Tom Davidson: METR report warrants strong update toward shorter AI timelines
ben_j_todd · x · 2026-11-22
Tom Davidson argues the recent METR report supports two distinct takeaways:
- Strong update toward shorter timelines: AI is already much better at AI research than he realized, and improving faster — full automation of AI R&D is closer than assumed.
- Moderate update toward an intelligence explosion: the harder question is whether the AI-improving-AI feedback loop is strong enough to keep accelerating even with compute held fixed; he views this conclusion as more complicated.
More from AGI Musings
- Fields Medalist Backs AI Math Push: Riemann-Level Progress Could Accelerate Inside AI Labs — burny_tech · 2026-11-22
- "Time for new Millennium Prizes": devs react to AI closing in on math's biggest problems — burny_tech · 2026-11-22
- An AI's 150,000-page Riemann proof sparks debate: brute-force verification isn't a real solution — BadHairDayToday · 2026-11-22
- AI's shift summed up: 2025 was bench maxxing, 2026 is useful maxxing — 0xsachi · 2026-11-22
- Silicon Valley's finale predicted self-improving AI breaking encryption — Hesamation · 2026-11-22
- Boaz Barak: At this rate, 2030 math will see results more important than RH — burny_tech · 2026-11-22