Evaluation of Large Language Models to Detect Team Mental Model Misalignments During Cardiac Surgery