LLMs fail 80%+ of differential diagnoses when patient data is incomplete, JAMA study finds
Mass General Brigham evaluation of 21 leading AI models finds failure rates above 80% at the open-ended start of clinical reasoning, with NHS-relevant implications.
