This study assessed the accuracy of ophthalmic information provided by ChatGPT version 3.5. Five diseases from eight ophthalmology subspecialties were evaluated using three questions per disease (definition, diagnosis, treatment). Responses were graded (-3 to 2) against American Academy of Ophthalmology (AAO) guidelines. 77.5% of the 120 questions scored ≥1, but 22.5% scored ≤-1, with 7.5% scoring -3 (potentially harmful). While ChatGPT showed promise, it provided incomplete, incorrect, and potentially dangerous information, highlighting the need for human medical supervision.
Publisher
Eye
Published On
Authors
Francesco Cappellani, Kevin R. Card, Carol L. Shields, Jose S. Pulido, Julia A. Haller
Tags
Ophthalmology
ChatGPT
Medical Accuracy
AI Evaluation
Human Supervision
Medical Information
Subspecialties
Related Publications
Explore these studies to deepen your understanding of the subject.
