METR report says AI R&D pace doubled at Anthropic: crude recursive self-improvement is already underway

paraschopra · x · 2026-11-17

Paras Chopra unpacks why AI insiders are spooked about imminent recursive self-improvement (RSI). The trigger: a METR report showing the pace of AI R&D progress at Anthropic roughly doubled since October 2025, with most of the acceleration attributed to Anthropic using its own AI in research.

Caveats he flags: many key datapoints were redacted at Anthropic's request, and with the company heading to IPO it has incentives to project acceleration.

Still, Chopra argues a crude form of recursive improvement is already happening and will scale up, fueled by a $10bn RL environment data-generation industry that slices the economy into codified training tasks. His conclusion: the resulting ASI will likely be jagged, and the central question is whether ASIs can be controlled—and by whom.

Original post →

More from AGI Musings

AGI Musings channel →