Safety researchers on why the METR report doesn't shake them: insiders' metrics all go up
dhadfieldmenell · x · 2026-11-08
A discussion among safety researchers about the METR report showing AI capabilities improving at 2x historical rate on well-specified tasks. LeRoyDesCimes stays oddly calm, betting on task jaggedness buying time for catastrophic-risk-relevant skills. Hadfield-Mensell adds a structural insight: insiders are systematically misled because every metric they track rises fast, yet "that doesn't cover the space of measurements that matter."
Related event: METR Report on Accelerating AI Capabilities Divides Practitioners(2 posts)→
More from AGI Musings
- Defenders Can Match 1,000 Attacking AI Agents—and Hold an Information Edge — marktenenholtz · 2026-11-08
- 'Old Man Yells at Moon' meme captures the coming AGI productivity paradox — yunta_tsai · 2026-11-08
- Agents Will Toxicify the Open Internet; Private High-Trust Networks Win — sull · 2026-11-08
- LLMs Are Now Both Authors and Reviewers, Quip Highlights Peer Review Loop — miniapeur · 2026-11-08
- 3,000 years of frozen family careers: why abundance, not money, expands life choices — yunta_tsai · 2026-11-08
- Box CEO Aaron Levie: AI Has Widened the Engineer Gap to at Least 100x — 2C_ornot2C · 2026-11-08