Tom Davidson
Article
Tom Davidson is a recurring person in the Astral Codex Ten archive, appearing 6 times across 6 issues between June 20, 2023 and February 12, 2026. The archive places it in contexts such as “by Tom Davidson. Davidson tries to model what some people (including me) have previously called “slow AI takeoff””; “When this happens, Tom Davidson and OpenPhil will be able to say”; “Tom Davidson’s Compute-Centric Framework report forecasts”. It most often appears alongside OpenAI, GPT-4, Yudkowsky.
Metadata
- Category: People
- Mention count: 6
- Issue count: 6
- First seen: June 20, 2023
- Last seen: February 12, 2026
Appears In
- Davidson On Takeoff Speeds
- Tales Of Takeover In CCF-World
- Sam Altman Wants $7 Trillion
- Links For April 2025
- Open Thread 405
- What Happened With Bio Anchors?
Related Pages
-
- OpenAI (4 shared issues)
-
- GPT-4 (3 shared issues)
-
- Yudkowsky (3 shared issues)
-
- AGI (2 shared issues)
-
- Ajeya Cotra (2 shared issues)
-
- Bio Anchors (2 shared issues)
-
- Bio Anchors (2 shared issues)
-
- Bio Anchors (2 shared issues)
-
- Biological Anchors (2 shared issues)
-
- China (2 shared issues)
-
- DALL-E (2 shared issues)
-
- Davidson (2 shared issues)
External Links
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.
I thought about this when reading What A Compute-Centric Framework Says About Takeoff Speeds, by Tom Davidson. Davidson tries to model what some people (including me) have previously called “slow AI takeoff”. He thinks this is a misnomer. Like skiing down the side of Mount Everest, progress in AI capabilities can be simultaneously gradual, continuous, fast, and terrifying. Specifically, he predicts it will take about three years to go from AIs that can do 20% of all human jobs (weighted by economic value) to AIs that can do 100%, with significantly superhuman AIs within a year after that.
When this happens, Tom Davidson and OpenPhil will be able to say “We drew a curve whose shape sort of corresponds to this process.” It probably won’t feel very reassuring.
Tom Davidson’s Compute-Centric Framework report forecasts a continuous but fast AI takeoff, where people hand control of big parts of the economy to millions of near-human-level AI assistants .
Inline links: Tom Davidson’s
[All numbers here are very rough and presented in a sloppy way. For the more rigorous versions of this, read Tom Davidson, Yafah Edelman, and EpochAI)
The article doesn’t explain why the board did such a poor job communicating their grievances, maybe it’s in the full book. It does sound like part of board’s problem was that they were leaning heavily on Mira Murati but she was playing both sides off against each other. 23: And the Forethought Institute has been putting out some great analysis lately, including Will AI R&D Automation Cause An Intelligence Explosion?, by Daniel Eth and Tom Davidson, and AI Enabled Coups: How A Small Group Could Use AI To Seize Power, by Tom Davidson, Lukas Finnveden, and Rose Hadshar. And here’s Davidson defending the coups paper on the 80,000 Hours podcast. 24: Agent Village is a sort of "reality show” where a group of AI agents has to work together to complete some easy-for-human tasks (currently: pick a charity and raise money for it) and you get to watch. 25: University of Austin promises approximately-automatic admission to anyone with a 1460+ on their SATs (or similar scores on other standardized tests). 26: Cremieux on birth order effects (X). His conclusion: “The birth order effect is social. It is driven by parental interactions and investments, and sibling interactions that are dynamic with respect to age.” 27: Claim from new paper, via Alex Tabarrok: “Prohibiting the FDA from regulating e-cigarettes reduced smoking attributable mortality by nearly 10% on average each year from 2011-2019 for a total savings of some 677,000 life-years, or approximately 1/3 the estimated benefit of early HIV/AIDS drugs through year 2000”. Related: FDA will not regulate lab-developed tests for the near future. 28: Bryan Caplan on Natal Con, the pronatalist conference in Austin. My strongest opinion on this is that they should either change the name or hold the next one in Natal, Brazil. 29: Am I living in a conservative filter bubble? I keep hearing how we need a “reckoning” over the government’s disastrous anti-COVID policies, but the latest YouGov polling suggests that large majorities of Americans continue to support those policies: 30: A California legislator proposed a bill that would ban OpenAI’s nonprofit → forprofit conversion, backed by a suspiciously specific interest group, the Coalition For AI Nonprofit Integrity. I assume this is either Elon Musk or our conspiracy; not sure which. But their plan was stymied when the legislature “amended” the bill to remove its entire text and replace it with unrelated text about airplane loans. The legislator apparently got cold feet after being warned it might inflict collateral damage on other companies, and because of the way the California legislature works it’s sometimes more efficient to turn doomed bills into other bills than to simply withdraw them. 31: EthnoGuessr is a GeoGuessr variant: it shows you pictures of an ethnic group, you click on the map where you think they’re from. Warning that if you play this too much you might get into race science. Their source, humanphenotypes.net, divides humanity into a hundred or so ethnic groups. Although they cite sources, I don’t understand the philosophical basis of the classification. Also, 100 images is so few that you start memorizing them after a while. I hope they move on to real pictures of real people in naturalistic situations. Remember, asking where someone is from ‘originally’ is a microaggression, but inferring it yourself based on their “mildly platyrrhine, high-rooted nose” is A-OK! 32: Farmkind has a new version of their calculator to determine meat offsets, eg how much do you have to donate to animal welfare charities to compensate for the animals you harm by eating meat. Does the average person really eat chicken 9x a week? 33: Not going to waste your time listing every bad thing Trump has done this month, but among the worst is sending innocent people to horrible Salvadorean prisons (including one person picked up because he had an autism awareness tattoo in honor of his brother, which they mistook for a gang tattoo), then refusing to bring them back. I have seen a couple of people defend denying immigrants due process; I assume they will not be moved by humanitarian arguments, but I think there are some more practical considerations: Zaid Jilani points out that if immigrants don’t get a right to due process, citizens also don’t get a right to due process, because the government can kidnap citizens, claim they’re immigrants, and the citizens can’t prove otherwise since they don’t get due process.
Inline links: https://substackcdn.com/image/fetch/$s_!WFgK!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6448a778-3290-4c85-9480-d4f78f9be960_916x255.png, Will AI R&D Automation Cause An Intelligence Explosion?, AI Enabled Coups: How A Small Group Could Use AI To Seize Power, here’s, Agent Village, promises approximately-automatic admission, Cremieux on birth order effects (X), new paper, Alex Tabarrok, FDA will not regulate lab-developed tests for the near future, Bryan Caplan on Natal Con, Natal, Brazil, the latest YouGov polling, https://substackcdn.com/image/fetch/$s_!HGBp!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd049433b-1cf5-442a-9d17-4be3a7846d61_760x370.png, Coalition For AI Nonprofit Integrity, “amended”, the way the California legislature works, EthnoGuessr, https://substackcdn.com/image/fetch/$s_!Y3AO!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbde5185a-2b93-4c07-904b-66d2c2311a53_1140x1033.png, humanphenotypes.net, a new version of their calculator, including, Zaid Jilani
5: Forethought (AI preparedness research org including Will MacAskill, Tom Davidson, etc) wants to hire more researchers. Offices in Oxford/Berkeley, slight bias towards people in these areas but remote work possible. Salaries £80,000 - £150,000 depending on qualifications and seniority. Must be, uh, good at research, I think this looks more like academic philosophy or economic modeling than like training LLMs, but it’s pretty vague. Learn more and apply here.
Inline links: Learn more and apply here
In 2023, Tom Davidson published an updated version of Bio Anchors that added a term representing the possibility of recursive self-improvement. The new calculations shifted the median date of AGI from 2053 → 2043. This doesn’t explain why our own timeline seems to be going faster than Bio Anchors: even 2043 now feels on the late side, and anyway recursive self-improvement has barely begun to have effects.
Inline links: an updated version of Bio Anchors