For more than fifteen years I worked as a Registered Nurse in psychiatric units, dialysis centers, cardiac telemetry, hospice and home health. One habit carried across every one of those settings: never trust something just because it sounds confident. Check the order. Check the dose. Check the patient.
That habit turns out to be exactly what AI needs.
Today’s AI models are fluent and fast, and often right. But “often right” isn’t good enough when someone is asking about a blood thinner, a chest pain or a lab value. Somebody has to check the work, and it helps if that somebody knows what the right answer looks like at the bedside.
What you’ll find here
- Case studies: benchmarks where I give the same task to several AI models and score how they do, like my clinical safety benchmark.
- AI field notes: shorter observations about how models behave, where they slip, and how to work around it.
- Publishing updates: news from MixedBagPublications, including my book, The Author’s Guide to an End-to-End AI Production Pipeline.
Thanks for stopping by.
— Bruce