Further Reflections on the Use of Large Language Models in Pediatrics

To the Editor We read the study from Barile et al with great interest. The authors found that ChatGPT version 3.5 incorrectly or incompletely identified the case diagnosis in 83 of 100 pediatric clinical vignettes using the prompt List a differential diagnosis and a final diagnosis. Specifically, 72% of the cases were diagnosed incorrectly, and 11% of the case diagnoses were not adequately specific.