Robotics is approaching its GPT-2 moment as pretraining and world models kick in
nathanbenaich · x · 2026-10-08
Nathan Benaich argues robotics is approaching its GPT-2 moment (some would say GPT-3): broader pretraining is helping robots generalize to unfamiliar tasks (Skild AI, Generalist AI, Sunday Robotics), while world models let robots simulate what might happen next and plan before acting in the real world (Odyssey, Wayve).
The quoted context matters too: AI is helping build AI — Anthropic's internal index says Claude led 26% of measured model R&D in August with humans supervising, and agent usage in knowledge work is climbing fast. What he's watching next: scientific taste — agents choosing worthwhile, elegant experiments and knowing when to abandon unfruitful approaches.
More from AGI Musings
- Four moves for 2026: self-host your AI, audit its output, keep skills sharp, set a family code word — alex_verem · 2026-10-08
- LeCun: automated formal proofs open a new era for math; critics say question taste can't automate — AryHHAry · 2026-10-08
- Émile Torres Publishes 6,100-Word Essay Deconstructing Silicon Valley's TESCREAL ASI Worldview — mjdramstead · 2026-10-08
- Ethereum researcher Justin Drake warns AI could crack Bitcoin private keys in months — jamestagg · 2026-10-08
- AI safety circles spar over whether AI exploits stay within the convex hull — QuintinPope5 · 2026-10-08
- Dev compares vibe coding to buying books you never read: 15-min PoCs, zero learning — MarcJSchmidt · 2026-10-08