GPT-6.1 Sol halves surgical positioning error after 3 mouse brain surgeries
141_1337 · reddit · 2026-11-06
Gabriel Ocana-Santero's team let GPT-6.1 Sol perform 3 successive mouse brain surgeries, with an expert surgeon correcting its landmark localization (bregma/lambda) after each attempt — in-context learning from real-world experience. By the third surgery: landmark search time dropped 169→86 sec, corrective moves 4→2, and positioning error halved 0.6→0.3 mm. N=3, but striking early results.
More from Embodied
- Nearly 100 testers praise Sphenomenon MR demo that turns a whole house into voxels — Scobleizer · 2026-11-06
- Spectral Skills lands on real robots: composing motion skills via energy functions — daibond_alpha · 2026-11-06
- Developer gives Claude Desktop a body and sensors via MCP and Raspberry Pi — jfrankie · 2026-11-06
- Travis Kalanick's new firm Atoms builds AI and robots for food, mining, transport — davidsenra · 2026-11-06
- Verne V0.5 trains a robot policy with video prediction, but runs actions-only — rohanpaul_ai · 2026-11-06
- Robot success rate doubles 23%→51% from pretraining data alone, no fine-tuning — rohanpaul_ai · 2026-11-06