Astral Codex Ten

Microsoft

46 min read

Microsoft

Article

Microsoft is a recurring organization in the Astral Codex Ten archive, appearing 14 times across 14 issues between February 22, 2022 and March 03, 2026. The archive places it in contexts such as “current leaders of … Microsoft”; “The last big antitrust case involved Microsoft”; “Bill Gates Microsoft”. It most often appears alongside Google, Anthropic, OpenAI.

Metadata

  • Category: Organizations
  • Mention count: 14
  • Issue count: 14
  • First seen: February 22, 2022
  • Last seen: March 03, 2026

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.

Links For February
February 22, 2022 · Original source
8: Economist: Why Brahmins Lead Western Firms But Rarely Indian Ones. Brahmins are the highest Indian caste; in India they tend to be academics/lawyers/etc, but in the US they are disproportionately likely to become CEOs (including the current leaders of Google and Microsoft). Article theorizes that this is a combination of more business-related Indian castes having better networking within India (so motivated Brahmins tend to go abroad), Brahmins being good at the traditional academic pathway that lends itself well to immigration, plus maybe affirmative action against them in India. Here’s a rebuttal I link to out of duty, but I’m not sure it’s worth wading through the woke outrage to get to the two or three mildly interesting facts (Brahmins started immigrating before India’s affirmative action really ramped up, and they might have a first-mover advantage from building immigrant communities earlier).
Inline links: Why Brahmins Lead Western Firms But Rarely Indian Ones, Here’s
Highlights From The Comments On Billionaire Replaceability
September 22, 2022 · Original source
The last big antitrust case involved Microsoft. When IBM got sick of antitrust fights, they decided to outsource the operating system for their PCs. This was like throwing a monopoly bouquet at a wedding and Bill Gates was the bridesmaid who jumped highest and snatched the prize. His plan was to leverage this operating system monopoly into an internet monopoly, and the scheme was working before the Clinton administration sued. The reason I am writing this on Substack and not some Microsoft comment board is because of an antitrust lawsuit.
I’m still not sure about this line of thought. How is this situation? Do we think Mark Zuckerberg wouldn’t have founded Facebook, or Bill Gates Microsoft, if he could only get $1 billion? Can people really tell the difference between $10 billion and $100 billion? Has Jeff Bezos even spent $10 billion?
OpenAI's "Planning For AGI And Beyond"
March 01, 2023 · Original source
And so on . . . Meanwhile, in real life, OpenAI released ChatGPT in late November, helped Microsoft launch the Bing chatbot in February, and plans to announce GPT-4 in a few months. Nobody thinks society has even partially adapted to any of these, or that alignment researchers have done more than begin to study them. The only sense in which OpenAI supports gradualism is the sense in which they’re not doing lots of research in secret, then releasing it all at once. But there are lots of better plans than either doing that, or going full-speed-ahead. So what’s OpenAI thinking? I haven’t asked them and I don’t know for sure, but I’ve heard enough debates around this that I have some guesses about the kinds of arguments they’re working off of. I think the longer versions would go something like this: The Race Argument: Bigger, better AIs will make alignment research easier. At the limit, if no AIs exist at all, then you have to do armchair speculation about what a future AI will be like and how to control it; clearly your research will go faster and work better after AIs exist. But by the same token, studying early weak AIs will be less valuable than studying later, stronger AIs. In the 1970s, alignment researchers working on industrial robot arms wouldn’t have learned anything useful. Today, alignment researchers can study how to prevent language models from saying bad words, but they can’t study how to prevent AGIs from inventing superweapons, because there aren’t any AGIs that can do that. The researchers just have to hope some of the language model insights will carry over. So all else being equal, we would prefer alignment researchers get more time to work on the later, more dangerous AIs, not the earlier, boring ones.
Reading even further between the lines - at this point it’s total guesswork - OpenAI’s corporate partner Microsoft asked them for a cool AI. OpenAI assumed Microsoft was competent - they make Windows and stuff! - and gave them a rough draft of GPT-4. Microsoft was not competent, skipped fine-tuning and many other important steps which OpenAI would not have skipped, and released it as the Bing chatbot. Bing got in trouble for threatening users, which gave OpenAI a PR headache around safety. Some savvy alignment people chose this moment to approach them with their latest ideas (is it a coincidence that Holden Karnofsky published What AI Companies Can Do Today earlier that same week?), and OpenAI decided (for a mix of selfish and altruistic reasons) to get on board - hence this document.
Inline links: What AI Companies Can Do Today
Book Review: Paper Belt On Fire
March 23, 2023 · Original source
Probably it depends on the angle or something. Lindh was the only American to find Osama bin Laden in the early 2000s - he went to lots of jihadi training camps in the process of learning how to jihad, and Osama happened to be at one of them. The lesson, Walker says, is that if you want to find people who are hard to find, you need to steep yourself in their culture, truly understand them, become one with them. Good founders are hard to find. But he and Strachman went to dozens of dingy college dorms, math competitions, group houses, and hackathons, looking for people with the right sort of talent. After pooh-poohing IQ (“Marilyn vos Savant is listed as having the highest recorded IQ, and what does she do? She writes a column for a Sunday supplement in the newspaper”) he lists some of his own preferred metrics for judging would-be Thiel fellows and founders: Polytropon - a famously untranslatable Greek word (“of-many-turns”? “always-has-a-trick-up-his-sleeve” “clever bastard”?) used to describe Odysseus. Edge control - willingness to constantly surf the boundary between order and disorder Crawl-walk-run - ability to scale from a tiny startup to a big company. …and several others, including “tensive brilliance” and “Friday night Dyson sphere”. He and Danielle searched the country for people with these qualities, annoying colleges (he was banned from MIT after showing up too often to convince their students to drop out) and doing various stunts (on October 31 2017, the 500th anniversary of Luther’s theses, he nailed a list of anti-formal-education theses to the doors of the admin buildings of top colleges (“Our commercial printer had misunderstood our request and printed them on seven-foot-long scrolls. They were ridiculous . . . but it turned out for the best.”) At one point, he negotiated with a brilliant 21 year old who may have discovered a transformative diabetes therapeutic, but the hidebound conformist novelty-hating establishment refused to work with him just because he liked the Marvel Cinemat - okay, fine, he may have legally changed his name to “Tony Stark”. Still, Gibson saw past his eccentricities, helped him start his company, and gave him sage advice (he should introduce himself to other investors as “Anthony”). Skip through several more chapters of everyone hating Gibson and telling him he was wrong and refusing to give him money and cheating him out of the money he already had, and the payoff is Luminar. One of the dropouts they cultivated founded a beyond-cutting-edge lasers-for-self-driving-cars company which went public at $3 billion. 1517 made $200 million from the deal - it sounds like they had only ever raised about $25 million, so their investors must have octupled their money on that company alone. Everyone involved is now very rich, and Gibson considers his anti-education thesis on the way to being proven. The book ends with a newly-resourced Gibson continuing his quest to figure out whether and why the CIA killed his father, but it’s slow going. If any of you know a guy named Albert van Dam in Amsterdam, or how to convince Swiss banks to reveal secret account information, get in touch with him. II. A common pattern: I assert something. Everyone yells at me and tells me I’m wrong and stupid, sometimes in very colorful language. I wait, time proves me right, and I write an essay gloating educating people about this. The median comment is “of course this is true, nobody ever denied this was true, why are you wasting our time with something obvious?” I hate this and I try to avoid doing it to other people. This is too bad, because I’m tempted to say: obviously talented dropouts can start good companies. We’ve known this at least since Bill Gates dropped out of Harvard in 1975 to start Microsoft. But also, obviously they can. Brilliant and driven people can succeed whether they get a college education or not. If Bill Gates had stayed an extra two years at Harvard, he probably would have taken a few more advanced math classes not really related to programming software or running a company. So why should we even have as a hypothesis that he couldn’t start Microsoft successfully without doing that? Still, Gibson adequately proves that lots of people hated him and were sure he would fail. Either we should read this backwards - learn that there was once a time when pro-college messages were even stronger than now, so strong that people thought it was literally impossible to succeed without every single day of a four-year college application - or the critics were trying to get at something deeper they were bad at expressing. For example: what, exactly, is Gibson’s alternative to the education system? The back-of-book-blurb says Paper Belt On Fire is about “how higher education and other institutions must evolve to meet the dire challenges of tomorrow” - but evolve how? What exactly has been proven here? A few of the very brightest young people, hand-picked by an expert young-person-picker and given $100K, can become billionaires or make great discoveries without a college degree. What are the implications? Suppose you are an average college student with an average level of talent and motivation. Should you drop out and try to create a company for Peter Thiel? Based on how many average-talent people Thiel rejects, even he doesn’t think you should do that. And if you don’t have a good answer to this question - the one relevant to 99.9% of education system inmates - have you really launched a challenge to the educational system? Gibson doesn’t address this question, but I predict he would admit that, fine, he doesn’t have an alternative to the education system in the sense of “educate people this way rather than that way”. He just wants less formal education, and has proven this will work fine. True, he’s only proven it for a tiny subset of ultra-talented people. But “billionaire tech founder” is a hard job - if it wasn’t, more people would do it and reap the $1 billion reward. Proving that people can become billionaire tech founders without college degrees implicitly suggests they can be successful middle managers or budget analysts without college degrees. So the sort of companies that need middle managers and budget analysts should also consider hiring people without degrees, and the sorts of average-level-of-talent-and-motivation people who want these jobs should consider skipping college. Would this work? Probably. It worked in the early 1900s, when only 5-10% of Americans had college degrees but the country seemed about as dynamic and successful as it does now. It worked for people like George Washington, Abraham Lincoln, and Thomas Edison, none of whom went to college. It works in other countries - for example in the UK where young doctors skip undergrad and go straight to medical school, and whose patients get about the same outcomes as in the US. It works for people with impractical degrees like philosophy, who are constantly getting jobs in (and doing well in) fields that don’t require you to compare Locke vs. Leibniz’s perspective on a priori truths. So this would work if everyone agreed to do it at once, which they won’t. The way college gets you is adverse selection. Suppose that tomorrow, you - a smart and hard-working person who could easily get a college degree - decline to do so, because you appreciate Peter Thiel and Michael Gibson’s anti-institutional perspective. The pool of people without college degrees is now, to a first approximation: 200 million people who weren’t smart to get in, rich enough to afford it, or motivated enough to finish.
Inline links: his company, read this backwards
The Question Of Separatism
May 19, 2023 · Original source
When people argue against separatism, they often tout the benefits of being large. A Canada that would be split in two would mean smaller markets, and a weaker political counterweight to the United States. (Not to be mean to Canadian readers, but this argument seems delusional to me — I don’t think Americans currently see Canada as a political counterweight of any significance.) It would certainly be less prestigious. Large size, Jacobs says, is associated with power, and we admire power. We love slogans like “unity makes strength.” But after the medium-sized country of Sweden-Norway became the two smaller countries of Sweden and Norway, they both did well. Small size is less powerful, but it has its own advantages, such as nimbleness and ability to fail non-catastrophically. Small size also allows more diversity in cultural and economic matters, and here Jacobs waxes philosophical, pointing out that favoring diversity over uniformity is a recent, post-Enlightenment idea that has not yet been fully embraced in politics. We can see analogs everywhere. Europe, split into numerous small countries from the Middle Ages onward, became far more advanced than China, which has been unified more often than not. The city-states of ancient Greece and Renaissance Italy are seen as golden ages of Western civilization, even if they weren’t part of larger political units and therefore constantly went to war with one another. In business, large companies are impressive and powerful, but people always complain that Google or Microsoft have become stagnant and that the best place to work is tiny startups of about 2 cofounders and 4 employees. In biology, humans are more successful than numerous larger animals, and in terms of raw numbers, small animals like rats or insects are the most successful of all. Jacobs’s point isn’t that smaller is always better. Her point is that the converse statement, “bigger is always better,” is false — despite how intuitive it feels for political entities. Just like we don’t view a small nation like Switzerland or Singapore as a failure of unity, we (and in particular, Canadians) shouldn’t see the secession of a place like Quebec, if it’s done peacefully and democratically, as a failure either. Still, some people in online reviews of the book complain that this argument is a bit thin, especially considering that it serves as the foundation for the later chapters (which are more directly about late 1970s Quebec politics). Sure, small is beautiful, but large states are great for stability, peace, markets, whatever. If the potential benefits of small national size are Jacobs’s strongest argument, then we can breathe a sigh of relief and go back to agreeing that separatism is bad. Pointing out the widespread bias in favor of unified political entities does seem valuable to me, but okay, fair enough. Does Jacobs have deeper reasons why separatism might be a good idea in general? Yes, and for this we go back to the second half of Cities and the Wealth of Nations. Why Nations and Empires Fail Our breathing rate is regulated through a feedback mechanism. Too much carbon dioxide in the blood, or too little oxygen, and the brain stem commands the diaphragm to accelerate breathing. Once the levels are back to normal, the brain stem receives this feedback and slows breathing down again. Now, Jacobs asks, imagine an impossible creature: ten people, all doing their own thing, but whose breathing is somehow regulated by a single brain stem. The feedback the brain stem receives is a consolidated average of everyone’s carbon dioxide and oxygen levels, and the breathing rate the stem decides on is applied to all ten people, regardless of whether they’re sleeping or playing tennis. This, to put it mildly, wouldn’t work. This creature is an analogy, representing a nation. The ten people are its individual cities, and the breathing rate is the cities’ economies. If it sounds like a stupid analogy, that’s because it is: “I have had to propose a preposterous situation,” writes Jacobs, “because systems as structurally flawed as this don’t exist in nature; they wouldn’t last.” Nor do they exist in machines we design; they wouldn’t work. But “nations, from this point of view, don’t work either, yet do exist.” The feedback mechanism that fails to work properly in a nation is currency. A currency always fluctuates according to the exports and imports of the area where it circulates. Let me use the Republic of Venice and its ducat as a toy example, because the coins look nice: Whenever Venice produces something (like salt) and sells it abroad, foreigners need ducats to buy the exports, so the demand for ducats increases. When Venice buys something from abroad, it needs to use foreign currencies, so the demand for ducats decreases. Add up everything that Venice exports and imports, and you get either a trade surplus (more exports than imports) or a trade deficit (more imports than exports), which determines the value of the ducat relative to other currencies. In both cases, a negative feedback loop restores balance over time, just like our brain stem does with carbon dioxide levels. A trade surplus, and therefore a strong ducat, means that when foreigners want Venetian salt, it’s expensive. So Venice’s exports decrease, while imports increase, since Venetians can use their valuable ducats to buy stuff cheaply from abroad. Conversely, a trade deficit makes exports a bargain for foreigners and imports expensive for Venetians. This feedback loop is great. It’s exactly what a city needs to trigger the crucial import replacement process. When exports decrease and a trade deficit begins (maybe because Constantinople found a cheaper source of salt somewhere else), the weak ducat means that Venice is less able to afford the resources and manufactured goods it used to import. The people of Venice don’t want to have less of those goods, though, so they figure out ways to produce some themselves — that is, they do import replacement. Later they will be able to export the output of the newly expanding industries too, strengthening the ducat and continuing the cycle. Currencies, Jacobs explains, function as automatic tariffs (to protect local industry from foreign imports) and automatic export subsidies (to encourage local industry to export). They are “automatic” because of the feedback mechanism. Just like an accelerated breathing rate, they take effect exactly when they are needed — and no longer. … Or so they should, except that import replacement, as we discussed, is a city process. Whereas most currencies are national or supranational. National currencies work well for city-states, like the Republic of Venice or today’s Singapore. But in large nations, which, remember, are not the fundamental unit of economic life, they mess everything up. Take a city like Detroit. When Detroit’s exports (primarily cars) decrease, Detroit gets no feedback about this, because its currency is the United States dollar, and the United States dollar’s value depends on much more than Detroit. It depends on other cities whose foreign exports might be increasing at the moment. And on rural regions that are selling resources like oil abroad. Also, trade between Detroit and other cities that use the United States dollar — i.e., American cities — is structurally unable to provide any feedback whatsoever. So Detroit doesn’t get the signal that it should buy less stuff from other cities and replace the missing imports with local production. Instead, it just declines. Jacobs hypothesizes that this issue of national currencies is at the root of every large country’s economic troubles. It is why nations and empires always centralize everything into one large city, whether that’s Paris, London, Tokyo, or Toronto, or ancient Rome: that city, being the largest, is simply the only one for which national-level currency feedback works fine. The rest of the nation or empire, then, declines. But of course, nations and empires don’t accept this. They care about the economic well-being of their peripheral regions, sometimes out of genuine concern for the people there, sometimes out of fear that they rebel or hold independence referendums. So nations and empires will embark on every possible solution to reverse the decline. All of their solutions will look like good ideas at first, and yet fail at helping the peripheral regions. Worse, these solutions will weaken the cities, thereby destroying the only real wealth of the country and bringing untold hardship for everyone. Eventually the nation or empire will disintegrate, as nations and empires always do, and always will. Jacobs calls these false solutions transactions of decline. She identifies three types, and, content warning, you might not like some of them depending on your political sensibilities. Sustained military production is a transaction of decline. Permanent military bases and garrison towns are a special kind of settlement: they import a lot and export nothing. Superficially, producing weapons and supplies for the military seems like a good deal for some cities — Jacobs gives the example of Seattle, which, before Microsoft and Amazon were a thing, depended mostly on making military aircraft. But because nobody in a military base ever tries to replace those weapons and supplies with their own production, the trade is sterile in terms of economic development. In a sense, the wealth is slowly “drained” from cities. Large empires are especially prone to this: eventually all of their wealth is destined to the military just to keep the empire together.
Inline links: https://substackcdn.com/image/fetch/$s_!YeXj!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6fe681e6-b571-4f4c-aa90-313f77fc1dbb_750x359.png
Sustained military production is a transaction of decline. Permanent military bases and garrison towns are a special kind of settlement: they import a lot and export nothing. Superficially, producing weapons and supplies for the military seems like a good deal for some cities — Jacobs gives the example of Seattle, which, before Microsoft and Amazon were a thing, depended mostly on making military aircraft. But because nobody in a military base ever tries to replace those weapons and supplies with their own production, the trade is sterile in terms of economic development. In a sense, the wealth is slowly “drained” from cities. Large empires are especially prone to this: eventually all of their wealth is destined to the military just to keep the empire together.
Your Book Review: Lying for Money
May 26, 2023 · Original source
Even a heuristic like "don't invest in companies that fake product demos" won't allow you to avoid the "false negatives," as Davies points out, as many companies that present fake demos go on to create functional products and be worth billions of dollars, so if you consider that disqualifying criteria, you would have had to say no to Microsoft in 1983, when they faked a "live" product demo for an interface manager that didn't actually exist yet. (Given that Microsoft's split-adjusted share price has risen by approximately 325,000% since its IPO in 1986, investing in Microsoft is one of the more profitable things you could have done in the 80's.)
Links For September 2023
September 28, 2023 · Original source
41: AI company Anthropic announces partnership with Amazon (including $1.25 - 4 billion investment). This was predictable: the story of the AI industry so far has been that from 2015 - 2020, a few true believers founded early startups that ate up the talent and gained the institutional knowledge. Now that AI is the Next Big Thing, the big tech companies are trying to catch up, having a hard time, and choosing to partner with the prescient early startups instead. The early startups are finding they can’t keep scaling without more money and data, forcing them to accept the big tech companies’ offers. First it was DeepMind + Google, then Open AI + Microsoft, and Anthropic was the last holdout but has acknowledged economic reality. The safety movement is concerned that Amazon might have enough power to steamroll over Anthropic’s safety-conscious culture; this did happen with DeepMind and Google, didn’t with OpenAI and Microsoft, and my guess is Anthropic held out for a good enough deal (and had enough bargaining power) that it won’t happen there either.
Inline links: AI company Anthropic announces partnership with Amazon
42: Related: one joke I keep hearing is that Anthropic will single-handedly put FTX back in the black - FTX was one of Anthropic’s biggest early investors, and Anthropic’s valuation keeps jumping by billions of dollars. Could this be literally true? I think not yet: this article explains that FTX has $16.9B in liabilities and $9.5B in remaining assets, for a debt of ~$7.5B. We don’t know what stake they had in Anthropic, but they were lead investors in Series B, Series B is usually 25-40% of stock, I’m going to estimate about 25%. Amazon offered to pay $4 billion for some unknown stake in Anthropic; if it’s 49% (the same as Microsoft in OpenAI) that values the company at $8 billion. So FTX has $2 billion worth of stock, less if it’s been further diluted. That’s only enough to take care of about a quarter of their debt. Will Anthropic go up 4x in the next few years? OpenAI is already seeking (though hasn’t yet gotten) a valuation of $90 billion and it doesn’t seem unreasonable for Anthropic to be a third as valuable as OpenAI, so who knows?
Inline links: FTX was one of Anthropic’s biggest early investors, this article, $90 billion
Sam Altman Wants $7 Trillion
February 13, 2024 · Original source
The basic logic: GPT-1 cost approximately nothing to train. GPT-2 cost $40,000. GPT-3 cost $4 million. GPT-4 cost $100 million. Details about GPT-5 are still secret, but one extremely unreliable estimate says $2.5 billion, and this seems the right order of magnitude given the $8 billion that Microsoft gave OpenAI.
Inline links: GPT-2, GPT-3, GPT-4, one extremely unreliable estimate
(Sam Altman is working on fusion power, but this seems to be a coincidence. At least, he’s been interested in fusion since at least 2016, which is way too early for him to have known about any of this.)
Inline links: Sam Altman is working on fusion power
GPT-6 will probably cost $75 billion or more. OpenAI can’t afford this. Microsoft or Google could afford it, but it would take a significant fraction (maybe half?) of company resources.
SB 1047: Our Side Of The Story
October 10, 2024 · Original source
You can’t see it in the screenshot, but the first stock is NVIDIA, the second TSMC, the third Alphabet, and the fourth Microsoft. On average they went up about 0.5%, on a day when the NASDAQ as a whole also went up about 0.5%.
OpenAI Nonprofit Buyout: Much More Than You Wanted To Know
March 13, 2025 · Original source
Defendants’ secretive venture included an effort by Microsoft to “exploit” OpenAI. As part of this, Altman established a vast network of for-profit entities in which both he and Microsoft hold significant ownership stakes. Further, OpenAI and Microsoft have several contractual arrangements, including, for example, Microsoft’s agreement to supply raw materials to OpenAI and OpenAI’s granting Microsoft an exclusive license to its technology. Defendants have ensured OpenAI’s board is full of directors fully aligned with Altman and Brockman’s interests, such that independent directors constitute only a minority of the board.
ACX Grants 1-3 Year Updates
June 18, 2025 · Original source
Codebuff, an AI coding startup I probably can’t take full credit for all of this just from giving them $20K in seed funding, but I continue to appreciate everything they do for this community and the world. 35: Further S’s Political Career This person didn’t win their election, but has since pivoted to AI safety and works in a well-regarded AI policy think tank. 36: Seeds Of Science, A Journal Of Non-Traditional Research No update received, but this was a public journal and it is easy to follow their work, see their website and Substack. They published two dozen articles of widely varying quality through 2023 and 2024, then closed in 2025. A remnant of the original vision survives as a science blogging aggregator. This was about my median expectation for this grant, but it was very inexpensive and I decided to take a chance on it anyway. 37: Good Science Project, Working To Improve Federal Science Funding No update received, but they have a public Substack discussing their progress. Their proposals for NIH reform have influenced Congress and made government agencies pay more attention to scientific integrity. 38: Advising Developing Countries On How To Grow Their Economies With our initial ACX grant, we piloted the Growth Teams model in Rwanda, helping the government jumpstart the export-oriented call center (BPO) industry. Since 2022, that effort has contributed to the creation of 2,000 formal jobs and the emergence of some of the country’s largest private employers. We’ve since expanded to Tanzania, Malawi, and the Indian states of Goa and Meghalaya. To refocus the global development discourse on broad-based economic growth, we co-organized the Growth Summit with the Center for Global Development and the Charter Cities Institute, and have published articles in leading outlets including Stanford Social Innovation Review, ProMarket, and the Global Prosperity Institute. Our work has attracted support from Open Philanthropy, Schmidt Futures, and Mulago Foundation, and our advisors now include economists Lant Pritchett, Stefan Dercon, and Kunal Sen. 39: Help Luca De Leo Get Started In AI Safety Research No update received, but Luca now runs the AI safety group at the University of Buenos Aires, Argentina. 40: Typist For Saharon Shelah This was another ACXG+ Grant, funded by an anonymous outside funder and not listed in the original announcement. Saharon is a prolific and influential Israeli mathematician, but many of his discoveries are hand-written in an unpublishable format. This grant funded a typist to help make his results suitable for publication. According to this page, they have made over fifty new papers and preprints available. Second Cohort: One Year Updates 41: Lead-Acid Battery Recycling In Nigeria The Nigeria field research was a major success. We spent most of September doing field research in multiple major cities in Nigeria, and got a good sense of the used lead-acid battery supply chain. This field research served as the foundation for expanding our project, and has been very impactful in shaping our ongoing research. We published our findings from Nigeria, which were shared with Nigerian government regulators and global NGOs working on lead poisoning. The grant also gave us the on-the-ground experience we needed to both fully understand and credibly engage with groups, both in Nigeria and globally, on the ULAB issue. In the meantime, beyond continued research, we’ve also launched a dashboard (trade.leadbatteries.org) for analyzing global lead trade data. Right now, we’re: Launching two studies (one RCT, one environmental analysis) in Nigeria in collaboration with local universities to develop a more rigorous understanding of lead pollution due to low-standard ULAB recycling in Nigeria Collaborating with a non-profit incubator to launch an NGO focused on demand-side solutions Beginning a partnership with a West African environmental regulator to scale cheap air monitoring technology to quickly identify and reduce lead pollution from low-standard smelting If any of this sounds interesting to you, please sign up for our Substack (leadbatteries.substack.com) or send us an email at hugosmith@uchicago.edu! 42: Compensation For Kidney Donors The End Kidney Deaths Act (H.R. 2687 / EKDA) is a groundbreaking ten-year pilot program designed to save lives and reduce healthcare costs. It provides a refundable tax credit of $10,000 per year for five years, a total of $50,000, to living kidney donors who donate to a stranger, helping those who’ve waited the longest on the transplant list. Between 2010 and 2021, 100,000 Americans died while qualified and waiting for a kidney. The EKDA aims to change that trajectory. Within ten years of its passage, up to 100,000 Americans could receive a life-saving living donor kidney which typically lasts twice as long as a deceased donor kidney. This would not only save lives but also save taxpayers up to $37 billion. The legislation has been reintroduced in the House, and we have a committed Republican Senate lead. Now, we need a Democratic Senator to co-lead and help move this bipartisan effort forward. Time is short, and we are racing to pass the bill this Congressional session. 36 organizations already support the EKDA. Join the movement and help end preventable kidney deaths. Visit EndKidneyDeaths.org to help us get to the finish line. Elaine and her org have been working extremely hard on this; you can read a Vox article on their campaign here. If you want to sign up for her email list and get updates any time there is a representative you can contact or meeting you can join in, go here. 43: Genetic Hack To Prevent Suffering In the estimate of multiple team members, the ACX grant was “worth it” - it likely had a counterfactual net positive impact, even though we had to pivot from our initial fast-track plans for developing the precision anti-suffering therapy. We identify three primary streams of value: a) reducing uncertainty in the emerging field through early exploratory research, helping with the identification of dead ends and promising R&D trajectories; b) a wide range of downstream effects (beyond the “raising awareness” cliché), including talent mobilization and rekindled interest in suffering abolitionism as a distinct cause area; and c) certain developments that cannot yet be publicly disclosed. In December 2024, Marcin Kowrygo (Acting CEO & volunteering contributor), David Pearce (Director of Bioethics), Aatu Koskensilta (President), and a few other team members decided to leave The Far Out Initiative. They look forward to collaborating and applying their experience to advance the suffering abolitionist lineage in the spirit of open science, public good, and thoughtfully decentralized governance. Feel free to reach out to us at suffab at protonmail dot com to discuss collaboration opportunities! I wrote a post profiling the Far Out Initiative here. Unfortunately there were some internal disagreements, and the people ACX Grants was closest to left the organization. I plan to continue to monitor whatever they do next. 44: Advocate For Pandemic Response Team At FDA This team prefers has asked me not to discuss their progress publicly, but you can probably guess what their lives are like right now, and your guess would be correct. 45: Anti-Mosquito Drones We developed a cheap sonar that is able to detect, track and classify the ultrasonic echoes of mosquito wings at more than three meters. I believe it’s a world first! We also have control algorithms that take the sonar data and output control commands that both ram into mosquitoes and avoid the walls of a simulated environment. Our current work is on integrating both components on a real drone, and we expect to be able to kill mosquitoes by June. We’ve also made an internal impact study (napkin-sized) that shows we’ll be more cost-effective than ITNs in urban to periurban environments. So, we’re super excited with what comes next and can’t wait to share the videos of our first interceptions! More information [in the video below] and on our website, https://tornyol.com 46: Tarbell Fellowship For AI Journalism No update received, but they have a public website. I can’t find the Voices program in particular, but the overall fellowship completed their first class of seven fellows and is working on their second. 47: Germicidal UV Lamp Study The research has successfully demonstrated the ability of off the shelf ozone scrubbers to mitigate the ozone production of far-UVC lamps, is now available as a preprint (https://chemrxiv.org/engage/chemrxiv/article-details/67e4cde76dde43c9084d88b7). The paper has been submitted for publication and is currently undergoing peer review. Any ideas you have for potential funders we can approach to help execute our six-year plan to accelerate far-UVC would be appreciated https://blueprintbiosecurity.org/introducing-project-air/ 48: Technological Solutions To Animal Welfare Challenges Directly because of Innovate Animal Ag's work, the first U.S. egg producer publicly announced in the New York Times their adoption of in-ovo sexing technology, eliminating the need to cull day-old male chicks. The initial in-ovo sexing machine began operating in the U.S. at the end of 2024, with the first eggs from these hens expected on shelves in mid-2025. External evaluations estimate our work accelerated U.S. adoption of this technology by over seven years, meaning that once fully implemented, more than 2 billion chicks will have been spared. In addition to continuing to support the rollout of in-ovo sexing in the US and globally, we're now exploring other technologies and paths to impact. Current promising projects include developing humane slaughter methods for fish and advocating for USDA approval of a poultry vaccine against bird flu. They add: If you ever meet folks that are interested animal welfare and are partial to more technocratic and practical solutions, please continue to pass them our way, or connect them directly to me. 49: Assurance Contract Website www.Spartacus.app is an ACX grantee that created a platform to help solve coordination and collective action problems. It enables the creation of campaigns that build critical mass through conditional commitments, which only activate when a sufficient number of people join, converting risk and uncertainty into a higher probability of successful outcomes. They are currently facilitating several projects that leverage conditional commitments, including a dominant assurance contract interface for fashion pop-ups, accelerating a community business association's membership drive, and helping an AI safety organization organize petitions and events, among others. They have pivoted from an emphasis on high-stakes coordination problems requiring anonymity (because they occur too infrequently) to a broader range of more common use cases and have successfully run small-scale campaigns, but are still working toward product-market fit. Despite resource constraints and split time commitments that have impeded faster progress, they remain dedicated to the project's growth and success. You can follow its progress on X or Substack, or email Jordan directly here. 50: Cause Prioritization @ Center For Exploratory Altruism Research Moderately good progress on a salt reduction policy advocacy project we funded; informal commitments have been made by the Ministry of Health, and we're awaiting the publication of a formal administrative order. The official description sounds maximally generic, but this is an EA charity with a broad mandate whose current thesis is that dietary guidelines in developing countries can have outsized effects in saving lives. They’re making some progress on a salt reduction campaign in a developing country they prefer not to name publicly. 51: Mark Webb Studying Land Reform The purpose of this project was to identify specific farmland that could be acquired and transferred to the farmers already working the land. This has been difficult to achieve. I have been able to connect with other charities and landless farmers, and was able to interview a number of people about what their situation looks like, as well as what it would look like to them personally if they owned, rather than rented, their farmland. All this was immensely helpful in pushing this long-term project forward, even if I was unable to identify a specific plot of land that could be used to try the experiment. I intend to continue this project. If you have any insights or connections, I am interested. 52: More AI Advocacy In Australia Good Ancestors is focused on AI safety policy in Australia. Middle powers might be a useful path to influence as the US and China focus on racing, rather than safety. The ACX grant helped us give testimony about AI safety to the Australian Senate alongside Google, Microsoft and Facebook (We were the only nonprofit to give oral evidence to the inquiry. We also engaged government on other AI-related issues, including cybersecurity, biosecurity, consumer law and automated decision making (https://www.goodancestors.org.au/ai-safety). We’re currently working to inform voters about where parties stand on AI safety for the election, ahead of engaging on a likely Australian AI Act in 2025 (https://www.australiansforaisafety.com.au/). This is the same Australian lobbying organization we founded in Year 1, after a change in name and leadership. I continue to be excited about AI safety in middle-tier countries for a few reasons. First, these countries have some power in international organizations to set international standards. Second, companies will usually comply with any not-excessively-burdensome regulation set by any country with a significant market. Third, AI safety is underfunded by the standard of government programs, so Australia setting up a national AI Safety Institute would significantly expand the field. It’s kind of crazy that ACX Grants tier levels of money can have significant effects at this scale, but GA continues to do a great job and we continue to be proud to support them. 53: Campus For African School Of Economics At Zanzibar Charter City The ACX grant helped launch the first research center at the African School of Economics-Zanzibar, which is a main anchor of the Fumba Town charter city project in Zanzibar. This research center is called the Africa Urban Lab (AUL), focused on rapid urbanization across Africa. The AUL launched its first Diploma program in Urban Development with 38 students in our first cohort (now graduated!), including mayors, and deputy mayor, a director of a national Ministry of urban development, and many others. We published our research framing papers for the AUL's research agenda. We raised funding to launch an Urban Expansion Program that's now selecting 15 African cities to support in implementing urban expansion planning on the urban periphery. We held two Public Talks by renowned cities scholars and practitioners. We received additional funding from Emergent Ventures and from the Templeton Foundation. And we've partnered with 8 universities across the region, and with one of these universities (Ardhi) we'll be working with them to update their urban planning and urban economics curriculum (amplifying AUL's impact beyond our own organization). A longer update from end of 2024 is here: https://www.aul.city/blog/reflecting-on-africa-urban-lab-s-inaugural-year-2024-highlights) 54: Online Training Program For Health Workers In Developing Countries To date, over 11,000 health workers in Nigeria have completed our course on basic, life-saving newborn care. ACX funding was catalytic for helping us secure government approvals and complete an evaluation of the impact of our training on health workers' clinical practices. The evaluation shows that birth attendants provide better birth care after taking the course. We fed the evaluation results into an updated model, which suggests the program is 24 times more cost-effective than direct cash transfers (a widely recognized benchmark for cost-effectiveness). The program is likely to become even more cost-effective as we scale up. https://healthlearn.org/blog/updated-impact-model 55: Smartphone Pupillometry To Diagnose Neurological Conditions We have continued to expand our work in the smartphone pupillometry space and the development of our application, PupilScreen (https://www.apertur.ai/). We have expanded our pilot/research program to include new sites across the United States (Missouri, New Jersey, Kentucky, USAC racing, PitFit driver performance training in Indiana) and the world (Nepal, Taiwan, South Africa). We continue to publish at the leading edge of the pupillometry literature as well looking at concussion (https://neuro.jmir.org/2024/1/e58398 and https://pubmed.ncbi.nlm.nih.gov/39682632/), cerebral vasospasm (https://pubmed.ncbi.nlm.nih.gov/39128501/), and stroke (https://pubmed.ncbi.nlm.nih.gov/39674431/ and https://pubmed.ncbi.nlm.nih.gov/39561861/). Currently, we are raising a $3 million seed round via a SAFE to fund the expansion of our work into the hands of healthcare workers and the general public. We will first focus on traumatic brain injury for clinical use and develop a neuro-monitoring wellness application utilizing our technology for the general public. They add: “We would welcome connections to anyone that you think might be interested in supporting our work further by investing in our $3M seed round of funding.” 56: Mike Saint-Antoine’s Biology Tutorial Videos Since getting the grant, I've continued to make Youtube tutorials as planned. One series that I'm especially proud of is about how to make a neural network in the Julia programming language completely from scratch, with no imports, up to the point of being able to solve MNIST (https://www.youtube.com/playlist?list=PLWVKUEZ25V97tNULapu07DhWv6_W4NfpE). Also, a college student in Pakistan came across my videos and invited me to give a virtual Zoom-lecture to her department, so I ended up teaching a 6-hour "Python-for-Biologists" workshop to more than a hundred college students in Pakistan over Zoom. So that was pretty awesome. Also, lately I've been teaching some in-person classes too, mostly at Fractal University in NYC, and I also recently organized a day-long, in-person Beginner Python class for people in my local area (Philly suburbs) who wanted to learn some basic programming. I'm having a lot of fun with this project, and am grateful to Scott and the grant funders for their generosity! 57: Conceptual Boundaries Workshop On AI Safety The workshop was completed successfully; you can read a writeup here. 58: Apart Research To Incubate AI Safety Scientists No update received, but they have a public website, and you can see their impact metrics here. They seem to be in urgent need of more funding. 59: Primer On How To Achieve Political Change No update received and I can’t find anything about this. 60: Research IVF Clinic Success Rates We've built a predictive model that estimates the odds of having a child at different IVF clinics across the country while controlling for factors like patient age and infertility differences that can falsely make some clinics look better than others. We found that an average patient can increase their odds of having a kid by 43% just by going to a top 10% clinic. Patients unlucky enough to go to a bottom 10% clinic will reduce their odds of having a kid by 40%. Next month, we're adding several more clinics, 2023 data, additional procedural controls, and donor/gestational carrier models, which should push our accuracy beyond state-of-the-art models in this space and better isolate clinic impact on patient outcomes. We've launched ivf.clinic, a website where patients can access personalized IVF reports and browse our clinic rankings (though we're still squashing some bugs). Currently, we're expanding our research to include comprehensive insurance coverage and pricing data across clinics nationwide. If anyone has insights on automating the collection of IVF clinic pricing information, I'd love to hear from you at scelarek@gmail.com. 61: Replicate Study On Brain Wave Synchronization For Speeding Learning We have acquired and configured the OpenBCI UltraCortex Mark IV 8-channel EEG headset and a clinical-grade Biosemi 32-channel EEG system. We’ve implemented the required components for the experimental pipeline (computing alpha from EEG, flashing bright white light, presenting stimulus images). We are currently putting them together into a single system that we’ll use to collect the data from several participants. We are aiming to gather data on several participants in late June / early July and complete the pilot of the replication in July 2025. If you’d like to be a participant in the study, [they might announce a link once they have it]. 62: Advocate Repeal Of Interstate Runaway Compact No update received and I can’t find anything about this. 63: Animal Welfare (Especially Fish) In Turkiye Future For Fish asks companies to sign up to FFF's fish welfare commitment, which requires producers to certify their facilities and enforce specific standards for stocking density and harvest. Luckyfish, İlknak, Divan (35 restaurants, 17 hotels) and NG Hotels (5 hotels) have signed and published FFF's fish welfare commitment with İlknak publishing the commitment on their website. Kılıç published its first sustainability report detailing fish welfare policies, including enforcing a maximum stocking density of 10 kg/m³ and confirmation of electrical stunning practices. Longer version with some caveats: https://manifund.org/projects/improving-fish-w From the longer document, these commitments involve things like reducing overcrowding, or stunning fish before killing them. Over 30 million fish were affected just from their single largest commitment, and they say 100 fish are helped per dollar spent. 64: More Georgism Advocacy Lars and Will used the 2021 grant to co-found ValueBase. Will remained with the company, and Lars left to do advocacy work at the Center For Land Economics. Here’s their summary of how things are going: [Our] organization transitioned leadership with Greg Miller, a former Program Analyst at the US Department of Housing and Urban Development, and Lars Doucet, author of Land is A Big Deal and Co-Founder of Valuebase, working full time and Joe Caissie stepping aside. This transition happened naturally as the next career transition for each respective person. Since then, progress has been made on pushing forward legislation. Maryland had two bills introduced to give Baltimore and counties the ability to enact split-rate taxes. One of the bills passed the state senate and would allow Baltimore to enact land value taxes within one mile of rail corridors–this contains 50% of Baltimore’s land value. However, the legislative session ended. We expect the bill to revive next session. The Center for Land Economics has been actively working to help efforts to get this bill passed the line. At the same time, we have uncovered systematic undervaluing of vacant land in assessments. We are writing a report on the assessment issues in Maryland with actionable steps to resolve them.
Inline links: Codebuff, website, Substack, survives, a public Substack, in Rwanda, Growth Summit, Stanford Social Innovation Review, ProMarket, Global Prosperity Institute, Saharon, this page, eadbatteries.substack.com, here, here, a post profiling the Far Out Initiative here, https://tornyol.com, a public website, https://chemrxiv.org/engage/chemrxiv/article-details/67e4cde76dde43c9084d88b7, https://blueprintbiosecurity.org/introducing-project-air/, our way, connect them directly to me, www.Spartacus.app, X, Substack, here, https://www.goodancestors.org.au/ai-safety, https://www.australiansforaisafety.com.au/, https://www.aul.city/blog/reflecting-on-africa-urban-lab-s-inaugural-year-2024-highlights, https://healthlearn.org/blog/updated-impact-model, https://www.youtube.com/playlist?list=PLWVKUEZ25V97tNULapu07DhWv6_W4NfpE, here, public website, here, in urgent need, https://manifund.org/projects/improving-fish-w
Why AI Safety Won't Make America Lose The Race With China
November 26, 2025 · Original source
Compute: America is far ahead. We have better chips (thanks, NVIDIA) and can produce many more of them (thanks, TSMC). Our recent capex boom, where companies like Google and Microsoft spend hundreds of billions of dollars on data centers, has no Chinese equivalent. By the simplest measure - total FLOPs on each sides - we have 10x as much compute as China, and our advantage is growing every day. A 10x compute advantage corresponds to about a 1-2 year time advantage, or an 0.5 - 1 generation advantage (eg GPT-4 to GPT-5).
Open Thread 415
January 05, 2026 · Original source
Some people have argued that you have to find a way to join an AI company, because AI company employees will form the new ruling class, with everyone else as serfs. I disagree. The main thing an AI company employee has that you don’t is AI company stock. But you can buy stock in Google, you may soon be able to buy stock in OpenAI and Anthropic, and even if not, you can get indirect exposure to these companies via stock in Amazon and Microsoft. I don’t recommend putting all your money in these stocks. But there’s no fundamental difference between a Google employee having 75% of their money in Google stock because they didn’t cash out their equity vs. you having 75% of your money in Google stock because you’re crazy and fail at diversification. So either put 75% of your money in Google stock or don’t (I recommend don’t), and don’t worry about how you need to join an AI company or be left out of the future oligarchy.
Mantic Monday: Groundhog Day
March 03, 2026 · Original source
Framed this way, the Pentagon’s actions sound devastating. Anthropic relies on compute to train and run its AIs. Most of this compute is in data centers owned by Amazon, Google, and Microsoft. At least Amazon and Microsoft have contracts with the US military. If they had to drop Anthropic, it would make it impossible for the company to stay a frontier AI lab.
The lawyers who weighed in seem to think that Anthropic’s interpretation of the law is correct, and Secretary Hegseth’s interpretation confused. In some situations, this might be cold comfort - how much does it help to be right about the law when the government is wrong? But in this case, it probably helps a lot. Amazon, Google, and Microsoft are all big Anthropic investors - each owns about a 10% stake - and have multi-billion dollar AI compute contracts. Together, the three tech giants must have at least $100 billion riding on Anthropic’s success. They also have good administration connections and great lobbyists, and even Hegseth isn’t stupid enough to pick fights with them all at once. So probably they send their lobbyists to have a talk with Hegseth about what the “supply chain risk” designation actually entails, Hegseth enforces the letter of the law, and Anthropic is barely affected. At least this is the story the prediction markets are going with: