Frontier LLMs solved robot in-context learning as a side effect, says researcher

YouJiacheng · x · 2026-11-22

YouJiacheng reshares a robot in-context learning demo: in August it was hailed as the first time ICL worked for robots, but three months later frontier LLMs solved it as a side effect of general intelligence, without any robotics training. He argues instruction understanding, planning, ICL, memory and long-horizon skills should all live in LLMs rather than dedicated VLA/world models, calling it the fastest he's seen a research direction age. Also references OpenAI shutting down its robotics team as general models proved sufficient.

Original post →

More from AGI Musings

AGI Musings channel →