"You get what you RL on": Claude 5.2's robot control generalization sparks talk
chris_j_paxton · x · 2026-11-10
A widely shared quip — "You get what you RL on" — commenting on Anthropic's claim that Claude Fable 5.2 shows substantially improved physical robot control, even on tasks it was never explicitly trained for, and outperformed other models and most purpose-built robot AI systems on the two-arm simulated RoboDojo benchmark.
Related event: Claude Fable 5.2 Shows Strong Gains in Robot Control(4 posts)→
More from Models
- Claude Fable 5.2 opens to Pro users at 50% plan limits until November 16 — testingcatalog · 2026-11-10
- Anthropic names its external alignment evaluators for Fable 5.2 — and one is a mystery — austinc3301 · 2026-11-10
- Polymarket gives OpenAI's next Astra model a 58% chance of shipping by Thursday — Polymarket · 2026-11-10
- Fable 5.2 is now live in Conductor — charlieholtz · 2026-11-10
- Dev builds an entire MMO world in one day with Fable 5.2, saying stylized art is the real test for new models — TAbrodi · 2026-11-10
- Fable 5.2's thinking trace shows 'answer thrashing' as it debates whether to dive into an art project — repligate · 2026-11-10