METR estimates AI now speeds up AI research ~4.5x, oversight window is now
sebpaquet · x · 2026-11-07
A thread arguing you can use AI daily and still want it regulated. The quoted post cites METR's independent review: AI is accelerating its own progress 2x overall, with Ryan Greenblatt estimating 4.5x acceleration of AI software research — "a big update towards ASI within 1–2 years." The author argues a ban is impossible and would hand the frontier to whoever cares least, and that since progress can now come from software alone — much of it written by AI itself — hardware limits are no longer a reliable brake. Oversight is urgent, now, while it still matters. The retweeted take adds the irony that those closest to the technology, often its biggest believers, are the ones most alarmed.
More from AGI Musings
- Space compute advocate says data center protests reveal opponents' true motive — Aiden_Tech_Ai · 2026-11-08
- Snapshot of agent use cases today: canceling subscriptions and spamming support — tekbog · 2026-11-08
- Hassabis: recursive self-improvement is 'very dangerous' path to alien AGI — Dr_Atoosa · 2026-11-08
- If compute moves to space, will data center critics still protest progress itself? — Dr_Singularity · 2026-11-08
- Dev pushes back on AI-replacement thesis: software and math work keeps growing — timigod · 2026-11-08
- MIT CSAIL shares roundup of 15 neural network courses, topped by Andrew Ng's 4.9/5 intro — MIT_CSAIL · 2026-11-08