Tag
This study compares zero-shot prompting of LLMs against supervised baselines for detecting shared decision-making in pediatric clinical encounters, finding that supervised models outperform zero-shot approaches and highlighting critical data leakage issues in evaluation pipelines.