Stanford trial says GPT-4 beat doctors when used alone, but hurt non-experts
jonc101x · x · 2026-07-29
A Stanford doctor’s trial found GPT-4 could outperform physicians in some tasks
Dr. Jonathan Chen, who leads AI education for medical students at Stanford, reportedly ran a study on real doctors and found a striking pattern:
- Doctors with internet access performed well.
- Doctors using GPT-4 performed about the same.
- The model on its own beat every doctor in the study, including those who had it open in front of them.
- Non-medical participants did worse with the model than when they simply guessed.
He also says he replayed a real end-of-life conversation he had with a patient’s wife and concluded the chatbot handled the counseling better than he did. When he showed that transcript to 300 physicians, they applauded — and he had to remind them they were applauding a chatbot.
More from AGI Musings
- Asimov on Intelligence: High IQ doesn't equal true smarts; the auto-repair man's wisdom — burkov · 2026-07-30
- Musk's Real Claim: Optimus Isn't Just a Product, It Sells Human Hours Back — r0ck3t23 · 2026-07-30
- Lab-Grown Diamonds Threaten Botswana's Economic Miracle: From Diamond Boom to Technological Obsolescence — KarlMuth · 2026-07-30
- Compute Surge: 10 Major Scientific Breakthroughs AI Could Unlock by 2028 — Annual_Judge_7272 · 2026-07-30
- Combating Bio-Vulnerability: Will Mind Uploading Be the Ultimate Solution? — yeastsplainer · 2026-07-30
- AI Copyright Moats Fail, Prompting Shift to Government Protection — RexDouglass · 2026-07-30