Most clinicians now use AI. Few have been trained on how it fails. Medical AI Competence teaches the skills that close that gap — how large language models work, where they break, and how to use them safely for your patients.
Fabrication isn't random. It's what a model does when you give it room to guess. The same model that invents a citation for a vague request will flag its own uncertainty when you ask it to. The fix isn't a better model — it's a better prompt.
"How do I treat preeclampsia at 28 weeks?"
Open-ended, no context, no rules. The model fills the gaps with whatever sounds right — including confident detail it cannot support.
"I'm caring for a patient with preeclampsia at 28 weeks of gestation. What are the evidence-based options for managing her, and what factors decide between expectant management and delivery? Use the Clinical Evidence Prompt (CEP): answer with reliable clinical evidence, identify missing clinical information, separate evidence from inference, state certainty, cite factual claims, and say clearly when evidence is limited or you do not know."
Adds it to any clinical question. It forces the model to mark what is evidence, what is inference, and where it isn't sure — the difference between a guess and an answer you can check.
Ready-made, copy-and-paste prompts that do exactly this: force the model to mark evidence, flag its uncertainty, and hand you real numbers. Free to save and share.
Ten modules, a 10-question assessed examination, and a certificate of completion at 70% or above. Designed for physicians, residents, nurses, midwives, and advanced practice providers. No technical background assumed. Free.
Start the course See the competenciesAlready know the basics? Go deeper with the full practicum →
Start with the foundation, go deep with the practicum, learn to build your own tools, and point your patients to the plain-language companion. Same evidence standards across all four.
Ten short modules on how large language models work, where they fail, and how to use them safely in care. A 10-question assessed examination and certificate. Free, ~60 minutes.
Start the foundation → Level 2 · PracticumThe hands-on deep course: 51 lessons across 9 modules plus AI setup, drilled on real, de-identified examples. A randomized final examination and a verified certificate. Free, ~3–4 hours.
Enter the practicum → Hands-on · BuildBuild your own clinical tools by directing AI — no coding required. Ten modules from mindset to a real tool you publish yourself, the rules that keep it safe, plus a 10-question assessment and certificate. Free, ~60 minutes.
Start building → For patients & the publicA plain-language companion on using AI wisely for your own health: what to ask, what to verify, what never to type into a chatbot. Built for everyone — share it with your patients.
Visit AIHealthCourse.com →AI is already writing notes, answering patient questions, and suggesting differentials. Most clinicians treat it like a search box — type a question, accept the answer. Competent use rests on two skills, in this order:
The quality of an AI answer is set before the model writes a word — by how you define the clinical question and engineer the context: the patient's situation, the decision point, the constraints, what you already know, and what you need back. This is the core craft of clinical AI use, and it occupies most of the curriculum: defining the question, choosing the right tool, framing context, and building stepwise workflows.
Even a well-framed question can return an answer that is confident, fluent, and wrong. The three failure modes that do most of the damage: fabrication — citations, doses, and studies that don't exist, formatted exactly like ones that do (you just saw it above); sycophancy — present a wrong premise with confidence and the model validates it; and stale confidence — a guideline replaced last year, summarized with today's certainty.
If the pace of AI feels impossible to keep up with, that's because it is. No clinician can track every model release, and none needs to. Feeling behind is nearly universal — including among the people who teach this. Competence is not consumption. It is a small, durable set of skills — framing the question, supplying the context, verifying the output — that holds steady while the models change underneath it. That is what this course teaches.
A sequenced curriculum — from how the technology actually works, through verification and failure recognition, to ethics, governance, and the standard of care. Each module is free, takes a few minutes, and ends in a real assessment.
Prediction, non-determinism, and why confident text is not truth.
→ 2Match AI use to the task: counseling, documentation, differentials, workflow.
→ 3Fit the model to the clinical job, privacy needs, and setting.
→ 4Check facts, references, doses, and calculations before clinical use.
→ 5Frame the patient context and decision point. Better prompts, stepwise workflows.
→ 6Detect hallucination, outdated guidance, bias, sycophancy, and overconfidence.
→ 7Let AI handle retrieval so you can focus on judgment and interpretation.
→ 8Readability, plain-language explanation, informed consent, and correcting misinformation.
→ 9Protect PHI, document responsibly, understand medicolegal risk.
→ 10Safe implementation, team oversight, and readiness for changing standards.
→New: the AI terms glossary. 110 plain-language definitions of the AI vocabulary clinicians keep tripping over: from hallucination and PHI to agentic testing, plus the features unique to Claude and ChatGPT. The companion reference to the course. Open the glossary →
Amos Grünebaum, MD, is Professor of Obstetrics & Gynecology and Maternal-Fetal Medicine at the Zucker School of Medicine at Hofstra/Northwell and Senior Ethics Consultant at Northwell Health. In more than fifty years of clinical practice he has delivered over 10,000 babies and published more than 175 peer-reviewed papers.
He was writing about large language models in clinical medicine before most of medicine had tried one — and he has spent the time since building tools, courses, and frameworks that turn AI enthusiasm into AI competence. He publishes ObGyn Intelligence at obmd.com and maintains a free suite of evidence-based clinical tools at tools.obmd.com.
Grünebaum A, Chervenak J, Pollet SL, Katz A, Chervenak FA. The exciting potential for ChatGPT in obstetrics and gynecology. Am J Obstet Gynecol. 2023;228(6):696-705. doi:10.1016/j.ajog.2023.03.009. PMID 36924907. Published online March 14, 2023.