A short, hands-on tutorial built around a real H&P with 22 deliberate errors — the kind of mistakes a tired intern, or a confident model, will quietly make at 2 a.m.
You already know how to write an H&P. This teaches you how to supervise a model writing one — what to trust, what to catch, what to throw out.
How LLMs work — tokens, context, hallucination — and how to prompt them like you'd sign out to a colleague.
Find the planted errors. Compare what different models and different prompts catch — and what they miss. 15 minutes of close reading you'll actually remember.
What just happened, why models fail in predictable ways, and how to choose one for the task in front of you.
Most learners skip the reading. Don't — the prompting chapter is what makes the exercise work.
The model landscape will change again next quarter. The way you supervise it shouldn't have to.
Tell the model who it is, who you are, and what counts as a good answer — before you paste anything.
Verify numbers, dates, doses, and pronouns. Treat fluency as suspicious, not reassuring.
Match latency, reasoning depth, and privacy posture to the clinical question — not to the brand on the tab.