OpenAI's o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors | Hacker News

A Hacker News thread questioned a study claiming OpenAI’s o1 diagnosed 67% of ER cases correctly, versus 50–55% for triage doctors.

Commenters argued the result may depend heavily on benchmark design and may not reflect real-world clinical practice.

Many noted that medicine is more than pattern recognition, involving judgment, context, and patient interaction.

The discussion also weighed whether specialized medical AI could eventually beat humans in narrow diagnostic tasks.

Overall, the thread highlighted both the promise and the risks of deploying AI in healthcare.