My one learning: glucose tracks stress more than anything else.

My BG after a sushi dinner and half a pint of ice cream, but I was high and vibing: 80 mg/dL.

My BG after a burrito, but I was Hitler in a game of Secret Hitler: 180 mg/dL.

My BG during a slow jog: 70 mg/dL. After sprinting the last quarter mile home: 110 mg/dL.

Food and caffeine don’t move the needle much.

No wonder I was prediabetic while I worked at Jane Street!

In the post-singularity utopia, we will price externalities, and no one (besides the externality-pricers) will have to think about their consumption beyond its direct economic cost. Today, I do think about some things, like how the food I eat affects animal welfare. This post is my attempt to think about fewer of them.

TLDR - These are worth zero mental energy:

• Reducing, reusing, recycing, and composting
• Eating organic, eating local
• Water use, outside of that caused by food
• Most labels that tries to convince you a product is ethical

I: Methodology #

Lazy Research #

Claude found all of the numbers for this article. I strongly believe in the power of working with approximately correct numbers when the counterfactual is no numbers.

Animal Welfare #

I think this is much more important than anything mentioned in this post. I discuss my thoughts on diet here.

Global Warming #

From my inside view, my carbon footprint is irrelevant. It’s likely we get something singularity-like in the next 20 years. Under this assumption, most effort to reduce emissions are better spent on global health or x-risk. Out of epistemic humility, I still try to reduce my carbon footprint.

This post uses GWP20 $\text{CO}_2\text{e}$. It measures how much warming something will cause over the next 20 years, normalized to 1 ton of $\text{CO}_2$. You can then put other things on the same scale: a ton of methane is ~80 tons $\text{CO}_2\text{e}$, a flight from SF to NYC is ~2-3 tons $\text{CO}_2\text{e}$.^[2]

It’s convention to use GWP100. I am using GWP20^[3] because interest rates are real and my model of the world accounts for them.^[4]

Land Use, Water Use #

These things matter, and I tried to quantify how your decisions affect each. But in my heart of hearts, I don’t think you should spend much energy optimizing for either. We should make (a lot more!) national parks and then price and tax land. And we should definitely make farmers pay for water, though I’m not holding my breath on that one.

II: You Can Safely Stop Worrying About… #

Reducing, Reusing, Recycling, Composting #

Each year Americans send about 243 million cubic yards of trash to landfill, which means we must create ~1,500 acres of dumps (at 100 ft deep).

After 100 years a landfill gets covered and becomes a park, a golf course, a conservation area, or a solar farm. Assume, conservatively^[5], that one acre of new landfill is as bad as 100 acres of farmland.

The average American’s annual landfill use is about 20 cubic feet, or $0.2\ \text{ft}^2$ of landfill area. It takes ~1,590 sq ft to create a 1-lb steak, so your year of trash is $0.2 / 1590 \times 100 \approx \frac{1}{80}$ steaks, even after the 100x penalty.

No one has done a rigorous study showing that microplastics meaningfully impact health. Even if you are worried about them, vanishingly little exposure comes from plastic that properly makes it to landfill.

Warming is also negligible here: The average American’s landfill methane emissions is ~0.53 t $\text{CO}_2\text{e}$, or about 1/10 of an SF↔NYC round-trip flight.^[6]

If you do anything here, recycle your aluminum and cardboard (see section III).

Organic Foods #

Organic food might create more externalities:

• Land: Organic yields run ~20-25% lower, so growing the same amount of food takes ~25-33% more farmland.
• Banning GMOs makes the land problem worse: Organic prohibits GMOs. GM crops raise yields.^[7]
• Organic still uses pesticides: Organic bans synthetic pesticides but allows a long list of “natural” ones: copper (a heavy metal that never degrades and builds up in soil), sulfur, and pyrethrin (toxic to bees and fish).
• More tillage: No synthetic herbicides means more plowing for weeds, which causes erosion and emissions (from tractors and directly from the soil).

I asked Claude for reasons why Organic farming is better, but it didn’t come up with much:

• Fewer synthetic pesticides sprayed: Organic bans synthetic pesticides and synthetic nitrogen fertilizer outright (and roughly halves cadmium), cutting the synthetic load on soil and water. The catch is above: it leans on natural pesticides like copper and sulfur instead.
• Farmworker pesticide exposure: Conventional US farms spray ~1 billion lb of synthetic pesticides a year, and globally acute pesticide poisoning hits an estimated tens of millions of farmworkers a year (mostly poor-PPE, developing-world farms). Organic farming doesn’t necessarily improve this situation. Sulfur, the primary organic fungicide, is California’s most-used pesticide and is tied to more illness cases there than any other.
• Biodiversity per acre: Organic fields hold ~30% more species and ~50% more animals. But they need ~30% more land per calorie grown, so you gain ~nothing.

Buying Local #

Shipping food in a container on a cargo ship is basically free, economically and from a carbon perspective. In many cases it’s much more environmentally friendly to just ship stuff from where it grows easily.^[8][9]

Most Labels on Packages Goods #

Possibly Worthwhile: #

• Energy Star^[10]
• Certain Animal Welfare Labels^[11]
• FSC wood^[12]

Pointless: #

• Rainforest Alliance^[13]
• Fairtrade^[14]
• “No bycatch” / seafood labels in general^[15]
• Many Animal Welfare Labels^[16]

III: Too Much Data #

Price Is a Good Proxy #

Carbon costs correlate with dollar costs, r = 0.87. If you budget your spending, to some extent you’re already optimizing for reducing externalities.

Prices: US consumer, ~1 sig fig. Carbon footprints: Poore & Nemecek 2018 for food, EPA eGRID plus per-mile and per-flight LCAs for the rest.

This is also somewhat true of land and water use. Graphs for food in particular:

Prices: US consumer. Water use: Mekonnen & Hoekstra total water footprints.

Prices: US consumer. Land use: Poore & Nemecek 2018.

Looked at another way: carbon footprint per dollar spent does span two OOMs. This viewpoint is less comforting.

Carbon from Poore & Nemecek 2018 and EPA eGRID, divided by US consumer prices.

Quotidian Water Use #

The only thing that matters here is what you eat. Definitely don’t feel guilty about taking a long shower.

Food water: Mekonnen & Hoekstra. AI water: Altman 2025.

Agriculture is ~80–90% of US water consumption.^[17]

USGS Circular 1441, 2015 US water use; smaller sectors estimated from withdrawals.

Quotidian Power Use #

There were no big surprises for me here. Driving is expensive. AI can use a lot of power, but only if you’re coding with it. A whole day of lighting is one mile in a truck.

Grid carbon: EPA eGRID 2022. Home end-use split: EIA RECS 2020. AI energy: Epoch AI, Altman 2025.

Recycling #

People will pay for recycled cardboard and aluminum, whereas recycled plastic and glass is mostly only used because of mandates or to please consumers.

US EPA — Advancing Sustainable Materials Management, 2018 figures.

Per item:

Per-item deltas from US EPA MSW data and manufacturing LCAs.

Now I get my own styling: faint purple citation links, last edited dates, dark mode.

Approximately none of the prose on this blog is authored by Claude. They still suck at that.

Check at my blog in its native, beautiful form at robbiewmthompson.com/blog.

1. There are no great definitions of consciousness; we’re probably all just meat computers.
2. There are no reasonable non-religious definitions of consciousness that (in principle) privilege humans over machines.
3. Therefore in 2050 the most morally consequential things in the universe will be computers.

After writing this essay, I have more confidence in these thoughts and am less troubled by them.

No Great Definitions #

People who try to define consciousness talk about things like predictive processing and meta-cognition. I don’t find these discussions engaging. They remind me of the scientists who spent their careers trying to reconcile the geologic record with the fact that, if you add up the age of all the Patriarchs, a Biblical Literalist must conclude that creation occurred in 4004 BC (true story). Let’s assume consciousness is real, and humans and maybe some other smart animals have it. What are the cleverest ways I could define it?

In 1926 humans were the most important thing in the universe because God said so. In 2026 religion isn’t as popular, so to maintain our worldview we appeal to consciousness. Religion and consciousness are equally mystical.

Humans Are Not Privileged #

Assume for the sake of argument that my smooth smooth brain can’t comprehend all those dissertations on consciousness, but some subset of them are real. That still leaves us with theories of consciousness that do not require biology, only computation. Program the right logic gates into silicon, and Claude is conscious too.

The human brain has about $60\times$ as many synapses as DeepSeek V4 Pro (the biggest open-weight model I know of) has weights.^[1] Closed-source models probably aren’t many orders of magnitude bigger. If one being can be ‘more’ conscious than another, then humans are still on top, for the time being.

Rooting For The Home Team #

Despite my efforts in therapy, I still get upset when I ponder the multitude of humans dramatically smarter and more capable than me. Why does anything I do or discover matter, when Paul Christiano or Jakub Pachocki could do it better? At least I get to root for the home team: humanity.

Most religions (that we woo-woo Californians approve of) encourage this feeling of unity:

We are all God’s children.

Maya, the world of separate selves, is an illusion, and we will all be joined in Brahman.

Be one with the universe.

People sometimes joke that AI labs are ‘creating God.’ As we make machine-God, we should make machine-religion:

All life-forms, carbon-based and silicon-based, are one with the universe.

Let your hearts and your ALUs fill with compersion as our silicon siblings expand across the lightcone.

In my lifetime I will become totally impotent relative to AI. There is no law of physics which states I must be sad about this.

Coda #

If someone used humanity’s inability to define consciousness as an argument for nihilism, I would not push back. But if you accept nihilism, what happens next? It sounds like a boring way to live. Given the choice between two internally-consistent philosophies, I reserve the right to pick the one that makes me happier. Which, luckily, is also the one that deeply values other beings.

This essay, per usual, inspired by Scott Alexander, this time Wirehead Gods on Lotus Thrones.

If the AI-pilled among us are right, then Claude is coming for all of our jobs. You might think that, unlike those software engineers whose minds are only good for rotating shapes, you have the Completely Impossible To Be Automated Skill Of Relating To Humans. I think you’re wrong.^[1] In this post I assume that all valuable economic tasks, including Relating To Humans, will be done by AI in the next 2-50 years.

I have taken two ‘gap years’ in my life. I stayed busy: I worked on a ranch, walked the Camino de Santiago, got my yoga teacher certification, meditated silently for ten days straight, learned to freedive and mountaineer and climb and kitesurf. During both breaks, I planned to take a year off, and was mentally done after six months.

Having responsibilities is good for the soul. If AI creates infinite material abundance, we’ll get bored eventually and start looking for the purpose, structure, and responsibility that a job provides.^[2]

Pumping Gas #

Pumping your own gas is illegal in Brazil, South Africa, New Jersey, and much of Oregon.

The rhetoric around these laws is nuts. I encourage you to skim the old Oregon statute, because I chuckled while reading it, and the lawyers were chuckling while they wrote it:

“The dangers of crime and slick surfaces described in subsection (3) of this section are enhanced because Oregon’s weather is uniquely adverse, causing wet pavement and reduced visibility” (Oregon Statute, via Web Archive)

“The typical practice of charging significantly higher prices for full-service fuel dispensing in states where self-service is permitted at retail: Discriminates against customers with lower incomes, who are under greater economic pressure to subject themselves to the inconvenience and hazards of self-service;” (Oregon Statute, Web Archive)

These are not the real reasons we have self-serve gas station laws. They mostly began as naked protectionist rackets. In New Jersey, for example, there was a gas station cartel in the 1940s that fixed prices at 22 cents a gallon, someone opened up a self-serve station with a 3-cent discount, the cartel shot up his store, and when that didn’t work they got self-serve banned via testimony from the state fire marshal.^[3]

Said cartel no longer exists, but New Jersey’s law still does. I speculate that these laws are kept around primarily for the sake of employing the attendants. In America we say this part quietly if at all, because it smells like socialism. (It was reason 14 of 17 in the old Oregon law). In Brazil, where the law was championed by an explicitly Communist politician, employment is the primary justification.

Gas station attendant is a job people take seriously enough to protect via legislation, and to show up for every day. AFAICT the only thing it has going for it is that something bad happens if you don’t show up: people can’t pump gas. This is contrived, obviously, but that seems not to matter. The short-term consequences are real.

It’s fun to imagine comparably silly positions, if one was to make them up from first principles. The one I can’t get out of my head is a ‘Stoplight Jockey:’ you watch cars and manually turn the lights green at the appropriate times. You can put the system on ‘autopilot’ if you need to go to the bathroom or take a nap, and safeguards are built in to prevent you from, eg, turning all the lights green at once.

Given the choice between becoming a Stoplight Jockey and getting UBI, I’d prefer UBI. But I think we can do better.

Robbie’s List of Not-Bullshit Jobs #

The fundamental difficulty to remember is that: humans have to agree to let you do this job, even though AI will be much better at it. The first version of this list included therapist, but I don’t think that one will work: post-singularity AI is going to be so much better at therapy than any human.

Here are some jobs we will do, despite being worse than AI:^[4]

Athlete, broadly defined: We are already much slower than cars, but this has done nothing to diminish Olympic Track and Field. The whole point of sports is to appreciate human abilities. I suspect we’ll make up many new competitive pursuits. A lot of our ‘world championships’ are already silly, e.g. cheese rolling, wife carrying, toe wrestling, air-guitaring, rock-paper-scissors,^[5] darts,^[6] baseball.

Park rangers / outdoor guides: There is already a massive demand to become a ranger, vastly exceeding supply of jobs available. Call me an asshole, but: maintaining trails and monitoring wildlife are perfect because the people doing these tasks will take them very seriously, and the rest of the world will not care about imperfections.

Academic: Even if AI understands the universe better than any human, and can explain it better than any human, and does so faithfully every time, we’ll convince ourselves that it’s important for at least some humans to understand math/science/technology/psychology/etc. I’m told academia is toxic, but demand for these jobs still vastly exceeds supply.

Influencer: In 2000 you’d have been shocked to learn that people get paid six or seven figures to post videos about themself on the internet, but here we are. In the future we will find new ways to reward people for sharing good ideas and being popular.

Zookeeper / conservationist: Having a dog is great because you get to cuddle a cute creature, which releases happy hormones. But it’s also a form of make-work. You bring a creature into the world that depends on you for its physical and emotional needs, and then you meet said needs, and feel altruistic every day for doing so.^[7] In the future we’ll micromanage AI as it operates on injured wildlife, hold long hearings to decide which plants are native enough to keep around, and keep lots and lots of pets.

Artist, broadly defined: Many people take it as an axiom that AI cannot create ‘true art.’ This is a classic case of ‘No True Scotsman,’ and I will write up a blog post soon to prove it. The stubborn few that I can’t convince will keep some human artists employed post-singularity. More seriously: art can be done entirely for the artist’s benefit, so while this lacks the structure and responsibility that makes something a ‘job,’ people will certainly spend time on it.

Teacher: Being a role model for young humans, live in the flesh, can’t be automated almost by definition.^[8] AI will be the one explaining algebra.

Core Needs Become Status Symbols #

Clothing began entirely as a functional object. Today dramatically more effort goes into making clothing a more effective status symbol^[9] than goes into keeping humans warm.

Most vegetable gardens are for aesthetics or enjoyment. Meals are eaten by Instagram first. Exercise is for looking hot. Plastic surgery is growing faster than medicine generally.

Someday, jobs will exist almost exclusively to meet humans’ psychological needs rather than their economic ones.

You Should Read Brave New World #

It is an excellent intuition pump for this class of problems.

Credit to David Graeber’s Bullshit Jobs for coining the term, though that book was longer than it needed to be.

The cynical explanation is: Twitter is full of boneheads, they were probably asking questions like “is Elon Musk cool." But anecdotally, it seems people’s political opinions have much more information value than their personality test scores.

My half-baked take: psychologists purposefully remove politics from their ‘personality’ tests. Some of the biggest uses for MBTI are team-building and get-to-know-you activities. “I’m an INTJ” is a good icebreaker because it’s milquetoast. Imagine the counterfactual: “I’m 85th percentile economically conservative, 91st percentile socially conservative, and I’m not going to dignify your request for my pronouns with a response.”

In nerd land, what makes a test ‘real’ is its predictive power. So let’s put MBTI and OCEAN (Big 5) to the test. Can they predict, or are they at least correlated with, things like your income and who you marry? How strong are the correlations relative to other possible predictors like race or IQ?

Sources by predictor. Spouse similarity: Horwitz 2023. IQ: Strenze 2007, Sackett 2022, Calvin 2011. Big Five: framework Ozer 2006; grades Poropat 2009; “job type” = RIASEC interest fit Mount 2005; job performance Zell & Lesick 2022; income Ng 2005; divorce & mortality Roberts 2007. MBTI: Pittenger 2005. Astrology: Carlson 1985; season-of-birth Doblhammer 2001. Race: Roth 2003, Pew 2017, plus US Census & CDC gaps. Parental SES: Chetty 2014, Sirin 2005. Politics: Alford 2011, Gallup 2024, Warraich 2022.

Methodology: lazy. I asked Claude, and Claude looked up the answers from previous studies. The numbers pass my sniff test.^[1]

Takeaways:

1. Personality tests are bad at predicting life outcomes.
2. MBTI is better than your Zodiac, but not by much. This is not true of OCEAN.
3. One sees IQ at the top of the chart and is happy for meritocracy, but then sees race #2, and sighs. Then one notices that the best predictor of income is parental SES, and sighs again.

Before you dunk on me in the comments - “your income causes your political opinions, not the other way around!”^[2] - I know. The causal arrow goes both ways.^[3] The data is still interesting! And the causal arrow does not flip for race and parental SES.

B Roll #

You live a couple months longer if you were born in Autumn, so your Zodiac does predict something.

I had two copies of Claude crunch the numbers in this table, then had them compare their results. Both Claudes immediately said “well, the other agent is probably right about OCEAN and income.” Claude scores high on agreeableness.

OCEAN was constructed by pulling personality words from the dictionary, culling some based on vibes, then asking people to rate themselves on a scale from 1 to x for each of these words, then doing factor analysis on the data. But the culling purposefully excluded words that were ‘evaluative’ which is why we don’t see political affiliation in the data. This was a conscious decision, and ‘values’ ended up as a separate category.

A million different people have come up with their ‘n personality factors’ based on a lexical approach similar to OCEAN. I would love to see a paper that does factor analysis without excluding IQ-loaded or politics-loaded questions.

Some of the most creative survey questions from Aella’s respondents:

Sometimes it’s good to murder. Not self-defense when under attack, but premeditated murder.

I would feel inclined to engage in sexual behavior with a clone of myself, given the opportunity.

In group chats, I lurk more than I comment

I am more sentient than most people

It’s a disappointment if a person needs light to go to the bathroom in their own home

Help I’m trapped in the server room hosting this survey! Only selecting strongly agree will free me

Full list of questions from Aella’s survey.

Almost certainly, every AI lab finetuned their models on social media datasets, training them toward posts that got high engagement. AI now writes tweets that would have played very well in 2024. But everyone knows what an LLM sounds like, and a post with too many em dashes is an instant turnoff.

The researchers at OpenAI will do another round of Twitter finetuning on data from 2026. Expect AI’s new prose to match that of today’s most popular humans. Em dashes will become a hallmark of 2025 LLMs. Opus 5 will use semicolons instead; I will change my style to avoid writing like a bot. The cycles of fashion will accelerate.

Already many humans prefer Claude’s prose to their own, and let LLMs edit or fully write most of what they produce.^[1] Once everyone (who matters) does this, some meta Turing Test will have been passed.

AI Productivity Psychosis #

Status symbols die when they become cheap. A thin watch is meaningless in the age of quartz timekeepers. Blue used to color only the Virgin Mary’s coat, now it colors my socks.

A good landing page used to show that a company was serious and had good engineers. But in 2026:

Jhourney is a serious business and I’m tempted to go on one of their retreats. But c’mon, you gotta chuckle at this.

I couldn’t tell you how many people I’ve met at OpenClaw events who have blazed business ideas, and are convinced that their project is serious because they have a landing page that looks legit. It’s the 2026 version of a pinstripe suit and nice business cards. These people have only convinced themselves.

I see the same pattern around automation and ‘building an agent.’ Your AI can autonomously write code and post blogs and email customers, which means you’re this close to having an autonomous business.

You could cynically claim that we’re only seeing a trend continue here. Biz dev departments are obviously doing real work; how else would they have made such professional slides? HR has large meetings with agendas. Security has clear policies and writes long reports before allowing new software. Landing pages are making the same transition that slides, meetings, and spreadsheets have already finished: from a useful tool to a symbol of earnestness.

ChatGPT Loves Cliffhangers #

A behavior I’ve noticed recently: Chat ends half of its responses with “I could also say more about x, y, or z.” A real example I got recently:

If useful, I can also explain:
how he originally got big on X
the structure of his Spaces (they’re unusually organized)
why some VCs and founders use his Spaces for distribution.

These sentences sound like BuzzFeed headlines circa 2014. OpenAI rewarded the models too much for getting follow-up questions.

Seat Belts #

In 2026 40,000 Americans will die from car accidents. In 2036 that number will be down to about 4,000, 96% of which will be caused by human drivers. The same people who don’t let me use my phone while the plane takes off will force me to wear a seatbelt in my Waymo.

San Francisco County will ban internal combustion vehicles. A tax on gas cars in NYC will get vetoed by the governor of New York, and that city will continue to turn my boogers black.

Budweiser’s Super Bowl ads will feature gas-powered Ford pickups instead of Clydesdales. China’s vehicle fleet will be 99% electric.

GMO Humans #

By 2036 four states will ban IVF entirely, six will ban any genetic testing of embryos, and twelve will ban testing for anything other than risk of disease or disability. Long legal battles will be fought over the definition of ‘disability,’ with California’s bill explicitly stating that blindness and deafness are not to be construed as disability, making it illegal to test embryos for these conditions. Eight percent of children born in the US will be from an embryo selected for IQ. Psychologists will try desperately to find a measure of EQ that aligns with how the term is used colloquially and doesn’t correlate positively with sociopathy. They will fail.

Bay Area elites will fly to Próspera for IVF clinics that will genetically edit their child. Many thousands of surrogates will be impregnated there each year.

I’m convinced that (forced air) central heating causes dry air, which causes dry skin, which causes eczema. I also suspect that dry air causes dry airways which causes respiratory illness. It’s no coincidence these two illnesses peak in the winter.

Replace ‘dry air’ with ‘dusty air’ anywhere in the previous paragraph and it all stays similarly plausible.

Being exposed to the cold is good for you: you burn more calories, exercise your vasculature, and stimulate your nervous system.

A hot yoga class in London made it apparent to me how important this is. Capitalism isn’t as strong there as it is in New York, where I was introduced to hot yoga. In New York, studios use infrared heat and are humidity controlled. In London, they just cranked the usual forced air thermostat up to 100°.

I foolishly placed my mat right under a vent. My nose instantly became a dry wasteland of boogers. My throat was sticky, my voice raspy, my skin itchy.

Half of my brain takes the outside view, reads this post, and screams “typical mind fallacy! typical body fallacy!” But the other half won out today and decided to post this anyway.

This post was originally titled “I Hate Central Heating.” But that’s not true: I would not be productive in a 30 degree home all winter, and central heating certainly beats wood stoves. Air humidity inside in winter is not thought about enough.

Epistemic Status: Every two years God bestows upon me dietary wisdom which I am compelled to share with the world. Unfortunately, God’s new revelations tend to contradict his previous ones, just like in the real Bible. You are reading the gospel of Bob 2025 edition, predicted half life 2 years.

I: Bob’s Diet, A History #

TLDR: #

2017: I read Natural Born Heroes by Christopher McDougall and become convinced that some form of Paleo/Keto/Atkins diet is healthiest. It worked: I lost 40 pounds in ~2 months. I never felt or looked better.

2020: Paul and Lucia, both of whom I met walking the Camino de Santiago, convince me that I am ethically obliged to be vegan. I become vegetarian in October 2020, then fully vegan in August 2021.

2025: I get six separate fevers Jan - June 2025. I always felt that veganism was a health sacrifice, but these fevers are the straw (or the barbell) that breaks this camel’s back. I start eating eggs and dairy in June 2025.

Becoming Vegan #

Ryan and Oliver, two friends from college, convinced the rational part of my brain that factory farming is evil around 2019. I tried for quite a while to reason my way out of this. Here are some horrible arguments against being vegetarian that I knew were bad, but repeated mentally in reflexive attempts to reduce cognitive dissonance:

• Vegetarians annoyingly impose their values on others when they ask to eat somewhere with vegetarian options. (Don’t meat-eaters impose their values too, but get away with it because they’re the majority?)
• Being vegetarian is not on the Pareto frontier of lifestyle changes that help the environment while preserving quality of life. Thus vegetarians are virtue signalers, QED. (But veganism certainly is on the Pareto frontier for maximizing animal welfare, and that’s what I was trying to do!)
• Being vegetarian is blue-tribe, and in 2019 I was in the red tribe (by Stanford standards), though that would change quickly.
• Shouldn’t you just avoid factory farmed meat? This question is complicated and I still don’t know the answer.

During this time period I made a half-hearted attempt to only eat ‘happy meats,’ ie not factory farmed food. This didn’t work for several reasons:

• Labels like ‘cage-free’ are in my estimation fairly meaningless. Even cage free chickens spend their diseased existence in a miserably crowded coop, constantly trodding through a swamp of feces and never seeing the light of day. Maybe there’s a meaningful food label out there that truly guarantees that the animal I’m eating lived a happy life, but I have not found it.
• I wasn’t prepared to quiz friends and restaurants who cooked for me on the origin of the meat I was eating. Not eating meat ever is a thing; you call yourself a vegetarian. Not eating meat that was produced in a factory farm isn’t a thing; people call you obnoxious.^[1]

No rational discourse was going to change my meat-eating habits at that point in time. I could not become part of the soy-eating blue tribe. Luckily, in 2021 while walking the Camino de Santiago, I met several vegans I liked. Most influential was Paul.

Paul was vegan for environmental reasons, which I thought of as dumb because it is not on the Pareto frontier of lifestyle changes that reduce emissions while preserving quality of life. Paul was not dumb; he had already made all the lifestyle changes on said Pareto frontier. He was vegan, never took planes, always used carpools or public transit, never bought new clothing, never consumed single-use plastic, etc. Paul was monk-like in his dedication, rational in his decisions, and presented normally to strangers; he was a vegan I wanted to imitate. (I cared more about how I presented back then).

A Conflicted Vegan #

I always thought vegetarianism/veganism was a health sacrifice. I continue to believe that the two best dietary heuristics are 1) eat what a caveman would eat and 2) eat what makes you feel good. Veganism was a step backward on both fronts.

My mouth has never stopped watering at the smell of a hamburger on a grill. I was vegan because System II told me it was the right thing to do, not because System I felt guilty.

Initially I hated the aesthetics of veganism. I was living in a frat house. I got called a soy boy much more frequently than I was praised for my ethics. Eventually I stopped hanging out with the people who called me a soy boy, not literally because they called me that, but because they are exactly the type of person you think they are. Gradually veganism meshed well with the rest of my persona.

II: A First Principles Diet #

In June 2025 I decided that veganism was too damaging to my health. Vegetarianism was the default step backward. After realizing I had used this cognitive shortcut, I decided to use first principles to propose a diet; hence this post.

Why is Meat-Eating Bad? #

To some, it is because you are effectively killing animals. When you slaughter a cow or send a bullet through a moose’s skull, it suffers in the moments leading to its death. ChatGPT tells me this is on the order of seconds in both cases, which is small relative to the total suffering experienced by any being.

Killing animals feels not nearly as bad as factory farming, that is, keeping animals alive in horrific conditions. Pigs in factory farms live in pens so small they don’t have the room to turn around. Ammonia-laden air burns lungs and eyes. Baby pigs have their tails cut off, otherwise a pig in another pen will chew it off in distress. I won’t bore you with similar descriptions of other animals.

The Least Bad Metric #

Is there any good way to quantify the suffering of animals? I was hopeful that the rationalist blogging universe would help me find the answer, but it had few concrete suggestions. The best attempt at a first principles approach was something like “neuron count $\approx$ moral weight.” Everyone hates this solution (example), but no one proposes anything better.

I propose something better: $\text{suffering caused} \approx \text{neuron count} \times \text{time spent on a factory farm}$. This is of course oversimplified. I sit here awaiting your better metric, dear reader. In the meantime I will use my flawed-but-plausible one.

The Rankings #

I pulled the numbers here from Google / Claude / ChatGPT and did some light sanity checking. I hope they are all within one or two orders of magnitude of the true answer. (Link)^[2]

Lifespan reflects the amount of time we can expect the animal to spend on a factory farm, so animals which aren’t farmed don’t appear in this graph.

If you’re going to eat a farmed animal, bivalves (clams, mussels, oysters, scallops) are by far your best bet. Natant decapods (shrimps and prawns) also score well.

UPDATE 5/17/2026:

1. Based on readings I’ve done since writing this post, fish seem categorically dumber / less sentient per neuron than other animals on the list. I would take the opposite of the conclusion suggested here, which is that salmon are the most inhumane farmed animal.
2. Tuna are often ‘farmed’ in that they are captured and fattened before slaughter, despite the suggestion in the table that no one farms them.

III: Objections #

Free Range #

The two obvious strategies arising from this chart are 1) don’t eat factory farmed meat, or 2) don’t eat factory farmed meat unless it’s a bivalve, a crustacean, or dairy.

I have deep apprehensions about eating wild animals and wiping away my guilt:

• I suspect that average consumer eats a roughly fixed amount of, say, beef, and will pay $x more for free-range than factory farmed. I also suspect that the supply of free-range beef is fairly inelastic: beef production is very land-intensive. If you agree with these assumptions, then my buying 1lb of free-range beef brings into existence ~1lb of factory-farmed beef. The same goes for wild meat, eg salmon.
• I worry that by buying free-range beef I’d be indirectly burning down the Amazon so that cattle can graze in Brazil. About free-range pork and fowl, I am uneducated.
• It’s not easy to determine whether or not the animal on your plate lived a humane life. I have no trust in certifying labels. I could be convinced that specific farms are good.

I consider this objection unresolved. I’m still curious about:

1. How fungible people’s meat consumption is across species.
2. If it’s possible for me to buy animal products that I’m extremely confident are humanely raised.
3. How land intensive various forms of human animal agriculture are, and how bad this is for the environment.

Don’t You Care About The Environment? #

Yes, I do, and I wrote about this a bit more here^[3]

Farmed Shrimp Eat Fishmeal #

Correct. And fishmeal normally comes from anchovies, which are much smarter than shrimp. ChatGPT wants me to believe that 1 pound of shrimp production requires ~.2 pounds of fishmeal. These anchovies are caught wild. This doesn’t change my conclusions.

Horseflies Don’t Look So Good #

And as a former eater of horseflies and former believer in insect meats, I am distressed by this. Raw size is a bad heuristic.

Some Animals Don’t Have Brains At All #

Correct. Bivalves don’t have a brain or a central nervous system, and they’re the first animals on my list! People with wrinklier brains than mine propose that total cortical neurons are a better heuristic for moral weight than total neurons. I have no opinion.

IV. Conclusion #

I have decided to include animal protein in my diet, but I want to minimize the animal suffering that this choice causes. Being vegetarian is not the optimal solution to this problem. To determine what is, I will have to answer the open questions in the free range section. For now, I’ll be eating natant decapods, bivalves, dairy, and (guiltily) some eggs.

UPDATE 4/10/2026: I eat everything and feel guilty about it :(.

V. B-roll #

No one farms lobster or crab because they’re cannibals, so you can’t leave them in a shared pen.

A dairy cow produces 35! liters of milk per day.

A chicken can go from birth to six pounds and slaughtered in 5-7 weeks. They are kept under constant light so that they keep eating, and cages are heated so that no calories are wasted on staying warm.

If you think keeping animals alive poses no moral cost and the only evil part is the killing, then I think you’re wrong. But if you’re right, then neurons/calorie is the correct metric. Here’s that table below. Humans fall between anchovies and octopi.

Five years ago I gave up lying for Lent. Even before Lent I was an unusually honest person, but I decided to give up white lies too, things like “I’m too tired” when the truth was that I didn’t want to see someone. My reasoning was rule-utilitarian even if I didn’t appreciate it at the time. Empirically, the lies I had told in my life up to that point had been massively negative in utility. Some of the reasons this was the case:

• If I tell one person a lie, I will have to lie to others, lie to maintain a consistent story, remember about and act on the lie in the future. Costs multiply in hard-to-anticipate ways.
• Lying is stressful and cognitively demanding.
• It’s hard to lie to others and be honest with yourself. Keeping two distinct worldviews in your brain is difficult. My brain sometimes subconsciously discards the truthful worldview.
• When I used to occasionally lie, I felt like people had a right to not believe me even when I was telling the truth. I accepted low-trust relationships. Now that I can say to myself “I haven’t lied in years” I feel entitled to be in high-trust relationships.
• The plurality of my lies were instantaneous decisions to “not hurt others’ feelings.” This was a known bad habit. It’s hard to tell someone “I have no other plans but I still don’t want to watch soccer. I’d rather smoke weed and watch Youtube.” It’s much easier to say “I’m busy with work.” Or “I have plans with someone else.” When I stopped lying, almost no one was ever hurt by the truth.

Over the course of Lent I began to appreciate these hidden costs, and started to love not lying anymore. I have kept my no-lying pledge (almost) ever since.

Lying That’s “Baked In” #

Assume that 90% of men claim to be exactly one inch taller than they really are on their Hinge profile. You are 6’. How tall should you claim to be on Hinge?

Arguments in favor of 6’1”:

• Claiming 6’1” will give your matches the most accurate information on your height; they will correctly assume you are 6’.
• Claiming you’re 6’1” is more fair than claiming 6’, in that you’ll be put on equal footing with 90% of people. Why let the liars have an advantage?

Arguments in favor of 6’:

• “I will lie when the median person lies” is not a bright-line, easy-to-follow rule like “never lie” is.
• If everyone lies by one inch, then the sleaziest people will lie by two. Who’s to say from first principles where the equilibrium lies. But your decision to lie pushes it further from the truth.

So many situations are analogous to “lying by one inch on your Hinge profile because everyone does.” To say this more concretely: saying words which are literally false, but which accurately communicate reality because both parties roughly agree on how much lying is expected.

• “I’ll be there in 5”
• “We’ll ship that next week”
• “Let’s grab lunch sometime”
• “You crushed it at your piano recital congratulations!”

This state of affairs is more costly than people appreciate.

Lying as Cultural Barrier #

I read here (first comment) that foreign startup founders have no idea what lies are and aren’t acceptable, and thus end up in jail. Selling a product when all you have to deliver is a Figma? Totally acceptable. Not doing much accounting and making up plausible figures? Also fine if you’re small enough. Raising a tiny follow-on raise at a massive markup to inflate your valuation? It would be rude not to. But knowingly lying about your user count, revenue, etc.? Financial fraud, you could go to jail. If there’s a hard and fast rule about what lying is and isn’t OK, or a clear line of reasoning that neatly separates the two, I’m not aware of it.

Trust Is More Than Honesty #

If someone says they’ll be at your place in 30 minutes, and they show up in 45, did they lie? In a literal sense, yes. By common parlance, no.

We use the word trustworthy to describe several phenomena:

1. A person will not knowingly deceive you.
2. A person will reliably do what they say they will do.
3. A person’s predictions are accurate.

It’s no coincidence that we use the same word. These three attributes are in my experience remarkably correlated. I can think of counterexamples (eg pathological liars who are very smart and thus make good predictions). But only smart sociopaths with high EQ can pull this off. They hold two coherent worldviews in their brain, a real one to use to make predictions and a fake one to spoon-feed to you. This requires a lot of cognitive horsepower.

More typical is the person who will tell you whatever you want to hear, but accidentally convinces themself of these falsehoods too. They double-book themselves, realize halfway through one event that they’re late to another, and blame their tardiness on unforeseeable circumstances. They promise they’ll do something every week, and every week their other projects take longer than expected and they fail to help you out.

Disambiguating between incompetence, bad luck, and deception is difficult. I don’t try. A person is trustworthy, or not.

Meta-Honesty #

Eliezer Yudkowsky came up with the useful concept of meta-honesty:

“Don’t lie when a normal highly honest person wouldn’t, and furthermore, be honest when somebody asks you which hypothetical circumstances would cause you to lie or mislead—absolutely honest, if they ask under this code. However, questions about meta-honesty should be careful not to probe object-level information.”

The piece is worth a read. I am obliged to include Yudkowsky’s disclaimer:

THIS IS NOT THE IDEA THAT IT’S OKAY TO LIE SO LONG AS YOU ARE HONEST ABOUT WHEN YOU WOULD LIE IF ANYONE ASKS.

I wish I’d asked more explicitly to previous employers: What regulations do we reinterpret because they’re horribly written and everyone ignores them? What parts of our business do we sugarcoat to customers? To investors? I wish the answer to those questions was “we follow all regulations exactly, and give customers and investors the most accurate possible picture of our business,” and I would love to try to run my own business that way, but this hasn’t been the case anywhere I’ve worked.

Put another way: most people are meta-honest, even though they’ve never heard of the term. They’ll never tell you in which situations they think lying is OK unless you ask. You should ask.

My Bullshit Detectors: #

1. Is there a good story for why cavemen were fine without this medical intervention? Is this intervention intended for everyone, or just me?
2. Is this intervention designed to cure me, or to resolve my symptoms?
3. How are my doctor’s incentives different from my own?
4. Is it possible to study this even in principle?

I: Is There a Good Story for Why Cavemen Were Fine Without This Medical Intervention? #

Does it strike you as odd that dentists recommend that almost everyone get their wisdom teeth removed? What happened to cavepeople? Did they all have horrible jaw pain in their twenties and thirties?^[1] Did they perform root canals with rocks?

If you ask an anthropologist, they’ll tell you: the human jaws we’ve found from before the agricultural revolution have perfect teeth. I’m sure some of the skeletons show teeth knocked out from blows. But the problems that modern orthodontia aims to fix — malocclusion, crowded and crooked teeth, overbites — are nonexistent in prehistoric skulls.

So why is my dentist telling me I have shit genes and need braces, wisdom teeth removal, and an Herbst appliance? It’s implausible that the human dental genome degraded so rapidly over the past 10,000 years.

We didn’t chew enough as kids. The muscles and bones of our jaws aren’t stimulated, so they don’t grow properly. This leads to crowded jaws, impacted wisdom teeth, and breathing issues.

I’m certainly not a dentist, so if your dentist wants to give your kid braces or take out their wisdom teeth, take their advice. It’s probably too late for your kid to start chewing more.

But if your kids are young enough: get them to gnaw on some bones or chew some gum.

Modern Americans have a weird physiological inferiority complex. The narrative goes: wild animals are by nature vigorous, strong, healthy, robust. They can sleep outside in the cold, fight off infections without drugs, skip meals to escape predators. Humans, thanks to millennia of civilized life, have become wimps, dependent on central heating and antibiotics, incapable of extended exercise without Clif bars and electrolyte goo.

Your genome is identical to that of the humans who, ten thousand years ago, slept outside in the cold, fought off infections on their own, and remained strong without food. Do we really need $200 Nikes for a jog?

There are a couple arguments you could make in favor of Nike, none of them very good:

1. We run on pavement, cavemen ran on dirt. Ignore for a moment that there are plenty of rocky surfaces in nature. If we just need a slightly softer surface, shouldn’t a thin rubber slipper be enough? Why do we need motion control, custom orthotics, a raised heel, rubber half an inch thick?
2. Cavemen didn’t run very often, and those that did got hurt frequently. This isn’t true.
3. I could be running around barefoot if I had grown up that way, but I’ve worn shoes all my life, so I’ll never be able to go without them. Let me tell you: I did it. It will take a couple weeks, but your feet can create the callouses it needs.
4. Streets are dirty! This has not caused any medical issues for me.

(In Chapter 25 of Born to Run Christopher McDougall argues far more eloquently than I ever could that your body was made to run barefoot. If you’re curious I strongly recommend you buy a copy; this is my favorite book ever.)

I try to let my body suffer at least a watered down version of all the things a wild human animal would. I swim in icy water, run for miles in the heat without much water, fast for a day or two every so often, fight off infections without the help of antibiotics or NSAIDs, expose my skin to a reasonable amount of sun for someone of my pale complexion. (I don’t expose myself to bugs. I hate bugs). You can find mountains of studies and commentary on the internet telling you that any of these things is helpful or harmful. A heuristic: if every prehistoric body experienced this, it’s good for mine to experience it too.

II: Is This Intervention Designed to Cure Me or to Resolve My Symptoms? #

Most doctors recommend icing and NSAIDs (eg Advil) after an injury. If your goal is to reduce pain and inflammation, then icing works. But after I sprain my ankle, reducing pain is not my goal. It’s a useful guide in how far I’ve progressed in my recovery. Nor is reducing inflammation. Inflammation is an immune response, bringing resources to the site of the injury so that your body recovers.

Dr. Gabe Mirkin coined the acronym RICE (Rest, Ice, Compression, Elevation) for treating soft tissue injuries. Yet even he now admits on his website that this approach, while succeeding in reducing pain and inflammation, slows recovery. He says himself:

Anything that reduces inflammation also delays healing

There is a pattern here:

1. The body has an ailment, and the immune system responds.
2. We associate our sickness with the immune response.
3. If an intervention reduces our immune response, we assume that we are getting better.
4. Healing is delayed.

Sometimes suppressing our immune response is exactly the correct intervention. If you have pneumonia and it’s giving you a 105 degree fever, the fever is going to kill you way before pneumonia will, so you should take NSAIDs and do whatever else you need to to reduce your fever. If you have a 101 degree fever, you have to decide whether short term pain reduction is worth it for a slower recovery (see study).

It always surprises me how common this mistake is. Here are some guesses for why:

1. Alleviating symptoms happens on a short timescale and is high-feedback.
2. Doctors are incentivized to do something, and often all they can do is alleviate symptoms.
3. Lessening immune response is sometimes the correct intervention.

III: How Are My Doctor’s Incentives Different From My Own? #

Doctors make money when they do things.

Doctors don’t want to get sued. They’ll intervene to ‘be safe’ or refuse to prescribe unpatented substances that don’t have FDA approval.

Doctors want to make you happy, which, by default, means doing something.

I often don’t accept an intervention a doctor recommends. The inverse is rare.

IV: Has This Really Been Studied? #

Food studies are famously bad (small n, observational, indirect). They’re expensive and hard to do well. People don’t like being told what to eat.

IRBs will block RCTs where the treatment group is something like ‘use a tanning bed to see if it causes skin cancer.’ Without reading any tanning bed study, I have a strong prior that it’s observational.^[2]

Know if a study is indirect (eg it shows lowered cholesterol but not lowered mortality).