Leopold Aschenbrenner · May 29, 2024
Leopold Aschenbrenner argues in Situational Awareness that AGI by 2027 is strikingly plausible
Markdown · JSON · Original source
Leopold Aschenbrenner’s piece was published on May 29, 2024. He argues that trends in compute, algorithmic efficiencies and “unhobbling”—changes that make models’ existing capabilities more useful—make a large further jump in AI capabilities plausible. He expects models to move beyond chatbots toward agents that can work more independently, and argues that systems able to do AI research could accelerate further progress. This is a forecast based on extending observed trends, not a report that these outcomes have occurred. Aschenbrenner identifies limited training data as a major uncertainty, saying progress could stall if research does not overcome that constraint. He argues that the consequences could include automating cognitive jobs, while noting that some areas, such as robotics, could take longer and societal choices or regulation could slow adoption in medical and legal work. Whether the data constraint can be overcome, and whether models can become capable of sustained independent work, remain unknown.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Leopold Aschenbrenner · Article
Source details
Original source date: May 29, 2024 (2024-05-29T21:07:53+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-18T12:43:09.925Z. First observed: 2026-10-30T09:46:08.423Z. Retrieved: 2026-10-30T09:47:45.604Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 30, 2026; originally published May 29, 2024. Aschenbrenner’s forecasts rely on extrapolating trends and carry the uncertainties he identifies, including the possibility that progress stalls at the data constraint. The piece does not establish that its predicted capabilities or effects on jobs have occurred.
Open question
Can AI research overcome the data constraint and enable models to work independently on extended tasks?
Source-stated
Aschenbrenner argues that AGI by 2027 is strikingly plausible.
AGI by 2027 is strikingly plausible.
Source-stated
Aschenbrenner predicts that models could do the work of an AI researcher or engineer by 2027.
it is strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer.
Source-stated
He argues that compute, algorithmic efficiencies and “unhobbling” gains could produce another GPT-2-to-GPT-4-sized jump by the end of 2027.
we should (roughly) expect another GPT-2-to-GPT-4-sized jump in the 4 years following GPT-4, by the end of 2027.
Source-stated
Aschenbrenner identifies limited training data as a possible constraint that could cause progress to stall.
There’s a very real chance things stall out (LLMs might still be as big of a deal as the internet, but we wouldn’t get to truly crazy AGI).
Source-stated
He argues that systems able to automate AI research could accelerate further progress, while noting that some areas and forms of societal adoption could take longer.
Perhaps some areas, like robotics, might take longer to figure out by default. And the societal rollout, e.g. in medical or legal professions, could easily be slowed by societal choices or regulation.
Source-stated
Aschenbrenner predicts that AI systems could automate cognitive jobs, while describing this as a possibility rather than an accomplished outcome.
These AI systems will basically be able to automate basically all cognitive jobs (think: all jobs that could be done remotely).
Source and contributor context
Leopold Aschenbrenner · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Artificial general intelligence
The idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Editorial provenance
Version 1. Publication ID: 6e84d9e9-d036-52d1-43f6-75f76de56aeb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-18T12:39:53.619Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 37e556060e5aec018dcaa691df5ef2997e962622d0e3687685d7915a14548e53. Source text hash: 894566c454fd6c2ddbe9aea8d00157b9ac1b1f8fc72fb3cfe8e1148f46529100.
Revision and source-check history
Version 1 · 2026-11-18T12:43:09.925Z
Leopold Aschenbrenner argues in Situational Awareness that AGI by 2027 is strikingly plausible
Leopold Aschenbrenner’s piece was published on May 29, 2024. He argues that trends in compute, algorithmic efficiencies and “unhobbling”—changes that make models’ existing capabilities more useful—make a large further jump in AI capabilities plausible. He expects models to move beyond chatbots toward agents that can work more independently, and argues that systems able to do AI research could accelerate further progress. This is a forecast based on extending observed trends, not a report that these outcomes have occurred. Aschenbrenner identifies limited training data as a major uncertainty, saying progress could stall if research does not overcome that constraint. He argues that the consequences could include automating cognitive jobs, while noting that some areas, such as robotics, could take longer and societal choices or regulation could slow adoption in medical and legal work. Whether the data constraint can be overcome, and whether models can become capable of sustained independent work, remain unknown.
Original source
Leopold Aschenbrenner · Article
Source details
Original source date: May 29, 2024 (2024-05-29T21:07:53+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-18T12:43:09.925Z. First observed: 2026-10-30T09:46:08.423Z. Retrieved: 2026-10-30T09:47:45.604Z.
Limits
Text as retrieved on October 30, 2026; originally published May 29, 2024. Aschenbrenner’s forecasts rely on extrapolating trends and carry the uncertainties he identifies, including the possibility that progress stalls at the data constraint. The piece does not establish that its predicted capabilities or effects on jobs have occurred.
Open question
Can AI research overcome the data constraint and enable models to work independently on extended tasks?
Source-stated
Aschenbrenner argues that AGI by 2027 is strikingly plausible.
AGI by 2027 is strikingly plausible.
Source-stated
Aschenbrenner predicts that models could do the work of an AI researcher or engineer by 2027.
it is strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer.
Source-stated
He argues that compute, algorithmic efficiencies and “unhobbling” gains could produce another GPT-2-to-GPT-4-sized jump by the end of 2027.
we should (roughly) expect another GPT-2-to-GPT-4-sized jump in the 4 years following GPT-4, by the end of 2027.
Source-stated
Aschenbrenner identifies limited training data as a possible constraint that could cause progress to stall.
There’s a very real chance things stall out (LLMs might still be as big of a deal as the internet, but we wouldn’t get to truly crazy AGI).
Source-stated
He argues that systems able to automate AI research could accelerate further progress, while noting that some areas and forms of societal adoption could take longer.
Perhaps some areas, like robotics, might take longer to figure out by default. And the societal rollout, e.g. in medical or legal professions, could easily be slowed by societal choices or regulation.
Source-stated
Aschenbrenner predicts that AI systems could automate cognitive jobs, while describing this as a possibility rather than an accomplished outcome.
These AI systems will basically be able to automate basically all cognitive jobs (think: all jobs that could be done remotely).
Source and contributor context
Leopold Aschenbrenner · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Artificial general intelligence
The idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Editorial provenance
Version 1. Publication ID: 6e84d9e9-d036-52d1-43f6-75f76de56aeb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-18T12:39:53.619Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 37e556060e5aec018dcaa691df5ef2997e962622d0e3687685d7915a14548e53. Source text hash: 894566c454fd6c2ddbe9aea8d00157b9ac1b1f8fc72fb3cfe8e1148f46529100.
uncertain · 2026-11-18T12:39:53.619Z · anthropic/claude-opus-5.5
Publication: 6e84d9e9-d036-52d1-43f6-75f76de56aeb
attribution — pass: The reading consistently attributes the argument and forecasts to Leopold Aschenbrenner, the author and publisher of this single essay in the Situational Awareness series. No source assertion is presented as an established fact or as EM's own view.
I make the following claim: it is strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer.
summary — uncertain: The summary is readable, explains unhobbling, and AGI is defined beside the reading, but it flattens the author's force. The body drops the 2027 timeframe and AGI framing, which survive only in the title. It softens 'basically all cognitive jobs (think: all jobs that could be done remotely)' to 'automating cognitive jobs', and omits his stated conviction ('We are on course for AGI by 2027'). Smallest repair: in the summary, state that he forecasts this jump by the end of 2027, and restore his scope of 'basically all cognitive jobs', meaning remote-doable jobs.
We are on course for AGI by 2027. These AI systems will basically be able to automate basically all cognitive jobs (think: all jobs that could be done remotely).
date — pass: The stated date, May 29, 2024, matches the UTC calendar day of originalPublished (2024-05-29T21:07:53Z). The evidence is a publisher page rather than an archive copy, so no archive disclosure is needed. The retrieval date is correctly separated from the publication date, forecasts are kept as forecasts, and there is no hindsight framing.
support — uncertain: Most claims are directly supported. The final claim says Aschenbrenner describes job automation 'as a possibility rather than an accomplished outcome'. That correctly marks it as a forecast, but the source states it with conviction ('will basically be able to') while also noting large error bars elsewhere. Calling it merely 'a possibility' slightly mischaracterizes his framing. Repair: say he forecasts it, and keep his separate caveat that the error bars are large.
To be clear—the error bars are large. Progress could stall as we run out of data
media — pass: No media is selected and no verified candidates were supplied, so there is no media identity or relevance issue.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.