Astral Codex Ten

Paul Graham

32 min read

Paul Graham

Article

Paul Graham is a recurring person in the Astral Codex Ten archive, appearing 9 times across 9 issues between January 21, 2021 and June 18, 2025. The archive places it in contexts such as “Paul Graham was the most cited name”; “Is Paul Graham some weird kind of basilisk”; “claims, similarly argued, to those of Paul Graham”. It most often appears alongside Eliezer Yudkowsky, Twitter, Bitcoin.

Metadata

  • Category: People
  • Mention count: 9
  • Issue count: 9
  • First seen: January 21, 2021
  • Last seen: June 18, 2025

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Still Alive
January 21, 2021 · Original source
...but I was also grateful to get some emails from journalists trying to help me understand the perspective of their field. They point out that reporting is fundamentally about revealing information that wasn't previously public, and hard-hitting reporting necessarily involves disclosing things about subjects that they would rather you not know. Speculating on the identities of people like Deep Throat, or Satoshi Nakamoto, or QAnon, or that guy who wrote Primary Colors, is a long-standing journalistic tradition, one I had never before thought to question. Many of my correspondents brought up that some important people read my blog (Paul Graham was the most cited name). Isn't there a point past which you stop being that-guy-with-a-Tumblr-account who it's wrong to dox, and you become more like Satoshi Nakamoto where trying to dox you is a sort of national sport? Wouldn't it be fair to say I had passed that point?
With all due respect to these reporters, and with complete admission of my own bias, I reject this entire way of looking at things. If someone wants to report that I'm a 30-something psychiatrist who lives in Oakland, California, that's fine, I've had it in my About page for years. If some reporter wants to investigate and confirm, I have some suggestions for how they could use their time better - isn't there still a war in Yemen? - but I'm not going to complain too loudly. But I don't think whatever claim the public has on me includes a right to know my name if I don't want them to. I don't think the public needs to know the name of the cops who write cop blogs, or the deadnames of trans people, or the dating lives of sexy cyborgs. I'm not even sure the public needs to know the name of Satoshi Nakamoto. If he isn't harming anyone, let him have his anonymity! I would rather we get whatever pathologies come from people being able to invent Bitcoin scot-free, than get whatever pathologies come from anyone being allowed to dox anyone else if they can argue that person is "influential". Most people don't start out trying to be influential. They just have a Tumblr or a LiveJournal or something, and a few people read it, and then a few more people read it, and bam! - they're influential! If influence takes away your protection, then none of us are safe - not the random grad student with a Twitter account making fun of bad science, not the teenager with a sex Tumblr, not the aspiring fashionista with an Instagram. I've read lots of interesting discussion on how much power tech oligarchs should or shouldn't be allowed to have. But this is the first time I've seen someone suggest their powers should include a magic privacy-destroying gaze, where just by looking at someone they can transform them into a different kind of citizen with fewer rights. Is Paul Graham some weird kind of basilisk, such that anyone he stares at too long turns into fair game?
Your Book Review: How Children Fail
June 11, 2021 · Original source
And he makes similar claims, similarly argued, to those of Paul Graham and Eliezer Yudkowsky, that the strategies that lead to nominal success in school are often the ones that stop at superficial understanding of the subject--hacks to be able to get to the correct answer quickly, without ever really looking at the problem.
Inline links: Paul Graham, Eliezer Yudkowsky
Links For March 2023
March 10, 2023 · Original source
7: I still haven’t read Garett Jones’ The Culture Transplant yet, but I’m seeing a lot of good discussion. Via Paul Graham, here’s a graph of migration-adjusted tech history score 1500 (ie how advanced a region was in 1500, adjusting for the fact that eg Australia is mostly inhabited by English people and should count as England rather than as the Aborigines) vs. income per person today (actually 2005):
Inline links: Paul Graham
Your Book Review: The Laws of Trading
July 21, 2023 · Original source
We need a framework for thinking about these trades. Lebron’s first law states that we must know ourselves and our motivations for trading before we trade. We tell ourselves many stories, but someone with intellectual honesty – the person with the most alignment between their motivations and actions – will take money from the person who didn’t go through the work to understand their own motivations. There is a reason that Citadel and other hedge funds pay millions of dollars to trade with retail. They know why they are trading: to maximize profit. And the dilettante who “trades for fun” will be eaten alive by a firm with a much better model of a) the world and b) the dilettante themself. Why did I write this book review? To test my intellectual mettle. I could easily have posted this book review elsewhere, but no, I wanted to see how I stack up against other ACX Book Review contest participants. Similarly, this is often the reason people get into trading. One motivation that Lebron explicitly calls out is intellectual validation. You can toil in obscurity for years as an academic. But in trading, there is a quick feedback loop. If your P&L showed $10M last year and the guy sitting next to you showed $8M, you have demonstrated who is “cleverer” and established a clear hierarchy. What lessons here transfer to our daily lives? Like Paul Graham, Lebron encourages us to keep our identities small. He gives the standard decision-making advice to write down your framework and reasoning for why you made a decision at a specific point in time, in order to avoid biases after the fact. This section of the book contained good general advice, but nothing that will be particularly new for the median ACX reader. 2: Adverse Selection You’re never happy with the amount you traded. Now we start to get into the good stuff. Financial markets are an information aggregation mechanism, relying on multiple parties’ beliefs and recursive Bayesian updates of an individual actor’s beliefs based on the beliefs of others2. Market mechanics demonstrate Bayesian beliefs in action. The following quote is quite long, so skip over it if you don’t want to dive deep into the psychology of making a market. I retained it in full because this is quite literally the best description I’ve ever seen of the Bayesian dance between two market makers: “You are a market maker in South African mining companies. Through years of effort and continual improvement, you have built a trading model for the company Veldt Resources. You walk into work one day, ready to set up your trading for the day. It's a stock that doesn't trade much, and usually there are only two market makers: you and another (we'll call her Jo). She's sharp, and she competes well to trade against customer orders that come in. Your model has Veldt valued at 54.35 ZAR (South African rand). You're going to start quoting the stock, so you're about to turn on your machine making a market 54.25 - 54.45 (1000x)3. Before you turn on, you check the current market and notice that Jo has already turned on and she's making her market 53.50 - 54.00 (2000x). If you were to turn on your machine, your market would cross her market, and you would buy 1000 shares from her for 54.00. You now need to make a decision. Whose model do you believe more, yours or Jo's? If you believe yours, you should turn on your machine, trade at 54.00, and expect to make money. If you believe Jo's model, you should adjust your own model parameters to match her market and turn on, making a similar market to hers. What to do? As with many dichotomies, this is a false one. And as with many decision processes, Bayesian reasoning lights the way… …Jo presumably believes Veldt is worth around 53.75 (the average of her bid and offer). But how confident is she in her belief? The width of her market can give you a clue. It's 0.50 ZAR, whereas yours was going to be 0.20 ZAR wide. All other things equal, you should think that Jo only has 40% (0.20/0.50) of the confidence in her fair value as you do in yours. On some absolute scale of confidence, you can say you had a belief-strength of 100 in your fair value of 54.35 (before seeing Jo's market), and Jo has a belief-strength of 40 in her fair value of 53.75 (before seeing yours). And it turns out the weighted average of these two beliefs is quite a reasonable way to combine them: 100/140 * 54.35 + 40/140 * 53.75 = 54.18. Your updated fair value, having seen Jo's market, is thus 54.18 ZAR. This procedure is a quick, heuristic, and reduced version of Bayesian belief-updating, and a good reference on the subject is A.L. Barker's 1995 paper. After updating, you now believe that the stock is worth 54.18. Assuming your trading costs, risk limits, and return requirements are satisfied, buying 1000 shares for 54.00 is a good trade. Naively, you might just put out a 54.00 bid for 1000 shares, trade with half the 2000 share offer, and hope to collect your expected-value ZAR. In practice, however, you might be able to make even more. If Jo is making a 0.50 wide market, maybe she'd be willing to sell lower than 54.00. It's conceivable that if you put out a 53.90 bid for 1000 shares, Jo will sell at that price, and you collect an extra 100 ZAR! Of course, Jo could react differently. She could see your bid and use that information to change her market, in much the same way you did before turning on. These are difficult decisions, ones where experience with the product and the market make a big difference in being able to eke out a little extra edge. Let's play it safe however and pay 54.00 for 1000 shares. You trade, and Jo reacts by immediately canceling her market. This is not an uncommon occurrence in illiquid stocks, especially in emerging markets, so you're not too surprised. You wait a couple of minutes, mentally visualizing Jo in front of her six monitors, evaluating her trade and her model. Finally, she turns back on. Her new market is 53.50 - 54.05 (10000x)! You reason that Jo has seen that someone (you) disagrees with her valuation of the stock. Jo is a good Bayesian like you, and so she has incorporated that information into her model and updated her beliefs about the fair value of the stock. Her updated belief is that she now wants to sell even more stock, at a marginally higher price. Clearly, she almost entirely discounts the information you've communicated to her with your trade. How should you react? It seems fairly clear that, assuming Jo is not a crazy or incompetent market maker (usually a fair assumption), your trade was a bad one. You bought 1000 shares, when in retrospect, you would have wanted to buy much less, probably zero. Imagine instead that Jo had turned back on with a market of 54.00 - 54.50 (1000x). Her reaction now clearly indicates the information you gave her with your trade is valuable, and she has adjusted her beliefs accordingly. Your trade was probably a good one. Don't you wish you had bought all 2000 shares on offer? No matter what Jo's reaction is, you will be unhappy with your trade. Note that Jo will be unhappy too, since retrospectively she should have either made her initial market bigger or smaller. Welcome to the joyous world of trading!” Whether or not you make money, you have regrets! If you profited, you could have made more. If you lost money, you shouldn’t have made the trade at all. Like death and taxes, you can’t avoid adverse selection. Lebron continues to highlight a few areas of trading that have adverse selection problems. First, IPOs. If you buy the stock in an IPO, you expect the share price to “pop” on the first day of trading. However, if others also have this expectation, the round will be oversubscribed. You can only get the quantity of shares that you bid for when the market doesn’t think the shares will go up. So if you are able to get the shares that you want, the IPO is likely a dud. See also: Venture Capital fundraising. Second, powerful entities that change the rules of the game while you’re playing. Exchanges nullify “erroneous” trades. Brokerages limit buying. Anyone who tried to buy GameStop stock on Robinhood on January 28, 2021, knows this form of adverse selection all too well. Lebron also highlights “special trades”, in which you should throw the “normal rules” out of the window. This advice generalizes to other areas of life: “The normal rules do not apply. If you remove yourself from our usual routine, if you think hard and clearly about the specific situation, maybe you can do something good. Perhaps even great. Others will be paralyzed by inaction, but perhaps you won’t be. Crises can be opportunities.” 3: Risk Take only the risks you’re being paid to take. Hedge the others. In trading, as in life, you can make the right call in expected value terms but still lose due to randomness. Some of that randomness is avoidable. Some of it is not — and can be accounted for by hedging. Here, Lebron encourages us to rely on multiple risk measures and actively seek to understand the risks that we might be subject to. That’s all well and good in the world of finance, with derivatives contracts. But how might this apply in other areas of life? If you work for a publicly traded company and are compensated in stock, sell your shares as soon as you receive them. This is not because I don’t expect the share price of Microsoft/Meta/Apple/etc. to go up. The stock may very well outperform the market. But you are not being compensated for the added risk that you take on here. Your employment prospects at Microsoft/Meta/Apple/etc. are highly correlated with the share price. When the share price is down is when layoffs happen. Former Enron employees can chime in here. Similarly, it makes sense to hedge anything that is outside of your control. Let’s say you’ve decided the crypto bear market of 2023 is a great time to start a new crypto company. Your success depends on things within your control, such as: Your idea
Inline links: pay millions of dollars to trade with retail, keep our identities small, 2, market makers, 3
23: Room Temperature Superforecaster
August 01, 2023 · Original source
…but Metaculus is much lower, probably because the other two are asking if any replication will be positive, and Metaculus is asking if the first replication attempt will be. It’s bad news that these numbers are so different, and suggests a high chance that this stays confusing and comes down to finicky resolution criteria. Still, this has gotten lots of people checking the prediction markets, including Paul Graham: …and around 500 others, according to the Manifold Active Users graph (source): Aside from headline numbers, I’ve also appreciated prediction market comment sections as a good place to stay up to date on the latest developments (including a link to this thread) Elsewhere In Forecasting NYPost: Blind Mystic Baba Vanga Makes Terrifying Nuclear Disaster Prediction For 2023: A blind mystic who allegedly predicted 9/11 is said to have foreseen a nuclear disaster that will ravage Earth before the end of 2023. Baba Vanga, a blind Bulgarian woman, is rumored to have predicted some of the biggest events in world history. She died more than a quarter of a century ago, but many of her predictions are said to have come true long after her death. Now, her followers claim that Baba Vanga foresaw a devastating nuclear disaster that will unfold this year. Big if true. In what sense did she predict 9/11? Another article gives the exact text of the 1989 prediction: “Horror, horror! The American brethren will fall after being attacked by the steel birds. The wolves will be howling in a bush, and innocent blood will be gushing.” This is a 1989 prediction! If you’re calling airplanes “steel birds” in 1989, you’re just hoping that people forget you lived when airplanes already existed and then get impressed with you for predicting them. Come on! (you could argue that the second half is about Assistant Secretary of State John Wolf and Deputy Secretary of Defense Paul Wolfowitz howling for war with Iraq from within the Bush administration, but Ass. Sec Wolf played a minimal role in the war buildup so I think if you are being very strict in your interpretation there was really only one wolf involved.) Anyway, Vanga’s other predictions for 2023 include: Earth’s orbit will change
Inline links: Paul Graham, source, this thread, Blind Mystic Baba Vanga Makes Terrifying Nuclear Disaster Prediction For 2023, https://substackcdn.com/image/fetch/$s_!yIPJ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc81ce0d2-e9c7-41b9-be38-0182d21f0381_1024x682.jpeg, the exact text of the 1989 prediction
X Fact Check: Does Gender Integration Moderate Politics?
February 15, 2024 · Original source
Click to go to original tweet. Although there has long been a weak tendency for men to be more conservative and women more liberal, it seems this has grown rapidly in recent years. Paul Graham proposed an explanation: Click to go to original tweet. …and commenter Nevin Climenhaga proposed a potential test: Yes! Let’s do it! Using the 2022 survey data, I compared: Men with at least one brother, but no sister
Inline links: https://twitter.com/paulg/status/1750881800174420216, https://twitter.com/NevinClimenhaga/status/1751175197242400806
Links For January 2025
January 17, 2025 · Original source
42: Paul Graham has a new essay on the causes of and possible responses to wokeness. He says the short-run cause was the rise in social media and group chats, medium-run cause was the student protester generation of the 60s growing up and taking power within academia, and the long-run cause is that people will always have an urge to virtue-signal and the fall of traditional virtue-signaling categories (sexual purity, religious orthodoxy) left an unfilled niche. He recommends as a response our usual rules around religious pluralism - everyone can have a religion, but you shouldn’t bring it to work or demand orthodoxy from your employees. I think this is mostly right, but our tolerance around religion has always gotten awkward when religion has any real-world/political implications (eg the headscarf in France, Christian schools not wanting gay employees, teaching creationism, etc) and since wokeness is made entirely of real-world and political implications, our religious norms aren’t yet well-adapted to deal with it. Also, a lot of our religious pluralism norms are “just do the neutral, non-religious thing”, but wokeness thrives precisely by challenging what “neutral” is: if a studio releases ten films, and they all have white protagonists, is that “neutral”, or is it a surrender to the opposite “religion” of racism? If a woke employee demands that the studio have more films with black protagonists, were they the first one to defect, or just responding to a previous defection? If they claim it’s a business decision (“we’ll do better with minority demographics if we have some minority characters”), then it takes an active effort beyond just applying regular pluralism norms to “accuse” them of “wokeness” and mount some kind of response.
Inline links: a new essay on the causes of and possible responses to wokeness
Choose Nonbook Review Finalists 2025
June 03, 2025 · Original source
...Growth The Drum Major Instinct The Emperor Of All Maladies The Internet That Might Have Been The Life's Work Of Banerjee, Duflo, and Kremer "The Origins Of Wokeness", by Paul Graham The Metaethics of Joy, Suffering, and AI The Men Are Not Alright The Pebble, Jewel of the 1960 World Series The Russo-Ukrainian War The Sermon On The Mount The Sermon On...
ACX Grants 1-3 Year Updates
June 18, 2025 · Original source
Minnesota and Virginia also have legislation to enable cities to implement land value taxes. We are monitoring these efforts. There are a few other cities we are operating in. We have helped another organization prepare for a meeting in Tennessee by doing impact analysis of land value taxes in the city. We have presented to city officials in the City of South Bend who have expressed support for land value taxes. Finally, we are in conversation with a State Senator in Colorado who is a champion of land value taxes. Meanwhile, we have soft launched and developed the OpenAVMKit, which uses a unified schema to do assessment accuracy reports and automated valuation methods for any property tax data given. Valuation of land is the key binding constraint to successful implementation of land value taxes. We plan to be the leaders in this space with strong benchmarking capabilities and a repo that can enable the open-source community to make the best automated valuation methods. Along with these efforts, we have expanded the movement. We have posted to the Progress and Poverty Substack growing the subscriber base to around 5,000 subscribers. We have spoken to over 25 local advocates interested in working on land value taxes in their local communities. Yet, there is a long way to go. We need to start earning income through technical assistance contracts as our grant funding expires. We need to continue pushing for a state to implement, and we need to be prepared to tell the success story for when they do. 65: EN’s Work On Bacteriophage Therapy Our project is aimed at pioneering phage therapy in Nigeria, where limited resources/infrastructure have historically held back research in this field. Starting from the ground up, we are establishing the foundational systems needed to support a robust phage research ecosystem. So far, we’ve isolated 34 bacteriophages targeting Pseudomonas aeruginosa, an essential step toward building a comprehensive phage bank. This began with collecting a wide range of clinical Pseudomonas isolates, which we are now characterizing alongside the phages through genome sequencing and phenotypic assays including studies on phage stability across pH, temperature, and salinity ranges. Our long-term goal is to develop a phage-based hydrogel for treating diabetic wounds. On the regulatory front, we have secured approval from the Attorney General to register our nonprofit organization, the Centre for Phage Biology and Therapeutics. Additionally, we’re expanding into vaccine development; following a research stay in Prof. Roderick's lab at the University of Waterloo, we have initiated the design of a phage-based universal Salmonella vaccine aimed at covering all major serotypes—an urgent need underscored by Africa’s reliance on external vaccine sources during the COVID-19 pandemic. I have signed an MTA agreement with Roderick to use his phage-based vaccine platform patents to enable us to design vaccines against any common disease affecting us. This is only the beginning, but we are proud to be laying the scientific and institutional groundwork for homegrown phage innovation in Africa. Emergent Ventures funded EN before we did and deserves a lot of credit here also. 66: Create An Artificial Kidney For an implantable artificial kidney, the first essential component is a hemofilter designed to emulate the glomerulus. Critical requirements for this hemofilter include high permeability (to maximize flow for a given area), selectivity (specifically, the retention of albumin), and robust blood compatibility (ensuring sustained function over time). Our initial strategy focused on using negative surface charge to reduce fouling. I began by testing polyelectrolyte (PE) coatings on 24nm pore membranes featuring a negative terminal charge, similar to the glomerular barrier. These initial static tests, assessing platelet adsorption in whole blood, yielded positive outcomes for some polyelectrolytes, indicating potentially desirable blood compatibility. However, static test setups are not truly representative of dynamic in-vitro conditions and don't provide data on key parameters like permeability, fouling progression, or changes in membrane selectivity. To address these limitations, I designed and built a blood filtration setup. This system sustains human whole blood in circulation for 20 minutes, allowing us to analyze all the aforementioned parameters, as well as platelet activation markers. This has resulted in a fairly high-throughput system for evaluating any surface coating. I'm pleased to report this setup has been accepted for presentation at this year's European Society for Artificial Organs (ESAIO) conference. I am also currently working on a full manuscript, as I believe this system offers a viable way to partially replace animal experiments in our early-stage research, requiring only 1.2ml of human blood per run. Working with a PhD student (hired to support both this research and work on membrane substrates), we have continued testing these PE coatings, alongside PEG coatings, on our membranes. Here, we're finding that optimization of the coating layer is crucial. With the current PE coatings, we observe a permeability drop of about an order of magnitude compared to the base membrane, making them unsuitable for an implantable device in their present form. This is likely due to the specific nature of the initial PE layer, which we can modify. We also suspect there may be ingress of PE into the pores, meaning we're not achieving just a surface coating (our goal), but rather a very thick coating, which would explain the flux loss. Optimizing the coating process to control penetration depth is now a primary focus of my ongoing work. I am currently aiming for a flux of 20ul/min (as this is cap introduced by the protein gel layer anyway) but for it to be at this 'steady state' permeability without drop in permeability. I am also imaging the membranes after contact with SEM to see if there is indeed any platelet adsorption etc. Tugrul has the dubious honor of maybe being "the only person to climb a 4000m peak with severe kidney failure". To raise money and awareness for his artificial kidney project, he is running Climb Against Time, where he will climb 41 mountains over 4000m (13000 ft) this summer. He is looking for donors and climbing partners. 67: Add Tardigrade Genes To Human Cells The goal of this one was to make hybrid cells that are more resilient for research and certain medical applications. They report: The grant was to synthesize vectors for the expression of humanized tardigrade proteins that can be targeted to different areas of the cell. All the vectors were designed, generated, and transposed into human cells. The proteins all localize successfully (e.g. they match the designed target), with one exception (we are still working on validating it). We've done some stress testing with the trangenic cells, but haven't reached firm conclusions yet. We've further generated some multigene designs but have not yet transposed them into cells, but should shortly. We're hoping to submit a manuscript on the first round later this year. 68: Teach Forecasting To EU Policy-Makers The original project didn't work out, but our grantee (who still prefers to remain anonymous) is now working with an EU think tank pursuing the same agenda, and has been teaching forecasting workshops to policy-makers for the past two months. 69: Platform For Single-Cell Imaging They ended up unable to accept this grant and returned the money. 70: Open Source Polygenic Predictor For EA/IQ They have an update here. They think they have a predictor that can explain 12% of variance in intelligence, and they’re working on validating it and creating an easy-to-use website. 71: Improve Flu Vaccines The grant mainly funded agent based modelling to demonstrate the benefit of pre-existing immunity to pandemic influenza if and when a future pandemic occurs (academic publication will result). The original proposal was to attempt to influence the WHO influenza strain selection process. After attending WHO meetings and a global influenza conference, I believe this is not feasible. Stakeholder feedback was the potential short term negative effect on vaccine hesitancy is believed to outweigh the less tangible future benefit. Given the conservative nature of decision makers, pandemic vaccines are likely to remain research only. There are still green shoots of research into pandemic preparedness/prevention that I am continuing to work on. I'm working under the "Australians for Pandemic Prevention" brand of Good Ancestors, another group that ACX funded in 2024. 72: Scenario Analysis For Developing World Agricultural Programs In addition to the research and analysis funded by the grant, I’ve learned to code with LLMs and have built an MVP of the project. The app is being considered for further development by staff at a large international organization. 73: Further C’s Political Career C’s political career is going well, but he continues to think it wouldn’t be strategic to give more information publicly at this time. Lessons Learned I'm most impressed with our lobbying/advocacy organizations. In particular, Good Ancestors has gotten the Australian government to sign onto an international AI safety declaration, partner with various x-risk-related organizations, and (possibly) extend charity tax deductions to some EA causes that previously didn't have it - I think this on its own goes a substantial way to paying back the cost of all ACX Grants. Coalition to Modify NOTA has a kidney donation bill in front of Congress that the (very illiquid) prediction markets give a 45% chance of passing; if it works, it could save thousands of lives. The Georgists are partly responsible for bills making land value taxes slightly easier to implement in a handful of states. Good Science Project seems to have significantly improved science. Are lobbying organizations a better bet than other types of nonprofit (within the constraints of ACX Grants)? I'm not sure. It could just be that lobbyists are (naturally) better at playing themselves up and sounding successful than (for example) scientists, or that politicians are good at people-pleasing and make people feel heard and encouraged in a way that might not change overall policy later. Also, I recently talked to some grantmakers who funded a lobbying organization that superficially seems excellent, but they expressed concern it was net negative (!) by taking away oxygen and spotlight from potentially more effective orgs. So I am encouraged but wary. Animal welfare organizations were another standout success. Again, I don't know how to think about this - while I think our grantees were exceptional, there's also an issue where the scale of animal welfare challenges is so great, and work on them so neglected, that lots of organizations can save a million chickens here, or a million fish there, without particularly making a splash. On the one hand, this is exactly what effective altruism should be doing - exploring grants that are very high in linear utility even if they don't feel satisfying. On the other, they're unsatisfying - and also hard to assess retroactively. How many chickens should a good animal welfare grant save? Any realistic number will both be overwhelmingly large in absolute terms and far too small in relative terms. I'm most ambivalent about our science grants. Many of them say they are successful and can point to published papers which explain the science they did. But it's hard to judge whether anything useful has changed based on the science getting done. I know it's important to fund basic research and not just last-mile technology startups, but it's hard for a mini-grants program like this one to evaluate these kinds of abstract interventions. One disappointing result was that grants to legibly-credentialled people operating in high-status ways usually did better than betting on small scrappy startups (whether companies or nonprofits). For example, Innovate Animal Ag was in many ways overdetermined as a grantee - former Yale grad and Google engineer founder, profiled in NYT, already funded by Open Philanthropy - and they in fact did amazing work. On the other hand, there were a lot of promising ACX community members with interesting ideas who were going to turn them into startups any day now, but who ended up kind of floundering (although this also describes Manifold, one of our standout successes). One thing I still don't understand is that Innovate Animal Ag seemed to genuinely need more funding despite being legibly great and high status - does this screen off a theoretical objection that they don't provide ACX Grants with as much counterfactual impact? Am I really just mad that it would be boring to give too many grants to obviously-good things that even moron could spot as promising? Someone (I think it might be Paul Graham) once said that they were always surprised how quickly destined-to-be-successful startup founders responded to emails - sometimes within a single-digit number of minutes regardless of time of day. I used to think of this as mysterious - some sort of psychological trait? Working with these grants has made me think of it as just a straightforward fact of life: some people operate an order of magnitude faster than others. The Manifold team created something like five different novel institutions in the amount of time it's taken some other grantees to figure out a business plan; I particularly remember one time when I needed something, sent out a request to talk about it with two or three different teams, and the Manifold team had fully created the thing and were pestering me to launch a trial version before some of the other people had even gotten back to me. I take no pleasure in reporting this - I sometimes take a week or two to answer emails, and all of the predictions about my personality that this implies would be correct - but it's increasingly something that I look for and respect. A lot of the most successful grants succeeded quickly, or at least were quick to get on a promising track. Since everything takes ten times longer than people expect, only someone who moves ten times faster than people expect can get things done in a reasonable amount of time. In almost every case where I thought to myself “this is a cool idea, but I don’t know how it’s going to really pay off, as opposed to reaching a cool intermediate accomplishment and then stagnating”, this was a correct criticism, and I should have taken it more seriously. But I can’t rule out that these were good in vague and hard-to-measure ways that I should take more seriously. This one is really self-serving, but in general when people were good communicators (or even bloggers) and wowed me with the writing-composition of their application, they turned out to be a good bet. And when people were hard to understand and annoying to communicate with, even if their ideas seemed good, they were less likely to pan out. Overall Thoughts The total cost of ACX Grants, both rounds, was about $3 million. Do these outcomes represent a successful use of that amount of money? Very naively, startups originating from ACX Grants have about $50 million in value1. If ACX Grants is equivalent to a pre-seed funder, and pre-seed funders usually get ~5%, then if we were VCs we would have a portfolio worth $2.5 million. About 1/5 of ACX Grants were attempting to be market-valued startups, so if we assume the charitable portion did about as well as the startup portion, then the charity portion is “worth” $10 million. There’s some reason to expect this is too high, since much of the startup value came from one successful outlier. But there’s another reason to expect this is too low, since we were aiming at charity rather than market cap, and any actual market cap that our grantees got was an unexpected side effect. I’m treating this as a sanity check rather than as a real number. It’s harder to produce Inside View estimates, because so many of the projects either produce vague deliverables (eg a white paper that might guide future action) or intermediate results only (eg getting a government to pass AI safety regulations is good, but can’t be considered an end result unless those regulations prevent the AI apocalypse). Because we tend towards incubating charities and funding research (rather than last-mile causes like buying bednets), achieved measurable deliverables are thin on the ground. But here are things that ACX grantees have already accomplished: Improved the living/slaughter conditions of 30 million fish.
Inline links: Progress and Poverty Substack, Climb Against Time, here, 1