Have AI timelines suddenly moved?

Why it matters

A METR report on AI R&D acceleration at Anthropic, followed by responses from Ajeya Cotra, the AI Futures team, and Buck Shlegeris of Redwood Research, suggests Anthropic's internal capabilities growth has sped up since late last year, with the gains largely credited to AI building AI. Epoch AI capability indexes show a clear trend shift around October of last year, coinciding with a major Opus release that many saw as making AI coding agents the default. Crucially, the overall frontier has not accelerated: competing labs like OpenAI hold a stable pace, and Anthropic merely leapfrogs rivals rather than breaking the long-run trend. Anthropic attributes the change heavily to AI-driven gains, and METR assumes most of the acceleration comes from AI building AI, but external observers report far smaller AI-driven productivity gains, and Anthropic's pace did not visibly rise afterward despite much greater internal AI use. The evidence warrants genuine skepticism.