Abstract

Aim: To assess the efficacy of ChatGPT in delivering precise insights into amblyopia queries.
Methods: Answers to 14 common amblyopia-related queries were generated twice from ChatGPT. Responses were evaluated by 5 ophthalmologists, and the accuracy and quality of responses were rated on a Likert scale. The median and range were estimated and compared. Agreement between the five evaluators and reliability were estimated. Results: Very high-quality answers were delivered for 3 questions (median 4.5), adequate for 9 (median 4). and unsatisfactory for 2 (median 3.0). Out of 140 evaluations, 12 had very good ratings, 91 were good, 28 were acceptable, and 9 were poor. Cronbach's α of 0.810 shows a strong agreement. Mean scores in the second and first sessions were 4.20.5 and 4.10.6, respectively, indicating a potential learning effect. Conclusion: ChatGPT provides reliable information. Inaccuracies highlight the need for thorough evaluation and understanding of its limitations.

Full Text