What generic AI gets wrong about lab data, and what it takes to get it right

General-purpose AI models treat lab data the same way they treat any other text, tokens in and tokens out. That breaks down fast in clinical hematology, where reference ranges, patient context, laboratory quirks, and temporal patterns decide whether a reading is meaningful or noise. This talk breaks down where generic AI fails on blood data specifically, what a domain-built model like BloodGPT does differently, and the wider lesson for building trustworthy AI on any medical signal that is not language.