METR report suggests AI is already speeding up R&D 2x, a big update toward 2027 AGI timelines
birchlse · x · 2026-11-06
Key points
- Prominent AI safety researchers (Ryan Greenblatt, Eli Lifland, austinc3301) say METR's report on AI R&D automation (RSI) is a major update toward short timelines: a reported 2x speedup in benchmark progress maps to a 3-5x multiplier under the AI 2027 model, modally pulling a key milestone forward to March 2027.
- Greenblatt admits he was seriously wrong: he previously expected only 1.15-1.3x RSI speedup by 2026, and notes private evidence existed that would have changed his mind.
- Caveat: the public report redacts most underlying evidence, so conclusions beyond the observed 2x depend heavily on trust in Anthropic. The posters express frustration that Anthropic paired heavy redactions with no meaningful progress on pacing interventions.
Related event: METR Assessment of Anthropic's AI R&D Acceleration Shakes Community(83 posts)→
More from AGI Musings
- METR publishes risk assessment on whether AI is accelerating AI R&D — GaryMarcus · 2026-11-07
- Benedict Evans: Big tech only wins when a new business extends what it already does — Aiden_Tech_Ai · 2026-11-07
- 'You just have to believe in straight lines': the Opus 4.5 take-off meme — scaling01 · 2026-11-07
- Creator dilemma in the AI age: impostor syndrome and the dilution of creative work — sull · 2026-11-07
- "If we started today, we wouldn't have nearly as many people": AI-era headcount reflection goes viral — bigdata · 2026-11-07
- Design engineer automates his own $56K/yr drafting job with an AI drawing tool — luisdans · 2026-11-07