Performance of the Artificial Intelligence large language models ChatGPT 3.5, Gemini (Google Bard), ChatGPT 4.0, and Gemini 2.5 flash in surgical subspecialty questions of Brazilian medical residency exams

Key Points

• AI models like ChatGPT and Gemini show promise in enhancing medical education, especially in surgical subspecialties.
• While these models perform well in exam scenarios, they do not replace the need for clinical decision-making skills taught in medical schools.
• The assessment included prominent models capable of answering surgical questions relevant to Brazilian residency exams.
• The integration of AI in medical training could reshape educational methods, though caution is needed regarding their limitations.

Abstract