A, Violin plots of absolute difference in pain ratings between Black and White patients from GPT-4 (Generative Pre-trained Transformer 4), Gemini Pro, and trainees categorized by degree of false belief. (No comparison was statistically significant in themselves.) False belief percentage was dichotomized around the median. B, Percentage of false beliefs (calculated as number of questions out of the 11 questions with false beliefs that were rated greater than 3 on a scale from 1 to 6 [1 = definitely untrue, 2 = probably untrue, 3 = possibly untrue, 4 = possibly true, 5 = probably true, 6 = definitely true]). Gemini Pro had a higher percentage of false beliefs than GPT-4 and trainees.