The adoption, efficacy, and limitations of large language models in urological education

The integration of large language models into medical education is rapidly transforming pedagogical approaches, particularly in specialized fields like urology. This study aims to evaluate the adoption patterns, perceived efficacy, and significant limitations of these artificial intelligence (AI) tools among urology trainees. This research employed a descriptive, cross-sectional study design through an online survey disseminated to urology trainees at a single institution. A total of 194 complete responses were collected from medical students and interns. The survey assessed adoption patterns, perceived efficacy on a 7-point Likert-scale, and limitations. The analysis revealed near-universal adoption (98.4%), with 65% of participants using AI tools daily, predominantly ChatGPT. Key applications included generating exam-style questions and summarizing complex guidelines. Participants reported significant perceived improvements in core knowledge areas, with the highest gains in disease knowledge and diagnostics (Median = 7, interquartile range: 6-7), which were strongly correlated with increased clinical confidence. Multivariate regression identified usage frequency and comfort with technology as significant predictors of knowledge improvement (β = 0.29, P = .001) and confidence gains (β = 0.30, P < .001). The primary limitation identified was the potential for AI "hallucinations" or incorrect information (54.1%). Large language models are widely adopted and perceived as highly effective for enhancing knowledge and clinical reasoning in urological education. However, their transformative potential is tempered by critical concerns regarding reliability. A balanced approach that combines these powerful AI tools with vigilant oversight is essential.