GPT-6 Astra does robotics ICL out of the box — but GPT-3.5 already could, paper notes
philfung · x · 2026-09-11
A researcher showed GPT-6 Astra performs physical in-context learning with robots without training: feed a recording of a human doing a novel task, prompt it to drive a robot arm the same way, and it works on the first pass.
Skeptics point to Berkeley's RoboPrompt paper ('In-Context Learning Enables Robot Action Prediction in LLMs'), showing even GPT-3.5 did ICL out of the box: RoboPrompt lets text-only LLMs predict robot actions with no training by selecting keyframes, converting end-effector actions and object poses into text, and building structured ICL prompts — outperforming zero-shot and ICL baselines in both simulation and the real world.
Related event: GPT-6 Astra learns robot tasks from a single human video(2 posts)→
More from Embodied
- Stanford teams complete 100+ functional neurosurgery procedures with imaging and robotics since Dec 2024 — drjfm_stanford · 2026-09-11
- Tomo Voice preview lets you call your AI companion, sometimes it calls you first — Jackyhuang · 2026-09-11
- Apple's Live Rewind Watch feature instantly gets the Black Mirror comparison — IgorCarron · 2026-09-11
- GPT-6 Astra solves a Rubik's Cube with robot hands — ZeYanjie · 2026-09-11
- Unitree's Playbook: Massive Scale, Low Costs, and R&D Only Where It Cuts Prices — Afinetheorem · 2026-09-11
- YC Demo Day: Live Drone Tracking Used by Ukraine's Military and Self-Improving Voice Agents — ycombinator · 2026-09-11