OpenAI's o1 beat ER doctors at triage. 67% vs 50-55%

Harvard trial: o1 correctly diagnosed 67% of emergency-room patients from triage notes alone. Triage doctors hit 50-55%. The catch: the model also confidently wrong-diagnoses 1 in 8. Net: helpful as a check, dangerous as a replacement.

Guardian: theguardian.com/technol…