Anthropic's capability acceleration hit 1.9x the earlier rate since Oct 2025, per METR
Miles_Brundage · x · 2026-11-07
METR Evals' AECI analysis shows that since October 2025, AECI scores for Anthropic's models have grown at 1.9x the earlier rate (95% CI: 1.5–2.8). METR cautions that analysis choices affect this estimate, and how benchmark scores map to real-world capabilities remains unclear. AI safety researcher Miles Brundage quote-posted it with a "very hard to sleep" image, signaling concern about the acceleration.
More from AGI Musings
- AI Exposes the Economy's Inefficiency: The Real Threat Is Getting Paid for Mediocrity — TinfoilTricorn · 2026-11-07
- basedjensen: recursive self-improvement is already here and bans are unenforceable — basedjensen · 2026-11-07
- Musing: Much of ML's progress may just be rediscovering good discipline — yunta_tsai · 2026-11-07
- Geoffrey Hinton's 47-minute AI masterclass promises to reshape how you think about AI — HeyAmit_ · 2026-11-07
- AI video characters may feel conscious sooner than LLMs, argues viral post — flowersslop · 2026-11-07
- WIRED archive analysis: 60.5% of AI stories use harm framing vs 18.6% in early internet era — emollick · 2026-11-07