Accuracy of a generative artificial intelligence model in a complex diagnostic challenge.

Generative AI (Gen AI) may help clinicians consider difficult diagnoses, but important limitations remain. In this research letter, the Generative Pre-trained Transformer 4 (GPT-4) model included the correct diagnosis in 64% of differential lists and ranked it first in 39% of challenging cases, performing similarly to existing diagnostic support tools. Limited transparency and real-world validation underscore the need to study bias, knowledge gaps, and clinical usefulness before widespread Gen AI adoption.