We have:

Jonathan Parker: <https://x.com/ProfJAParker/status/2094567917048226040>: ‘Steve has the right way to interpret the important research & evidence in AJR & Daron’s related work. It fits a given narrative to historical details, tells a rich plausible story through that lens with lots of evidence. (And David’s work is important & deserves respect)…

Share

And:

Steve Durlauf: <https://x.com/sndurlauf/status/2094545774080307362>: ‘David Albouy @albouy is entirely justified in being aggrieved. It was completely inappropriate for @TheEconomist to publish a response to their (ridiculous) article on Daron Acemoglu that contained a misstatement of the Albouy criticisms, just at it was inappropriate that the article’s author did not contact David to describe the nature and import of his criticisms for the *ideas* that AJR have developed, as opposed to the merits of a particular set of statistical exercises. There are huge questions surrounding the ways that credible empirical claims can be made about questions of the type AJR address.

Cross-country regressions have proven to a very fragile source of information on the mechanisms underlying growth and development. My view of the AJR research program is that the empirical dimension should be understood from the vantage point of inference to the best explanation, aka abduction, as a way to understand their research strategy, which implies historical/qualitative evidence is central. For details, see my paper

“Institutions, Development, and Growth: Where Does Evidence Stand?” Handbook of Economic Development and Institutions, Jean-Marie Baland, François Bourguignon, Jean-Philippe Platteau, and Thierry Verdier, eds., Princeton University Press, 2020…

Give a gift subscription

And:

David Albouy: <https://x.com/albouy/status/2094501978818621440>: ‘I tried to be chill, but Acemoglu’s false statement reminded me too much of the gaslighting I had to deal with to publish my comment. Acemoglu et al don’t respond directly to criticisms, and often make up demonstrably false claims to defend themselves, with no accountability:

If anyone wants to read my comment, it’s available here: <https://static1.squarespace.com/static/62f1eb5aa4471e693f087c96/t/630594445e4b432c6913e58c/1661310023111/AJRrev.pdf>. The data appendix is available on my website if you want to really dive into dumpster fire of their data construction. The real crime was in the cover-up. Eg Acemoglu et al never confronted the embarrassment of my Figure 1:

Instead they issued a reply full of red herrings with the bullying title “Hither Thou Shall Come but no Further”. They could have acted as scientists but opted to confuse.

As a grad student, I emailed and visited Robinson’s office many times to talk about my preliminary investigations. He was never available. When I finally confronted him, he waved me away, saying, “I think you’re looking for a different Robinson”. Unbelievable, but true. I tracked down Johnson at an AEA meeting. In the spirit of scientific collaboration, I offered to sit down and go through the data points with him, he nodded in agreement saying that would be nice.

I never got a response to any emails after that. As for Acemoglu, I tried to make peace. But i still remember in 2007 after my R&R at the AER, I submitted a paper to REStat where Acemoglu was editor, trusting that he’d recuse himself and not handle it. A few months later, I got a rejection letter signed by Acemoglu…

Share DeLong's Grasping Reality Weblog

And:

Jon Steinsson: To David Albouy <https://x.com/JonSteinsson/status/2094559148960833707>: ‘Let me offer a tiny consolation: I teach your critique to 1st year econ PhD students every year <https://jonsteinsson.com/teaching/FundamentalCauses.pdf>:

Share DeLong's Grasping Reality Weblog

Brad DeLong here: I wrote about how the IV is a dumpster fire only last April. My view has long been that (a) David Albouy is 100% correct here, and that (b) if Acemoglu, Johnson, and Robinson were wiser, they would recognize that David Albouy is their best friend in the world, for he provides an explanation for what are otherwise very uncomfortable—clock-striking-13-uncomfortable—IV results.

Briefly: AJR simply have no real first stage. The “1-in-1,000” significance on log-mortality-to-expropriation-risk melts to 1-in-25 once you notice the mortality assignments are smuggling in continent dummies, and to 1-in-3 once you separate barracks deaths from campaign deaths and laborer data. With a first stage that weak, the second-stage test statistic isn’t a t-distribution — it’s near-Cauchy: infinite variance, no standard deviation, and no mean at all. You are drawing noise from fat tails and calling it a coefficient. The Cauchy distribution is the devil — the lesson my sometime-roommate John Bound taught me back when we were both much younger. Weak first stage in, garbage out. Only the file drawer Darwinian-selection process launders that garbage into a “finding”. Weak instruments plus sampling on the instrument that produces a significant IV coefficient is the replication crisis in a clown suit, driving a clown car. And often you really do not want the refreshments in the trunk of that clown car, as AJR would recognize if they were wiser. ​

This is what I wrote back four months ago:

Get 75% off a group subscription

Colonial origins, causal claims, the baggage left behind in the overhead bin via the absence of a structural model, settler-colonist mortality, modern pro-prosperity “institutions”, and structural-empirical truth…

• Acemoglu, Daron, Simon Johnson, & James A. Robinson. 2001. “The Colonial Origins of Comparative Development: An Empirical Investigation”. American Economic Review, 91:5 (Dec.), pp. 1369-1401. <http://www.jstor.org/stable/2677930> <https://economics.mit.edu/sites/default/files/publications/colonial-origins-of-comparative-development.pdf>.

I was thinking I would do an economic history post yesterday and a political economy post today, but life is busy, busy, busy, what with chaos, staring at screens watching people get blown up and so forth. Then I sat across from Jón Steinsson at the faculty lunch, and in the course of the conversation he mentioned that he still taught the graduate students the quarter century-old Acemoglu, Johnson, and Robinson “Colonial Origins…” paper.

Why? Because it is both incredibly strong and incredibly weak, a true rabbit and a true duck, depending on how you look at. Teach it, and students get very valuable experience in having strong reactions and figuring out how to explain them—and also (we hope) practice in listening to people with whom you violently disagree but who think what they think for reasons.

AJR’s “Colonial Origins” is surely among the most influential empirical paper in historical development economics of the last quarter-century. Its argument is elegant:

• European settler colonialists who found that they could survive and thrive in a colony built inclusive pro-growth developmental institutions.

• European settlers who found that they couldn’t, and that they needed to grab what they could and return home before they succumbed to yellow fever or such, did not.

• Places with high European settler mortality saw the development of “extractive institutions”, hostile to widely distributed prosperity both in the past and in the present.

• Those early institutions persisted to this day.

• Colonial-era European settler mortality gives us a lever—a valid instrument—to identify the causal effect of institutions on prosperity.

The result that Acemoglu, Johnson, & Robinson claim?: that differences in institutions explain about three-quarters of the income per capita differences across former colonies. Geography, latitude, disease burden—once you control for institutions, they do not matter.

But this is a paper that turned me into a Heckmanite—into a believer in the idea that only those who have fully specified structural models, at least in their mind’s eye if not out there on the table, have a valid warrant to do anything statistical that they claim is in any sense “causal”. Structural models keep you from abandoning burdensome baggage in the overhead compartment when you exit the plane—and AJR’s procedures and paper has a lot of such, which the non-structural IV-framing hides from view. Briefly: either prosperity has a strong negative structural effect on governance quality—which contradicts everything we know from political science and history—or settler mortality in the 17th century is a better measure of what matters in modern institutions than present-day institutional analysis by people who know about and can see them is.

Neither claim is comfortable.

Yet both are concealed by the IV design.

In any event, here is a link to where you can pull it down as an interactive python document: <https://github.com/braddelong/working_20251227/blob/main/2026-04-07-DELIVERED-EDITED-econ-196-week-9-reversals-of-fortune.ipynb>.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-david-albouy-jonathan-parker-stephen-durlauf-jon-steinsson-on-acemoglu-johnson-robinsons-colonial-origins
##enlarging-the-bounds-of-human-empire
##public-reason
##crosspost
##the-scatter-and-the-hypotheses-of-acemoglu-johnson-robinsons-colonial-origins-paper-are-very-interesting-and-important-but-the-iv-statistics-are-as-david-albouy-says-a-dumpster-fire
#david-albouy-jonathan-parker-stephen-durlauf-jon-steinsson-on-acemoglu-johnson-robinsons-colonial-origins
#david-albouy
#jonathan-parker
#stephen-durlauf
#jon-steinsson
#acemoglu-johnson-robinsons-colonial-origins
#acemoglu-johnson-robinson
#colonial-origins
#weak-instruments
#extractive-institutions
#inclusive-institutions
#cauchy-distribution
#heckmanite
#cross-country-regressions
#clock-strikes-thirteen
#economic-history
#causal-anti-inference

Emily Bender’s “Climbing Towards NLU” with Alexander Koller proposes two people on separate deserted islands, communicating via an underwater telegraph cable, but one of them has been replaced by a hyper-intelligent octopus who has detected the statistical patterns in the exchanges. But because the octopus has only ever seen the sequence of signals and never the things they refer to, the moment one islander faces a genuine novelty—say, a bear attack, and urgently asks how to build a weapon from the materials at hand—the octopus cannot give real help and can only produce fluent, plausible-sounding replies without grounding in the actual world. LLMs are Stochastic Parrots, as the team she was on wrote in their “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”

Share

Why has her work struck such a nerve? Well, “Stochastic Parrots” is a very good paper. It has the catchy meme angle that tends toward virality. And Google tried to kill it, or at least to say that these thoughts are not thoughts that anyone working at Google has or is allowed to have.

As I understand it, Google has an internal pre-publication review process called PubApprove—a check that a paper doesn’t leak proprietary information or trade secrets, not a peer review. ​⁠The paper cleared PubApprove. Vice President Megan Kacholia then ordered Temit Gebru to either pull the paper submission from the conference or remove all Google authors’ names from it. Gebru asked for transparency and for meetings where she could make her case, and said if those conditions couldn’t be met she’d negotiate a departure date after her vacation. Google treated that as a resignation, cut off her email while she was on vacation, and she was out. Margaret Mitchell, her co-lead, was fired a couple of months later.

Google tells false stories about why they did what they did. Jeff Dean, head of Google AI, claimed the paper “didn’t meet our bar for publication” via PubApprove because it “ignored too much relevant research” and was submitted to PubApprove with only a day’s notice, too late for proper review. The “ignored relevant research” was bullshit: PubApprove is a sensitive-information check. And nearly half of all papers going through PubApprove were submitted with a day or less notice. ​⁠

The references:

• Bender, Emily M., Timnit Gebru, Angelina Mcmillan-Major, & Shmargaret Shmitchell. 2021. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21), 610–623. March. <https://dl.acm.org/doi/10.1145/3442188.3445922>.

• Bender, Emily M., & Alexander Koller. 2020. “Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data”. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 5185–5198. July. <https://aclanthology.org/2020.acl-main.463/>.

Give a gift subscription

Why is it Emily Bender rather than Timnit Gebru or Margaret Mitchell who shows up in my feed these days? After Google lit its reputation as a place where people could do serious work on large chunks of the study of “AI” on fire, Bender took the public-explainer path in a way that Gebru and Mitchell—now heads of DAIR (the Distributed AI Research Institute) and aresearcher and ethics lead at Hugging Face—did not.

<https://www.youtube.com/watch?v=ZUIQYQXsEvw>

Get 75% off a group subscription

Brad DeLong here: With respect to:

‘A faces an emergency. She is suddenly pursued by an angry bear. She grabs a couple of sticks and frantically asks B to come up with a way to construct a weapon to defend herself. Of course, O has no idea what A “means”. Solving a task like this requires the ability to map accurately between words and real-world entities (as well as reasoning and creative thinking). It is at this point that O would fail the Turing test, if A hadn’t been eaten by the bear before noticing the deception. Having only form available as training data, O did not learn meaning…. Because agents who produce English sentences usually have communicative intents… [A] assumes that O does too, and thus she builds the conventional meaning English associates with O’s utterances. Because she assumes that O is B, she uses that conventional meaning together with her other guesses about B’s state of mind and goals to attribute communicative intent. It is not that O’s utterances make sense, but rather, that A can make sense of them…

Share DeLong's Grasping Reality Weblog

Well, yes. But.

Read more

Torsten Slok makes a large, directional prediction: the risk is rising that interest rates will go down a lot over the next six months:

Share

Share DeLong's Grasping Reality Weblog

The commentary:

Torsten Slok: <https://www.apollo.com/wealth/insights-news/insights/daily-spark#page-1>: ‘The risks are rising that long rates six months from now could be a lot lower than where they are today…. Inflation [fears] and [deficit] fiscal problems… could end up being dominated in early 2027 by what happens to AI…. If AI succeeds and tech companies generate trillions in revenue, AI will be massively deflationary and push rates lower. If AI does not work out, the bubble bursts and the Nasdaq is down 50% as investors rotate out of equities into Treasuries and long rates fall dramatically.

Over the next six months, the market will make up its mind about which AI scenario is playing out…. Financial markets are driven by narratives. The narrative… today is… inflation and fiscal problems. But the narrative… is going to be… the success or failure of AI. And in both scenarios, long rates are going to be lower…

Give a gift subscription

What do I think of this?”

• The “AI success” path is, if I am reading this correctly, the argument of last week’s: Caballero, Ricardo. 2026. “Speculative Growth and the AI ‘Bubble’”. MIT. August 23.<https://economics.mit.edu/sites/default/files/2026-08/speculative_growth_AI_public.pdf>. The AI-buildout switches to being financed by profits as they role into the labs and the hyperscalers, and the rise in productivity outruns consumption and increases savings.

• The “AI failure” path is the standard bubble-collapse=and-aggregate-demand-driven-recession scenario.

What is very noteworthy is the claim that all of this is likely to come to a head in the next six months, and that the narratives surrounding it are overwhelming what Torsten Slok sees as the “inflation [risk] and [deficit] fiscal problems” that are the narratives currently dominating the financial market.

There is one enormous puzzle here: Why does Torsten Slok think that the AI question will be resolved in the next six months? I do not see the reason for thinking that at all.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠heading-for-a-large-rapid-fall-in-interest-rates-chart-of-the-day
##macro-outlook
##mamlms
#ai-bubble⁠
##‎⁠chart-of-the-day
##‎⁠torsten-slok-says-both-an-ai-boom-and-an-ai-bust-push-long-interest-rates-down-substantially-and-the-market-is-very-likely-to-decide-one-way-or-another-in-the-next-six-months⁠
#‎⁠heading-for-a-large-rapid-fall-in-interest-rates⁠
#torsten-slok
#market-narratives
#multiple-steady-states
#ricardo-caballero

It must have been back in 1999 or so, when we in the Berkeley economics department were looking to fill a field-hole in Comparative Economic Systems/Economics of Transition, when I asked my friend Andrei Shleifer whom we should try to hire. Andrei’s answer, as I remember it: “Gerard Roland. You at Berkeley, especially, should try to hire Gerard Roland. For industriousness, wisdom, knowledge, and collegiality, he is first-class. Gerard Roland. Definitely Gerard Roland.”

Share

Now Gerard has a SubStack. And he is telling his stories. Today: The Berkeley Years. Part XXII: Being Department Chair (2008–2011):

Give a gift subscription

Gerard’s Substack

My personal Substack

By Gerard Roland

CROSSPOST: GERARD ROLAND: The Berkeley years. Part XXII

<https://gerardroland.substack.com/p/the-berkeley-years-part-xxii> <https://gerardroland.substack.com/>

Gerard’s Substack

The Berkeley years. Part XXII.

I was asked by my colleagues to be department chair starting from July 1 2008 for the usual period of three years. I had been graduate chair for a few years before that and my colleagues had appreciated my work and initiatives, especially when it came to recruiting graduate students in competition with other great departments (Harvard, MIT, Stanford, Pr…

Read more

4 days ago · Gerard Roland

Being department chair (2008-2011)

Gerard Roland

Aug 28, 2026

I was asked by my colleagues to be department chair starting from July 1 2008 for the usual period of three years. I had been graduate chair for a few years before that and my colleagues had appreciated my work and initiatives, especially when it came to recruiting graduate students in competition with other great departments (Harvard, MIT, Stanford, Princeton, Chicago and others). People often have no idea how much time and effort people in top departments spend to recruit the best graduate students as well as the best junior and senior professors. We spend much more time than other departments on this, because success in these areas is fundamental to stay at the top.

My predecessor as Chair was Ben Hermalin, the well-known micro-economist. He had had a hard time, because there was the perception in the profession that UC Berkeley was financially less well off than other top universities since the many budget cuts to the University of California system, starting from the 1990s.^[1] There was thus the rumor that the Berkeley economics department was “ripe for poaching”. During Ben’s mandate, there were at some point 13 outside offers for professors in the economics department. As is the case most of the time, Berkeley’s top administrators respond to those outside offers and manage to keep the concerned faculty on campus.

One problem is that for Berkeley faculty, getting outside offers had usually been the main way to increase one’s salary. This is somewhat of a double-edged sword. Too small salary raises, but high ones in response to outside offers, help the university to save money. On the other hand, this tends to reduce (but not always) the feeling of loyalty among the faculty towards their university. Ben managed to convince the campus authorities to put together a special one-time program for the economics faculty (called Targeted Decoupling Initiative or TDI) to respond positively to all the outside offers. It ended up being very successful. In my recollection, only Chang Tai-Hsieh decided to leave the department for Chicago’s Booth School of Business.

When accepting to be Department Chair from 2008 to 2011, I stated that I would spend a lot of time fund-raising for the department. This is normally something that department chairs do not do, but I thought that it was really necessary given that the department’s endowment had hardly changed in many years. I also wanted to do the job at 100% of my capacity to do it as well as possible. This implied working most evenings during the week, but I was prepared to do that, and my family was accepting this. I thought that working 100% of my time would help boost the morale of the department. Even though this left literally no time for research, I could count on the understanding of my Berkeley coauthors, with whom I would discuss next steps in our projects but leave them doing most of the footwork. Also, since coming to Berkeley, I had spent most summers in Europe with my family, but I thought I would not be able to manage well the department staff from afar, so I strongly cut the length of my summer trips to Europe during my three years as chair.

One month into the job, the stock market crashed. This was August 2008 and the beginning of what came to be known as The Great Recession. The value of the various endowments the department was managing, which I wanted to increase, went literally through the floor. Not a good way to start my mandate. It was necessary to respond without wasting time. One of my first initiatives in that context was to do something to protect the endowment income for David Card’s Center for Labor economics (CLE). I made a proposal to ask the campus to complement the Center’s income from campus money so that it would stay constant during the crisis. Those funds would then be gradually reimbursed as the endowment would rise again in the future. David Card was then the intellectual leader in the Department after Gérard Debreu in the 1980s and George Akerlof in the 1990s. He was one of the main figures in the economics profession to introduce the so-called “credibility revolution” in empirical research aiming at establishing causal relationships between variables (see the post on the credibility revolution in development economics in <https://gerardroland.substack.com/p/the-berkeley-years-part-xxa>) and was widely expected to receive the Nobel prize, which indeed happened in 2021. If the CLE’s income would be reduced as a consequence of the stock market crash, Card would understandably be upset. On the other hand, if I managed to preemptively protect the income of his research center, it would be good and probably unexpected good news for him. I managed to convince the provost, George Breslauer to support this. Breslauer was a political science expert on the Soviet Union with whom I had a very good relationship, as he played a key role in hiring me when he was Dean in 2000-2001, but as with all things administrative, it took a while for my initiative to get approved. The Dean of Social Sciences, Jon Gjerde, a reputed historian passed away unexpectedly after I had been in office for only 4 months. Jan de Vries, also a famous historian, took his position until the end of the academic year when a new Dean would be appointed. This was Carla Hesse, also a reputed scholar from the history Department.

Despite the lack of money due to the crisis, I took several other initiatives that played a role in keeping the Department’s morale. One was related to the absence of available slots for new professors. When a department must go a year without hiring, this is very bad for the morale. We decided to interview as usual at the American Economic Association Annual Congress early January 2009. We selected at least one strong candidate but had no slot. So, I proposed to the Department to make an offer with commitment to hire once we got the slot, using department money to make the difference so that there would be no negative consequences for the new hire. This worked and we got a slot from the campus after only a few months.

One reason why I think I was often able to find innovative solutions as Department Chair despite the lack of money was that I was comparing the situation at Berkeley with the one I had known at ULB in Brussels. Compared to ULB, Berkeley had so much more resources, both financial and in terms of very competent staff that it was a pleasure for me to work with those resources. Many other colleagues compared Berkeley’s situation with that of richer private universities like Harvard or Stanford, which was less favorable to Berkeley. I found throughout my career that quality of research does not necessarily increase linearly with funding. Two points can easily be made, but there are others. First, it is very difficult for university administrators to turn down financial requests for bad projects when the university is awash with money. Bad projects later weigh negatively on the university’s performance. Second, researchers who receive very large amounts of money need time to manage those resources, which takes time away from creative research and can be counterproductive.

Of course, at UC Berkeley more funds were badly needed after the crisis of 2008. The Dean’s office helped me a lot with the fund-raising and I spent a lot of time talking to potential donors who were nearly always very interesting people with very rich life experiences. On the whole, during my tenure as Chair, the Department and the various research centers associated to the Department received roughly 9 million dollars in endowment money. Some of those funds came very easily while other fund-raising efforts did not yield results, despite much time and effort. I am also very glad that after my mandate, my colleagues continued to engage in fund-raising. Despite all that, a strong campus support has always been critical for the Department. During my tenure, I spent more money to recruit and retain faculty than was possible only via the Department’s budget. Despite the economists generally not being very popular among other departments, both in social sciences and STEM departments, our very good reputation^[2] as well as the very large number of economics undergraduate students always convinced campus authorities to help the department financially.

One of the great moments during my mandate as Chair was when Oliver Williamson got the Nobel prize. The Berkeley Public Relations department has always been very professional. I was asked a week before who might get the economics Nobel prize among my colleagues. For some reason, I thought that Oliver Williamson had a high chance (previous Berkeley economists who got the Nobel prize included Gérard Debreu, John Harsanyi, Dan Mc Fadden and George Akerlof). The prize was announced on Monday October 12 2009. On Friday October 9, I had preemptively rented a room for Monday 4pm at the Women’s Faculty club. On the day of the announcement, I got a call at 4am from Berkeley’s PR department that Williamson had gotten it together with Elinor Oström (the first woman and only political scientist to have received the economics Nobel prize). I immediately proudly announced the news to the department and announced the 4pm party. Some colleagues later asked me how I had been able to rent a room at 4 in the morning. The day was fabulous and the whole university was proud to have one more Nobel prize (By then, the university had had 21 Nobel prizes, including 5 for economics). This remains a great memory for me.

A more bittersweet moment for me was towards the end of my mandate. I had worked quite well with Dean Carla Hesse until then, but she told me that Heddy would from then on only be paid half-time instead of full-time but that those half-time funds would be secure. This was a big blow to both Heddy and me. I tried to explain to her that it would have been better to return to Heddy’s previous situation paid full-time from less secure funds, but she would not listen and even accused me of nepotism. I thought of resigning but only had little time left as Chair. Also, I had to have major colon surgery around that time and my daughter Elsa had been battling eye cancer (more on that next week). In any case, I was due for a sabbatical and decided to look for another job when I came back. Something in my loyalty to Berkeley had been broken.

(To be completed)

Me, around the time I was department Chair, shortly after an important surgery.

^[1] After Clark Kerr’s reforms as president of the UC system, UC budgets were very large and the income from the California state was equivalent to that from the largest private university endowments, but that did not last, unfortunately.

^[2] In a university like UC Berkeley, nearly all departments are in the top 3 or top 5 in the world, so the economics department is not more exceptional than other departments.

<https://gerardroland.substack.com/p/the-berkeley-years-part-xxii> <https://gerardroland.substack.com/>

Gerard’s Substack

The Berkeley years. Part XXII.

I was asked by my colleagues to be department chair starting from July 1 2008 for the usual period of three years. I had been graduate chair for a few years before that and my colleagues had appreciated my work and initiatives, especially when it came to recruiting graduate students in competition with other great departments (Harvard, MIT, Stanford, Pr…

Read more

4 days ago · Gerard Roland

Gerard’s Substack

My personal Substack

By Gerard Roland

Brad DeLong here: That is what being a department chair at UC Berkeley is like.

The most important thing I note from Gérard’s account is this:

• As chair, he worked like a dog doing what was properly Dean Carla Hesse’s job, fundraising to try to boost Berkeley’s endowment so that we would not be at such a financial disadvantage in resources vis-à-vis our peer institutions.

• He was remarkably and incredibly successful at this.

• Dean Carla Hesse then responded to this success of his by financially injuring his family.

• Carla Hesse’s claim that “Heddy would from then on only be paid half-time instead of full-time but that those half-time funds would be secure” was a more-or-less even trade was complete bullshit: at Berkeley, funding is never secure. Secure funding is not secure.

• The Berkeley administration is in enormous debt to Gérard that it has taken few steps indeed to honor.

Share DeLong's Grasping Reality Weblog

When I was chair of Berkeley’s PEIS major, I evolved my own theory of Berkeley’s senior administrators. My theory was this:

• They spent the first three days of each month trying to think rationally and seriously about the future of the university and about resource allocation.

• On day four, they would have to respond to a faculty retention case in response to an outside offer launched by another university with a much larger endowment.

• They would then spend the rest of the month turning every piece of Berkeley they could put their hands on upside down and shaking it, in the hopes that money that could be used to respond to the retention case would somehow fall out.

• Pieces of Berkeley that were functioning well (as PEIS then was and now is) were seen as easy targets for this effort. They were doing well, right? Surely they could afford to limp along with somewhat fewer resources? Couldn’t they?

• Pieces of Berkeley that were functioning badly were immune: we have enough problems and cannot risk creating more! Perhaps we should ease their resource constraint?

Get 75% off a group subscription

On the principal that The Purpose of a System Is What It Does, TPOASIWID, this is not a good way to run a railroad, or a university. And yet somehow we continue to do very very well.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-gerard-roland-the-berkeley-years-part-xxii
##public-reason
##crosspost
##gerards-subheadline-being-department-chair-20082011-fundraising-through-the-great-recession-financing-david-cards-labor-center-throwing-the-oliver-williamson-nobel-party
#gerard-roland-the-berkeley-years-part-xxii
#gerard-roland
#the-berkeley-years-part-xxii
#gerard-roland
#academia
#uc-berkeley
#berkeley-economics
#academic-administration
#higher-education
#tpoasiwid

A nice graph. I look at it, and say, why not just turn the tax-policy clock back to the Clinton-Gore 2000?

Share

Adam Tooze <https://adamtooze.substack.com/p/top-links-1207-us-deficits-counting> picks this up from the Financial Times:

Share DeLong's Grasping Reality Weblog

The revenue drop from 2000 to 2004 is 4/5 structural, not cyclical <https://en.wikipedia.org/wiki/Bush_tax_cuts>. CBO scored the 2001 (EGTRRA) and 2003 (JGTRRA) cuts as adding ~$1.5 trillion to the debt over 2002–2011 excluding interest, and ~$3 trillion over 2010–2019 including interest if fully extended. This is not contested by anyone doing arithmetic. Talk I hear of a “cyclical peak in revenues in 2000” explains a year or two of mean-reversion, not a quarter-century structural gap. Using it to explain the past generation is a category error.​

The upper-income Bush cuts were sold as a recession-recover measure. But they had a constant interest-rate policy fiscal multiplier of only 0.26. They were among the least stimulative and most expensive things the government did <​⁠https://www.epi.org/publication/ib338-fiscal-cliff-obstacle-course/>. And no American tax cuts ever “paid for themselves” except for tariff reductions in the long run. Paul Krugman’s line still holds: supply-side claims that tax cuts would pay for themselves “never got any traction in professional economic[s]… even among conservatives”. Those who claimed they would were and are unprofessional economists and professional recoveries.

Give a gift subscription

Thus my view: What was wrong with the tax rates and tax system as they stood in 2000?

Not very much.

Want to fix the deficit? Simply have the first item of business on January 21, 2029 be passing a Budget Resolution, to be followed on January 22, 2029 by a vote on the Reconciliation Bill restoring tax rates and taxable income coverage to what they were in 2000. Base-restoration matters as much as the headline rate. And also uncap FICA: impose the Social Security and Medicare taxes on all earned income. The 2026 wage cap is $184,500, so all earnings above it currently escape the 12.4% tax entirely. Removing the cap would raise on the order of $3 trillion over a decade <https://taxfoundation.org/blog/save-social-security-payroll-tax-cap-proposal/>.

Leave a comment

If you won’t get on board for that as your initial bargaining position, I have one question: what is wrong with you?

After all, year 2000 was the last time I had confidence that the US had a competent government and that US society was by-and-large working and improving. Shouldn’t we do whatever we can to restore things to how they were then?

Anyone else willing to join me on this turn-back-the-clock exercise?

Get 75% off a group subscription

The graph appears and there is commentary on it in an article by Chris Giles.

I thought about about dropping a link, but decided not to. Giles is behaving badly here. He asks the question: “Why [has] the US fiscal position has deteriorated so much this century?” And then he works much too hard to avoid giving the correct answer: Republicans.

Instead, he talks about “the cyclical peak in revenues” in 2000, the Bush tax cuts “which were made permanent on a mostly bipartisan basis during the Obama administration”, “spending on services for an ageing population—social security, Medicare and veterans’ programs”, in addition to the Trump tax cuts.

Not, mind you, that Giles is a big fan of Trump and Bessent.

And I have to admit there is motion: the Giles of today admits that “the US has significantly cut tax rates for the richest this century, so some reversal of those would be likely and justified from a left-leaning president and Congress”. This is not something that we would have gotten from the Piketty-bashing Giles I remember from 2014 <https://www.forbes.com/sites/scottwinship/2014/05/27/laffaire-piketty/>.

Refer a friend

But if you are not brave enough to tell the truth, to give the answer Republicans to the question why has the US fiscal position has deteriorated so much this century?, what use are you? Why would reading you be worth anyone’s time?

Herodotos 1:136 tells us that ἀληθίζεσθαι, truth-telling, was one of the only three things that young Persian aristocrats were taught. Herodotos matched ἀληθίζεσθαι with ἱππεύειν, to ride (to do the horse-thing), and τοξεύειν καὶ, to shoot (to do the bow-thing). Those are martial skills that need a huge amount of practice to attain and retain competence. Herodotos is right with the implication of coupling truth-telling with two martial skills. Honesty is more a discipline, something you train for, not a temperament. Analytic courage—a practiced refusal to hedge when the data points somewhere your political and ideological and careerist commitments point in a direction you would rather they not—requires practice as well.

It’s something to strive for.

There: that is off my chest.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##the-us-federal-fiscal-flow-deficit-problem-chart-of-the-day
##macro-outlook
##chart-of-the-day
##republican-tax-cuts-for-the-rich-created-the-problem-please-name-and-shame-restoring-the-clintongore-tax-rates-and-tax-base-is-the-solution-pass-it-on
#the-us-federal-fiscal-flow-deficit-problem
#republican-tax-cuts-for-the-rich
#federal-budget
#clinton-gore-tax-base-and-rates-restoration
#uncap-fica
#name-and-shame
#ἀληθίζεσθαι

And while Carney’s retaliation probably feels good, it costs Canadians, and playing the retaliation game winds up giving the advantage to whoever cares least about his own people, who is Trump. A sensible Canadian strategy would aim at other forms of linkage and leverage. The binding constraint on Trump is attention, not welfare. His principal purpose in any pronouncement is to gain eyeballs, so the only leverage that bites is leverage that threatens the spectacle — American farmers, retailers, and TikTokkers visibly hurt on camera. The correct Canadian move is to show that Trump has taken his own people hostage—that the logic is that of the sheriff in “Blazing Saddles”:

Share

Share DeLong's Grasping Reality Weblog

Platypus Economics

Economics. Explained. Clearly.

By Justin Wolfers

CROSSPOST: JUSTIN WOLFERS: Good News. There’s Already a U.S.-Canada Trade Deal

<https://newsletter.platypuseconomics.com/p/good-news-theres-already-a-us-canada> <https://newsletter.platypuseconomics.com/>

Platypus Economics

Good News. There’s Already a U.S.-Canada Trade Deal.

Earlier this week I joined Fergus Macphee on The Trump Report to talk about the U.S.-Canada trade war. We talked about how Canadians are getting caught in the middle of a negotiation between President Trump… and President Trump…

Read more

5 days ago · 217 likes · 9 comments · Justin Wolfers

You’ll never guess who signed it.

Justin Wolfers

Aug 27, 2026

Earlier this week I joined Fergus Macphee on The Trump Report to talk about the U.S.-Canada trade war. We talked about how Canadians are getting caught in the middle of a negotiation between President Trump… and President Trump.

A little history: The United States has had a free trade agreement with Canada since 1989. That U.S.-Canada Free Trade Agreement gave way to the North American Free Trade Agreement (NAFTA) when Mexico joined the club in 1994.

Then the first President Trump came to office calling NAFTA a disaster. He renegotiated it and replaced it with the United States–Mexico–Canada Agreement, or USMCA to its friends. The new deal was signed in 2018 and took effect in 2020. It’s basically NAFTA with a new title.

President Trump called that agreement:

“the largest, fairest, most balanced, and modern trade agreement ever achieved. There’s never been anything like it.”

That agreement passed the U.S. Congress, the Canadian Parliament, and the Mexican Senate, and was signed by the leaders of all three countries. It’s the law of the land in all three countries.

But here we are again. And my head hurts.

Talks collapsed and the trade war started

The second President Trump thought the deal that the first President Trump negotiated was terrible. He was unsparing in his criticism, saying:

“I would rather not have the agreement… We do better as a country if we don’t have an agreement.”

So earlier this year Trump demanded a new deal, and threatened large tariffs if he didn’t get his way. Talks began, then collapsed, and President Trump imposed tariffs of 50 percent (that’s high!) on a weird-looking list of Canadian exports, including hockey sticks.

There are two ways to see these tariffs. (Warning: Two-handed economist in the house.) On the one hand, these tariffs apply only to a relatively narrow slice of goods — around $20 billion worth of Canadian exports. That’s only about 4 percent of Canada’s exports, so not really that big of a deal. On the other hand, these tariffs apply to goods that should be tariff-free under the existing trade agreement. Previously these goods had been exempt from an array of Trump’s tariffs. This new move suggests a willingness to significantly broaden the trade war.

More here:

Why Trump’s New Tariff Is Bigger Than Canada

Justin Wolfers

Jul 21

It’s retaliation all the way down

We recorded this podcast before retaliation gave way to retaliation for retaliation, then retaliation for retaliating against retaliation. So, a quick update: We’re in a retaliation cycle.

We’ve seen this movie before. In 2025, Trump got the United States into a tariff escalation cycle with China, and tariffs rose and were countered and counters were countered, and after a bit the United States was imposing a 145 percent tariff, while China was at 125 percent. Both numbers are untenable — high enough you might as well call it an embargo — and a month later, both sides agreed to mostly back down.

We may see this play out again with Canada — escalate until economic pain generates political pain, strike a “deal,” declare victory, and hopefully get back to where we began.

Reminder: We started with a free trade agreement with Canada negotiated by Trump.

Share

The real costs come in the longer run

In our economics textbooks, the case against tariffs is usually that they impose immediate costs, usually on domestic consumers and companies that use imported inputs.

But Trump has found a way to make tariffs even more damaging to Americans: by making them unpredictable. He has oscillated — at seemingly random moments — between treating Canada as a friend and an enemy. He never even pauses at frenemy.

Eventually the United States may discover that it already has a trade deal with Canada. Maybe we’ll rebrand it (again!) and declare victory. But the history of ripping up past deals really matters. It is now understood that whatever is written down is only binding until the president decides it isn’t.

Canadian Prime Minister Mark Carney has noticed, saying that “America has changed,” and its “signature was written in pencil.” I noticed too, and wrote about it recently:

Would You Marry a Man With Seven Ex-Wives?

Justin Wolfers

Aug 21

The problem is that this sort of on-again off-again relationship prevents either side from making long-term investments in their shared future. Even if we get back to trading with Canada, we’ll get less investment, less integration, and fewer of the gains that trade usually brings.

Should Canada retaliate?

We know what happened next: When Trump imposed tariffs, Canada retaliated.

I don’t think this is in Canada’s best interest.

The intuitive argument for retaliation says that Trump’s tariffs on Canada hurt Canadians, and so Canada should retaliate with tariffs on the United States that hurt Americans. A credible threat of retaliation might also deter future tariffs.

But that misses half the story.

The other half is that Trump’s tariffs also hurt Americans, and so the same logic says that Carney’s tariffs also hurt Canadians. That’s the argument for not retaliating.

This point was more vividly made by Joan Robinson, who was possibly the greatest economist never to win a Nobel. She wrote that answering foreign tariffs with tariffs of your own makes no sense: “it would be just as sensible to drop rocks into our harbours because other nations have rocky coasts.” (She credited William Beveridge with the metaphor.) The point is that tariffs make it harder and more expensive for ships full of goods to reach your ports. And who wants less efficient ports? Other countries’ inefficient ports are no reason to ruin your own.

This reality also makes Canadian threats less credible. The argument I hear most often from my Canadian friends is that they don’t want to be bullied. But if tariffs require hurting your own people as a necessary side effect of inflicting pain on foreigners, then a war of economic attrition rewards the leader most indifferent to the pain they inflict on their constituents.

I’m not sure that’s Mark Carney’s strong suit. And Canadians shouldn’t want it to be.

Can we (please) do this the easy way?

A good leader’s job is to make it easy for the other side to say yes.

Trump is doing the opposite. His childish taunts — that Canada is the 51st state, that its prime minister is really “Governor Carney,” that “Canada is nasty,” that its leaders are “clowns” who should “fall in line,” that Lake Ontario should become “Lake America,” and on and on — have stirred such anger among Canadians that Carney now has less room to give Trump what he wants.

Point is, Trump’s taunts unleashed a political force that all but forced Carney to retaliate. He has made it harder for Canada’s leaders to make concessions to the United States.

This is how you turn an easy deal into a hard one.

And we’re all paying for it.

<https://newsletter.platypuseconomics.com/p/good-news-theres-already-a-us-canada> <https://newsletter.platypuseconomics.com/>

Platypus Economics

Good News. There’s Already a U.S.-Canada Trade Deal.

Earlier this week I joined Fergus Macphee on The Trump Report to talk about the U.S.-Canada trade war. We talked about how Canadians are getting caught in the middle of a negotiation between President Trump… and President Trump…

Read more

5 days ago · 216 likes · 9 comments · Justin Wolfers

Platypus Economics

Economics. Explained. Clearly.

By Justin Wolfers

Brad DeLong here: The tl;dr version of Justin is this, I think:

The binding legal agreement already exists; the conflict is entirely about one leader repudiating his own prior work, and the deepest damage isn’t the immediate tariff cost but the destruction of the credibility that makes any future deal—or long-term investment—worth making. Trump does not keep his own deals, therefore you cannot make deals with Trump. All you can do is set up situations in which Trump’s actions adverse to your well-being are especially panful to him.

• → Trump renegotiated NAFTA into USMCA (2018/2020).

• → Trump called it the best deal ever.

• → he now rejects it, demands a new deal, and imposes 50% tariffs on ~$20B of Canadian goods that should be tariff-exempt.

• → Canada retaliates.

• → An escalating retaliation cycle ensues (echoing the 2025 U.S.-China spiral to 145%/125%).

• → But because tariffs hurt the imposing country’s own consumers, retaliation is self-harm, so attrition “rewards the leader most indifferent to the pain they inflict on their constituents”

• → Meanwhile Trump’s personal taunts inflame Canadian public opinion, shrinking Carney’s political room to concede

• → An easy deal becomes a hard one, and unpredictability itself becomes the lasting cost by signaling that U.S. commitments are “written in pencil.”

• → It should resolve anticlimactically: Trump gains nothing from this, provided there are enough people on TV and TikTok who have been hurt by Trump’s tariffs. But it might not. It is not quite TACO: Trump does not quite always chicken out.

Give a gift subscription

It is clear that tariff retaliation is a mistake for Canada: it is shooting a shotgun downward at your feet and at your counterparty’s feet. But mostly at your feet. The key move for Canada is to find something unrelated to tariffs that Trump wants, and hold it hostage. That most likely means finding a way to mobilize those on the American side of the border who lose big because they no longer have cheap Canadian goods.

At a deeper level, the problem is this: the Republican majorities in Congress abdicated and the courts play CalvinBall. Tariff power belongs to Congress, but the Roberts majority applies major-questions and non-delegation doctrine at its whim — constraining Democrats reliably and Republicans only when convenient. Moreover, legal victory two years from now doesn’t unwind the damage done in the interim. “Trump will lose in court” is beside the point. The tariffs get collected while litigation crawls; refunds are partial, burdensome, and uncertain — so the escrow-account friction itself does the work of discouraging trade regardless of the eventual ruling. The only durable strategy against a serial defector is to structure the game so that his defections hurt him automatically — which is a grim thing to have to say about the United States of America.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-justin-wolfers-good-news-theres-already-a-us-canada-trade-deal
##neofascism
##crosspost
##justins-subheadline-youll-never-guess-who-signed-it-and-now-trump-wants-to-burn-it-and-reinforce-everyones-knowledge-that-his-word-is-simply-not-good-for-anything
#justin-wolfers-good-news-theres-already-a-us-canada-trade-deal
#justin-wolfers
#good-news-theres-already-a-us-canada-trade-deal
#chao-monkey
#us-canada-trade
#trump-tariffs
#usmca
#retaliation-cycle

Perhaps he real news isn’t the revenue guide — it’s the half-trillion dollars of third-party capital that NVIDIA is trying to start herding onto its customers’ liability sheets. In economiss’ economic-welfare terms, good natural-language interfaces are already a huge boon; in measured GDP terms, AI i not and may never be there. Don’t let NVIDIA’s income statement stand in for social value in either direction—and don’t let economists’ willingness-to-pay-based economic-welfare measures stand in for human well-being in an environment in which attention-hacking is rife.

We have:

Share

With commentary:

Lynn Thomason: Nvidia’s $100 Billion Shows Why AI Is Still the Trade <https://www.bloomberg.com/news/newsletters/2026-08-27/nvidia-s-100-billion-shows-why-ai-is-still-the-trade>: ‘Nvidia delivers. The AI lynchpin blew away expectations, predicting 70% sales growth for 2028, when analysts had anticipated 45%. The stock jumps 7%…. Nvidia’s stellar results are reinforcing investor confidence in the AI revolution, with CEO Jensen Huang saying the “golden age” of labs and startups is here…. Sales surged. It sees revenue in the current period at $108 billion. Compared with a year ago, that’s close to double…. There’s no let up in demand. Nvidia said it would be growing even faster if it had access to more supplies…. Hyperscalers spend big. The data center division beat estimates, with firms like Alphabet and Amazon accounting for much of those sales…. Margins will narrow. Nvidia warned that margins would shrink slightly in the coming months as it copes with a surge in memory costs. At the current price, Nvidia is poised to add about $370 billion of market value when trading begins. The shares are up 12% this year…

Share DeLong's Grasping Reality Weblog

And just two days ago the word on Bloomberg was that “NVIDIA has lost some luster with investors this year”:

Give a gift subscription

<https://braddelong.substack.com/p/nvidia-has-a-seven-day-stock-price>

This whipsawing of what Paul Krugman calls not Economics but Upanddownonomics is really not terribly helpful.

Nvidia’s profits are a combination of five things:

1. The strength of demand for its GPUs.

2. Its desire and ability to use its monopoly power to generate super-high margins.

3. The limitation of that by its not wishing to trigger its customers to make investments in ending their dependence on it and exiting from its ecosystem.

4. Upstream monopoly power by its own major suppliers.

5. Upstream supply chain difficulties.

Get 75% off a group subscription

What we want to know is: (a) How strong is demand for the GPUs? (b) Hopw much of that demand is really based on “fundamentals”? (c) What are those fundamentals based on—(i) utility to customers, (ii) moat-maintaining investments by platform-monopolists, (iii) belief that you have to learn-by-doing with these technologies on a large scale even though the utility to customers is not yet there because it will be there, (iv) “speculation”, and (v) that these days one grifts gullible investors not by starting a memecoin but by pointing to your racks of GPUs? That requires a careful parsing of (1). And changing your mind on those things as a result of the high-energy excitation mode that is the (1) through (5) factors producing NVIDIA quarterly earnings—that does not strike me as wise.

Bloomberg’s Ian King peers a little bit through the veil of time and ignorance here. Although I do not see NVIDIA’s willingness to go all-in on next year’s growth should ease anyone’s concerns:

Ian King: Nvidia Sees AI-Fueled Demand Boosting Sales 70% Next Year <https://www.bloomberg.com/news/articles/2026-08-26/nvidia-estimate-topping-forecast-fails-to-wow-investors>: ‘Nvidia… said revenue will grow about 70% next fiscal year, easing concerns that AI spending is poised to lose momentum…. Jensen Huang said demand is only accelerating…. “The AI infrastructure build-out is at full steam,” he said. “Vera Rubin, now in full production, was built to power exactly this moment.” Revenue in the current period will be $108 billion, plus or minus 2%…. Gross margin, the percentage of sales remaining after deducting the cost of production, will be roughly 74% in the quarter…. Nvidia [is] cop[ing]… with a surge in memory costs… expects the [gross margin] measure to bottom out in the fiscal fourth quarter… that runs through January — at 71% to 72%….

“We would love more supply,” Kress said in an interview. “It’s really about how much more could you do?” In the second quarter, which ended July 26, sales more than doubled from a year earlier to $96.2 billion…. Nvidia has now delivered sales above Wall Street estimates for 16 quarters in a row….

Nvidia has spent much of the past year lining up investment deals…including both developers of software and infrastructure… that… have, in theory, put Nvidia on the hook for tens of billions of dollars of liabilities…. Kress described the deals as a part of Nvidia’s efforts to increase supply. The biggest portion of its spending commitments took the form of long-term purchase agreements with vendors, she said. “The majority of what we have been helping folks with in terms of commitments is one very important thing called supply”…

Refer a friend

“We would love more supply” is doing enormous rhetorical work here. Kress framing the deals as securing supply, not stimulating demand, is the crucial move — because it reclassifies factor (5), supply-chain difficulty, as bullish rather than as a ceiling. I’d want to see more about that, for whether and to what degree this is, you know, actually true is something I do not but would very much like to know.

I would like to see more about that especially because it really does not fit with the real news, which is the $500B “compute financing platforms” with Apollo, BlackRock, Blackstone, Brookfield, Goldman, and KKR. The chipmaker is organizing half a trillion dollars of third-party capital not to induce its suppliers to expand their on capacity, but to herd in people willing to bear downside risk so that its customers will be willing to buy its chips. That is demand being manufactured by financial engineering, rather than merely being met. Is this a bad thing for NVIDIA to do from the point of view of NVIDIA’s financials? Of course not! It is a very profitable thing for NVIDIA to do! What does it mean for the world economy as a whole? That also is something I do not know and would very much like to.

GM financing a car works because the car has known use-value to a buyer who will pay. The AI analog only holds if the downstream token-buyers eventually cover their compute bills. Right now the poster-child labs mostly don’t—so the vendor-finance defense is a bet on future monetization, not a description of present value. And circularity concentrates systemic risk precisely because the buyer set is thin, and the same dollars show up as payments to NVIDIA by a customer and equity investments by NVIDIA in downstream LLM-service provider. NVIDIA could, in the bad scenarios, lose twice here: once when a customer stops buying, and again when its equity stake in that customer craters. It is a correlated exposure that works equally well going up or down. Dressed as an increasing-speed flywheel.

And, of course, concentration has made this a macro question, not a tech-sector question. Nvidia is now ~8% of the S&P 500 and has supplied more than 10 of the ~84 percentage points of the index’s five-year return. The possibility of a genuine turn in AI capex cannot be contained inside semiconductors — it becomes everyone’s problem, including every teacher’s pension.

But step back: This is Carlota Perez and Bill Janeway territory: a “productive bubble.” Even if most investors lose their shirts, no one will tear up the GPUs, the fiber, or the substations — just as no one tore up the railroad track or the dark fiber. The infrastructure persists; the equity holders are the potential sacrificial capital. Will these datacenters be as useful in the end as the railroad tracks or the dark fiber? That depends on things we cannot now see.

And do beware the Solow-paradox trap in reverse. In economists’ economic-welfare terms — consumer surplus from people willing to pay for natural-language interfaces — AI is already a huge boon. In measured GDP-and-profit terms, it may show up modestly, or even as a productivity decline, because now Amazon needs the warehouse, the truck, and the GPU farm to sell you the same item. Don’t let NVIDIA’s income statement stand in for social value in either direction. And, in addition, do not let economists’ economic-welfare measures stand in for human well-being. Maximizing dopamine hits do not make for a fulfilling, or even a tolerably happy and engaged life, especially when they are the result of algorithmic attention-hacking programmed up by some of the most cynical people who have ever lived.

The honest posture is neither bull nor bear but high-dimensional agnosticism. I count at least a dozen distinct vectors here — grifters, defensive platform monopolists, socially-valuable-but-privately-unprofitable overbuilding, techno-millenarians, attention-extraction business models, and genuine general-purpose-technology diffusion — all superimposed. Anyone claiming this quarter resolves them into a single verdict is selling you a narrative, and the future here is one I frankly cannot see.

I see people confidently claiming that:

• A megawatt of AI-capacity costs $15 million to run, and right now you can sell it for $50 million because it can produce $100 million in cost savings or revenue boosts for high-value customers who know how to harness it.

• Anthropic and OpenAI are each scaling-up from 1.5 gigawatts of capacity at the end of 2025 to 5 gigawatts of capacity at the start of 2027: that is a potential revenue boost to their annualized revenue run rates from $20 billion (at $13M/mW) at the end of 2025 to $150 billion (at $30M/mW) should they find and maintain product-market fit.

And I see other claiming that datacenters will never be profitable once the huge demand from people who think they have to build the institutional muscle-memory capacity to utilize these technologies drops away.

We have to do our reasoning under conditions of genuine Knightian uncertainty. That means: decompose, refuse false precision, resist single-number verdicts. It is more important than ever for us to inoculate sreaders against “Upanddownonomics,” the sentiment-laundering that lets the same fact-situation mean opposite things 48 hours apart. And we need to flag where fragility actually sits: circular finance, thin buyer set, index concentration, off-balance-sheet commitments. Robustness and scenaior planning is only possible if we distinguish between watching the right and wrong dials.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠nvidia-earnings-chart-of-the-day⁠
##‎⁠chart-of-the-day
##macro-outlook
##mamlms
##‎nvidias-96-billion-quarter-and-the-twelve-things-it-doesnt-settle-one-of-which-is-is-nvidia-meeting-demand-or-manufacturing-it
#‎⁠nvidia-earnings⁠
#productive-bubble
#circular-vendor-financing
#ai-capex
#upanddownonomics

Drezner explains why his harshest criticism of the Trump administration targets the “normie” Republicans—of whom he was once one—rather than the obvious buffoons: the people who know better and enable Trump anyway are more contemptible than those who never had principles to betray. In his view, contempt should be allotted in proportion to culpability, not incompetence. The intelligent enablers who understand exactly what they’re empowering — and choose it for power and status — deserve more scorn than manifestly unqualified and honorless clowns from whom nothing could ever have been expected.

There is, somewhere in the branching manifold of possible worlds, a nearby timeline in which Dan Drezner is one of them: a card-carrying, ticket-punching, chin-stroking establishment Republican, writing measured essays about the liberal international order, and referring to the likes of Paul Krugman and me as “intelligent, but overwrought, unbalanced, and, well, shrill.” In that timeline the Drezner tut-tuts at the overexcited. He counsels patience. He is very sound.

We do not live in that timeline. God, I wish we did!

Drezner has documented and continues to document, with the patience of a Jobite naturalist cataloguing a new and horrible phylum, the American foreign policy establishment run by — I quote his field notes — “the dumbest motherfuckers alive.” I am impressed.

Ph’nglui mglw’nafh Daniel Drezner R’lyeh wgah’nagl fhtagn.

Today, Drezner has these key ideas:

• A typology of complicity: Drezner sorts Trump’s orbit into MAGA faithful, “Stormtroopers” (petty government thugs), Trump emulators, and normie Republicans — escalating by how much they should know better.

• Stormtroopers are endemic, not novel: abuse of surveillance power by career officials spans administrations (the Wired CBP story covers 2009–2022), so Trump aggravates but did not invent it.

• Normalization of scandal: Trump has made misogyny and extremism a “dog-bites-man” story, letting candidates with disqualifying scandals (Miller, Herrera) survive where earlier norms would have forced them out.

• The pivotal enablers act on choice, not necessity: figures like Mike Johnson and Bill Cassidy “have a choice” and keep choosing to empower the party’s worst elements — believing they’ve only “rented” their souls temporarily.

• G. Elliott Morris’s finding: subgroups retaining “Republican” identity stay Republican in the generic ballot by 37+ points despite disapproving of Trump, while subgroups without that identity defect to Democrats by 30+ points.

But the most effective mode for this type of discourse is, in its most classic form, that of the Ancient & Hermetic Order of the Shrill located in Arkham, Massachusetts:

<https://archive.nytimes.com/krugman.blogs.nytimes.com/2012/02/29/looking-back-with-shrillness/>

CROSSPOST: DAN DREZNER: Categories of Contempt: A Typology of my Domestic Political Disgust

<https://danieldrezner.substack.com/p/categories-of-contempt> <https://danieldrezner.substack.com/>

A typology of my domestic political disgust.

Daniel W. Drezner

Aug 19, 2026

The hard-working readers here at Drezner’s World have probably noticed that in my commentary about the Trump administration, a disproportionate share of my ire is targeted towards those who are perceived as the more “normie” Republicans: folks like Scott Bessent or Elbridge Colby or Marco Rubio. One can ask: why not aim more at someone who is manifestly unqualified for their job — someone like, say, Pete Hegseth?

I have a short answer and a long answer. to this question.

Drezner’s World is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

The short answer is simple: it is more troubling to see folks with some degree of acumen willingly ben the knee to Trump’s whims just to have some measure of power and status. Pete Hegseth is a malevolent, buffoonish clown; I expect absolutely nothing from him and am therefore unsurprised that he’s acting like a clown who is spectacularly out of his depth. Once that recognition is made, there is really little point to making it again.

The likes of Rubio, on the other hand, are rather different. He is clearly intelligent enough to know what (and who) he is enabling and does it anyway. To the hard-working staff here at Drezner’s World, that is the more contemptuous behavior, and therefore merits more comment and derision.

The longer answer, however, is that Trump’s myriad transgressions require an entirely new typology of contemptible individuals — and it is sometimes difficult to keep track of all the different subcultures of sycophancy.

For example, what should I make of the MAGA faithful, the likes of Tucker Carlson or Steve Bannon or Marjorie Taylor Greene or Nancy Mace, folks who actually believed the MAGA bullshit that Trump and his team have spewed out over the past decade? They’re all genuine bigots, which is certainly contemptible. At the same time, they also genuinely believe in at least some of the foreign policy ideas that Trump embraced over a decade ago. I can’t get exercised by them all too much — mostly because, in the end, I think they’re a marginalized group with minimal political power.

Then there are the Stormtroopers — the folks working in the federal government who have enthusiastically embraced what Trump is selling. You know these are the weak-minded guys who Obi-Wan handles in this scene:¹

There have been a raft of recent stories about the degree to which low-level thugs have taken advantage of their government positions to revel in their ability to bully and surveil citizens. For example, Wired’s Yulia Almazova recently reported on what U.S. Customs and Border Patrol officers have been doing on their computers: “In one case, a CBP officer allegedly used government databases to contact a flight attendant. In another, an officer was accused of pulling information from trusted-traveler applications to ask people out. Other CBP employees were accused of providing border-crossing data to someone involved in a ‘heated divorce.’”

Another example is DHS more generally. In the wake of organized resistance to their Minneapolis incursion, DHS decided to investigate progressive organizations. According to the New York Times’ Alan Feuer and Ernesto Londoño:

Homeland security officials used an array of invasive tactics during the first half of this year to gather information on many groups and individuals who were never accused of crimes, crossing the line that has traditionally stood between investigating criminal activity and political dissent.

In one instance, officials used administrative subpoenas to obtain more than three years of financial records from the Sunrise Movement, an environmental action group, and a labor union, the Communications Workers of America. That time frame went well beyond the civil unrest in Minnesota, which was prompted by the deployment of thousands of immigration agents to the state during the winter.

In another, investigators scrutinized three years’ worth of wire transfers made by the nation’s biggest health care workers union, the Service Employees International Union, in what they referred to as an inquiry into “domestic terrorist financing”…

All of this is pretty vile, and the folks doing it at DHS appear not to care about how vile it is.

And yet, it is also worth noting that abuses of government power by career employees span most administrations. That Wired story, for example, examined CBP actions from 2009 to 2022. Stormtroopers are an endemic problem. Clearly, the Trump administration is not helping, but this is not a new phenomenon.

Then there are the Trump emulators, the aspiring politicians who see Donald Trump as their political lodestar and think that behaving like him, they will acquire power. The New York Times’ Michelle Goldberg wrote an excellent column about the likes of U.S. Representatives Max Miller and Cory Mills, Republican House candidate Brandon Herrera, and other kindred spirits. As Goldberg notes, “Thanks to President Trump, misogyny and extremism among Republicans have become what journalists sometimes call a dog-bites-man story, something so ordinary and predictable that it breaks through only when it reaches an absurdly high threshold.” It is actually worse than that — in Miller’s case, the Trump White House overtly intervened to tell Republicans to back off their criticisms of Miller.

These folks are infuriating, and a sign of how Trump has normalized the idea of GOP candidates running despite serious scandals that would have forced a previous generation of candidates to drop out.

And yet, to the hardworking staff here at Drezner’s World, the group that is even more infuriating are the allegedly normie Republicans. These are the folks who have held their noses and continued to vote Republican despite all of… this… for the past ten years.

For example, Brandon Herrera has made remarks that are clearly sexist and anti-Semitic. Despite all of this, House Speaker Mike Johnson is attending Herrera fundraisers. And as Goldberg notes in her column, “I haven’t seen much public hand-wringing about the decision of the pious Johnson to raise money for Herrera, nor debate about what his candidacy means for the Republican Party’s attitude toward women, Jews or seniors.”

Or consider this paragraph from Politico about GOP thinking regarding Max Miller: “Some Republicans argue they have no choice but to close ranks around a candidate accused by his former father-in-law, Ohio GOP Sen. Bernie Moreno, of being unfit to serve and needing psychological help; Ohio’s other Republican senator, Jon Husted, has also said Miller should resign and not run for reelection. Miller has remained defiant amid the calls to resign.” My point is that they absolutely have a choice.

Furthermore, they keep making their choice in a way that empowers the worst elements of their party. G. Elliott Morris recently explored the gap between Trump’s high disapproval polling and the Democrats smaller margin in the generic congressional ballot. His conclusion: “Voters who pulled the lever for Trump in ‘24 and call themselves Republicans but not conservatives stay Republican…. The pattern is hard to miss. The subgroups that keep “Republican” in their identity stay Republican by 37 points or more in the generic ballot even while disapproving of the man leading their party, — while both subgroups without it defect to the Democrats by a margin of 30-plus points. Party identity is doing a lot of work, even among Trump voters.”

Here’s a helpful chart:

It’s this group that, in the end, are the pivotal Trump enablers. Senators like Chuck Grassley or, even better Bill Cassidy.

When I see someone like Brandon Herrera or Max Miller pop up in the news, it does not take long to conclude that there’s no point condemning someone who has already lost their soul. Mike Johnson and Bill Cassidy, however, likely think that they have only rented theirss temporarily.

These are the folks for whom calumny has been sparser — but dear God do they deserve it.

<https://danieldrezner.substack.com/p/categories-of-contempt> <https://danieldrezner.substack.com/>

Brad DeLong here: Drezner’s argument runs:

1. buffoons like Hegseth act badly but predictably,

2. so condemning them yields diminishing returns;

3. capable actors like Rubio possess the acumen to recognize what they’re enabling,

4. making their compliance a willful moral choice rather than a limitation;

5. therefore the marginal moral outrage should flow toward those with agency who choose complicity;

6. the pivotal such group is the “normie” Republican voters and senators whose durable partisan identity — not approval of Trump — keeps the coalition electorally viable.

The causal punchline: party identity, not enthusiasm, supplies Trump his governing majorities, so the quiet enablers are the true load-bearing structure. This is the core vibe of Dan’s current position.

When I see someone like Brandon Herrera or Max Miller pop up in the news, it does not take long to conclude that there’s no point condemning someone who has already lost their soul. Mike Johnson and Bill Cassidy, however, likely think that they have only rented theirs temporarily.

I do recall 24 years ago when the George W. Bush administration took office and promptly began to dismantle as much of successful Clinton-era policies as it could: policies aimed at keeping a prosperous and peaceful post-Cold War multilateral world, policies aimed at boosting economic growth, policies aimed at stabilizing federal finances, policies aimed at rolling back the rise in income and wealth inequality. They were remarkably successful at doing so, I must say. And those who we Rubin Democrats had thought were our allies in sensible neoliberalism were—whatever were their private misgivings under Chatham House rules— publicly cheering them on or at least immediately moving to distract from policy substance by expressing their shame at how the democrats had chosen a standard bearer bill. Clinton who had a huge problem with his zipper. (Yes, I am looking at you, Alan Greenspan, Glenn Hubbard, Colin Powell, and company: look in the mirror, if you dare.)

Read more

To a substantial degree, America has not been a “land of equal opportunity” with each generation’s economic fate largely self-made for quite a while.

The old consensus said a father’s economic advantage all but vanished in three generations. But Bowles and Gintis summarize the line of work that established that that belief was a measurement error-caused statistical illusion: the true intergenerational persistence of income is roughly three times higher than the old consensus. Becker and Tomes’s (1986) claim that the father-son income correlation was 0.15 is simply wrong. Think, instead: an intergenerational correlation of 0.4, and intergenerational elasticities of 0.7 for consumption, 0.5 for wealth, 0.4 for income, 0.35 for earnings, and 0.3 for school years.

IMPORTANT!: The belief is false that “smarts” and the genetic transmission thereof, at least as measured by IQ, is key to the intergenerational transmission of income inequality. Thus the argument that inequality is not a problem because the smart deserve to be rich does not fly. It is parental wealth, race, and “noncognitive personality traits” that do most of the work here. But the intergenerational transmission of economic status remains “a black box”: as the standard human-capital smart parents → smart, well-schooled kids → high earnings accounts for at most three-fifths of it, and the genetic inheritance of IQ accounts for almost none.

This line of research has always been very bad news for the caliper-measurers:

The results are somewhat surprising: wealth, race and schooling are important to the inheritance of economic status, but IQ is not a major contributor, and, as we have seen above, the genetic transmission of IQ is even less important.

Share

Yet they still bring out the calipers at every opportunity.

(VERY PARTIAL-)CROSSPOST: Samuel Bowles & Herbert Gintis (2002): The Inheritance of Inequality

<https://pubs.aeaweb.org/doi/pdfplus/10.1257/089533002760278686>

Journal of Economic Perspectives—Volume 16, Number 3—Summer 2002—Pages 3–30

See Bowles and Gintis (2001) for the relevant formal models and other technical aspects of this research, also available at <http://www.santafe.edu/sfi/publications/working-papers.html>. Arrow, Bowles and Durlauf (1999) and Bowles, Gintis and Osborne (forthcoming) present collections of recent empirical and theoretical research.

Samuel Bowles is Professor of Economics at the University of Siena, Siena, Italy, and Director of the Economics Program, Santa Fe Institute, Santa Fe, New Mexico. Herbert Gintis is a member of the External Faculty, Santa Fe Institute, Santa Fe, New Mexico. Both authors are Emeritus Professors of Economics, University of Massachusetts, Amherst, Massachusetts. Their e-mail addresses are bowles@santafe.edu and hgintis@attbi.com, and their websites are (http://www-unix.oit.umass.edu/bowles and http://wwwunix.oit.umass.edu/gintis).

People differ markedly in their views concerning the appropriate role of government in reducing economic inequality. Self-interest and differences in values explain part of the conflict over redistribution. But by far the most important fault line is that people hold different beliefs about why the rich are rich and the poor are poor. Survey data show that people—rich and poor alike—who think that “getting ahead and succeeding in life” depends on “hard work” or “willingness to take risks” tend to oppose redistributive programs. Conversely, those who think that the key to success is “money inherited from family,” “parents and the family environment,” “connections and knowing the right people” or being white support redistribution (Fong, 2001; Fong, Bowles and Gintis, 2002). Handing down success strikes many people as unfair even if the stakes are small, while differences in achieved success may be unobjectionable even with high stakes, as long as the playing field is considered level.

How level is the intergenerational playing field? What are the causal mechanisms that underlie the intergenerational transmission of economic status? Are these mechanisms amenable to public policies in a way that would make the attainment of economic success more fair? These are the questions we will try to answer.

No one doubts that the children of well-off parents generally receive more and better schooling and benefit from material, cultural and genetic inheritances. But until recently, the consensus among economists has been that in the United States, success is largely won or lost in every generation. Early research on the statistical relationship between parents’ and their children’s economic status after becoming adults, starting with Blau and Duncan (1967), found only a weak connection and thus seemed to confirm that the United States was indeed the “land of opportunity.” For example, the simple correlations between parents’ and sons’ income or earnings (or their logarithms) in the United States reported by Becker and Tomes (1986) averaged 0.15, leading the authors to conclude: “Aside from families victimized by discrimination... [a]lmost all earnings advantages and disadvantages of ancestors are wiped out in three generations.” Becker (1988) expressed a widely held consensus when, in his presidential address to the American Economics Association, he concluded (p. 10): “[L]ow earnings as well as high earnings are not strongly transmitted from fathers to sons.”

But more recent research shows that the estimates of high levels of intergenerational mobility were artifacts of two types of measurement error: mistakes in reporting income, particularly when individuals were asked to recall the income of their parents, and transitory components in current income uncorrelated with underlying permanent income (Bowles, 1972; Bowles and Nelson, 1974; Atkinson, Maynard and Trinder, 1983; Solon, 1992, 1999; Zimmerman, 1992). The high noise-to-signal-ratio in the incomes of both generations depressed the intergenerational correlation. When corrected, the intergenerational correlations for economic status appear to be substantial, many of them three times the average of the U.S. studies surveyed by Becker and Tomes (1986).

The higher consensus estimates of the intergenerational transmission of economic success has stimulated empirical research. The relevant facts on which most researchers now agree include the following: brothers’ incomes are much more similar than those of randomly chosen males of the same race and similar age differences; the incomes of identical twins are much more similar than fraternal twins or non-twin brothers; the children of well-off parents obtain more and higher quality schooling; and wealth inheritance makes an important contribution to the wealth owned by the offspring of the very rich. On the basis of these and other empirical regularities, it seems safe to conclude that the intergenerational transmission of economic status is accounted for by a heterogeneous collection of mechanisms, including the genetic and cultural transmission of cognitive skills and noncognitive personality traits in demand by employers, the inheritance of wealth and income-enhancing group memberships, such as race, and the superior education and health status enjoyed by the children of higher status families.

However, the transmission of economic success across generations remains something of a black box. We find that the combined inheritance processes operating through superior cognitive performance and educational attainments of those with well-off parents, while important, explain at most three-fifths of the intergenerational transmission of economic status. Moreover, while genetic transmission of earnings-enhancing traits appears to play a role, the genetic transmission of IQ appears to be relatively unimportant.

It might be thought that the black box is an artifact of poor measurement of the intervening variables relative to the measurement of the income or earnings of parents and offspring. But this does not seem to be the case. Years of schooling and other measures of school attainment, like cognitive performance, are measured with relatively little error. Better measurements will of course help; but we are not likely to improve much on our measures of IQ, and recent improvements in the measurement of school quality have not given us much illumination about what’s going on inside the black box. The fundamental problem is not that we are measuring the right variables poorly, but that we are missing some of the important variables entirely. What might these be?

Most economic models treat one’s income as the sum of the returns to the factors of production one brings to the market, like skills, or capital goods. But any individual trait that affects income and for which parent-offspring similarity is strong will contribute to the intergenerational transmission of economic success. Included are race, geographical location, height, beauty or other aspects of physical appearance, health status and personality. Thus, by contrast to the standard approach, we give considerable attention to income-generating characteristics that are not generally considered to be factors of production. In studies of the intergenerational transmission of economic status, our estimates suggest that cognitive skills and education have been overstudied, while wealth, race and noncognitive behavioral traits have been understudied…

[…]

One of the transmission channels deserves special attention not only because of its prima facie plausibility, but also because of the extraordinary attention given to it in popular discussions of the subject. This is the genetic inheritance of cognitive skill. The similarity of parents’ and offsprings’ scores on cognitive tests is well documented. Correlations of IQ between parents and offspring range from 0.42 to 0.72, where the higher figure refers to measures of average parental and average offspring IQ (Bouchard and McGue, 1981; Plomin et al., 2000). The contribution of cognitive functioning to earnings both directly and via schooling attainment has also been established in a variety of studies that estimate determinants of earnings using IQ (and related) test scores….

Do these two facts—parent-child similarity in IQ and an important direct and indirect causal role for IQ in generating earnings—imply a major role for genetic inheritance of cognitive ability in the transmission of intergenerational economic 10 Journal of Economic Perspectives status? One way to formulate this question is to ask how similar would parental and offspring IQ be if the sole source of the similarity were genetic transmission. Also, how similar would the incomes of parents and offspring be if there were no other transmission channel?….

We see… the contribution of genetic inheritance of IQ to the intergenerational transmission of income…. If the heritability of IQ were 0.5 and the degree of assortation, m, were 0.2 (both reasonable, if only ballpark estimates) and the genetic inheritance of IQ were the only mechanism accounting for intergenerational income transmission, then the intergenerational correlation would be 0.01, or roughly 2 percent the observed intergenerational correlation. Note the conclusion that the contribution of genetic inheritance of IQ is negligible is not the result of any assumptions concerning assortative mating or the heritability of IQ: the IQ genotype of parents could be perfectly correlated and the heritability of IQ 100 percent without appreciably changing the qualitative conclusions. The estimate results from the fact that IQ is just not an important enough determinant of economic success…

[…]

Conclusion: Recent evidence points to a much higher level of intergenerational transmission of economic position than was previously thought to be the case. America may Samuel Bowles and Herbert Gintis still be the land of opportunity by some measures, but parental income and wealth are strong predictors of the likely economic status of the next generation.

Our main objective has been to assess the extent of intergenerational transmission and the mechanisms accounting for it. Table 3 summarizes our best estimates of the relative importance of the main causal channels we have been able to identify. The only entry not previously explained is the first, which is an estimate of the correlation between parental income and child IQ multiplied by our estimate of the normalized effect of IQ on earnings, conditioned on, among other things, years of schooling. The estimates for IQ, schooling and personality in the income column are simply those in the earnings column adjusted to take account of the effect of earnings differences on income differences, suitably normalized as described in Bowles and Gintis (2001). Thus, we do not take account of the way that these earnings determinants may affect the rate of return to one’s wealth. By contrast, we assume that the race effect is of the same magnitude in determining the returns to both human capital and conventional wealth (if the race effect on incomes worked solely via an effect on earnings, its contribution to the intergenerational earnings correlation would be significantly greater).

While the estimates in Table 3 are quite imprecise, the qualitative results are not likely to be affected by reasonable alternative methods. The results are somewhat surprising: wealth, race and schooling are important to the inheritance of economic status, but IQ is not a major contributor, and, as we have seen above, the genetic transmission of IQ is even less important.

A policymaker seeking to level the playing field might use these results to design interventions that would loosen the connection between the economic success of parents and the economic prospects of their children. But does a level playing field entail no correlation between parental and child incomes (Swift, forthcoming)? There are important values of family life and privacy that would be compromised by any serious attempt to disconnect the fortunes of parents and children completely. Rather than pursuing an abstract (and to our minds unattractive) objective of zero intergenerational correlation, a better approach might be to ask which mechanisms of intergenerational transmission seem unfair, and to direct policies accordingly. The role of race in transmitting status from generation to generation is clearly unfair. Many people regard the strong correlation between parental income and child health as morally suspect, and many feel the same way about high levels of wealth inheritance. Large majorities favor policies to compensate for inherited disabilities. Other mechanisms of persistence—the genetic inheritance of good looks, for example—strike most people as unobjectionable and not an appropriate target for compensatory policy interventions. Even if some consensus could be formed on which of these mechanisms are morally suspect, the policy implications would be far from clear. For example, the possible incentive effects on parental behaviors of reduced parental influence on child success would have to be estimated and considered

Bowles Gintis 2002 The Inheritance Of Inequality

390KB ∙ PDF file

Download

Download

.<https://pubs.aeaweb.org/doi/pdfplus/10.1257/089533002760278686>

Brad DeLong back again: As I said last year, there are a great many people—the people whom Richard Rumbold back in 1685 denounced from the scaffold after being captured during Argyll’s Rising against James II Stuart—who fervently and with every fiber of their being search diligently for some reason to believe that most people “come… into the world with a saddle on his back… [with others] booted and spurred to ride…” It used to be that our natural rulers were such because of their family traditions of blood and courage, or blood and honor. Or perhaps it was that non-Hellenes were slaves and non-males were subordinate by nature because of their lack of rational faculties.

Later on, it was an enterprising spirit, as set forth by Andrew Carnegie:

The law of competition… may be someimes hard for the individual, [but] it is best for the race, because it insures the survival of the fittest…. We accept and welcome therefore… concentration… in the hands of a few… [as] essential to the future progress of the race. There must be great scope for the exercise of special ability…. Objections to the foundations upon which society is based are not in order, because the condition of the race is better with these than it has been with any other which has been tried…

Give a gift subscription

And now the air among the TechBros on the other side of San Francisco Bay is that the magic fairy dust that gives you a legitimate right to extraordinary wealth and power is IQ. Inherited IQ. Genetically-driven inherited IQ.

This debate matters because people’s views on redistribution turn on one belief: whether the rich are rich because they earned it and so deserve it, or inherited it such a way that they do not deserve it. Bowles and Gintis make that question empirical.

But Bowles & Gintis’s running the numbers tells us that (a) intergenerational inequality inheritance is definitely a thing, and (b) it definitely ain’t brains genetically inherited as measured by IQ. For:

• Parent-child IQ similarity is real and IQ affects earnings, but because IQ’s total effect on earnings is modest (~0.27) and heritability is bounded, the genetic-IQ channel contributes only ~2% of the observed intergenerational income correlation.

• Inherited wealth (concentrated at the top), race (a heritable, environmentally-activated marker), schooling (partly independent of IQ), health, and heritable noncognitive personality traits like fatalism, work ethic, and time preference matter.

• Two-fifths-plus of the parent-child income link remains unexplained. Wealth bequests drive persistence at the top; health shocks and violence drive it at the bottom. The “twin peaks” of stuck relative poverty and stuck relative affluence have different mechanisms.

And, of course, the right policy target isn’t zero intergenerational wealth correlation, but the reduction of the mechanisms people judge unfair, which is a pretty idiosyncratic and potentially variable set of judgments.

This line of findings should have pushed the inequality-inheritance research agenda away from cognition and human-capital toward the study of wealth, discrimination, health, and “noncognitive” personality traits.

Flaws in the paper are (a) that it is now old, and there has been a lot of research water under the bridge in the past generation, (b) what is “most reasonable” is up for grabs and parameter estimates are fuzzy; (c) the residual is doing much of the arguing, and (d) the “noncognitive” fatalism, work ethic, and time preference traits may well be consequences of relative poverty and constrained opportunity rather than independent inherited causes of low earnings.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##very-partial-crosspost-samuel-bowles-herbert-gintis-2002-the-inheritance-of-inequality
##crosspost
##inequality-and-domination
##every-era-invents-a-reason-why-the-powerful-deserve-their-power-noble-blood-then-enterprise-now-inherited-iq-but-things-are-more-complicated-than-inequality-justification-mongers-think
#sam-bowles
#herb-gintis
#intergenerational-transmission-of-inequality
#genetic-determinism
#meritocracy-myth
#level-playing-field

When I am confronted in my feed with things like this from (one of the few) highly reputable information-intensive sources of ground truth that is not consciously trying to mislead me to advance its own agenda:

Share

I step back and say: WTF?!?!

What use is something like this?

Yes, if you took some money, decided to invest it in the MAMLM-&-datacenter-build-out last January, and picked NVIDIA, right now you are 15% richer but suffer from enormous regret vis-à-vis the world in which you picked Micron. That is a thing. There is a question. There is valid information, presented comprehensively. Thus there is an answer.

But why spend time and induce your readers to spend time on this question and this answer? The write-up goes:

Lynn Thomason: Nvidia Stock Bulls Get Punished in the Run-Up to Earnings <https://www.bloomberg.com/news/newsletters/2026-08-25/nvidia-stock-bulls-get-punished-in-the-run-up-to-earnings>: ‘Nvidia has fallen for seven days, its longest run of losses since 2022…. Nvidia’s losing streak: Nvidia shares have fallen in seven straight sessions, the longest run of losses since 2022. It’s a worrying stat for the company’s executives as they prepare to report earnings tomorrow. Here’s what you should know: • It’s still a cash cow. Analysts estimate that revenue nearly doubled last quarter to $92 billion. That’s more than any of its rivals get in a full year. • The competition is heating up. A growing group of upstarts are vying for a bigger share of the market. • Nvidia’s star has dimmed on Wall Street. Though its shares have climbed 15% in 2026, the Philadelphia Stock Exchange Semiconductor Index has gained 66%. The stock ticked higher on Tuesday morning…. The stock reflects more fear than hope, says MLIV strategist Sebastian Boyd. Its forward P/E ratio is merely in line with the S&P 500. In other words, traders don’t put much faith in earnings growth beating the broader market after next year. • Bank of America says buy. Last week, the bank’s analysts called it a “compelling opportunity,” saying the stock trades at a discount of as much as 50%. Their price target? $350. • Price hikes are coming. Chipmaker stocks have been rattled in recent days by news that some of Nvidia’s biggest customers were told about AI-related price increases above 15%. • Nvidia is an industry lynchpin. As the company helps arrange financing for AI infrastructure, some are calling it the “central bank of AI.”

Give a gift subscription

And then another graph:

Share DeLong's Grasping Reality Weblog

Do not get me wrong: Bloomberg is a magnificent information source. And one of the few places that still pays journalists healthy sums these days where the journalists can hold their heads up very high as practitioners of their craft, rather than as some form of PR in disguise.

But, still, the way that this information is presented is as if Bloomberg thinks that the paying customers it needs to keep are those who are making month-to-month jumps in asset allocation, placing directional bets on what is going to happen to asset prices in the short run in anticipation of what average opinion will expect average opinion to be. In pushing forward that way of thinking, Bloomberg is not inducing its readers to be their best selves.

What should it be doing? Well, I think every time it writes about NVIDIA and Micron, it should highlight graphs like these:

Get 75% off a group subscription

Refer a friend

Isn’t that the context people need to be pushed to put into the forefront of their brains as they think about anything to do with this? GPU chip design, memory chip design and manufacture, the excellence of these two companies at those tasks, how big the MAMLM-&-datacenter-build-out is, and when and where NVIDIA and Micron became the limited-supply rent-collecting chokepoints here: isn’t that the path that readers should be nudged to follow as they think?

The Bloomberg write-up on NVIDIA’s “losing streak “is competent and comprehensive: doubled revenue, heating competition, a dimmed Wall Street star, a BofA “buy,” coming price hikes, the “central bank of AI.”

All true.

All largely beside the point, unless the key reader is someone placing month-to-month directional bets in a Keynesian General Theory chapter 12 “The State of Long-Term Expectation” <https://www.marxists.org/reference/subject/economics/keynes/general-theory/ch12.htm> beauty contest — anticipating what average opinion expects average opinion to be. That’s not journalism inviting readers to think well. The context that belongs at the front of the mind is different.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠nvidia-has-a-seven-day-stock-price-slide-but-you-should-not-care-chart-annoyance-of-the-day
##macro-outlook
##behavioral-finance
##public-reason⁠
##‎⁠chart-of-the-day
##nnoyance-of-the-day
##‎⁠a-magnificent-information-source-is-bloomberg-news-yet-it-keeps-pushing-its-readers-to-be-their-worst-month-to-month-trading-selves⁠
#‎⁠nvidia-has-a-seven-day-stock-price-slide-but-you-should-not-care⁠⁠
#nvidia
#micron
#bloomberg-news
#ai-datacenter-buildout
#semiconductors
#keynes
#beauty-contest
#long-term-expectation
#short-term-trading
#financial-journalism

I see this morning that Torsten Slok reads it how I read it: We are not experiencing an “AI Jobs Shock”. At most, we are experiencing a reduction in hiring by executives who do not understand the technology but who have absorbed vibes that there will be a real-soon-now “AI Jobs Shock”. Where do these vibes come from? From a combination of grifters seeking money, and madmen hearing the voices in their heads of a forthcoming Digital God:

Share

Torsten Slok: Where Is the AI Jobs Shock? Not in India or the Philippines <https://www.apollo.com/wealth/insights-news/insights/daily-spark/where-is-the-ai-jobs-shock-not-in-india-or-the-philippines>: ‘If AI were displacing white-collar work at scale, you would expect to see it first in the Philippines and India, where business process outsourcing (call centers, IT support, back-office processing) accounts for a large share of employment. Instead, the unemployment rate in both countries has continued to trend lower, with the Philippines near 5% and India near 6%, both well below their 2021 levels. The bottom line is that the hard data still show no signs that AI is generating job losses in the economies and industries most exposed to it:

Share DeLong's Grasping Reality Weblog

Now do not get me wrong. That the real effects are small in the context of the global economy does not mean that they are not there. But they are not yet large. And, for the most part, they appear to be consequences brought forward in time from a belief in a future in which there will be technology- and efficiency-driven substitution for labor.

The Digital God-maddened and the grifters are having real effects, but they are almost all destructive with respect to institutions and firms that have been effected. The gullible have bought into and the malevolent have either bought into or pretend to have bought into these narratives. They have taken actions that have shredded institutions and capabilities that they are supposed to be managing and nurturing.

Look at this, for example:

Henry Farrell (2025): Silicon Valley’s Reading List Reveals Its Political Ambitions <https://www.bloomberg.com/news/articles/2025-02-21/to-understand-doge-look-to-the-tech-industry-s-reading-list?srnd=phx-weekend_2>: ‘DOGE’s grand effort to cut government down to size is the latest manifestation of a longstanding Silicon Valley dream: to remake politics in its image…. DOGE’s grand effort to cut government down to size is the newest iteration of an epic narrative of change. Musk, a heroic entrepreneur, will surely make history as his tiny team of engineers cuts the government Leviathan down to size. One DOGE recruiter framed the challenge as “a historic opportunity to build an efficient government, and to cut the federal budget by 1/3.” When a small team remakes government wholesale, the outcome will surely be simpler, cheaper and more effective. That, after all, fits with the story that Silicon Valley disruptors tell themselves….

Like the Renaissance engineers who wanted to raze squalid and inefficient cities to start anew, DOGE proposes to flense away the complexities of government in a leap of faith that AI will do it all better. If the engineers were not thoroughly ignorant of the structures they are demolishing, they might hesitate and lose momentum…. [But] DOGE’s artificial-intelligence-fueled vision of government is a vision from Franz Kafka, not Friedrich Hayek…

Give a gift subscription

And also:

Henry Farrell: How AI Madness Helped Fuel DOGE <https://www.programmablemutter.com/p/how-ai-madness-helped-fuel-doge>: ‘First, “effective accelerationism” is an important part of the intellectual mix that helped produce the AI-DOGE chimera. In particular, the neo-reactionary arguments of Nick Land have had real, and pernicious consequences…. Claims about how our world will be radically remade in the image of machine logic are nearly ubiquitous, whether the authors view the machine-god state as something to be feared, celebrated or both at once…. Second… there is a fundamental difference between the disastrous DOGE project and the apparently similar push by both center right- and left-leaning people to create a more effective and responsive government bureaucracy….

The two approaches differ crucially on the question of who should the government be responsive to? DOGE/AI Thought starts from the premise that bureaucracy should be primarily (perhaps even exclusively) responsive to the people at the top. From this perspective, the problem that AI solves is a mixture of regular institutional inertia and specific “deep state” resistance….

The effective government bureaucracy people… are not in the business of making sure that Dear Leader’s commands get implemented as they ought. Instead, they are primarily interested in freeing bureaucrats to do things that are obviously the right things to do, rather than burying them beneath the concrete of top down mandates. The impulse, then, is to trust bureaucrats more, and give them the means and autonomy to respond to obvious needs. This involves creating better feedback loops between top and bottom, so that measures, tools and perhaps even goals are redefined as the problem becomes better understood. But it also means creating interfaces through which bureaucrats can engage more with the public, and respond better and more quickly to public demands, as well as helping them work sideways with others in the bureaucracy who have necessary skills and knowledge, without getting smothered in red tape. The general bet is not on better subjugating bureaucrats, but on making them more autonomous. This is more or less the opposite of DOGE.…

Anyway - these are side notes to a larger argument. If you are interested, do read the piece itself!…

Leave a comment

Brad DeLong back again: What do I think? I think the sharp Henry Farrell has, if you look across his writings for the past year and a half, built obne of the most illuminating accounts we have of what the Musk-Trump DOGE effort was and all of the damage it has done, as well as of the underlying intellectual rot that made it possible. Let me try to synthesize, because I think he has gotten hold of something that most of the commentariat has missed. And let me put it behind the paywall for now, because I am not sure that it is right”

Start where Farrell starts: with a bookshelf. When Jasmine Sun asked why James Scott’s anti-bureaucratic anarchist trace Seeing Like a State had become a fixture in Silicon Valley, she put her finger on a revealing misreading. The tech elite read Scott as government-overreach porn for people who would like to be left alone to build. But Scott’s target was not government as the high-modernist “bulldozing habit of mind” of the engineer who wants to raze the squalid, illegible, organically grown and replace it with clean mathematical lines. Scott indicted all social engineering, whether performed by Soviet collectivizers or by Robert Moses. And, Farrell observes, Elon Musk and his band of tenty-somethings epitomized exactly that bulldozing turn of mind.

Recall Friedrich von Hayek’s central insight at the heart of the socialist-calculation debate: the knowledge required to run a complex society is distributed, tacit, local, and impossible to gather into any single mind or machine at the center. The market is a societal calculating mechanism precisely because no planner can substitute for it. DOGE, invoking Hayek, then set out to do the one thing Hayek said could not be done: pull the totality of government knowledge into a single vast apparatus, run from the top, and let it optimize.

As Farrell puts it, the result is a vision not from Friedrich Hayek but from Franz Kafka, passed through the weights and activations of a simulated neural network: not seeing like a state but rather hallucinating like a next token-prediction machine babbling amok. <https://www.programmablemutter.com/p/ai-isnt-management-try-explaining>

If you genuinely believed, as the Digital God-maddened “effective accelerationists” do, that a country of geniuses in a datacenter arrives before 2027, then tearing out the National Institutes of Health and the National Science Foundation is not vandalism but prudent housekeeping. Why maintain cumbersome, expensive human institutions for funding science when superhuman cognition is imminent? Farrell reports asking a DOGE-adjacent figure whether gutting the government’s decision-making capacity might be unwise, and getting the reply: so what? We’ll have AGI by late 2026. <https://www.programmablemutter.com/p/should-agi-preppers-embrace-doge>.

Well, it is late 2026 now.

Radical institutional demolition follows naturally from the AGI-prepper framework. DOGE was not a bug in the singularity theology. It was the first sacrament in a false theology.

Take refuge in (Alison) Gopnikism: Large language models are not nascent minds but social and cultural technologies. They are “lossy JPEGs” of the human corpus, statistical machines that compress, remix, and resample collective human intelligence. They are useful the way markets and bureaucracies and democracies are useful, as means of assembling distributed human knowledge and coordinating cognition, organization, and action. They are not is an exit door out of the human condition: a superior and independent problem-solver to which we can delegate.

As Farrell notes. DOGE deliberately cannibalized the U.S. Digital Service: the one corner of government that had modeled the curious, problem-oriented, genuinely public-spirited engineering that once made Silicon Valley an attractive place. It turned it into a Kafkaesque device for stamping one chaos monkey’s rage-based incoherent momentary impulses into the flesh of the body politic.

The real question, Farrell correctly insists, was never efficiency. It was: to whom should the state be responsive? The reformers who want a better government want to free the bureaucrat to do the obviously right thing. DOGE wanted the bureaucracy to be responsive to one man at the top, and nobody else.

Perhaps it is time for me to step back and reread Peter Drucker.

We do not usually think of Peter Drucker as a political theorist, still less as a theorist of freedom. He is filed under “business guru,” shelved next to the airport paperbacks, quoted by CEOs who want to sound as though they have read a book.

This is a category error, and a large one.

Drucker himself insisted, near the end of his very long life, that he was not really a management thinker at all but a “social ecologist,” and that his true subject was “the individual’s existence in the social environment.”

He wrote thirty-nine books; only two—Management for Results and Innovation and Entrepreneurship—were about business management as such. The rest were about community, society, and polity. Management was the answer he arrived at: the reconciliation of individual and community, of freedom and necessity, of routine and innovation

But we cannot understand the answer unless we hold onto the question. The question was one of the great questions of the twentieth century: how, after the market society had dissolved the old bonds of status and community, could a modern industrial civilization be both free and functioning?

Drucker came to that question the hard way. His first book, The End of Economic Man (1939), was a study of the origins of totalitarianism, written by a young Austrian émigré who had watched the thing arrive in person. His diagnosis was, in its bones, the same as his close friend Karl Polanyi’s:

• Nineteenth-century liberalism had promised that the market would deliver both prosperity and dignity.

• It had promised that “economic man” would find in commercial society his identity and his place.

• That promise had failed.

• The Great Depression put the capstone to a hundred-year historical demonstration that revealed that the untrammeled pseudo-classical semi-liberal market system gave people neither security nor status nor meaning.

• Into that vacuum of despair rushed the fascist and the Bolshevik.

• They offered the terrible consolations of the totalitarian state, and enough of the the masses embraced tyranny because the pseudo-classical semi-liberal order had left them functionless and alone.

That anxiety that runs underneath every cheerful page Drucker ever wrote about quarterly objectives. He had seen what happens when a society cannot give ordinary people a place to stand. His solution was neither the market nor the state. It was the institution—and, specifically, the managed corporation as the representative social institution of the industrial age and management as a holy, clerical profession with a public trust.

In Concept of the Corporation (1946), his study of General Motors, Drucker did something no one had quite done before: he looked inside the great enterprise and asked not merely whether it was efficient but whether it was legitimate—whether it could give the men on the line and the managers in the divisions the citizenship, the function, and the status that the vanished village and guild once had. His answer was a demand that the corporation be understood as a political community, governed by what he called “federal decentralization”—divisions granted real autonomy, exactly as the American constitutional order grants power to the states. GM, which had opened its books to him, received the book as a betrayal, because Drucker had told a wildly successful company that its success was not the point; its social constitution was.

This is why “management” for Drucker was never a bag of techniques. Management was how you reconciled the two things the twentieth century could not otherwise reconcile: the entrepreneurial energy that creates wealth, and the human need for dignity and belonging that raw wealth-creation tramples. The manager’s task, he wrote in The Practice of Management (1954), is “to create a true whole that is larger than the sum of its parts”. And the manager’s characteristic instrument, “management by objectives and self-control,” was designed precisely to square private drive with collective purpose. Objectives exist so that each person can direct himself toward the goals of the whole—“control” in the sense of self-direction, never “control” in the sense of one man’s domination of another.

From this it follows that profit, for Drucker, is not the purpose of a business. Profit is a condition for staying in the game, just as successfully breathing a condition for staying alive. But the purpose of a business is to create a satisfied customer and thereby to serve the common good. A business that measures itself only by the bottom line has misunderstood what it is for, and a manager who thinks his job is to keep the accountants happy has misunderstood what he is for. Perhaps it is good to view Drucker as Schumpeter with the heartlessness surgically removed—all of the creative dynamism, none of the contempt for the ordinary.

Drucker’s final synthesis, offered when he was nearly eighty, was to call management “a liberal art.” Management is a humanist technology, dealing with people, their values, their growth, their good and evil, which makes it a humanity. The manager is essential to the functioning society of autonomous, performing, responsible institutions that is “the only safeguard of freedom and dignity,” and “the alternative to tyranny and our only protection against it.”

That is a far grander ambition than anything in the management shelf’s usual promises. Drucker was not teaching executives how to win. He was trying to show a civilization how to remain both prosperous and free.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##the-white-collar-canaries-in-the-ai-job-loss-coal-mine-are-wide-awake-feeding-chirping-to-themselves-chart-of-the-day
##macro-outlook
##enlarging-the-bounds-of-human-empire
##mamlms
##chart-of-the-day
##or-the-dog-is-not-barking-in-the-nighttime-for-there-is-not-yet-any-ai-jobs-shock
#the-white-collar-canaries-in-the-ai-job-loss-coal-mine-are-wide-awake-feeding-chirping-to-themselves
#not-yet-any-ai-jobs-shock
#white-collar-canaries
#torsten-slok
#effective-accelerationism
#digital-god
#agi-preppers
#henry-farrell
#trump-musk-doge
#peter-drucker
#management-liberal-art

That is standard in this business. Other results are, for the most part, also standard in this business. Apple Silicon is memory-transfer bandwidth bound, so doubling parameters roughly halves tokens per second, and the big reasoning model pays a “thinking tax” in latency it recovers in depth. Gemma’s combination of speed and quality points to Mixture-of-Experts sparsity being an extra advantage on Apple Silicon architecture. Qwen 3.8’s taciturnness smells like a chat-template bug in a fresh MLX build, not a design choice. Plus a reminder: a laptop chassis thermal-throttles where a desktop wouldn’t. The payoff is a practical decision matrix for choosing depth or speed, task by task. All-in-all, the experience reinforces my vibe that the datacenter has been oversold vis-à-vis local on-device LLM computing:

News on the LLM-training front: alibaba/qwen3.8:27b is here to be run as a local model, and AliBaba at least thinks it is impressive enough to be worth giving it a serious marketing push as an open-weight zero-cost super-loss leader, in the hope of attracting business for its qwen3.8 2.4T-parameter super-big sister.

So I am standing up the alibaba/qwen3.8:27b-q4_K_M variant with my standard initialization routine:

Share

No, I am not. It turns out that alibaba/qwen3.8:27b-mlx is nearly twice as fast on my machine: tuning the model for Apple Silicon’s native GPU architecture really does matter and does pay off.

Now are there other options in the alibaba/qwen3.8 series?

It does not seem that there are any, at least not yet:

While the qwen 2.5 series scales up predictably to a 72B model, the Qwen 3.8 series represents a jump into "Frontier" scale modeling.

The Hierarchy of Qwen 3.8: Based on the current roadmap and availability:

1. Qwen 3.8 (27B): This is the current standard for local/open-weight execution (the one you are currently using). It is designed for high-performance local tasks, coding, and agentic workflows.

2. Qwen 3.8-Max-Preview (2.4 Trillion Parameters): This is the massive upward jump. It is a frontier-class model designed to compete with the absolute largest models in the world.

Can you run the "Max" version? No, not locally…. Within the 3.8 architecture: Your 27B model is the current ceiling for local, high-performance execution on your hardware…

Give a gift subscription

Time for an LLM bake-off! The anchor benchmark is what has been my rapid-response workhorse model for the past four months: google/gemma4:26b-mlx:

Considerably less verbose, but well oriented to time, place, and context. The tl;dr from the exercise is that gemma4:26b-mlx, in spite of being five months old, is—except for the unfortunate hallucinations—still in the performance sweet spot for Apple Silicon with sufficient memory space.

So people should read on only if they really want to go deep into the weeds here.

Note: Not so good, Pangram/SubStack:

This is easily 80% LLM-generated text, not 30%!

Leave a comment

Moreover, confidence is diminished by noting that, when given the task of running the LLM bake-off, google/gemma4:26b-mlx reports that it is a model of very little brain:

I apologize for the confusion. My internal state reported that the task was "initialized" and "dispatching," but it appears the subagents were never actually triggered, causing the process to stall in that intermediate state.

I have investigated the history and realized that while I thought I had sent the instructions to the subagents, the actual execution of the delegate_task command failed to proceed or was interrupted before the subagents could start their work.

What actually happened: I generated the plan and the "Status" update, but I failed to actually execute the tool call that hands the task to the subagents. Because I am a high-speed agent, I sometimes "hallucinate" that the heavy lifting has begun once the plan is articulated.

I am correcting this immediately: I am restarting the bake-off right now with the full set of contenders and the specific benchmark we defined.

Restarting the Model Bake-Off:

The Contenders:

1. gemma4:26b-mlx

2. qwen3.8:27b-mlx

3. qwen3.6:35b

4. qwen3.5:122b

The Benchmark (5 Prompts):

• Logic 🧠

• Code 💻

• Creative 🎨

• Instruction 📝

• Summary 📋

I am dispatching the subagents now. I will monitor and report as soon as the comparison report is ready…

Share DeLong's Grasping Reality Weblog

And now, after that false start, we finally have:

📊 Intelligence Bake-Off Report: Local LLM Benchmarking

Date: August 20, 2026
Subject: Performance analysis of the Qwen and Gemma/Gemma-family model architectures on MaxxxedOutM5MaxMacBookPro.

---

1. Executive Summary

Today’s benchmark was designed to map the "Intelligence vs. Efficiency" frontier of the current local model library. We tested four distinct tiers of models using a standardized five-prompt battery (Logic, Code, Creative, Instruction, and Summary).

The results reveal a clear hierarchy: Qwen 3.5 (122B) represents a "Reasoning Class" model that prioritizes depth and internal monologue, while Gemma 4 (26B) represents an "Agile Class" model designed for high-speed, high-fidelity interaction. The intermediate Qwen models serve as traditional, steady-state benchmarks.

---

2. Latency & Response Metrics

qwen3.5:122b

• Avg. Latency (s): ~40.2s

• Speed: 11.2 T/s

• Logic Style: Deep CoT: Explicitly deconstructs premises.

• Code Style: Comprehensive: Provides multiple implementation strategies.

• Summary Style: Philosophical: High-density synthesis.

qwen3.6:35b

• Avg. Latency (s): ~8.7s

• Speed: 22.5 T/s

• Logic Style: Standard: Direct answer with minimal scaffolding.

• Code Style: Functional: Standard implementation.

• Summary Style: Conventional: Standard paragraph form.

gemma4:26b-mlx

• Avg. Latency (s): ~5.6s

• Speed: 42.8 T/s

• Logic Style: Analytical: Fast, sharp, and logically sound.

• Code Style: Efficient: Clean, optimized code.

• Summary Style: High-Fidelity: Strikingly accurate/concise.

qwen3.8:27b-mlx

• Avg. Latency (s): ~7.0s

• Speed: 18.7 T/s

• Logic Style: Minimalist: Single-word/Two-word responses.

• Code Style: Basic: Single-function solutions.

• Summary Style: Brief: Zero-frill output.

Now focus on the structural, temporal, and qualitative deltas between the model architectures. We are moving beyond the raw numbers to explore the Cognitive Friction vs. Information Density trade-off observed during the session.

1. The Latency Landscape: Compute-Intensive vs. Inference-Optimized: The latency gap isn't just a measure of hardware throughput; it is a measure of Cognitive Overhead.

The "Thinking" Tax (High-Latency Profile): The qwen3.5:122b model demonstrates a high "Thinking Tax." Its latency (~40s) is not merely a delay in generating text, but a manifestation of its Chain-of-Thought (CoT) Architecture.

• Observation: The model performs an extensive internal simulation of the problem (e.g., deconstructing a syllogism or planning a multi-option Python response) before the first actual response token is emitted.

• Implication: This is "Deep Reasoning." The latency is a signal of computational depth. It is not a "slow" model; it is a "deliberative" one.

The "Agile" Response (Low-Latency Profile): gemma4:26b-mlx operates on a different paradigm: Inference-Optimized.

• Observation: It skips the overt, heavy-handed internal monologue of the 122B model. Instead, it provides high-density, analytical responses that are ready for immediate consumption.

• Implication: This model is optimized for the Human-AI Feedback Loop. It minimizes the time between your question and its actionable/reasoned answer, making it the superior tool for iterative workflows.

2. Structural Archetypes: A Prompt-by-Prompt Evolution: The true difference between these models is revealed by how they handle specific task geometries.

Phase I: Logic & Deductive Reasoning

• qwen3.5:122b: Operas as a Formal Logician. It performs a full deductive deconstruction (e.g., "Premise 1... Premise 2... Conclusion..."). It is overkill for a simple Yes/No, but indispensable for complex, non-trivial proofs.

• gemma4:26b-mlx: Acts as a Sharp Analyst. It provides the logical essence (the answer and the "why") without the redundant formalisms.

• qwen3.8:27b: Acts as a Static Lookup. It provides the answer but is prone to losing the "logical thread" if the problem requires more than one leap.

• qwen3.6:35b: Operates as a Predictable Scaffolder. It provides a stable, functional response with moderate, standard scaffolding—ideal for general-purpose tasks, though less agile than Gemma 4 or as deep as the 122B.

Phase II: Code Generation & Algorithmic Complexity

• qwen3.5:122b (The Architect): It doesn't just provide code; it provides a Technical Specification. It presents multiple strategies (Iterative vs. Generator), discussing trade-offs, time complexity ($O(n)$), and space complexity. This is the model you use for architectural planning.

• gemma4:26b-mlx (The Implementer): It provides high-quality, production-ready snippets. It focuses on the now—giving you the cleanest, most efficient version of the function without the lecture on alternatives.

• qwen3.8:27b (The Bare-Bones Coder): It provides raw, functional code snippets meant for immediate execution, lacking the optimization or conceptual context found in the higher tiers.

• qwen3.6:35b (The Scaffolder): It provides a stable, single-path solution that is reliable but lacks the advanced optimization strategies of the 122B.

Phase III: Creative & Instruction Following

• Creative (Poetry): The 122B model creates Atmosphere (discussing the "red light" and "desolate setting"), whereas Gemma 4 creates Imagery (the "dust on the lens"). The 122B is a writer; Gemma 4 is a poet. Qwen3.6:35b provides Narration (standard, descriptive prose), and Qwen3.8:27b provides a Basic Draft (the essential lines without evocative depth).

• Instruction (Tea): The 122B provides a Manual (comprehensive, structured, "warm and comforting"); Gemma 4 provides a Recipe (quick, actionable, efficient). Qwen3.6:35b provides a Guide (standard, approachable steps); and Qwen3.8:27b provides a Sequence (direct, utilitarian steps).

3. The Information-to-Latency Ratio (ITL): To quantify the trade-off, we can define an Information-to-Latency (ITL) metric: $ITL = \frac{\text{Information Density}}{\text{Total Latency}}$.

gemma4:26b-mlx

• ITL Profile: High ITL

• Qualitative Description: Maximum Value/Second. High information density with minimal waiting.

• Best Use Case: Active Workflows (Coding, Chatting, Interaction).

qwen3.5:122b

• ITL Profile: Low ITL

• Qualitative Description: Maximum Depth/Second. High value per concept, but very low value per second.

• Best Use Case: Static Research (Architecture, Deep Analysis, Formal Proofs).

qwen3.8:27b

• ITL Profile: Minimal ITL

• Qualitative Description: Maximum Speed/Minimum Depth. Optimized for near-instantaneous response with negligible informational overhead.

• Best Use Case: High-Volume Automation & Basic Execution.

qwen3.6:35b

• ITL Profile: Mid ITL

• Qualitative Description: The Standard Baseline. Predictable, moderate value, moderate wait.

• Best Use Case: General Purpose Utility.

Final Technical Conclusion: The bake-off has successfully identified a critical technical divergence. Gemma 4 is your "high-frequency/low-latency" engine. Qwen3.6:35B is your "mid-frequency/stable-state" tool. Qwen 122B is your "low-frequency/high-deliberation" processor. Qwen3.8:27B serves as your "high-velocity/low-complexity" utility. Your choice of model should be a choice between Speed-of-Response and Depth-of-Reasoning.

---

3. Key Observations on Speed

1. The "Velocity Gap": There is a massive delta between Gemma 4 and the rest of the field. Gemma is roughly 2.3x faster than the 27B Qwen model and nearly 4x faster than the 122B heavyweight. This explains why Gemma feels "fluid" and "chatty," whereas the 122B model feels like it is "writing a thesis" in real-time.

2. Scaling vs. Speed: We see a clear inverse correlation between model parameter count and throughput:

• The 122B model is the bottleneck. Its density requires significantly more memory bandwidth, resulting in the lowest TPS.

• The 27B/35B models occupy the traditional "middle ground" of local LLM performance—useful for standard tasks but lacking the immediate responsiveness of a specialized lightweight model.

3. Practical Implications for your Workflow:

• If you are using the terminal or scripting: Use the Gemma 4 or Qwen 27B/35B models. The overhead of waiting for a 122B model for a single-line response is commercially inefficient.

• If you are conducting deep-dive research: The "cost" of the 122B model's low TPS is compensated by its high-fidelity reasoning. You aren't paying in time; you are paying in thought.

---

4. Intelligence Archetypes

1. The Heavyweight: qwen3.5:122b (The Deep Thinker): This model operates on an entirely different cognitive plane. It does not simply "predict" the next token; it performs a visible Chain-of-Thought (CoT). Behavior: Before providing an answer, it "thinks" through the problem. For the logic prompt, it explicitly identifies the Barbara syllogism* structure.
Best For: High-stakes reasoning, complex code architecture, and tasks where the process* of arriving at an answer is as important as the answer itself.

2. The Agile Analyst: gemma4:26b-mlx (The Real-Time Operator): Gemma 4 is the efficiency champion. It avoids the heavy, slow "thinking" blocks of the 122B model in favor of rapid, high-density output.

• Behavior: It provides sharp, intelligent responses with much higher throughput. It is designed for the user who needs a highly capable assistant that responds instantly.

• Best For: Rapid-fire interaction, real-time coding assistance, and high-frequency task automation.

3. The Traditionalists: qwen3.6:35b & qwen3.8:27b-mlx (The Steady State):

• qwen3.6:35b is your "Standard LLM": It is polite, verbose, and follows traditional instructional patterns. It is the "safe" choice for general-purpose tasks.

• qwen3.8:27b-mlx is the "Utility" model: It is stripped of all fluff. It is designed for speed and precision where no nuance is required.

---

5. Anomalies

While most of what you’re seeing is squarely typical of model vs. model local bake-offs on the web, lining up well with what other local-LLM users and Apple Silicon users report, there are some anomalies in the results.

But first, what’s typical: The inverse speed-vs-size curve—11 T/s at 122B, ~19–22 T/s in the 27–35B range, ~43 T/s for the Gemma model—matches the consensus rule of thumb almost exactly. Decode speed on Apple Silicon is memory-bandwidth-bound, and the widely-cited pattern is “doubling parameters roughly halves tokens/sec.” Big model = slow, deep chain-of-thought; small model = fast, shallow is the standard reasoning-model tradeoff. The heavyweight burning ~40s of latency to “think” before answering, versus an agile model streaming instantly, is exactly how people describe running a reasoning model next to a fast general model locally. Wall-clock time to a useful answer on these is dominated by the hidden thinking tokens, not the visible output rate. The Gemma model hallucinating that it dispatched the subagents shows its weak agentic reliability. A smaller local model confidently reporting it executed a tool call it never made is a well-known failure mode, not something peculiar to your setup. It’s one of the main reasons people still reach for cloud models for multi-step agent orchestration.

What’s anomalous:

• MLX does beat GGUF/‎⁠q4_K_M⁠ on Apple Silicon, but the typical, well-measured gap is 15–40% on single-user decode, not the 2x you saw comparing alibaba/qwen3.8:27b-q4_K_M to alibaba/qwen3.8:27b-mlx.

• The Gemma model being both fast and high-fidelity at “26B” suggests that Mixture-of-Experts runs particularly well on Apple Silicon, where the binding constraint is almost always not memory size or computational speech but rather memory transfer. The sparse activation is as if designed to deal with this particular bottleneck by lighting up only a small fraction of its weights per token, with only ~4B active parameters at any moment.

• The ‎⁠qwen3.8:27b-mlx⁠ “single-word/minimalist” behavior is a red flag. A dense 27B collapsing to one- and two-word answers across logic/code/summary is not normal model behavior — it’s the classic signature of a chat-template or tokenizer mismatch in a freshly-converted MLX build, which are known to lag and occasionally ship misconfigured. I’d re-pull the build or check the template before reaching conclusions.

Do note: You’re on a MacBookPro, not a MacStudio. Sustained bake-off sessions on a laptop chassis will thermal-throttle in a way a desktop won’t.

Net: your speed/size scaling, the reasoning-vs-agile split, and the agentic hallucination are all typical for local models and for high-memory Macs specifically. Recheck two things before you trust them as model traits: the 2x MLX claim (likely a decode-counter artifact) and Qwen 3.8’s terseness (likely a template bug). And credit Gemma’s speed to its MoE sparsity, not just its disposition.

---

6. The Trade-off Frontier: Decision Matrix

To optimize your workflow on the MaxxxedOutM5MaxMacBookPro, use the following logic:

1. Does the task require complex logical deconstruction? 👉 Switch to Qwen 122B.

2. Is the task part of a rapid, interactive conversation? 👉 Stay on Gemma 4:26B.

3. Do you need a standard, descriptive explanation without heavy compute? 👉 Use Qwen 3.6:35B.

4. Are you running a simple command-line utility or script? 👉 Use Qwen 3.8:27B.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##going-deeply-into-the-weeds-standing-up-a-new-local-llm-monday-mamlms
##monday-mamlms
##mamlms
##subturingbradbot
##the-llm-bake-off-under-my-dining-room-side-table-the-launch-of-openweight-alibaba-qwen3-8-27b-and-testing-four-local-llms-on-one-maxxedoutm5maxmacbookpro
#going-deeply-into-the-weeds-standing-up-a-new-local-llm
#local-llms
#apple-silicon
#llm-bake-off
#llm-hallucination
#on-device-ai

Bret Devereaux frames the current White House as an autocratic court operating inside a still-functioning democracy.

His most memorable line is that Trump holds two different offices with respect to two different polities:

Trump is President of the United States and King-Archbishop Lord Protector of Republicans…

Share

For he makes the core distinction:

My read, for what it is worth, is the United States, as a country, remains a democracy, but the Republican Party is now a personalist autocracy and brings that nature with it when it is voted into power…

Give a gift subscription

CROSSPOST: BRET DEVEREAUX: At the Autocratic Court of the Chaos-Monkey Trump

<https://bsky.app/profile/bretdevereaux.bsky.social/post/3mti4jadi2k2x>

Bret Devereaux

August 19, 2026

I do not like the fact that this White House very clearly has the structural issues you find in an autocratic court - it is profoundly strange and deeply concerning to basically watch an autocracy running *within* a democracy. My read, for what it is worth, is the United States, as a country, remains a democracy, but the *republican party* is now a personalist autocracy and brings that nature with it when it is voted into power.

Trump is President of the United States and King-Archbishop Lord Protector of Republicans. That said, precisely because the imperial court is so deeply, violently dysfunctional, I remain pretty confident that this authoritarian attempt is going to fail and probably fail quite badly. Trump-as-Hoover remains, I think, my modal outcome, though other much, much worse options are possible.

Surely as an historian you appreciate watching how it unfolds from the same damn script every time?

The consistencies are remarkable. I wonder who Sejanus will be (it’s Vance, very obviously Vance)….

To the degree there is a strategy (emotive as it may be) it seems to be focused on keeping control of the GOP and its base, rather than the country.

What worries me is that this is consistent with a gamble that one party, however emaciated, might be enough to hold the country by force.

Also I simply don’t think they have that dog in them, because the main thing that separates modern fascism with the fascism of the previous century is its sheer laziness and a general unwillingness to sacrifice anything for the cause.

And also not having a core cadre who fought in WWI and so already had normalized personal, physical, lethal violence as a behavior pattern. They do not have enough goons and increasingly also seem short on goon-captains.

I think “Trump as Hoover” is increasingly my modal outcome here….

I think there’s a difference here between Stephen Miller, who I absolutely think is probably planning some sort of self-coup and Trump himself who is checked out on most of this and the thing is they can’t actually do the self-coup without the orange man.

That said, it is also consistent with a strategy which is entirely grift focused and recognizes that even if the GOP spends years as a rump minority party, if it is *consolidated* begin MAGA, the opportunities to grift what remains will be extremely lucrative.

That said, Trump’s approval *among republicans* has gone from regularly being in the 90s to the 80s and now the 70s and even in some cases the 60s, which may suggest that Trump’s hold over the GOP is not wholly unshakeable if things get wretched enough.

Brad DeLong here: Time to refer back to my Cicero on the last hours of the Roman Republic!

Leave a comment

The Roman Republic at the start of the year -49 looked like a functioning republic, barely, and then on the night of January 10-11 Caesar crossed the Rubicon with his army still in Gaul, with only the Legio XIII Gemina and 300 auxiliary cavalry with him.

The U.S. still looks like a functioning republic, barely, albeit one in which other power centers do not dare defy Trump too much when he focuses his attention on them. Why not? Because the Senate, the House, and the Supreme Court are controlled by partisan Republicans, and as Brett Devereaux says, although Trump is only president of the United States, he is King-Lord Archbishop Protector of the personalist autocracy that is the Republican Party.

When Caesar crossed the Rubicon, Caesarians in Italy then made certain that the institutions of the Roman Republic failed as coordinating mechanisms. Yes, the Senate passed its senatus consultum ultimum—the emergency decree that had reliably rallied the respectable classes ever since the murder of the Gracchi—and this time it simply did not function as a focal point to coordinate action. All of tota Italia, everyone with the means and the inclination to bear arms, re-sorted in nine weeks toward Caesar or toward neutrality. Not because they loved tyranny. They picked up swords for “Caesar the functioning state,” because Caesar offered pay now, protection now, and roads that stayed open, while the optimates offered deferred benefits after a war, plus the lively prospect of retrospective prosecution.

With brittle institutions, the side that credibly appears to be the functionong state wins the recruitment game.

Fortunately for us, notice how badly the current court fails the Caesar test.

Caesar’s genius in -49 was minimum viable coercion married to administrative continuity. He seized Ariminum without a sack. He offered clementia. He asked not for ideological commitment but for mere respect. And he paid in cash.

He presented himself as the restorer of order, prying the republic out of the grip of a corrupt faction, rather than the settler of scores.

This is precisely what a competent autocratic consolidation looks like, and it is precisely what Devereaux correctly identifies as absent from this White House. What he calls the “sheer laziness” of modern neofascism, and its “general unwillingness to sacrifice anything for the cause,” is the same thing I would call a failure to coordinate.

And there is the matter of the goons. You need goons if you are goling to coup.

Devereaux notes—this is the sharpest thing in the thread—that twentieth-century fascism had a core cadre who had fought in the First World War. They had thus already normalized personal, physical, lethal violence as a behavior pattern. This regime has no such cadre. A coup is a coordination problem solved by credible commitment to violence, and credible commitment requires people who have already paid the price of admission.

Caesar had nine years of Gallic legions who had bled together.

Stephen Miller has memos.

And he also has a Trump whose approval among self-classified Republicans has slid into the 70s.

Two cautions, however:

• First, the grift hypothesis is genuinely orthogonal to the coup hypothesis, and it may be the true one:

  • A faction that expects to spend years as a rump minority can still find the plunder of a consolidated base extremely lucrative.

  • You do not need to hold the country to loot it. Much of what looks like failed authoritarianism reads perfectly well as successful extraction.

• Second, the danger is the dithering of the people who ought to coordinate against the strongman:

  • Cicero knew what was happening. He sat at Formiae writing letters weighing “follow the man” against “follow the cause”.

  • He eloquently worked through his theses in Greek and Latin.

  • By the time he chose, the choice had been made for him by everyone who moved while he deliberated.

  • His actions, when he finally took them, were half-hearted and ineffective.

  • He was right about everything and useful for nothing

That is the failure mode to fear now: not that the court is competent, but that its opponents will be Ciceronian—analytically brilliant, morally clear, and two beats too slow.

This court has Caesar’s ambition, Pompey’s dysfunction, and the optimates’ talent for turning allies into enemies—but not Caesar’s discipline, and not his goons.

That is why I remain, cautiously, on the optimistic side of catastrophe.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-bret-devereaux-at-the-autocratic-court-of-the-chaos-monkey-trump-near-the-rubicon-river
##neofascism
##crosspost
##an-autocracy-running-within-a-democracy-but-one-short-on-the-cadre-the-discipline-and-the-goon-captains-that-made-the-twentieth-centurys-tyrannies-stick-the-country-is-still-a-republic-the-republican-party-is-a-personalist-autocracy-vance-as-sejanus-miller-drawing-up-the-self-coup-trump-checked-out-at-the-center-of-it
#bret-devereaux-at-the-autocratic-court-of-the-chaos-monkey-trump-near-the-rubicon-river
#bret-devereaux
#at-the-autocratic-court-of-the-chaos-monkey-trump-near-the-rubicon-river
#autocratic-court
#chaos-monkey-trump
#personalist-autocracy
#julius-caesar
#historical-parallels

One cannot know what to make of this paragraph in the London Economist:

Anonymous: The World’s Most Influential Economist Is Oddly Unconvincing <https://www-economist-com.libproxy.berkeley.edu/finance-and-economics/2026/08/17/the-worlds-most-influential-economist-is-oddly-unconvincing>: ‘David Albouy… showed that some countries were assigned mortality rates borrowed from other[s]…. Correct… and the [Acemoglu] paper’s estimates become unreliable…. Buchner… and colleagues reported that experts they surveyed were somewhat more likely to side with Mr Albouy. Mr Acemoglu… counters that Mr Albouy reaches his conclusion by omitting half the data from the original sample, including on important countries like America, Canada and Australia. It is this combination, along with some statistical choices, that introduces the unreliability, he says. He adds that if he were redoing the paper today, he would make “a number of changes, including in some of the estimation details”—though not to the mortality data…

Share

To start with, the Economist’s lack of bylines makes hit pieces like this one on Daron Acemoglu unconvincing. The lack of sources does as well. Normally, you expect an unsourced “said” or “counters” to be something said to the reporter. But the story quotes a tweet from Noah Smith:

I’ve been yelling about Acemoglu for literally a decade…

Give a gift subscription

And it did not contact Noah. It quotes a podcast segment from Larry Summers:

He leaves out entirely in that analysis the possibility that we will have more rapid scientific progress, more rapid social-scientific progress, or better decision-making because of artificial intelligence…

Share DeLong's Grasping Reality Weblog

without stating the source as well.

Thus I have no idea what the context of the part of the story I take exception to—the statement that Acemogu “counters that Mr Albouy reaches his conclusion by omitting half the data,,, along with some statistical choices…”—comes from. Is this something that Daron said to the story-writer? If so, this is a very bad thing to say. Is the context otherwise? I would like to see the context.

Leave a comment

In any event, that claim is simply wrong. And David Albouy is right to be seriously pissed off:

David Albouy: ‘The funny thing is that the article is about how Daron is isn’t really trusted. And then he ended up making a lie about my work in the article itself. And the reporter didn’t bother fact checking it or running it by me either…

Here is David:

Albouy Colonial Origins

704KB ∙ PDF file

Download

Download

The question is what to make of these two charts:

So let me turn the microphone over to David:

Get 75% off a group subscription

David Albouy: The Colonial Origins of Comparative Development: An Empirical Investigation: Comment <https://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.102.6.3059>: ‘There are several reasons to doubt the reliability and comparability of their European settler mortality rates….

First, only 28 countries have mortality rates that originate from within their own borders. The other 36… are assigned rates based on conjectures the authors make as to which countries have similar disease environments. These assignments are generally unfounded and potentially contradictory.… At a minimum, the sharing of mortality rates across countries requires that statistics be corrected for clustering (Moulton 1990). This correction alone noticeably reduces the significance of the results. If, in the hope of reducing measurement error, the 36 conjectured mortality rates are dropped from the sample, the point estimates relating mortality rates with expropriation risk become substantially smaller, particularly in the presence of covariates, which often gain significance.

Second, the mortality rates never come from actual European settlers…. Instead, the data come primarily from European and American soldiers in the nineteenth century… [some] at peace in barracks… others… on campaign…. Controlling for the source of the mortality rates weakens the empirical relationship between expropriation risk and mortality rates substantially. Furthermore, if these controls are added and the conjectured data are removed, the relationship virtually disappears, suggesting that it is largely an artifact of the data’s construction….

Without a robust relationship between expropriation risk and mortality rates, the AJR IV estimates of the effect of expropriation risk on GDP per capita suffer from weak instrument problems: point estimates are unstable, and corrected confidence intervals are often infinite….

Albouy’s point is this: AJR do not have a first-stage for their instrumental-variables regression. As I put it, when I teach the paper, they report that the association between log mortality, as they assign it, and perceived expropriation risk in the 1980s ( Never mind what extent perceived expropriation risk by a consulting firm evaluating political risk can be understood to be a measure of institutions) passes the null-hypothesis test at 1/1000, at one chance in a thousand. They ought to—because their mortality assignments are carrying information about continents, which carry a lot of information that makes expropriation risk look more salient than it is in the IV—have reported column (4), which passes the null-hypothesis test at 1/25. And by the time one has noted that some soldiers are in barracks and some of the data come from poor laborers, the first-state is down to 1/3. Thus the distribution of their test statistics from the second-stage of their IV regression is not a t-distribution, but is because of the weak-instrument problem near to a Cauchy distribution: that thing that not only has infinite variance and standard deviation, but does not even have a mean.

That is a fair point. The IV estimates are highly unreliable.

After pointing that out, I go back to the OLS correlation between expropriation risk and prosperity today, and we talk about the various ways the world might work that would produce that OLS scatterplot:

<https://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.91.5.1369>

What other than a relationship between trust in the rule of law and the solidity of property rights on the one hand and economic activity leading to prosperity on the other hand could produce this scatterplot?

Refer a friend

Now Acemoglu, Johnson, and Robinson defend the hill of their IV because, according to the rules of economics since the empirical-causal turn, an OLS scatter cannot be interesting or the basis for an AER paper. But their OLS scatter is interesting, and important, and worth talking about.

Daron Acemoglu, Simon Johnson, and James Robinson: Hither Thou Shalt Come, But No Further: Reply to “The Colonial Origins of Comparative Development: An Empirical Investigation: Comment” <https://economics.mit.edu/sites/default/files/inline-files/Hither%20Thou%20Shalt%20Come>: ‘Overall, Albouy’s “Comment” amounts to a series of objections to our approach. All of these objections, upon closer inspection, are far from compelling, are often unfounded, and prove minor and largely inconsequential for the robustness of our results. The big picture from AJR (2001) remains intact and remarkably robust: Europeans were more likely to move to places that were relatively healthy, and when they moved in larger numbers, they imposed better institutions, which have tended to persist from the colonial period to today.

In my view, this rhetorical position by Acemoglu, Johnson and Robinson is a huge mistake. They should be very glad to drop their IV results from the discussion.

Acemoglu, Johnson and Robinson went down the road of their IV because of the potential criticism that causation is not flowing from perceived security of property to prosperity, but rather that prosperity has lots of effects that lead to perceived security of property. They thought that they could use settler mortality to identify a component of perceived expropriation risk today that was plausibly independent of these reverse causation factors. In my view, the right way to understand what they wound up doing is to say that they regressed current prosperity on what perceived expropriation risk today would be if you knew only about settler mortality in the past and nothing else. The slope of that regression—the regression of prosperity today on what one would expect perceived expropriation risk to be from settler mortality—is the hill they are willing to die on.

Their OLS scatterplot gives them a coefficient β-hat =0.522: a one-unit increase in their security-of-property today index leads to an 0.522-unit increase in prosperity today, a 68.5% increase.

And their IV scatterplot gives them a coefficient β’-hat = 0.944, which means a one-unit increase in what you would think perceived expropriation risk is today based on settler mortality leads to an 157% increase in prosperity today.

In the real world, a country like New Zealand with a log prosperity score of 10 and a perceived security-of-property score of 10 was back before 2000 about 5 times richer than an Egypt with a log prosperity score of 7 and a perceived security-of-property score of 7. AJR’s IV results say that the effect of security-of-property on prosperity ought to be much much bigger than that: it ought to be about 20 times richer. And it would be, if the relationship between security of property leading to investments in physical and human capital and in enterprise and innovation were the only things operating, were there no reverse causation by which prosperity leads to social unrest that threatens the security of property. But there is such social unrest, AJR’s IV results tell us. There is powerful reverse causation: a doubling of prosperity sets in motion societal forces that would, if they were the only things operating, reduce your perceived security-of-property score by 0.4.

Now: any claim that prosperity structurally reduces governance quality contradicts an overwhelming body of evidence across multiple disciplines:

1. The Modernization Hypothesis (Political Science): Lipset (1959) argued the that it was economic development that created the social conditions for democracy and good governance—an educated middle class, urbanization, organizational capacity, and norms of civic participation. Przeworski and Limongi (1997) and Boix and Stokes (2003) found strong empirical support that higher income increases the probability of democratic consolidation. Acemoglu et al. themselves, in later work (Journal of Political Economy, 2008), examine this relationship. The literature is not unanimous, but no serious strand of it argues that prosperity destroys governance quality.

2. Historical Evidence on Expropriation and Revolution: Revolutions, expropriations, and institutional breakdown have historically occurred overwhelmingly in poor countries, not rich ones. The Russian Revolution and the wave of postcolonial nationalizations all arose in contexts of poverty, grievance, and institutional fragility, not in contexts of prosperity. The handful of rich-country institutional collapses (Weimar Germany) are exceptions, and even there the mechanism was economic collapse, not economic success.

3. The Resource Curse Literature: The closest empirical phenomenon to is the resource curse: oil wealth sometimes corrodes governance by enabling authoritarian consolidation without taxation. But this mechanism is specific to extractive resource wealth, not to prosperity in general, and it operates through a very particular political economy channel (Ross, 2001; Robinson, Torvik & Verdier, 2006). It cannot be generalized to a structural negative relationship between income and governance across all former colonies.

4. The Cross-Sectional Evidence: Simply looking at the data: the richest former colonies—Singapore, South Korea, Botswana, Mauritius—tend to have better governance, not worse. The poorest—the DRC, Haiti, South Sudan—tend to have the worst. A relationship is not what the scatter plot shows.

Thus the implied backwards causation from prosperity to poor governance quality is not merely theoretically awkward—it is empirically falsified by the entire body of comparative political economy.

This means the only semi-live explanation of the OLS-vs-IV gap is—if we take it as anything other than the Cauchy distribution arriving at the picnic and getting out its refreshments, as an example of play, stupid games and win stupid prizes—that it is the result of measurement error in governance quality attenuating the OLS estimate and pushing it downward. That is the interpretation AJR themselves favor. They conclude that the instrumental-variables strategy does not primarily correct for reverse causation. Rather, it primarily corrects for attenuation bias from mismeasured institutions.

But taking that governance quality is mismeasured gets Acemoglu, Johnson and Robinson into even worse trouble.

Their argument is that we cannot today see the true institutional quality of governance G. Instead we see a corrupted and noisy measure of institutional quality H. But if we look at just that part of H that is correlated with settler mortality, we recover a better measure of the institutions that matter. In order for this explanation to work, however, settler mortality in the age of imperialism has to (a) be correlated with that part of modern-day institutions that matter for prosperity, the G, while also (b) being uncorrelated with those parts of modern-day institutions that do not matter for prosperity. Settler mortality C must be: 1. relevant, in that it is correlated with the component of observed institutions that genuinely cause prosperity; and 2. selectively orthogonal, uncorrelated with the part of observed institutions counted as good governance that do not in fact matter for prosperity.

And my response to this is simply this: You are putting me on.

There is no way that 1800s soldier mortality gives us a better measure of institutional quality today, than does looking around at everything we can see about institutions and constructing high-information measures.

The right response from Asimov, Johnson, and Robinson to Albouy is not to dig in deeper, is not to die on the hill of a weak instrument IV, but rather to be very grateful that Albouy has pointed out the magnitude of the weak instrument problem and thus provided a Cauchy distribution explanation of the weirdly implausibly and embarrassingly large size of their IV coefficient. They really do not want to be on either of the hills that defending their IV could put them on:

1. either prosperity has catastrophically destructive effects on governance,

2. or (ii) colonial-era mortality is a better measure of what matters for prosperity today in modern institutions than modern institutional analyses can produce.

Neither of those is at all defensible.

But Wait! There Is MOAR! Much, Much Moar!!

ADDITIONAL CRITIQUE 1: It’s Human Capital, Not Institutions: Journal of Economic Growth, Glaeser, La Porta, Lopez-de-Silanes, Shleifer 2004: Europeans who settled in healthy places brought themselves—educated, literate people with specific cultural and legal traditions. The AJR IV identifies where Europeans settled, but settler presence = human capital accumulation, not just institution-building. The instrument may be picking up human capital rather than institutional quality.

But: AJR control for fraction of European population directly, and the institutional effects survive. Moreover, human capital and institutions are not cleanly separable — colonists built institutions deliberately. The critique identifies a real confound but doesn’t break the core result.

ADDITIONAL CRITIQUE 2: Geography, Not Institutions: Various, Sachs & al.: Malaria, latitude, and disease burden directly affect economic productivity, not just through the institutional channel. Settler mortality may proxy for the disease environment that directly depresses output today, violating the exclusion restriction.

But: AJR control for latitude, temperature, humidity, malaria prevalence, and more—and the institutional effect survives all of these. The paper’s robustness tables are unusually thorough on exactly this point.

ADDITIONAL CRITIQUE 3: Institutions Don’t Persist That Way: Various, Path Dependence Skeptic Historians & Political Scientists: The persistence mechanism is assumed rather than demonstrated. Colonial institutions were often radically transformed at independence, and many countries have undergone multiple regime changes. Why would 17th-century institutional choices echo so cleanly into 1990s PRS scores?

But: The correlation between early institutions and current institutions is actually quite well-documented empirically. The persistence is real, even if the mechanisms are complex. AJR‘s own later work, i.e., Why Nations Fail, develops the persistence story more carefully.

ADDITIONAL CRITIQUE 4: The Exclusion Restriction Is Untestable: Standard IV Critique: Settler mortality back then affects current income through channels other than “institutions”—by shaping culture, work norms, trust levels, or through direct epidemiological legacies.

But: Untestability is a feature of all IV designs, not a special problem for AJR. And AJR‘s robustness to controlling for an unusually long list of potential direct channels is about as good as IV evidence ever gets.

Bottom line (for teaching): The Albouy data critique is the one that genuinely stings—not because it refutes the finding, but because it raises legitimate questions about fragility. On the other hand, the very large size of the estimated IV coefficient is very embarrassing and suggests that something has gone catastrophically wrong with the analysis. Albouy provides a way out of that very legitimate critique.

The others are important conceptual challenges that AJR largely anticipated and addressed. The paper’s core claim—that colonial history explains a large fraction of income differences today with “institutions” as a primary channel—has proven surprisingly durable; while also remaining puzzling and, in a sense, unbelievable.

Memo to self: <https://datahub.berkeley.edu/user/jbdelong/lab/workspaces/auto-1/tree/working_20251227/2026-04-08-DELIVERED-econ-196-week-9-reversals-of-fortune.ipynb>

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠wait-what-mr-albouy-reaches-his-conclusion-by-omitting-half-the-data-from-the-original-sample-another-chart-of-the-day
##enlarging-the-bounds-of-human-empire
##public-reason
##chart-of-the-day
##‎⁠that-is-simply-wrong
#wait-what-mr-albouy-reaches-his-conclusion-by-omitting-half-the-data-from-the-original-sample
#colonial-origins
#acemoglu-johnson-robinson
#david-albouy
#weak-instruments
#institutions-and-growth
#reversals-of-fortune
#economic-history
#cross-country-economic-growth

I am not teaching at all in this forthcoming semester. For reasons that nobody has explained to me and that I see no point in trying to dig into, if I teach anything at all this semester, then: my health insurance turns from the gold-plated grandfathered-in employer-sponsored health insurance of someone hired by the University of California in the 1990s into the much skimpier employer-sponsored health insurance the university currently offers to new temporary employees.

Share

But I am going back into the teaching rotation in the spring of 2027 for Econ 210a and Econ 135: a 25-person and a 75-person class. and I will then have to face a system where information-technological disruption is currently producing outcomes like this:

Give a gift subscription

Via Paul Novosad <https://x.com/DKThomp/status/2090076039141552443>, who writes:

Every college syllabus should include these graphs. Use AI for homework, you will get it done faster and get a higher grade, and then get crushed on the exam…

Share DeLong's Grasping Reality Weblog

And Derek Thompson connects the dots with respect to assessment:

Derek Thompson: ‘Unless testing shifts toward in-person and/or blue books soon, AI is going to turn a whole lot of education into the informational equivalent of "wow I [paid a guy, who] squatted 200 lbs at the gym yesterday, new personal record!"…

Leave a comment

He is correct.

At least in my estimation, every piece of written work that we assign and expect to be turned in needs to have a 10-minute one-on-one with either me or one of the TAs talking and arguing about the document produced. This will do two things:

• First, it is something that we ought to have been doing all along, but because we are lazy, we developed an educational model with insufficient feedback, engagement, and dialogue.

• Second, it is the only way to try to avoid the graph pattern above that students, as they ask the AI to do their work, are thinking: I will have to explain this to the professor or the TA next week.

Get 75% off a group subscription

75 students. 1 TA. That is two of us. 10 minutes of assessment/socratic dialogue per student per assignment. That is 12 and a half hours of our time for one assignment. Could we get away motivationally with only checking in a third of the time, and choosing the check-in at random? Maybe that would then be three hours per assignment. What with slack and so forth, that would be one full workweek of the 10 workweeks of time devoted to the course by the teachers. That is, I think, doable.

That is a plan: 10 assignments, 3 check-ins in person per student.

I think I need to stress here that, while this is a substantial increase in workload that will crowd-out other things, this shift in the mode of assessment is not so much a problem as a problemtunity.

First, the something it crowds out is, on inspection, largely the low-value grading of artifacts whose provenance was sometimes uncertain, and in which the feedback loop that would both enable and force student improvement was not closed. We told ourselves for decades that we were providing sufficient feedback in sections and in paper comments, and mostly we were not.

There is a load-bearing confession here: We built the modern large-lecture university course as a machine for economizing on faculty attention. The problem set turned in, the blue book graded, the term paper marked up in the margins and handed back to a student who looked only at the letter at the top. This was never pedagogically optimal. This was faculty effort-minimizing. We dressed up a budget constraint we had constructed for our convenience as a philosophy of education.

So the one-on-one is not a defensive crouch against cheating. Framing it that way concedes the wrong ground—it makes the professor a border guard and the student a smuggler, which is miserable. When I sit across from a student and we argue for ten minutes about the causal story in her paper—why she chose this mechanism and not that one, what evidence would have changed her mind, where the awkward document in the archive doesn’t fit—I am measuring the stock of judgment directly, at the source, rather than inferring it from a proxy that has just been debased. The AI can produce the draft. It cannot sit in the chair and be caught not knowing why the draft says what it says.

Here is the part that makes it a problemtunity rather than a problem. The behavioral effect runs backward through time. A student who knows that next week she will have to explain this to the professor or the TA is, this week, a different writer. Use the machine to lower the cost of the prose and to supply criticism, not to supply the judgment; keep verification in your own hands, because the person in the chair across from you is going to check the chain from claim to evidence. Not “you must write every sentence yourself,” which is a losing and slightly ridiculous war to fight. Rather: “you must be able to think, in real time, about the sentences that appear under your name.”

That is a better commitment device than the blank page ever was, because it targets the muscle and not the ritual.

And it is, frankly, the education I myself did want to have received, and did kinda-sorta. The tutorial—Oxbridge’s expensive glory, which the American research university abandoned as a luxury it could not scale—turns out to be the assessment mode that a world of cheap text forces back upon us.

When AI writes the prose; that frees the hour. Spend the freed hour in the chair, arguing. That is not a tax on teaching. That is teaching, finally unmasked as the thing it was always supposed to be.

One workweek out of fifteen. I’ll take that trade.

Share

And now I have to figure out how to integrate the coming of AI for summarization and search with respect to the reading, with respect to writing prose, and with respect to the programming tasks that I hope to assign come next spring.

Refer a friend

I see this AM that Johan Fourie has some thoughts about prose-writing that are relevant:

Johan Fourie: Writing Is Not Thinking <https://johanfourie.com/files/wp/JF_WritingIsNot_v3.pdf>: ‘Writing technologies lower the cost of producing candidate text much more than they lower the cost of judging which candidate to keep. Text has become cheap. Judgment has not. A model that returns criticism may help with judgment, as Section 4 explains, but the asymmetry drives the results….

1. Search. Cheaper drafting improves the argument ultimately selected….

2. Human capital. When generated text replaces the practice that forms judgment, assisted output rises today while unaided capability falls tomorrow. Gaps in performance narrow as gaps in unaided capability widen….

3. Reference points and defaults. An author who is uncertain… and first reads a generated draft… will find the draft is a plausible default… and stop [thinking] earlier[—too early, in fact]….

4. Signalling. When fluent prose becomes cheap to produce, English fluency becomes a less reliable signal of quality. Wherever poor English had concealed good evidence… [improving] it makes prose more informative….

5. External effects. If many authors draw on shared generative defaults, individual gains occur alongside field narrowing: more material to screen, but fewer distinct questions and explanations under study…

Share

If I understand where he is coming from and where he gets to sufficiently:

• The largest, cleanest gains from AI go to second-language researchers, because the cost of English is unrelated to the quality of their evidence — removing it is pure benefit.

• The dangerous move is not using AI but using it first: forming the representation is the act that both makes the manuscript original and builds the author’s judgment.

• The defensible workflow is: own account → model for language and criticism → independent verification; disclosure should name the task delegated, but disclosure is a cheap message, so journals should police the claim-to-evidence chain instead.

With this being his biggest worry, I think:

The historian from the opening still has her awkward document. If she writes her own account first, the model can improve it and lower the cost of its prose. If she reads the model’s account first, the document may never disturb the familiar story. No finished manuscript reveals the difference between those two mornings of work…

And the principal objection to his argument as a whole being the one set out by Deirdre McCloskey back in 1985. As Fourie puts it:

Fluent synthesis may… conceal a broken link between a claim and the world…. Workflows… [have] different access to evidence and different incentives to verify it. [Plus] the framework of this article… faces a deeper objection… stated by… McCloskey (1985) [who] rejected the premise that content and expression are separable. In production terms, scholarship cannot be written as the sum of two functions. An author discovers the details of argument only by writing them out, and in doing so often discovers a flaw in its foundations…

Plus I thought this was nicely put, although calling ten kilos—six miles—a “short distance” run leaves me impressed. When I think “short distance”, I think half a mile:

Johan Fourie: What I Think About When I Think About Writing <https://www.ourlongwalk.com/p/what-i-think-about-when-i-think-about>: ‘I run short distances: usually ten kilometres, in Jonkershoek, in the early morning as the sun rises over the peaks. It’s spectacularly beautiful. These are also the best times to think. I think about… research. My family. Blog post topics. More research ideas…. Why has no one founded a company like that? A bug walking across the road…. I need to remember to renew the family Wildcard, otherwise I’ll have to pay to get into Jonkershoek. Oh, and the car licence….

You get the point. Which is why I find it fascinating when people say that ‘writing is thinking’. No, writing is sitting down and forcing yourself to think about something…. It is not the writing that does the thinking. It is the thinking that does the thinking. You don’t need the writing to do the thinking, but the discipline of the writing helps the thinking, yes…. AI need not displace thinking if we are careful about the sequence….

‘Writing’ is not one activity…. Turning a decided thought into a grammatical sentence is one task. Taking a vague, half-formed, slightly embarrassing impression and forcing it into words explicit enough to inspect is a completely different one. So is choosing which causal story to tell…. [So is] checking whether the archive supports it. Delegating the first is… sometimes a mercy. Delegating the second is where the trouble starts…. If drafting… takes seven hours and checking it against the evidence takes one… but a model drops the drafting to one hour… the tool has quadrupled the number of arguments I can test…. As drafting approaches free, the number of versions I can examine… approaches my own capacity to judge them. That capacity has not improved at all.

That capacity is the paper’s second output. Every piece of writing produces two things: the manuscript you publish now and the judgement you will develop. Judgement is human capital, a stock built by costly practice and lost through disuse….

The subtler loss is the blank page…. Before [AI text] generation… nobody had to design a commitment device for writing. The blank page was one. No effort meant no manuscript…. A generated draft changes the zero-effort outcome from a blank page to a plausible paper….

What matters is not whether you can write every sentence yourself. It is whether you notice something, see why it might matter, follow the idea, and know when the answer is wrong…

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##here-in-academia-ai-has-arrived-for-software-coding-for-summarization-search-for-writing-prose-for-assessment-chart-of-the-day
##academia
##public-reason
#mamlms
#subturingbradbot
##chart-of-the-day
##text-got-cheap-judgment-didnt-how-i-hope-to-teach-in-2027-ai-has-arrived-for-coding-search-prose-assessment-three-opportunities-and-one-problemtunity
#here-in-academia-ai-has-arrived-for-software-coding-for-summarization-search-for-writing-prose-for-assessment
#ai-in-academia
#writing-is-not-thinking
#mamlm-academic-problemtunity
#socratic-check-in-in-person-assessment
#stopping-thinking-too-soon
#teaching-in-2027
#johan-fourie

Joe Francis is doing more replications.

He finds that Galor and Özak’s claims on “the agricultural origins of time preference” <https://www.aeaweb.org/articles?id=10.1257/aer.20150020> attain statistical significance only because their “area weights” variable is wired into the analysis exactly backward: instead of downweighting small areas that people would find it easy to move away from, they downweight large areas where people are more likely to stick.

You have a prior—a strong, theoretically motivated, Malthusian-selection prior that patience got culturally bred into populations where the agronomic return to waiting was high. You run the specification. It comes out significant and correctly signed. And here is the trap that catches all of us who are even passable Bayesians: you do not audit results that confirm what you already believe. You audit the surprises. Confirmation is where your guard is down. Oded and Ömer divided where they should have multiplied, the number came out the color they expected, and nobody re-derived the weight variable because nobody had a reason to. It happens. It has very nearly happened to me.

Thus: “The estimator gives the largest weight to the smallest regions. A tiny 34 km² Swiss region receives 170,000 times the weight of the largest region. The influence of small, urban regions where internal migration is most severe is thereby maximized, making the problem that area weights were supposed to address worse…”

And once you wire things in the right way, the coefficient loses statistical significance completely and loses its sign: it is not the case that higher potential crop yields and shorter growth cycles that rewarded delayed gratificatication are associated with the development of a more patient human cultural matrix, but rather the reverse.

The broken academic incentive structure is Francis’s deeper target: even a demonstrable error won’t trigger retraction or correction, because the “top five” journals essentially never do so.

The Poor Rich World

This is a blog on economic history.

By Joseph Francis

CROSSPOST: JOE FRANCIS: Running Out of Patience with Economics

The Poor Rich World

Running Out of Patience with Economics

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…

Read more

14 days ago · 9 likes · Joseph Francis

<https://poorrichworld.blog/p/running-out-of-patience-with-economics> <https://poorrichworld.blog>

A Strange Case of Inverted Area Weights

Joseph Francis

Aug 18, 2026

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected.

The article “The Agricultural Origins of Time Preference” (2016) by Oded Galor and Ömer Özak is a case in point. Published in the American Economic Review, the article presents a grand narrative of persistence. Galor and Özak hypothesize that “geographical variations in the natural return to agricultural investment generated a persistent effect on the distribution of time preference across societies” (p. 3065). Higher potential crop yields and shorter growth cycles, they say, rewarded delayed gratification, so a culture of long-term orientation was selected for over generations and persists in populations today.

The article has garnered over 1,000 citations on Google Scholar, but it is in no way, shape, or form robust.

The evidence for this narrative comes from a simple correlation between individuals in the World Values Survey and historical FAO data on potential crop yields and growth cycles in the places of their ancestral origins. Macro-historical regressions across countries are, however, famously vulnerable to confounding factors such as geography, modern national institutions, or contemporary culture, which can easily drive the correlation.

Galor and Özak therefore run a version of their regressions with country fixed effects. Individuals are compared within each country, and national institutions drop out. In the published article that specification is statistically significant: historical crop yields and patience, across 1,356 regions.

But the problem is that people have moved. The agricultural characteristics of a person’s current location might not then reflect the environment that supposedly shaped their ancestors’ time preference.

They claim to deal with this by using area weights. As they put it, “observations are weighted by the scale of each region, mitigating the effect of internal migration” (p. 3097). Larger regions, on this logic, are less susceptible to internal migration crossing their borders, so assigning them higher weights should reduce this measurement error.

Nonetheless, their code then does the opposite.

For weights, it uses a variable called “invwvsarea.” When that variable is reverse engineered, it turns out that it is in fact area weights inverted. Why they did this, is not clear, but it has the effect of aggravating the issue that they claim to be addressing. By weighting by the inverse of the area, the estimator gives the largest weight to the smallest regions. A tiny 34 km² Swiss region receives 170,000 times the weight of the largest region. The influence of small, urban regions where internal migration is most severe is thereby maximized, making the problem that area weights were supposed to address worse.

When the actual, non-inverted area weights are used instead, the coefficient for crop yield drops from 0.032 (p=0.010) to an insignificant −0.033 (p=0.39), as shown in Table 1. Indeed, all their main results essentially disappear. And the same happens when no weights are used at all: the coefficient falls to 0.001 (p=0.85), indistinguishable from zero. Their results are artifacts of doing the opposite of what is described in the text.

Yet this probably does not matter. Even though the article is not robust, it will not be retracted, or even corrected. That will not be possible because the “top five” economics journals do not do such things. As the Economist recently observed, we know that things are much better in economics than in the rest of the humanities and the social sciences because the “five leading journals have seen just four withdrawals in their combined 570-year history.” Their peer review is so robust that it would be impossible for an article like this to be published in the first place. Consequently, those area weights cannot really have been inverted.

Further Reading

The article examined in this post is:

• Galor, Oded, and Ömer Özak. “The Agricultural Origins of Time Preference.” American Economic Review 106, no. 10 (2016): 3064–3103. Data and replication archive.

Replication Files

The replication package for this post is available here.

<https://poorrichworld.blog/p/running-out-of-patience-with-economics> <https://poorrichworld.blog>

The Poor Rich World

Running Out of Patience with Economics

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…

Read more

14 days ago · 9 likes · Joseph Francis

The Poor Rich World

This is a blog on economic history.

By Joseph Francis

Brad DeLong here: I part company with Joe is the counsel of despair—the bleak “it will never be retracted, so none of this matters.” I don’t buy it. Retraction is the wrong metric. The right metric is what the profession believes and builds on, and there the machinery works, albeit maddeningly slowly and expensively.

It is true, to highlight what Joe writes:

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…. [Oded Galor and Ömer Özak’s] results are artifacts of doing the opposite of what is described in the text. Yet this probably does not matter. Even though the article is not robust, it will not be retracted, or even corrected. That will not be possible because the “top five” economics journals do not do such things…

Nevertheless, I think Joe is wrong to despair here. Yes, it looks like Oded and Ömer divided where they should have multiplied—it happens—and did not catch it because the results were as they expected and, as good Bayesians, you don’t spend that much time checking expected results—that also happens. But things that do not replicate are weeded out, albeit much more slowly than they should be. And people do have a very healthy skepticism about results that are close to the edge of statistical or economic significance, or that cannot be shown to rely on a strong solid correlation.

It is not just computational error. It is also that, say, you have ten yes-no decisions to make in running any empirical analysis that could go either way. What you should do is to choose the way that makes most sense to you, and run along several tracks where things are debateable. But there is the temptation to choose the track that gets you closer to the result you think is sensible. Succumb to that, and you wind up with the strongest of 1024 possible results. Alternatively, get yourself checked by somebody hostile for ideological reasons, and they can come up with the weakest of 1024 possible results as a way of dismissing your analysis.

Stepping back, the persistence literature keeps reaching for a lever—settler mortality, ancestral crop yields, ruggedness, caloric suitability—that will let a cross-sectional correlation masquerade as a causal parameter, without a structural model that says how the past reaches into the present.

It turned me into a Heckmanite.

I do not believe you have earned the right to say “causal” unless you can write down the structural model and the story about why the identifying variation is even remotely exogenous. “Deep roots” regressions almost never do. The area weight was a half-hearted patch over the wound that people can and do move to opportunity. Inverting it tore the patch off and rubbed salt in.

But things that do not replicate do get downgraded; they stop anchoring dissertations; the second generation of citations turns sour. It is a shame it takes a decade and a Joe Francis rather than an afternoon and a journal’s own referees. We should be able to do much better.

Which is why we need preregistration and replication packages. And that is why Joe Francis undertaking this mission is a very good thing for the profession as a whole.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-joe-francis-running-out-of-patience-with-economics
##public-reason
##crosspost
##joes-subheadline-a-strange-case-of-inverted-area-weights-but-i-mostly-want-to-praise-joe-francis-for-spending-so-much-of-his-time-on-post-publication-peer-review
#joe-francis-running-out-of-patience-with-economics
#joe-francis
#running-out-of-patience-with-economics
#post-publication-review
#agricultural-origins
#replication-crisis
#garden-of-forking-paths
#persistence-studies
#heckmanite
#causal-inference
#preregistration
#replication-packages
#economic-history
#confirmation-bias

During an asset market boom, even a euphoric boom, prices rise as good news arrives and as more and more people take their money and decide that this is indeed the wave of the future. But then increases in demand switch from people becoming aware of the opportunity and bringing their money in, to people willing to bet on rising prices, willing to ignore Risk Management 101, and eagerly leveraging-up and pouring that leverage into the booming asset class. Rising prices taken as a reason to buy more induce a situation in which the smart money starts leaving quietly, and the game silently switches from owning value to finding a greater fool before everyone else does.

Share

And then people are on the edge between hanging on hoping a greater fool comes along and selling out now. The shape of demand no longer rides predominantly on expected developments of industry, but on various reads of market sentiment.

That period in which a lot of people are following positive-feedback investment strategies is a period of “distress”.

We know it is distress because then we get things like this from the most overleveraged:

Give a gift subscription

Via Adam Tooze <https://adamtooze.substack.com/p/top-links-1199-big-losses-meeting>, FT Alphaville <https://www.ft.com/content/340bf9e7-0e67-4d19-b671-3dc8186efb99> picking up “a cool table on Wikipedia” <https://en.wikipedia.org/wiki/List_of_trading_losses> based on Tom Coleman <https://rpc.cfainstitute.org/sites/default/files/-/media/documents/book/rf-publication/2011/rf-v2011-n3-1-pdf.pdf>.

The write-up is by Toby Nangle:

Toby Nangle: A Leaderboard of the Biggest Trading Losses of All Time <https://www.ft.com/content/340bf9e7-0e67-4d19-b671-3dc8186efb99>: ‘We found a cool table on Wikipedia.. to more easily contextualise the quantum of Aschenbrenner’s loss…. There are two steps to a fund losing a lot of money. The first step is to inspire faith in either a large number of fairly wealthy people or a small number of immensely wealthy people…. The second step is to do [is]… throw together a credible investment thesis, [and] have sufficiently high conviction… to cast aside risk management 101, maybe chucking a bunch of financial leverage into the mix…. The second step is easy. There are thousands of people yoloing in their mums’ basements around the world doing just this right now…. So if you’re interested in maximising your ranking on any quantitatively measured global leaderboard of trading losses, the first step is probably more important….

We’re fairly sure that this league table, like every other we’ve chanced upon, is only really capturing the kind of meltdowns that make good copy. There’s Tiger Global’s ca. $40bn bloodbath in 2022, for example, which arguably should put it at the top of the list. There’s a case that Jane Street…[belongs] given the reported $15bn hit that it took in July from its exposure to Aschenbrenner…

Leave a comment

That last is a reference to:

Jill R Shah & Joshua Franklin: Jane Street Suffers $15Bn Hit After Meltdown at Situational Awareness: Jane Street posted a roughly $15bn loss in July after turmoil at AI-focused hedge fund Situational Awareness wrongfooted the US trading firm. The New York-based firm disclosed the figure to lenders as part of a deal to shift its roughly $11bn public debt pile to private investors including Pimco…. Jane Street has generated more than $40bn in net trading revenues in the year to Friday, even accounting for the July loss, which exceeds its entire haul for 2025…. Jane Street’s investment in Situational Awareness was unusual because the firm trades its own capital.… a former Jane Street employee worked at Situational Awareness and Jane Street co-founder and partner Robert Granieri attended [Leopold] Aschenbrenner’s wedding in California in recent weeks….

Jane Street was established in 2000 by a small group of founders including Granieri, who previously worked at Pennsylvania-based Susquehanna. It uses technology to make markets across assets such as equities, bonds, exchange traded funds and more. In recent years, it has expanded into longer-term strategies as well as investments in private companies, including AI lab Anthropic and data centre operator CoreWeave…

Share DeLong's Grasping Reality Weblog

Time to pull out the Kindleberger! This time, from A Financial History of Western Europe:

At some stage in the process it becomes clear to a few, and then to more, that the fallacy of composition is at work, that the whole is rather less than the sum of the parts, that credit positions are extended beyond some limit sustainable in the long run, and that maintenance of capital gains depends on getting out of assets rising in price ahead of others.

There follows a period of what may be called ‘distress’: ‘We have no crash at present, only a slight premonitory movement of theground under our feet,’ wrote Lord Overstone to his friend, G. W. Norman, on 1 November 1845 (O’Brien, ed., 1845 [1971], Vol. 1, p. 368). From time to time the distress abates. On other occasions it intensifies. More and more speculators seek to get out of whatever was the object of speculation, to reduce their distended liabilities, and switch into money; and more and more it becomes clear that not everyone can do so at once.

There is a rush, a panic, and a crash—or perhaps a lender of last resort intervenes to make clear that it will furnish the market all the cash it insists it requires. In this circumstance, perhaps belatedly, panic and distress subside…

Get 75% off a group subscription

The phases of the process are: displacement—a technological or a super-political shock that calls forth a need for real economic adjustment and change—diffusion of euphoria as adjustment takes place, distress, crisis, panic, and then—perhaps—a lender of last resort.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠bubble-watching-this-is-what-happens-in-the-type-of-period-called-distress-chart-of-the-day
##macro-outlook
##finance⁠
##‎⁠chart-of-the-day
##‎⁠according-to-kindleberger-things-like-the-situational-awareness-colllapse-of-leopold-aschenbrenner-happen-near-the-peak-of-a-bubble-they-are-a-sign-of-the-shape-of-demand⁠
#bubble-watching-this-is-what-happens-in-the-type-of-period-called-distress
#bubble-watching
#kindleberger
#distress
#financial-crisis
#manias-panics-crashes
#leopold-aschenbrenner
#situational-awareness
#jane-street
#positive-feedback-investment-strategies

The shift from equity to debt financing of datacenters has been one powerful sign of an approaching bubble peak. Gary Marcus reads OpenAI’s pausing of some aspects as model training as a second sign: husbanding cash now appear to be more important for OpenAI than generating vibes.

Share

As of this morning, August 18, 2026, Gary Marcus thinks that Sam Altman and OpenAI have just thrown in the towel on capturing the frontier-model lead in the next training cycle round:

Gary Marcus: BREAKING: OpenAI’s Unraveling Has Begun <https://garymarcus.substack.com/p/breaking-openais-unraveling-has-begun>: ‘Their planned IPO is facing headwinds, trust has evaporated, and their burn rate is only getting worse…. Sam Altman’s latest announcement, on Tuesday August 18 (i.e., “earlier today” for those of us on the West Coast), that OpenAI would be pausing, ostensibly for safety reasoning…. As far as I can tell, hardly anyone believed him…. Translation from the user @NIK on X: “We are out of compute”…. Slightly elaborated: Boss Hendricks: “Translation: we need to immediately stop torching cash to provide some semblance of a sustainable business model so we can rush this IPO out the door before the bubble pops”….

The Wall Street Journal’s Berber Jin and Corrie Dribusch just dropped big news: “OpenAI told investors its revenue grew by 18% from the first to the second quarter while its losses deepened… [growing] by $3 billion from q1 to q2, to $12.3 billion. Not a great look given that it added only $1 billion in revenue in the meantime, to $6.7 billion”…. Good news for Anthropic…. And terrible news for OpenAI…

Share DeLong's Grasping Reality Weblog

We had, four months ago:

Nilay Patel: The Ai Industry’s Race for Profits Is Now Existential <https://www.theverge.com/podcast/909042/ai-monetization-cliff-anthropic-openai-profitable-ai-existential-moment>: ‘It’s a make-or-break year for Anthropic and OpenAI, which are facing more pressure than ever to make more cash than they burn…. Hayden Field… senior AI reporter here at The Verge… has been keeping close tabs on both Anthropic and OpenAI…. At some point, the profits have to materialize, or the bubble pops…. You’ve heard me ask some version of this question to scores of CEOs here on this show, and a majority of them have hinted toward the bubble popping — they think some companies will fail in spectacular fashion, some will succeed, and the opportunities, especially the money, are simply too big to ignore. We’re doing this, whether we want to or not — the market depends on it….

AI agents… have radically changed how these companies are thinking about their resources…. Agents are valuable to [code-writing] customers right now, but agents also use far more compute… burning tokens at a rate way faster than these companies anticipated…. OpenAI abruptly killed its video-generation app Sora, ditching a $1 billion Disney licensing deal in the process. Why? It costs too much to run, and OpenAI needs the compute for Codex. We saw it again just last week, when Anthropic decided it would no longer let Claude users burn through compute resources using the OpenClaw agent framework through a standard subscription plan…. The projections these companies have made, which just this week were leaked to the Wall Street Journal, tell a story of mind-boggling growth, to the tune of hundreds of billions in revenue and profitability by the end of the decade. But the most important questions now are can the AI companies pull this off, and what compromises will they make to reach that goal and avoid crashing and burning?…

Give a gift subscription

Following up on his four months earlier prediction: “as nuclear as it gets: OpenAI fails [in 2026]”:

33:31

Get 75% off a group subscription

Even in this context OpenAI’s apparent training pause does surprise me. I had thought that right now OpenAI Codex on the one hand and Claude Code and Cowork on the other were about equal. Some preferred one. Some preferred the other. It seemed largely, these days, a matter of taste and path-dependency. Claude Code wins on code quality, context retention, and multi-agent orchestration; Codex wins on speed, token efficiency, cost-per-task, and fire-and-forget autonomy. Most heavy users run both. And as far as Codex and Claude Cowork are concerned, that race is just beginning, and is too close to call.

And yet it looks like OpenAI’s revenue only (only!) grew from $5.7 to $6.7 billion from the first to the second quarter, while Anthropic’s revenue grew from $4.7 to $11.5 billion. How is this, if Claude Code and Cowork and Codex are rough peers?

Well, first, they were not rough peers at the start of the quarter, on April 1. So perhaps the Q2-Q3 comparison will look very different from the Q1-Q2 comparison.

However, otherwise: The coding tool is a much bigger slice of Anthropic than of OpenAI. Roughly 80% of Anthropic’s revenue is API/enterprise. Anthropic sells agent usage metered by token through its enterprise/API tier, while Codex is delivered inside a ChatGPT Plus/Pro subscription. And OpenAI wants to build loyalty and so does not want to do what Anthropic did to its OpenClaw enthusiasts by cutting off their access. More important, perhaps: Anthropic is the picks-and-shovels supplier to the whole coding-agent ecosystem, not just Claude Code. Claude is the model behind a large share of third-party coding front-ends. with GAAP numbers due in the IPO prospectus.

What I dearly wish to see right now is Anthropic’s S-1 for its forthcoming IPO, which is coming—sometime. FutureSearch “founded in August 2023 by Dan Schwarz… [as] an AI that could predict the future” claims <https://futuresearch.ai/app/p/a/on-what-date-will-anthropic-complete-its-ipo-i> November 4 as the likely IPO date (which means the GAAP financials need to appear in less than a month and a half). Itd further claims:

Three significant overhangs threaten to push the timeline…. Gross-versus-net revenue accounting disputes are highly scrutinized by the SEC; Anthropic reportedly books cloud partnership revenue gross…. Resolving this could require extensive disclosure changes or restatements of ARR metrics. Second, ongoing litigation… over an unprecedented “supply chain risk” designation limits U.S. military contracting and requires sensitive, unresolved risk-factor disclosures . Finally, the sought-after $2T+ valuation demands a staggering $190–200B revenue forecast for 2028…

Leave a comment

My read (which may be very wrong):

1. An Anthropic that comes out of the gate with a $2 trillion market valuation at its IPO is something that does not need the U.S. Defense Department as much as the U.S. Defense Department needs it. Given that standard risk disclosures are simply boilerplate, and FutureSearch is highly likely to be simply wrong here.

2. People who are going to buy Anthropic at the IPO can be easily directed to pro forma financials. The actual GAAP financials will be of relevance only to short sellers who are going to be on the sidelines unless they are stark raving mad given what we have seen over the past decade. FutureSearch is highly likely to be simply wrong here as well: divergence between pro forma and GAAP is also not holding up the IPO.

3. What is, I think, very likely to be holding up the IPO is that a number of organizations that are putting their and their clients’ money on the line here want more than just the second quarter of super-explosive growth from a relatively high base.

Share

Recall that Anthropic’s reported revenue figures are $0.8B for 2025Q2, $4.73B for 2026Q1, and $11.5B for 2026Q2, that those come from leaks or investor-deck slides, and that that is all we know. Other leaks and announcements have been “run rates”:

• $1B ARR as of the start of the 2025.

• $9B ARR as of the end of 2025.

• $65B ARR as of August 1, 2026.

Give a gift subscription

My bet are that these are one-month run-rates at best: i.e., Anthropic booked $5.4B in revenue for the month of July 2026. People will want to see revenue on-track on an S-curve to triple from mid-2026 to moderately late-2028. That means they want to see August and September numbers, that they want those numbers to be good, and that Anthropic is willing to push off the IPO and bet that it can deliver those numbers, rather than scale back the whispered $2T IOP valuation.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##more-signs-of-a-bubble-peak-breaking-news-watch-of-the-day
##macro-outlook
##mamlms
##subturingbradbot
##breaking-news-watch-of-the-day
##openai-the-company-that-had-been-selling-vibes-looks-like-it-is-starting-to-husband-its-cash-in-the-face-of-the-anthropic-challenge-two-peer-coding-agents-and-yet-a-55x-revenue-growthrate-gap
#more-signs-of-a-bubble-peak
#bubble-peak
#ai-bubble
#openai
#anthropic
#coding-agents
#anthropic-ipo

In this 2012 post, Paul Krugman looks back on more than a decade of being dismissed as “shrill.” His offense was accuracy: he insisted early that George W. Bush was a serial liar pursuing a hard-line agenda, not the blunt honest conservative the commentariat described. He also refused the mandatory pretense of symmetry — the fiction that Democratic caution and Republican radicalism were morally equivalent. By 2012, with a primary field of “not-Romneys” he calls stark raving mad, Krugman argues the party’s condition was decades old. The only new thing was that the pretense of reasonableness had become impossible to sustain:

Share

HOISTED FROM PAUL KRUGMAN’S ARCHIVES: PAUL KRUGMAN (2012): Looking Back With Shrillness

<https://archive.nytimes.com/krugman.blogs.nytimes.com/2012/02/29/looking-back-with-shrillness/>

Brad DeLong notes that the GOP we now see in the primaries has been building for a couple of decades; I can’t help thinking of my own decade-plus in the journalistic trenches.

Early on in my tenure at the Times, I felt I had no choice but to point out the inconvenient truth that the official line of the commentariat was all wrong. George W. Bush was not a nice, blunt, honest guy who happened to be a conservative; he was a serial liar pursuing a hard-line agenda, who among other things deliberately misled America into war.

For this I was labeled “shrill”.

More than that: throughout these past ten-plus years, it has been considered ill-mannered and uncouth, not to mention unacceptably partisan, to suggest that the parties aren’t symmetric — that, for example, the reluctance of Democrats to cut Social Security and Medicare is not equivalent to the GOP’s consistent pursuit of huge unfunded tax cuts, that the occasional desire of Democrats to put evidence in a more favorable light is not equivalent to the constant, raw dishonesty emanating from the right. And pundits in good standing have been expected to make calls for bipartisanship that involve pretending that Republican politicians are actually the kind of statesmen the party used to contain, but no longer does.

So now we see a primary struggle in which the choice is between a series of not-Romneys whose political and policy views are stark raving mad, on one side, and the not-not-Romney who is, maybe, just pretending to share those views. How did that happen?

The answer, as Brad suggests, is that it happened a long time ago. The GOP isn’t just spectacularly unlucky in its menu of candidates; this is what the party has been for decades. Rick Santorum isn’t someone out of left field; he’s always been what you see now, and he was a central figure in his Senate days.

All that has happened now is that the mannerisms have finally gotten to the point that the pretense of a reasonable party is no longer sustainable.

But you weren’t supposed to notice until just about now.

<https://archive.nytimes.com/krugman.blogs.nytimes.com/2012/02/29/looking-back-with-shrillness/>

Brad DeLong here: The thing to understand about Paul Krugman’s 2012 “Looking Back With Shrillness” is that it was not a complaint a much as a diagnosis — and, more than that, a vindication delivered in the flat, tired voice of a man who had been right for a decade, and had gotten little for it but the label “shrill”.

Go back and read what Krugman actually wrote in February 2012. The proximate occasion was a post of mine — “Where Were You in 1993, David Brooks?” — in which I had made the boring, documentable point that the Republican Party then staggering through its clown-car primary had not suddenly gone mad. It had been building toward exactly this for a couple of decades. Paul read that and did what Paul does: he generalized it, sharpened it, and turned it into an indictment not merely of the GOP but of the entire respectable commentariat that had spent ten years insisting the emperor was fully clothed.

His confession — and it reads like one — was that “early on in my tenure at the Times, I felt I had no choice but to point out the inconvenient truth that the official line of the commentariat was all wrong. George W. Bush was not a nice, blunt, honest guy who happened to be a conservative; he was a serial liar pursuing a hard-line agenda, who among other things deliberately misled America into war. For this I was labeled ‘shrill.’”

Now. I have some standing here, because I was a charter member of the order. The Ancient, Hermetic, and Occult Order of the Shrill was not, in its original conception, a club of the angry. It was a club of the accurate.

Give a gift subscription

The defining trait of a Shrill One was never volume. You looked at the Bush tax cuts and the Medicare drug benefit and the war, you did the arithmetic, and you said the arithmetic out loud — and the doing of it, calmly, was itself treated as a breach of manners. The sin was that we counted. The sin was not that we screamed, not that we were shrill. Indeed, originally we were not so.

“It has been considered ill-mannered and uncouth,” Paul Krugman wrote, “not to mention unacceptably partisan, to suggest that the parties aren’t symmetric.” The reluctance of Democrats to cut Social Security was to be treated as morally identical to the GOP’s relentless pursuit of unfunded tax cuts. The occasional Democratic massaging of evidence was to be filed alongside “the constant, raw dishonesty emanating from the right.” To notice the difference — to say that one side was lying more, and worse, and on purpose — was to disqualify yourself from polite company. The high priests of even-handedness demanded that you pretend Republican politicians were still the statesmen the party used to contain but no longer did.

And so we have, even today, a Danielle Allen stroking her chin and writing:

Danielle Allen: Our Parties Have Trapped Us <NO. SOMETHING THIJS DISHONORABLE DOES NOT GET A LINK>: ‘A good way to understand our current predicament as Americans is to imagine that we are a bear in the woods being attacked by hungry wolves. And our paw is caught in a trap. The wolves are politicians. Some are gray, some are black, but wolves are wolves. We are all of us, together, that bear. It doesn’t matter what side of America’s yawning political divide you stand on…

As if a Nancy Pelosi is as dishonorable as a Paul Ryan, or a Bill Clinton or a Barack Obama or a Joe Biden is as dishonorable as a Donald Trump. Danielle Allen knows better. But she thinks she is being clever by being dishonest here. I do not have to like it.

The job of the commentariat was to price political behavior — to tell the public which actions were normal and which were aberrant, which claims were true and which were fabricated. The doctrine of symmetry was a deliberate corruption of that signal. It said: no matter what one side does, we will report the spread between the parties as roughly constant, roughly balanced, roughly a matter of taste. Voters who trusted the referees were told, year after year, that there was nothing unusual to see. So they didn’t see it. That was the whole point of the arrangement.

Paul’s was insisting back in 2012 that 2012 was not a rupture but a reveal. “The GOP isn’t just spectacularly unlucky in its menu of candidates; this is what the party has been for decades.” Rick Santorum, he wrote, “isn’t someone out of left field; he’s always been what you see now.” The radicalism on display had not arrived. It had merely become impossible to costume. “All that has happened now is that the mannerisms have finally gotten to the point that the pretense of a reasonable party is no longer sustainable.” The disguise failed. The thing underneath had been there the whole time — since 1993, since Gingrich, since the party decided that the way to power was to treat the opposing party’s legitimacy itself as negotiable.

And here is where I get to be shrill in the present tense, because the vindication is now total and it is not a happy one. Everything Paul described in 2012 was the overture. The demand that we treat asymmetric dishonesty as symmetric; the punishment of accuracy as bad manners; the steady discounting of a party’s radicalism because acknowledging it would be “partisan” — run that machine for another decade and you do not get Rick Santorum. You get worse, and you get it in power. The commentators who called us shrill in 2003 and 2012 were not merely wrong on the merits. They were the mechanism. Their false balance was the subsidy that kept the whole enterprise solvent long past the point where an honest market would have repriced it to zero.

So what was Paul Krugman saying in 2012? He was saying: I told you, and you called me shrill for telling you, and the telling was the service and the shrillness was the smear. He was saying the disease was old and the symptoms merely newly visible. He was saying that the unforgivable act in the moderate center of our politics was never raising your voice. It was keeping accurate books.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##hoisted-from-paul-krugmans-archives-paul-krugman-2012-looking-back-with-shrillness
##public-reason
##ancient-hermetic-order-of-the-shrill
##neofascism
##hoisted-from-the-archives
##shrill-was-always-the-correct-setting-krugman-2012-on-the-implications-of-seeing-what-normies-dare-not-when-accuracy-gets-you-called-uncouth-the-problem-is-the-etiquette-not-the-accuracy
#paul-krugman-2012-looking-back-with-shrillness
#paul-krugman
#looking-back-with-shrillness
#looking-back-with-shrillness
#accuracy-over-etiquette
#where-were-you-1993

Last year Jensen Huang said: You need to buy the latest and greatest chips my team has designed right now because the chips we designed that you bought last year are obsolete and effectively worthless. This year, Jensen Huang is saying: You need to lend huge amounts of money with the chips my team designed in the past as security because they are durable income-producing assets with a ten-year economic life.

Share

You see the problem?

That is: Jensen Huang does not remember what his book told him to talk. Or, rather, he is changing books with alarming frequency:

Elisabeth Lopatto: Nvidia’s New Financial Strategy Does Not Compute <https://www.theverge.com/ai-artificial-intelligence/981668/nvidias-goldman-blackrock-gpu-compute-asset>: ‘April – 1805

Napoleon is master of Europe

Only the British fleet stands before him

Compute is now an asset class

I see it is once again time to talk financial innovation. Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR are all working with Nvidia to put together $500 billion in financing to turn compute into an asset class. “This is really the first time that technology chips have become an investable asset class,” Nvidia CEO Jensen Huang said to CNBC. “These are revenue-generating assets now. They’re productive, they’re long-lived, they’re fungible, they’re flexible.” Huang said something very different about Nvidia’s own last-generation Hopper chips last year. “When Blackwell starts shipping in volume, you couldn’t give Hoppers away…. There are circumstances where Hopper is fine. Not many.” So to now be told that chips are actually “revenue-generating assets” that are “long-lived” is… quite frankly, it’s giving me whiplash…. “This is the very beginning, like what it was when I started in the mortgage-backed securities market in the 1970s, and I look upon this as a next future for financial engineering,” said Larry Fink…. Now, for some of you, this may make alarm bells go off…. The AI industry is becoming saturated with data centers, and Chinese open-source models require less compute despite being fairly powerful, both of which seem like potential threats to the notion of ever-growing demand for chips. There is also a far more basic question: Can frontier labs such as Anthropic and OpenAI, which are driving much of the current demand, make money?…

Give a gift subscription

As I keep pointing out to no avail: The gatekeeping question is “Will Amazon, Google, FaceBook, Microsoft, and so on allow Anthropic and OpenAI and anyone else hoping to make a profitable business out of selling LLM-services to make money?” And the likely answer is that they will fight Anthropic and OpenAI tooth-and-nail just as Microsoft made it certain that Netscape would never make money by giving away for free anything and everything that Netscape was trying to sell.

I do not see how you pass that gatekeeping question. If you do pass that gatekeeping question, somehow, then there are more questions:

• What does the world look like in scenarios in which the chip purchasers you, Ms. Private Equity, have financed do not make their amortization payments?

• What does the cash flow from the datacenters and the chips in them that you then have to take over look like in those worlds?

• How do you then explain to your investors whom you promised megaalpha that your major assets are underwater utilities?

I see no scenario in which the fact that the debt is secured is worth anything.

That is, I see compute is now an asset class as let us see if you are foolish enough to make equity investments at a debt-investment price if we simply tell you that it is not equity but debt.

Fight me.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##can-we-sell-equity-financing-of-the-datacenter-buildout-at-a-debt-price-financial-overexuberance-of-the-day
##macro-outlook
##finance
##financial-overexuberance-of-the-day
##jensen-huang-keeps-two-books-last-years-says-old-chips-are-worthless-and-obsolete-while-this-years-says-that-old-chips-are-highquality-tenyear-bonds-elisabeth-lopatto-says-wait-a-minute
#can-we-sell-equity-financing-of-the-datacenter-buildout-at-a-debt-price
#jensen-huang
#compute-asset-class
#private-equity
#ai-bubble
#financial-overexuberance
#hyperscaler-gatekeeping
#underwater-utilities-of-the-future