Surgical agents improve seeing with tools, not judging

maier_ak · x · 2026-11-17

In the same thread: turning the models into agents with a segmentation tool and a curated surgical library improves instrument and action recognition, but safety checklist and skill ratings do not improve, and extra evidence sometimes misleads.

Related event: NeurIPS 2026 Surgical Benchmark: 17 Multimodal Models Fall Short in Zero-Shot Surgery Tasks(5 posts)→

Original post →

More from Research

Research channel →