A Benchmark That Catches AI Blind Spots in the Lab

Imagine asking a brilliant art critic to describe a microscopy image — they're great with paintings, but they've never seen a cell under magnification. That's where today's AI models stumble: they ace natural photos but flunk the specialized visual language of working science.

This means the AI tools reshaping biotech need to actually *see* what scientists see — and right now, they don't, no matter how impressive they sound in press releases.