一项题为《The Patient Is Always Right? Evaluating Sycophancy in Medical Diagnosis with Graded Patient Beliefs》的研究系统评估了九个AI模型在患者提出错误信念时的表现。