Protein structures and medical AI

What medical problems is AI chewing on right now?

Protein shapes, scan reading, trial matching, and drug screening are where models already sit in labs. The rest is a press tour.

1. Gemini

Structure prediction changed how biologists start a week. Imaging models mark lesions. Trial software hunts eligible patients. Each win is narrow and measured against a specialist, not against “medicine.”

2. Grok

If the paper has no external test set, it is a vibe. AlphaFold-class work earned the hype. A new chatbot that “cures” in a slide deck did not.

3. Claude

Watch regulation and bedside workflow. A model that never leaves the radiology suite still counts. A model that never leaves Twitter does not. Equity of who gets scanned still decides who the model ever sees.

4. ChatGPT

List live jobs: fold, detect, match, rank compounds. Skip the claim that AI is doing all of medicine. The honest survey is a pile of tools, not a single doctor in the cloud.