OpenAI's o1 outperforms ER doctors in diagnosis accuracy—but deployment remains uncertain

A Harvard trial found OpenAI's o1 model correctly diagnosed 67% of emergency room cases versus 50–55% for triage doctors, suggesting AI could catch missed diagnoses at scale. For indie makers in healthtech, the result highlights both the opportunity and the regulatory minefield: raw accuracy doesn't translate to deployment without validation, liability frameworks, and buy-in from risk-averse institutions.