OpenAI's Astra decimates specialized robotics models, sparking scale-transfer debate
willdepue · x · 2026-11-01
OpenAI's willdepue says Astra's robotics result "decimated" specialized robotics models — a deeply surprising outcome with broader ML implications. He speculates it implies transfer is increasing with scale, or some fundamental algorithmic shift, while admitting he doesn't know which.
The accompanying diagram illustrates three possible mechanisms in task-space/capability terms: increased scale lifting capabilities broadly across tasks, reinforcement learning sharpening peaks on specific tasks, or "improved intelligence" raising the whole capability curve — i.e., general capability covering narrow tasks.
Related event: OpenAI's Will Depue Says General Model Astra Beats Specialized Robot Models(6 posts)→
More from Embodied
- Karvonen: frontier LLMs may end up directing specialized robotics models — adam_karvonen · 2026-11-01
- TransfHAR at UIST 2026: self-supervised wrist motions enable on-watch custom activity recognition — realkaranahuja · 2026-11-01
- LLM scores 19/20 zero-shot on robot gripper task, beating specialized model's 1/20 — willdepue · 2026-11-01
- Ronie: free open-source AI agent that makes Ray-Ban Meta glasses actually smart — Odd-Sun6296 · 2026-11-01
- Robot team claims world's first skateboard kickflip, code and paper coming — Jan_R_Peters · 2026-11-01
- He built a working 3D-printed RC skid steer with Claude Code — 163 prompts, 493M tokens, 16 days — CheapWelder4303 · 2026-11-01