AI Models Struggle with Slight Medical Question Variations, Experts Seek Solutions

A study reveals that large language models, despite high scores on medical exams, rely heavily on pattern recognition rather than true reasoning, as their performance drops significantly when answer options are subtly altered, raising concerns about their reliability in real clinical settings.