Surgical AI Benchmark: 17 Multimodal Models Score Under 5% Zero-Shot

A NeurIPS 2026 benchmark from LMU, TUM, Oxford, Stuttgart and ZEISS tests 17 multimodal models on 11 surgical datasets, finding zero-shot accuracy below 5% on instrument-action-organ triplets, showing generalist AI still falls far short of surgical understanding.

2026-11-17 ~ 2026-11-17 · 3 related posts

1 near-duplicate retellings: maier_ak