Distilling a frontier model into 4B params lifts robot arm success from 0 to 22.5%
ZeYanjie · x · 2026-11-17
Developer @ishanxnagpal distilled a large model (claimed as "GPT 6 Astra", unverified) into a tiny 4B parameter model to drive a robot arm. Before distillation, the small model couldn't finish a single task — at one point it decided the bowl right in front of it didn't exist. After, it went from 0 to 22.5% success on scenes it was never trained on. The author argues the future of robotics is agentic, with a thread promising more details.
More from Embodied
- CoRL fireside chat: Fei-Fei Li and AMD CEO Lisa Su on stage together — PeterStone_TX · 2026-11-17
- NeurIPS 2026 surgical benchmark: top multimodal models score under 5% zero-shot — maier_ak · 2026-11-17
- Musk retweets: Tesla Giga Berlin to build Cybercab — elonmusk · 2026-11-17
- Behavior Graphs turns raw robot policy rollouts into interpretable graphs — drmapavone · 2026-11-17
- Neuralink hackathon adds hands-free speech: iPhone calls, lock-screen controls, and a 'Hey Neuralink' Grok agent — XFreeze · 2026-11-17
- Six frontier models drive real robot arms into hazards; LEASH guardrail cuts harm to zero — 新智元 · 2026-11-17