METR probe finds Anthropic's capabilities progress 2x faster; 'dramatic acceleration' may already be crossed
RyanGreenblatt · x · 2026-11-07
Ajeya Cotra's post analyzes a METR deep-access investigation of Anthropic, which Ryan Greenblatt helped model. Key points:
- Anthropic's S1 says it "may imminently meet" its RSP threshold of "dramatically accelerated progress due to AI": overall capabilities moving 2x faster than the fastest sustained pre-AI rate;
- METR found that from Oct 2025 to Oct 2026, Anthropic's capabilities progress (measured by an internal Epoch Capabilities Index, AECI) ran about twice as fast as the prior twelve months;
- The acceleration is plausibly attributable to AI automating AI R&D rather than compute;
- Most supporting evidence was redacted but described as the majority of the most informative findings;
- Naive extrapolation of compounding acceleration implies superintelligence within a year — the author stresses this is extrapolation, not prediction.
Related event: METR Report Finds Anthropic's Research Progress Doubling Yearly(2 posts)→
More from AGI Musings
- Stanford's Anshul Kundaje: ML conference peer review is broken and unfixable — anshulkundaje · 2026-11-07
- Eric Drexler: AI options, not optimism — participants weigh options, not odds — sebkrier · 2026-11-07
- Musk lays out SpaceX vision of deploying up to 1 TW of space-based AI compute per year — XFreeze · 2026-11-07
- "AI media detection will become an arms race just like cybersecurity" — ZeroStateReflex · 2026-11-07
- tszzl: Nature destroyed itself five times — stop reading morality into it — tszzl · 2026-11-07
- Vitalik Buterin: 'Open weight is not open source' on AI models — wonderwomancode · 2026-11-07