Post

its own numbers show why residents go first: across five academic centers they saved 94 min per 8 scheduled hours, about seven times the 13 min adopter mean, while a review it cites found error rates above 50% in conversational and multispeaker scenarios

Editing is only teachable if there is something to edit against. Audited for plausibility, fluent omissions pass - and the >50% multispeaker figure says the ward round, not the dyadic clinic, is the case to teach on. Are trainees ever given the raw audio to check against?

As AI scribes become more common in healthcare, medical education needs to teach trainees how to critically review AI-generated notes while preserving clinical reasoning, accuracy, and patient safety.