Astral Codex Ten

AI 2027

7 min read

AI 2027

Article

AI 2027 is a recurring event in the Astral Codex Ten archive, appearing 5 times across 5 issues between April 08, 2025 and February 12, 2026. The archive places it in contexts such as “AI 2027 objects to the first bottleneck”; “AI Futures Project is the group behind AI 2027”; “AI-2027’s specific predictions for August 2025 appear to have happened”. It most often appears alongside OpenAI, ACX, China.

Metadata

  • Category: Events
  • Mention count: 5
  • Issue count: 5
  • First seen: April 08, 2025
  • Last seen: February 12, 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.

My Takeaways From AI 2027
April 08, 2025 · Original source
AI 2027 objects to the first bottleneck: smarter researchers can use compute more efficiently. In fact, we know this is happening; about half of all AI scaling since 2020 has been algorithmic progress, where we get better at using the compute we have. If we hold compute constant, but get 10x algorithmic progress (because of the intelligence explosion), then we get 5x overall AI improvement.
AI 2027 disagrees. Although the counter-objection is directionally correct, there are little ways intelligence can boost speed even when compute is held constant. How do we know? Partly through armchair attempts to enumerate possibilities - for example, even if you can’t speed up by adding more researchers, surely giving the same researchers higher serial speed has to count for something. And partly because we surveyed AI researchers and asked “if you had a bunch of AIs helping you but only the same amount of compute, how much faster would your research go?” and they mostly said somewhat faster. All these little boosts will compound on themselves in typical intelligence-explosion fashion, and when you game it out, you get a one-year-or-so takeoff to superintelligence.
Inline links: disagrees
When real-world researchers debate whether or not to implement neuralese, we hope they think “Hey, isn’t this the decision that doomed humanity in that AI 2027 thing?”
AI Futures: Blogging And AMA
April 24, 2025 · Original source
AI Futures Project is the group behind AI 2027. I’ve been helping them with their blog. Posts written or co-written by me include:
Inline links: AI 2027, their blog
AI 2027: Media, Reactions, Criticism - a look at some of the response to AI 2027, with links to some of the best objections and the team’s responses.
Inline links: AI 2027: Media, Reactions, Criticism
Curvy green dotted line is AI 2027's prediction; straight black dotted line is METR's measured seven month doubling time. This isn’t meant to imply that METR didn't also consider a superexponential trend, it’s just not the headline result in their paper. And speaking of expertise, the AIFP team have kindly volunteered to do an AMA (“ask me anything”, Q&A) here on ACX, this Friday, 3:30 - 6:00 PM California time. If you have any questions on the scenario, AI forecasting, or AI safety more generally, they can give you high-quality answers. I’ll make a separate post at the appointed time.
Links For October 2025
October 30, 2025 · Original source
6: Related: Checking In On AI 2027. “AI-2027’s specific predictions for August 2025 appear to have happened in September of 2025. The predictions were accurate, if a tad late, but they are late by weeks, not months.” But the early predictions were mostly straightforward extrapolation of benchmark improvements, with the later ones depending on a more controversial theory of recursive self-improvement, so the success of the early predictions doesn’t necessarily say much about the later ones. Related (X): OpenAI sets an “internal goal” of having an “automated AI research intern” and “true automated AI researcher” on approximately the AI 2027 timeline.
Inline links: Checking In On AI 2027, Related (X)
Open Thread 413
December 22, 2025 · Original source
1: Another charity fundraiser, this one for Lightcone Infrastructure. Lightcone is the group that does the hard work for many of the rationalist community resources you enjoy. You probably know them from the Less Wrong website and the Lighthaven campus. But did you know they also designed the websites for AI 2027, for Eliezer and Nate’s book, for AI Lab Watch, and (for some reason) for Deciding To Win, a renegade faction of Democrats who believe that, instead of supporting unpopular policies and losing, the party should support popular policies and win? And on the side, they play a big role in hosting ACX meetups, including letting us use their campus (if you’ve ever been to our Berkeley meetup location, that was them). They’re a rare intersection between “support effective altruist charities” and “support pillars of your your local community”. Donate here, or contact Oli if you have some kind of more complicated donation-related need.
Inline links: Lightcone Infrastructure, Less Wrong, Lighthaven, AI 2027, Eliezer and Nate’s book, AI Lab Watch, Deciding To Win, here
What Happened With Bio Anchors?
February 12, 2026 · Original source
AI 2027’s forecast for early 2026 (source).
Inline links: source
But its headline prediction - an AGI timeline centered around the 2050s - no longer seems plausible. The current state of the discussion ranges from late 2020s to 2040s, with more remote dates relegated to those who expect the current paradigm to prove ultimately fruitless - the opposite of Ajeya’s assumptions. Cotra later shortened her own timelines to 2040 (as of 2022) and they are probably even shorter now.
Inline links: current state, 2020s, 2040s, as of 2022
I think internalizing this lesson is more important than any sort of micro-calibrating exactly how much to believe in probabilistic forecasts. Once you understand that you can’t always just rely on your biases and sense that it would be inconvenient for things to get weird, you become desperate for real information. That desperation encourages you to seek any possible source of knowledge, including potentially fallible and error-laden probabilistic forecasts. It also encourages you to treat them lightly, as small updates useful for resolving near-total uncertainty into merely partial uncertainty. This is how I treat Bio Anchors’ successors - although right now a little more fallibility and error-ladenness might be genuinely welcome. AI 2027’s forecast for early 2026 (source).
Inline links: https://ai-2027.com/