
ChatGPT Aces Pediatric Diagnosis at a Rate of 17 Percent, Which Is Not a Bedside Manner
Ourtake
Getting 83 wrong out of 100 is a rough score for a tool that sounds sure of itself.
What happened
Researchers tested ChatGPT on 100 pediatric medical cases and scored its answers with two physician-researchers. It gave the correct diagnosis in 17 cases, was plainly wrong in 72, and partly right in 11. The researchers suggested that training on trustworthy medical literature could improve accuracy.