capital, AI in China, US-China competition, and the flow of human capital between the US and China

Han Feizi, columnist at Asia Times and the guest from the previous “Letter from Beijing” episode, is also in the room.

Letter from Beijing with Han Feizi: https://www.manifold1.com/episodes/letter-from-beijing-with-han-feizi-72

Audio-only version and transcript:

https://www.manifold1.com/episodes/letter-from-beijing-2-tsinghua-university-113

Chapter Markers:

• (00:00) - Welcome to Tsinghua University

• (02:47) - Gabriel’s Undergrad Journey

• (12:35) - Justin’s PhD

• (25:10) - Professor Alex on AI and Rankings

• (42:51) - Second Chances and Status Signals

• (46:48) - China’s Exam Ladder Explained

• (50:20) - Infrastructure and Tech Competition

• (01:17:18) - Semiconductors, EUV, and Wrap Up

Jaan Tallinn is a tech billionaire and founding engineer of Skype who leverages his wealth to mitigate existential risks from artificial general intelligence (AGI). He co-founded the Future of Life Institute and the Centre for the Study of Existential Risk, while making early foundational investments in frontier AI labs like DeepMind and Anthropic.

Chapter Markers:

• (00:00) - AI Risk Level Set

• (01:17) - Assessing Current AI Risk Levels

• (03:28) - Inside Self-Sustaining AI Scenarios

• (09:10) - The Global AI Race Dynamics

• (42:25) - Explaining the Techno-Capital Flywheel

• (45:34) - Insider Origins of AI Safety

• (56:06) - Race Politics and Public Fear

• (01:23:12) - Pop Culture, Movies, and Fame

• (01:30:15) - Big Questions for Humanity’s Future

Audio-only version and transcript: https://www.manifold1.com/episodes/ai-billionaire-on-existential-risk-jaan-tallinn

Dr. Alex Young, a statistical geneticist and assistant professor in the Human Genetics department at UCLA, joins Steve Hsu to discuss the cutting edge of genomic prediction. They cover his research on polygenic embryo screening in IVF (including the ImputePGTA method), family-based DNA analysis, missing heritability, and the implications of polygenic scores for traits like education and disease. Alex also discusses his recent battles with cancer.

https://x.com/AlexTISYoung

Chapter Markers:

• (00:00) - Alex Young Bio

• (06:36) - Biobank Era Genetics

• (10:49) - Missing Heritability Debate

• (27:18) - Embryo Selection Controversy

• (50:32) - Embryo Selection Backlash

• (53:42) - Mexico City Admixture Study

• (01:00:13) - Censorship Via Data Access Control

• (01:05:02) - Battle With Cancer and Circulating Tumor DNA (ctDNA)

Audio-only version and transcript:

https://www.manifold1.com/episodes/embryo-selection-and-frontier-genomics-with-dr-alex-young-111

Steve and Alf discuss the Iran War, emphasizing what it

reveals about modern missile and anti-missile technology, drones, and

the implications for a US-China conflict in the Western Pacific.

Links:

Seeking Truth from Facts

My personal Substack

By Alf B.

Audio-only version and transcript:

https://www.manifold1.com/episodes/iran-war-is-the-first-missile-war-crossover-with-seeking-truth-from-facts-podcast-110

Chapter Markers:

• (00:00) - Missile War Reality Check

• (01:49) - How the War Started

• (04:46) - Iran Outperforms Expectations

• (06:22) - Why Missile Defense Fails

• (14:25) - Ceasefire and Hormuz Brinkmanship

• (19:52) - Nukes and the JCPOA Fallout

• (33:44) - US Politics and Israel Lobby Aftershocks

Richard Ngo is an independent AI researcher and philosopher known for his work on AGI safety and alignment. He recently resigned from OpenAI, where he was a member of the Governance team focused on forecasting the capabilities and risks of advanced AI systems. His debut fiction collection is titled “The Gentle Romance: Stories of AI and Humanity”, published in December 2025. The book features 22 science fiction stories that explore the psychological and sociological impacts of advanced artificial intelligence.

On X: @RichardMCNgo

• (00:00) - Richard Ngo Origins

• (03:48) - DeepMind vs LLMs

• (09:24) - OpenAI Futurist and AGI Risk

• (39:17) - Machine God Tail Risk

• (45:20) - Weird Futures and Normies

• (51:28) - Alignment Research and Academia

• (01:16:25) - Doomers vs Skeptics

• (01:23:57) - Labs Governance Futures

• (01:39:37) - Doom Scenarios Society

Audio-only and transcript:

https://www.manifold1.com/episodes/dreamers-and-doomers-our-ai-future-with-richard-ngo-109

Matt Southey is an editor for an AI safety organization. He wrote his PhD dissertation on the philosophy of Nick Land. Matt's "A Brief History of Accelerationism" can be found here, and he recommends the second chapter of Crypto-Current as a good introduction to Land's usage of Kant. He and Steve discuss Accelerationism, AI, and Steve's recent meeting with Land in Shanghai.

Chapter Markers:

• (00:00) - Introduction

• (02:07) - China Trip: Shenzhen, Beijing, Shanghai

• (03:16) - Tsinghua University: Talent

• (09:02) - Robotics and AI research

• (24:01) - Shanghai and Nick Land

• (35:46) - Land’s Esotericism

• (37:19) - Accelerationism and Orthogonality

• (43:01) - Simulation Metaphysics and Physics

• (54:36) - Dark Enlightenment and Inevitable Complexity

Audio-only version and transcript:

https://www.manifold1.com/episodes/china-acceleration-and-nick-land-with-matt-southey-108

Recorded live in Shenzhen with Taylor Ogan, the founder and CEO of

Snowbull Capital, which invests in Chinese technology companies.

Taylor on X:

https://x.com/TaylorOgan

Previous episodes with Taylor:

• (00:00) - Shenzhen is the Technology Capital of the World, with Taylor Ogan

• (00:53) - Meeting in Shenzhen

• (02:40) - Greater Bay Area Explained

• (06:36) - Shenzhen Boom Stories

• (18:26) - China Tech Reality Check

• (36:49) - China Tech Leapfrogging

• (37:52) - Agentic AI on Phones

• (41:27) - Jobs Wealth and Governance

• (53:03) - Huawei Ownership and US Pushback

Audio-only version and transcript:

https://www.manifold1.com/episodes/shenzhen-is-the-technology-capital-of-the-world-with-taylor-ogan-107

–

This is the second episode of our series based on interviews for the documentary film, Dreamers and Doomers, about the SF Bay Area in the last days before AGI.

Steve interviews Jesse Hoogland, co-founder and executive director of Timaeus, an AI safety research org working on applications of Singular Learning Theory (SLT) for AI safety. SLT establishes a connection between the geometry of the loss landscape and internal structure in models. This connection is used to develop scalable, rigorous tools for evaluating, interpreting, and aligning neural networks. Jesse is one of the leading young minds in the new generation of AI safety researchers.

https://www.jessehoogland.com

• (00:00) - Jesse interview at FAR Labs, Berkeley

• (00:54) - Introduction

• (01:50) - From Physics to AI Safety

• (08:36) - AI Is Dangerous

• (26:08) - Funding, P(Doom), and Futures

• (56:21) - Trauma and Safety Vibes

• (01:00:39) - Asymptotic Guarantees Debate

• (01:03:54) - Mapping the Safety Tribes

• (01:26:09) - Timelines, AI Pause, and Failure Modes

Audio-only version and transcript:

https://www.manifold1.com/episodes/ai-doom-jesse-hoogland-of-timaeus