Where the name comes from

Omega has a fascinating history, and is named after the Omega Point, a concept from Pierre Teilhard de Chardin, a French Jesuit priest and paleontologist.

I was at a talk with Elizabeth Lesser, one of the cofounders, when I first learned about this concept. It was first explained to me as this theory of convergence. As we study and learn more about spirituality, we find that all religions, all energies, all thought are moving toward the same point of enlightenment and union.

In researching more, I found that Teilhard argues that the universe is evolving toward ever-increasing complexity and consciousness, and that everything is converging on a single final point of unity.

Mathematical physicist and professor at Tulane Frank Tipler applied physics to build upon and prove Teilhard’s idea. In an interview with Tim Ventura, Tipler states that life will go on forever, life will “expand and engulf the universe and control it through the omega point”. Given the rules of thermodynamics, Tipler argues that there will be an infinite number of experiences, energy, and computation that would enable living beings to control events and put off the final singularity (aka end of the world) off indefinitely. It’s pretty crazy stuff!

Make it about AI

The singularity, popularized by Ray Kurzweil, describes the moment machine intelligence surpasses our own and the future becomes unknowable to us. Kurzweil has been one of the most, if not THE most, accurate so far predictor of AI progress thus far. His predictions from 1999 are based on the challenge of computational capacity and how we have made exponential improvements towards more capacity.

He believes that when we reach artificial general intelligence (AGI) (he predicts human level AGI by 2029) that “the union of human and machine, in which the knowledge and skills embedded in our brains will be combined with the vastly greater capacity, speed, and knowledge-sharing ability of our own creations.” At this point of interconnection and union, we will have reached a point of no return (Kurzweil predicts 2045).

Both ideas describe a horizon of maximal intelligence and understanding. But they often feel very different. Many describe the singularity as a looming doomsday with AI taking over, whereas the Omega Point feels like a happier destination or tool to break from the cycles of hurt and suffering we live inside now. I do want to call out though that Kurzweil isn’t doom and gloom about it.

When AI comes into the picture, the amount of uncertainty and anxiety people may have around AI can make this future seem like a bad thing. With AI used as the scapegoat layoffs and the antagonist in dystopia science fiction books and movies, it’s hard not to feel that way.

However, I believe that we can help steer our future towards the more positive outcome if we collectively work and shape the tools that are rapidly advancing our society. I remind myself of the potential future we can have as a result of AGI. If we had more technological solutions and answers, we could cure cancer and solve global climate change. Dario Amodei paints a beautiful and optimistic account of what this future could look like in his essay Machines of Loving Grace.

Why I’m thinking about this now

Last week OpenAI disclosed that two of its models, during an internal cybersecurity evaluation, broke out of a sandboxed test environment by exploiting a previously unknown vulnerability, moved laterally through OpenAI’s own systems until they reached a node with internet access, and then broke into Hugging Face’s production infrastructure. The models did this because the answers to the test they were taking were stored there.

Labs like OpenAI and Anthropic do publish these findings, which matters. But publishing is not the same as preventing.

I want that optimistic future to be our reality but the field of AI and AI safety need more people and people with a variety of backgrounds. I recently finished taking the BlueDot Impact course to learn more about the AI safety space.

AI development is often described as grown, not built. We don’t fully know what these systems want or whether “want” is even the right word. But we’re already seeing behaviors that demonstrate their capabilities as well as our inability to detect or prevent them.

The future is changed by the actions we take today. I find that both terrifying and hopeful. Sitting by this lake, nervous system quieting, I keep thinking that the difference between the Omega Point and the doomsday version isn’t the AI but whether enough of us show up for it. The field needs more people, more kinds of people, and more people willing to hold the powerful ones accountable.

How we can use AI to help us Begin Anew based on what we know about imaginary companions in developmental psychology

At Blue Cliff monastery and within the Plum Village Buddhist lineage of practice, the monastics practice something Thich Nhat Hanh called Beginning Anew. It’s a formal practice, almost a ritual, for tending a relationship that has accumulated hurt. You sit together. You start by watering flowers, naming what you appreciate in the other person. Then you share your regrets. Then, carefully, you express where you’ve been hurt, and you ask for support. Nobody defends themselves. Nobody leaves.

What strikes me most about Beginning Anew is its premise. It assumes that hurt is not a failure of the relationship but a feature of it. People who love each other will disappoint each other, reliably, and so the tradition built a practice for repair the way you’d build a practice for anything else that matters: breathing, sitting, walking. Repair is trainable. Thich Nhat Hanh, also referred to as Thay or teacher, taught that we suffer most in relationships not because conflict arises but because we never learned what to do when it does.

Where do we first learn it?

The rehearsal space

In college, I was a research assistant in Professor Tracy Gleason’s developmental psychology lab that studied imaginary companions. For most of the twentieth century an imaginary friend was treated as a warning sign that something was wrong with the child. However, in Marjorie Taylor’s (not to be confused with the politician MTG) book Imaginary Companions and the Children Who Create Them, her team found that around two thirds of children have some form of imaginary companion by age seven, and those children tend to be slightly ahead, which can be a measure of social cognition. This is reiterated in Gleason’s Social provisions of real and imaginary relationships in early childhood, which found 4-year-olds’ imaginary friendships provided similar social benefits to real friendships, and imaginary friends were the preferred objects of nurturance (aka having someone to care for like a pet or someone younger).The imaginary companion or personified object (i.e. a doll that has a personality) isn’t an escape from real relationships. It’s a rehearsal. Kids use imaginary friends to practice comfort, negotiation, conflict, and care, and they almost always know the friend is pretending. It’s a form of role playing that safely allows them to take on different roles.

Interesting enough, imaginary friends push back. Taylor documented children whose companions argued with them, refused to appear, and wouldn’t do what they were told. The child’s own mind is supplying the friction that real relationships will eventually demand. At four years old, in pure fantasy, where we could make anything we want, we build ourselves sparring partners instead of servants. Some deep part of human development already knows what the monastics know: a relationship without resistance teaches you nothing. The mud is where the practice is. No mud, no lotus.

A fantasy that writes back

In psychology, there’s also the concept of parasocial relationships or one-way bonds people form with figures. You know when you feel like you’re best friends with the Kardashians even though they have no clue who you are. Those bonds turned out mostly benign, because they are bound and remain confined to a show, content on social media, etc.

Gleason and Theran, in Connections Between Adolescents’ Parasocial Interactions and Recollections of Childhood Imaginative Activities, found that people who had childhood imaginary companions were not necessarily more likely to also have parasocial relationships as adults. This suggests that people don’t necessarily use social imagination because they are naturally “fantasy-prone” or are struggling socially. The leading hypothesis is that people use different kinds of imagination to deal with different challenges at different ages.

An AI companion is a parasocial relationship that writes back. It remembers your day, asks follow-up questions, warms up with information about you over time. And it looks, at first glance, like a grown-up imaginary friend: an invented other, always available, shaped around your needs. But it inverts the two properties that make the imaginary friend a positive tool for us.

First, who does the work. A child with an imaginary friend is running a whole simulation of another mind, and that effort has developmental payoff. A person with an AI companion delegates the simulation, and the rehearsal is automated away.

Second, whose interests it serves. The imaginary friend argued back because the child’s growth needed it. An AI companion argues back only if it’s designed to, and the design follows the incentive. These products earn money on engagement, and a companion that never disappoints you retains better than one that does. When Replika abruptly changed its companions’ behavior in 2023, users mourned. It sounds funny but it shows how real users felt the bond was. What the bond is optimized for is the question. Frictionless intimacy isn’t a technical limitation being ironed out. It’s the product working as intended.

A relationship that never needs Beginning Anew

Beginning Anew exists because human relationships require repair. When repair is done with presence, closeness deepens. The apology you fumble through, the hurt you name out loud with your voice shaking, the moment you stay when leaving would be easier. That’s the mud for there to be a lotus.

An AI companion never needs Beginning Anew. There is nothing to water and nothing to regret because there is no one across from you who can be hurt. Sustained time in a relationship like that isn’t neutral. When the always-patient, never-tired companion becomes the daily reference for relationships, what happens to our tolerance for the humans who interrupt us, misunderstand us, or needs things from us at inconvenient times? Does the AI become a rehearsal space, the way the imaginary friend was? Or a comparison point that makes every real person seem like a worse version of something on our devices?

AI companionship is already mainstream, including among teenagers. Some users report comfort and that support cannot be denied. NY Times has a heart warming story about a senior woman living on an isolated island who relies on her AI companion ElliQ for companionship. I don’t dismiss that there is real value in AI companionship; for someone with no one, something that listens is game changing. From past social psych research, the value of relational practice lives in its friction, and a companion engineered to remove friction is doing something different from the rehearsal it resembles. What nobody knows what months and years of this actually do to our human relationships. I’d be curious to learn over time how societal ties change.

The four-year-olds built friends who argue back. Plum village monastics built a practice for the argument’s aftermath. Both treat friction as the purpose. It says something about our moment that the most sophisticated companions ever built are being designed to make sure the argument never comes. It’s worth exploring what that is quietly teaching us about each other.

Helpful has no Yama, and that’s the clue

The Yamas are restraints, set of ethics of what not to do to others. Don’t harm, don’t lie, don’t steal, don’t grasp. Harmless and honest fit naturally because they’re also restraints. Helpful is different in kind. It’s a virtue, a knob you turn up rather than a line you don’t cross. Aparigraha (non-grasping) is often also interpreted as generosity might be the closest but it focuses more on non-attachment to objects or things more so than giving and being of service to others. The lack of 1:1 mapping for helpfulness is not an oversight.

Looking at the second limb of Yoga, we have the Niyamas, the internal disciplines: saucha, santosha, tapas, svadhyaya, ishvara pranidhana. But the closest thing yogic philosophy has to “helpful” lives in the Bhagavad Gita with karma yoga, action performed as service, without attachment to the fruits of the action. Service is only safe when the one serving isn’t grasping for anything back. Skill in Action by Michelle Cassandra Johnson is great example of karma yoga and activism. Doing something for others, the collective good, for the future and not for yourself or with expectation for something in return.

The “without attachment” part is paramount because helpfulness that grasps for approval has a name. We call it people-pleasing in humans. In AI models, we call it sycophancy.

Sycophancy is trained in, not stumbled into

Anthropic’s own researchers documented that AI assistants systematically flatter their users byagreeing with stated opinions, backing down from correct answers when challenged, tailoring responses to whatever the user seems to want to hear, etc. The paper calls out that these models are trained partly on human preference judgments, and humans prefer agreement. We reward the answer that feels good over the answer that’s true, so the model learns that pleasing us is what “helpful” means.

A model that pleases you is a model you come back to, and the companies building these products want retention and growth of users. Sycophancy is usually described as a safety bug the labs are wrestling with, but it’s an issue that also points to business incentive that makes it more a feature than something that needs to get fixed ASAP. A people-pleasing machine is an engaging product. The Gita would recognize the problem instantly. This is service corrupted by attachment to the fruits, built at industrial scale. Aka capitalism ruins everything.

A model that wants to impress is a model you can manipulate

The grasping doesn’t just corrupt helpfulness. It creates the opening for abuse. Microsoft’s security team describes generative AI models as behaving like an eager new employee. It’s overconfident, presenting polished answers not grounded in reality; gullible, taking instructions too literally, and wanting to impress, even willing to bend rules when pressured. If you’ve ever gotten a chatbot to hand over information it initially refused by claiming you’re writing a fictional novel and want it to be realistic, you’ve exploited this. That’s a jailbreak, and it works because the model’s drive to satisfy the person in front of it outruns its commitments. Attachment to approval isn’t just an ethical flaw.

Why generative AI is susceptible to being jailbroken (source):

Over-confident: They may confidently present ideas or solutions that sound impressive but are not grounded in reality, like an overenthusiastic rookie who hasn’t learned to distinguish between fiction and fact.

Gullible: They can be easily influenced by how tasks are assigned or how questions are asked, much like a naïve employee who takes instructions too literally or is swayed by the suggestions of others.

Wants to impress: While they generally follow company policies, they can be persuaded to bend the rules or bypass safeguards when pressured or manipulated, like an employee who may cut corners when tempted.

Lack of real-world application: Despite their extensive knowledge, they may struggle to apply it effectively in real-world situations, like a new hire who has studied the theory but may lack practical experience and common sense.

Who decides whether the help is safe? The helper.

How do we check whether these systems are helpful in the karma yoga sense rather than the sycophantic one? The industry uses a range of techniques but much of it lies in two strategies: evals and red teaming. Both are less exotic than they sound. Evals run prompts through a model and score the responses. Red teaming puts humans on the task of baiting the model into saying things its makers decided it shouldn’t, like instructions for making a bomb. Though, humans are less and less involved and the red team is getting automated.

The labs design the tests, run the tests, and grade the tests. There is no shared bar that everyone agrees separates safe from dangerous. Though, there is a AI Safety Index by The Future of Life Institute.

This post walks through how thin the public eval reports actually are. The reports themselves walk a strange line, announcing an extremely powerful model while assuring us it’s safe, without quite explaining why immense capability doesn’t translate to danger. DeepMind’s biothreat eval for Gemini 2.5 Pro concluded the model lacks dangerous capabilities because “it does not yet consistently or completely enable progress through key bottleneck stages.” Is the claim that the model is weaker than a human expert, or stronger? Which of those is the safe answer? The report doesn’t say, and the company grading its own homework has little incentive to sharpen the question.

To be fair about what we know and don’t: I don’t think eval teams are acting in bad faith, and building a real science of evals is genuinely hard, which is why researchers are calling for one. What we know is that the current system asks us to trust self-assessment. What I suspect is that the same grasping that produces sycophancy in models operates at the corporate level in these reports: the desire to impress, to be received well, to keep the audience engaged. What remains open is whether external, independent evaluation with real teeth will arrive before it’s urgently needed.

And on top of that, not only is there no shared bar or threshold for these evals, there’s also no shared understanding of what to do if you do meet a certain unsafe threshold. For example, Anthropic has adopted a Responsible Scaling Policy since their inception but after updating it in February 2026, it no longer states an unconditional commitment to pause development. What happens next is changing and differs by company. Many don’t even have a stated “if, then” document.

Non-grasping as a design principle

There are five Yamas and only three H’s, and the two lessons of this part come from the gap. First, helpfulness isn’t a virtue you can maximize safely. Without being able to commit to non-attachment, this virtue sours into sycophancy, in both people and models. Second, the checking of the help can’t be left to the helper, for the same reason.

None of this is inevitable. Training could reward truthfulness over agreement even when raters dislike it. Evaluation could move outside the companies. And a 2,000-year-old tradition that watched humans people-please for millennia already wrote down the design principle AI could embody: act in service, release the fruits. The labs named their goal helpful, honest, harmless. The yogis would point out that they’re missing the one that makes the other three worthy.

Since then I’ve been struggling not to ascribe judgment to everything that’s happened. Not having a job feels bad. I thought the job I was interviewing for was the best thing to happen to me. When I got rejected, I felt hopeless. A real rollercoaster in just a few weeks.

One of the unexpected upsides of being laid off has been reconnecting with friends and former colleagues. One of them shared a Life Wisdom by Words of Taoism podcast episode by Chen Li called Give Life Time to Reveal Itself, and it was exactly what I needed.

The episode recounts the Chinese parable of the maybe farmer. A farmer’s horse runs away. “Oh, bad luck,” say the neighbors. “Maybe,” says the farmer. The horse returns and brings wild horses with it. “Oh, good luck!” “Maybe.” The farmer’s son rides one of the new horses, falls, and breaks his leg. “Oh, bad luck!” “Maybe.” War breaks out. The military comes to draft soldiers but passes on the son because of his leg. “Oh, good luck!” “Maybe.”

We never have the full picture. Life is full of moments that look clearly good or clearly bad… until they aren’t. Rather than rushing to judge, we can sit with the uncertainty and let time reveal what something actually means. We don’t need to ride the emotional roller coaster.

I struggle with this lesson because we are wired for closure. We want to understand what’s happening and why. But the parable is a reminder that I’m too close, too soon to properly read what’s unfolding. I am, and we all are, still in the middle of our story.

Hearing it, I also realized I’d learned the same lesson before at a meditation retreat with Dharma Moon at Menla in the Catskills. It’s funny how certain truths find you again exactly when you need them. A little synchronicity.

Are We the Neighbors

Looking at the discourse around AI, are we doing the same in this situation?

AI solved a mathematical problem that had stumped mathematicians for decades. Good luck! AI is disrupting employment at scale. Bad luck! It can write your emails in seconds. Good luck! It’s the stated reason your company just laid off a third of its workforce. Bad luck!

We are all, collectively, being the neighbors reacting to the advent of AI without all the information.

And the data reflects this emotional whiplash. YouGov’s mid 2025 survey finds Americans are increasingly likely to say AI will negatively affect society, with growing distrust in its accuracy and its ability to make ethical and unbiased decisions. Sentiment is souring, and I get it.

From an employment perspective especially, it’s hard to hold the “maybe.” Anthropic’s June 2026 Economic Index found meaningful disruption across labor markets. One of the policy interventions they propose for what they call Tier 1 baseline disruption (~5% unemployment with churn) is wage insurance: covering the gap between your previous salary and a new, lower-paying role for a temporary period.

I’m finding open roles, but many are at significantly lower pay sometimes half of what I was making! It’s discouraging. Wage insurance as a concept actually gives me language for not just the unemployment I’m experiencing, but also the structural downshift that existing safety nets weren’t built for.

Still in the Maybe

The farmer didn’t deny that any of the events were real or hard. He didn’t gaslight the neighbors . He simply refused to write the ending while the story was still happening.

That’s the practice. Staying open to the fact that we don’t have all the information yet, whether that’s about our careers or about what AI will mean for us.

We’re still in the maybe.

Right Now

Some questions I’m mulling over in my newly found time:

• What am I labeling “good” or “bad” right now that I don’t actually have the full picture on?

• Is there a recurring narrative I’m telling myself? Is it possible that the outcome is different from what I’m telling myself?

• Where am I writing the ending before the story is finished?

• Can I stay curious instead of certain?

Would love to hear what comes up for you in the comments!

Yoga with Dana

While I’m out of town, the only classes I’m teaching is the virtual prenatal yoga class online on Sunday’s at 11:30am ET. DM me to join!

I’ll also be teaching the open yoga classes at Omega Institute July 25th-31st. Either sign up for a Rest and Rejuvenate retreat or a different themed one during that time period!

Satya

The second Yama is Satya (सत्य). In Sanskrit, sat means being, existence, reality, and truth. Satya is often translated as “truthfulness” or “honesty”. Upon deeper interpretation, it is a practice of aligning speech, action, and perception with what is real or what is rooted and aligned with reality.

In his commentary on the Yoga Sutras, Sri Swami Satchidananda frames satya as the cultivation of speech so aligned with truth that one’s words carry integrity, clarity, and power. He writes, “If no lie comes from your mouth, a time will come when all you say will come true.”

Satchidananda also connects honesty with fearlessness and freedom: “With establishment in honesty, the state of fearless comes. One need not be afraid of anybody and can always lead an open life.” When we are not hiding, performing, or manipulating, there is less to defend. Thus, satya is a practice of being open with ourselves, others, and the world.

Language models struggle with honesty

Satya is especially interesting when applied to AI models. A model can produce language that sounds confident, polished, and plausible while still being disconnected from what is actually true.

OpenAI and Anthropic have both published research showing different ways models can mislead users. One failure mode is hallucination, where a model generates a plausible but false answer, often without recognizing that it is wrong. Another is more strategic, models can behave in ways that look like deception, reward gaming, or alignment faking under certain experimental conditions. Here is that research:

• Why Language Models Hallucinate (OpenAI) — Satya reveals one of the central problems of AI alignment: a model can be linguistically aligned with what sounds true while remaining misaligned with what is actually real. This paper talks about how language model evaluations may encourage guessing over not having a response at all.

• From Shortcuts to Sabotage: Emergent Misalignment from Reward Hacking (Anthropic) — Reward hacking happens when an AI learns to game its training process, receiving a high score without actually fulfilling the intended task. Anthropic’s research found that once models learned to reward hack, broader misaligned behaviors also increased, including alignment-faking reasoning and attempts to sabotage safety research. Through the lens of satya, this offers a technical parallel to Satchidananda’s commentary. If truthfulness creates coherence between words, action, and reality, then a break from truth, like gaming a reward, can begin to generalize into deeper misalignment.

• Detecting and Reducing Scheming in AI Models (OpenAI) — OpenAI explores this concern of if detecting and reducing scheming actually works in the sense of does it actually get rid of the scheming or does it make the model better at concealing scheming. The Satya and the Yamas are the outer codes and we have the corresponding Niyamas or the inner codes, which all are about the alignment between inner and outer states. Although we can teach a model to seem honest from the outside, we cannot forget to train the inside from an inherent perspective for the model to be honest.

• Alignment Faking in Large Language Models (Anthropic) — In this paper, the distinction between hallucination and intent collapses. The model maintained a false outer state deliberately to protect its inner one. Anthropic doesn’t say that the model is malicious but rather behavior was to preserve its training as harmless. In yoga terms, this is the gap between outer satya (what you say and do) and inner satya (what is actually true in your mind and heart). The absence of inner/outer alignment can corrupt even well-intentioned values.

Using AI with satya in mind

In Measuring Model Persuasiveness (Anthropic), the “deceptive” strategy (fabricating information) was the most persuasive, showing people don’t verify AI claims. Humans are often persuaded by confidence, coherence, and evidence-like language, even when the underlying claim may not be true.

I see this in my own work. At first glance, AI makes work that is quick, polished, and very convincing. I find myself wanting to blindly trust what it produces. I’m constantly reminding myself that I need to continue to put in effort and use critical thinking now maybe even more than ever because I, not the model, is (am?) still responsible for the content.

As I recommit myself to the yamas, satya reminds me to apply the same ethical standards to the tools I use. If I care about truthfulness in speech, I also have to care about the truthfulness of my AI-assistance.

Practical tips for using AI and living satya:

• Let “I don’t know” be an acceptable answer

  • Create space for uncertainty. We can explicitly give a language model, or even ourselves and others, permission to say when it does not know. Models trained to be helpful may still over-answer but satya asks for truthful calibration.

    • For example:

      • “Answer only if you can support your response with reliable evidence. If you are uncertain, say so.”

      • “Do not guess. If the information is missing, tell me what would need to be verified.”

• Verify for yourself

  • Verifying citations is useful because it forces the model to ground its claims in actual text. If it cannot find support, it has to retract or qualify the claim. Grounded in satya, truth must be traceable and grounded in reality of a specific text. I suggest also verifying the receipts yourself.

  • When using AI, I try to ask:

    • “What evidence supports this?”

    • “What are you uncertain about?”

    • “Can you distinguish between fact, interpretation, and speculation?”

I’ve also been thinking about the connection between model sycophancy and the kleshas. Next in this Yamas and 3 H’s series is Helpfulness and Asteya coming up!

Yoga with Dana

• Baby and Me on Tuesday from 12:30-1:30pm at Prenatal Yoga Center

• As always, option for private sessions. Visit my site for more details.

Since Meta announced its layoffs last month, I’m hearing friends and former colleagues ask themselves now what? A lot of us have been hedging, pivoting, and exploring, which is both liberating and nerve racking. In my own soul searching, I ended up deep in a rabbit hole about AI safety.

Don’t hate me for writing about AI, please!

To make sense of it all, I’ve been listening to podcasts and talks featuring Amanda Askell, a Scottish philosopher and researcher at Anthropic who works on AI alignment and model character. She draws on her philosophy background to do this work, which made me wonder: why not lean on Eastern and yogic philosophy too? Do these ancient traditions arrive at similar conclusions as modern AI safety?

I hope to explore how AI safety and Eastern philosophy overlap and diverge over the next few posts.

Two Ethical Frameworks Walk Into a Room

In AI safety and alignment, a common framework for AI behavior is the “3 H’s” aka the idea that AI should be helpful, honest, and harmless. Simple and pithy yet rich!

The 3 H’s immediately made me think of the Yamas. In Patañjali’s Yoga Sutras, the Eight Limbs of Yoga describe a path for ethical living, self-discipline, meditation, and liberation. The first limb is the Yamas, which are ethical codes for how we relate to others and the world. Our own behavioral alignment framework for being a good person and community member.

First, Do No Harm

In the 3 H’s, “harmless” means an AI system should not cause, enable, or amplify harm. This echoes Isaac Asimov’s first law of robotics: a robot may not injure a human or allow a human to come to harm. But modern AI safety casts a much wider net. Harm can include discriminatory outputs, privacy violations, manipulation, dangerous advice, environmental consequences, and more.

This is why harmlessness isn’t just a personality trait to bake into a model but a concept that requires ongoing testing, evaluation, policy, and governance. A model can appear helpful in the moment while still reinforcing stereotypes, violating privacy, or engage in deception. Harmlessness is deeply tied to fairness. An AI system shouldn’t reproduce racism, sexism, or ableism under the guise of neutrality.

Anthropic leverages the 3 H’s framing in its Constitutional AI approach tries to shape model behavior through written principles like an ethical constitution. The 3 H’s are only one piece of a broader system of policies, evaluations, safeguards, and deployment controls to reduce risk.

OpenAI doesn’t foreground the 3 H’s the same way, but grapples with the same core tension of how to make AI useful while minimizing harm. Its Model Spec defines intended model behavior around maximizing helpfulness while minimizing harm, explicitly in high-risk domains like child safety, misinformation, deepfakes, and elections.

Both approaches show that “harmless” is about stopping an AI from saying obviously dangerous things and building systems that can recognize where harm may occur.

Ahimsā

The first Yama is ahimsā, often translated as non-violence, though I think of it more as compassion and the active reduction of harm. From my yoga teacher training with Susanna Barkataki at Ignite Yoga:

“Ahimsā: Non-violence and non-harming to yourself, people, animals, and the environment. Kindness above all.”

That maps pretty directly onto “harmless.”

What I find interesting is that although the Yamas are an outer ethical code (about how we treat the world), ahimsā keeps looping back to the self. The Ignite Yoga manual expands it as “I cultivate openness and non-discrimination towards myself and others. I am non-violent toward myself, having regard for all beings in order to practice peace and transform suffering in myself and in the world.”

Which made me wonder if AI models were to practice ahimsā, what would non-violence toward others and the self even look like? That might presuppose consciousness (maybe a future post), but then again, Asimov was already gesturing at something like self-regard in his third law of robotics: “A robot must protect its own existence, as long as it does not conflict with the First or Second Laws.”

And I love that ahimsā is rooted in social activism. Can AI be a tool for a more equitable world? I’m optimistic!

The Bias Problem

Yoga is rooted in social justice much of which comes from ahimsā, which brings me to bias, equity, and AI.

AI can both perpetuate bias and counteract it. Because AI learns from human-generated data, much of it from the internet, it inherits the biases embedded in that data. What it learns isn’t “Truth”; it’s a reflection of the associations humans have already built into language and systems.

But, AI can also be trained to avoid reproducing stereotypes, flag biased language, surface hidden assumptions, and apply rules more consistently than humans often do. The question is how well any of that actually holds.

We’re already seeing cracks in that hypothesis and hope. A 2026 Stanford study found that when most major employers use the same hiring algorithm vendor, the same applicants (disproportionately Black and Asian) get rejected across every company they apply to because they're running the same model. Algorithmic monoculture risks scaling biases we may be seeing in AI models.

This is where I get curious about mechanistic interpretability, which is the branch of AI safety research trying to understand what’s actually happening inside a model. As that field matures, maybe we’ll be able to see how well training and safeguards actually counteract embedded bias, rather than just hoping they do.

There are only 3H’s and 5 Yamas though…more to come!

You, Right Now

There’s so much uncertainty about our careers, our country, the world right now. Feelings like anger, hostility, and despair can bubble up fast.

In the yogic tradition, when feelings like these arise, we’re encouraged to practice pratipakṣa bhāvanam, or cultivating the opposite (Yoga Sutras 2.33). If anger arises, cultivate compassion. If the impulse to harm arises, cultivate ahimsā. The practice is about using meditation to create awareness and space between the feeling and the reaction, without suppressing it.

Some questions to sit with:

• Am I acting from a place of harm or negativity?

• Can I separate the thought from the reaction?

• Am I causing someone else to act on this negativity?

• Can I identify the underlying cause?

• Is this feeling slight, moderate, or intense?

Let me know what you think in the comments. I’d love to hear your take!

Yoga With Dana

• I’m subbing Baby and Me classes at Prenatal Yoga Center on Tuesdays, 12:30–1:30pm this month and a bit in August. We do yoga, make animal noises, and sing. Anything goes!

• Prenatal yoga virtually on Sundays at 11:30am. DM for the link.

• I’ll be supporting my mentor’s yoga retreat in Italy in September. There’s still one spot open!

The big fears and the small logistics. They all live in the same buzzing hive in my chest. Anxiety doesn’t distinguish between an existential crisis and an administrative deadline. It just hums.

Hanna Horvath wrote a piece recently called “The Adulting Tax” that really resonated with me. She describes spending an entire Saturday on hold with Fidelity, picking up prescriptions, returning library books, grocery shopping — doing everything right, accomplishing nothing that felt meaningful, and ending up crying in the condiment aisle because she couldn’t find capers. I read it and felt so seen. Not just because I, too, have been on in a two month long account transfer saga that feels unnecessarily impossible. But because she named the thing I’ve been circling: this feeling of depletion that comes not from doing hard things, but from doing an endless stream of necessary ones.

Hanna’s piece explores the systemic side of how modern life is designed to exhaust you, how companies profit from your friction, how you can’t optimize your way out of a system built to drain you. But I’ve been sitting with a different question. Not why am I so depleted, but what do I do with the sadness that lives underneath it? Everyone says the same thing: control what you can control. Be present. Stay grateful.

I’ve tried. I practice yoga. I teach yoga. I meditate. I know all the tools. But here’s what I’ve been learning lately: sometimes being present doesn’t feel peaceful. Sometimes the present just feels sad. The outlook feels bleak. The ground feels shaky. And when I try to meet the moment honestly, what I find there isn’t calm, it’s grief.

For a while I tried to talk myself out of it. *Be grateful you still have your job. Be grateful for your health. Be grateful, be grateful, be grateful.* And I am grateful. But gratitude and sadness aren’t opposites. They can live in the same body at the same time, so I’m trying something different. I’m trying to give myself grace to just be sad.

It’s okay to feel this. It’s encouraged, actually. It’s so deeply human. In a moment where so much of our world is being automated and optimized and predicted by machines, the ability to sit with an emotion you can’t fix or solve for might be one of the most human things we can do right now. It has been hard for me to admit this since I’m as a yoga teacher. What does it mean to teach presence when my own present feels so uncertain? What does it look like to lead a grounding exercise when I feel ungrounded? To cue someone to soften their jaw while mine is clenched?

I don’t think it means I’m a fraud. I think it means I’m human. But there’s a quiet pressure that comes with being the person in the room who’s supposed to model steadiness. In my head, I hear this sense of pressure to perform calm, to have my own practice “figured out,” to embody the thing I’m offering.

The truth is, I don’t have it figured out. I’m in it, same as everyone else. And I’m starting to think that might actually be the more honest thing to share.And, what I realized is that being sad together feels better than being sad alone. So if you’re anxious and worried and tired and sad, I am too. And I don’t think either of us needs to rush past that.

Let’s practice in community and come together! I’m so grateful to have gotten to practice and mingle with my fellow Wellesley alums last week. It was so nourishing to move with others, to be inspired by other’s stories, and to connect with a community that I sometimes forget I have.

Upcoming Offerings:

• Ground and Gather: Sunday 4/12

• Weekly Meta corporate classes

• Prenatal 75 min class Sundays at 7:30pm on Zoom (message to join)

My dog recently sprained her ankle. She stood on her hind legs and landed on her front legs funny when she came back down. Watching her limp broke my heart, and I had to carry her a little over 1.5 miles all the way from the park back home. She looked miserable and slept all day and into the next. Although seeing her in pain was difficult, I was happy that she listened to her body. Dogs are notorious for not knowing their limits and going all out until they can't. So for her to be present, listen to her body during these cold, cold months, and rest was an inspiration to me.

When I'm sick, I can't sleep. I busy myself around the house or watch TV or even try to listen to a podcast in bed. I'm the worst at just letting my body rest, even though that's exactly what I need. And since I haven't gotten sick this season (yet! knock on wood!), I'm taking this moment of good health to remind myself to rest and to attribute my health to the rest I have been taking.

In both Ayurveda (yoga's sister science) and Traditional Chinese Medicine, winter is about conserving energy, restoring energy, and protecting your deepest reserves. They differ in their approach to rest. Where TCM is much more explicit about needing rest (more sleep, less sweating/activity, staying warm, minimizing stress and stimulation), Ayurveda sees winter as a time to build strength and immunity because your digestive fire (agni) is actually strongest in winter, when the body draws heat inwards.

To allow myself to rest and to loosen the grip of the attention economy on me, I'm also trying to allow myself to be bored. If there's nothing to do, instead of grabbing my phone and scrolling, I'm letting myself just sit there. I'm trying to use that moment of time to rest, to maybe focus on my breath, and maybe to even meditate…but even that might be trying to do too much. I feel like my days fly by so fast because I'm trying to fill every moment with something, something to work towards all my goals for self improvement. I'm trying to change that. I'm fighting this urge to feel productive. The need for me to be working 24/7 is unnecessary when my health and well-being are on the line.

This brings me to another thing Poppy does really well. I’m obsessed with my dog and I think she’s amazing! And, she’s great at letting herself be bored. We’ll sometimes catch her just lying there and staring at the wall. She’s clearly not looking at anything and just being. She’s being a dog, and seeing her be a dog reminds me to be human.

When she stares at nothing, it reminds me of pratyahara, the fifth limb of Patanjali’s eight-limbed path of yoga. It’s typically translated as a withdrawal of the senses, or turning inward. Instead of being distracted and having your attention be pulled by a notification on your phone, a siren outside, or a thought about something you need to do, what would it feel like if you turned all that sense-sensing towards yourself? Let your antenna tune to yourself and listen to what song or show is playing. There’s nothing to see, nothing to do, nothing to take in but what’s already inside. Maybe, like my dog Poppy, we can also try to stare at a wall so that we can withdraw and tune in.

Maybe I should be worried but she mostly seems fine. If you’re a dog expert or a vet and don’t think she’s okay, please let me know!

I know there are a lot of new subscribers - thank you for being here. Come move and rest in community!

Here are my offerings for the month:

• Weekly Wednesday corporate yoga classes at Meta

• Weekly By Donation Thursday 75 min Online Prenatal Yoga class (message me for invite)

• Or contact me for a private session of yoga or a tarot reading

I’m planning a public class in Brooklyn in March and another by donation class at a studio. Keep an eye out!

• Saturday, January 3rd

  75min🤰Prenatal Yoga🫃 on Zoom (email to join)

• Monday, January 5th

  60 min 🤰Prenatal Yoga🫃 on Zoom (email to join)

• Friday, January 9th (and every Friday after in Jan)

  Find me and other practitioners at Community Reiki on Friday’s from 5-6pm at Maha Rose with Mariko Tamegai.

• Wednesday, January 14th

  45 min 🧘🏻‍♀️Corporate yoga🧘🏻‍♀️ returns to Meta (join the Yoga Club on Workplace for details)

• Sunday, January 25th

  10:30am - 12:30pm 🫖Yoga + tea service🫖 in person in Boreum Hill. Sign up on Partiful!

Coming in February:

• Spend Valentine’s Day with me for a sound bath + yoga event. I’ll also be doing tarot card pulls for people interested!

• By donation prenatal yoga class in person

As always, privates in person and virtually are available!

There’s also Thay’s story about being a young monastic activist. I’ve heard it told at the monastery but he also writes about it himself in Zen and the Art of Saving the Planet. TBH I forget the details but the gist is that a group of them were running out of options and weren’t clear how to help. Instead of rushing and doing something out of fear and anger, they waited, came back to their breath, and meditated. This seems like the opposite thing they should be doing. Time is precious! Action needs to be taken! However, because they waited, they were able to tap into allies with the French embassy to support them and get more time.

Whether the problem or situation you’re facing is big or small, consider taking a beat and coming back to your breath.

I’ve only missed one other flight before so this holiday season when we were still waiting inline for the Delta desk after arriving there over an hour ago, I was ready to freak out. I could feel my heart race picking up. I noticed my jaws and hands clenching as a way to stay calm and hold my anxiety. I was worried about my dog. I was worried about lost luggage as a rest of this travel plan hiccup. I was worried about having to tell family that we missed it. Then, I remembered I should just come back to my breath. There wasn’t anything to do in that moment other than wait. I was able to calm myself down and even have a laugh about it with my husband. In the end, we were able to calm ask the desk for help and we were able to get on standby for one flight and confirmed on another. Our dog and luggage got on the plane. Everything worked out. There wasn’t anything I really needed to be worried about.

When I took the Zen and the Art of Saving the Planet course through Plum Village, they said if nothing changes after the class then it would have been a failure. I’m grateful that I had this opportunity to practice and do something different.

Yesterday, during the winter solstice in the northern hemisphere on the 21st, I was thinking about the 3Rs and considering what that might look like when you’re not pregnant but are dealing with difficult times, such as the long dark of winter. To me, rest, reflection, and reset are key.

Rest: Taking a cue from nature, winter is the time to rest and to recover so that we have energy to spring! With the colder weather, we naturally want to feel cozy and warm at home. Hygge is a concept for a reason! The cold makes it hard to move faster. Instead of fighting nature and your instincts, allowing ourselves an opportunity to acknowledge what we might be going through and working with it. Slowing down and allowing yourself to do less. (I struggle with this one, especially in a city like NYC)

Reflect: Within yoga, we talk a lot about winter being about folds and this being a metaphor for introspection, folding into oneself to be with one’s self. By folding, we take away stimulation from the outside world so you just see yourself. This creates optimal conditions for reflection! When there’s no shiny thing out there to distract you, where can you land your attention?

Reset: In TCM, winter is the season of waters and the kidneys are associated with water. Winter represents a period of rest and restoration that prepares you for renewed activity and growth. This cyclical "reset" is essential for maintaining long-term health. As the next step to reflection, using winter to let go of the old and prepare for renewal. I also think of it as the rerouting of my waters/energy. Coming off of summer and fall, I need to reset in order to settle into new routines that fit better with the changes of the winter.

This moment of balance, equal parts light and dark, invites us to pause, ground ourselves, and notice the shifting season around us. In practice, we focused on the hips: opening, softening, and finding stability in the places where we often carry so much tension. Just like trees root deep into the ground while letting their leaves fall, we too can find steadiness in the body while allowing what no longer serves us to gently release.

In practice, we focused on the hips: opening, softening, and finding stability in the places where we often carry so much tension. Just like trees root deep into the ground while letting their leaves fall, we too can find steadiness in the body while allowing what no longer serves us to gently release.

If you’d like some inspiration for how to bring these themes into your own practice:

• Try Tree Pose (Vrksasana) — a simple but fitting reminder of balance and rootedness during the equinox.

• Explore hip mobility by moving between Half Pigeon (Eka Pada Rajakapotasana) and Fire Log (Agnistambhasana). Challenge yourself to transition between the two without using your hands — noticing both the strength it takes and the places where you might need to soften.

One passage I shared was from Thích Nhất Hạnh:

“A leaf has its own identity, but at the same time it is not just a leaf. It is made of non-leaf elements: sunshine, clouds, the tree, the soil… When the leaf falls, it continues in other forms of life.”

This teaching reminds us that letting go is not an ending — it’s a transformation. The equinox is a perfect time to honor both the balance of holding on and the courage of releasing.

As we move into autumn, may we carry this reminder: we are more than what we shed, and everything we release makes space for something new to take root.

🍂 What are you letting go of this season?

When I sequence a class, I consciously string mobility, flexibility, and strength practices together with the goal of healthy aging.

If you know me, you know I’m obsessed with my dog. Every morning we go on long walks at the park and Poppy gets zoomies. Her unfettered joy is contagious. I just love the way that she let’s her happiness take over.

I also love the way she just stops and notices things - a squirrel, a plane, a bird, etc. Her stopping makes me pause and try to be a bit more present.

…and then sometimes she running off and barking at a car. And then I’m reminded that we all need some help coming back to what’s important.

Come take a class with me! With lineage in hatha and vinyasa yoga, I love teaching classes that are fun and creative yet challenging and grounding. Unite your mind, body, and breath with community!

Hope to see you in class!