Distillation
A technique for training a smaller, cheaper model to copy the behavior of a larger, more capable one, so you get much of the quality at a fraction of the running cost. The large model acts as a teacher and the small model as a student. This is increasingly how the AI tools people actually use are built, because the biggest models are expensive to run at scale. For a healthcare audience the point is that a small fast model in a product may be a distilled version of a frontier model, and it can carry both the strengths and the blind spots of its teacher.
An IT team is choosing between a hosted summarization model and a smaller distilled version that runs on the hospital's own servers, keeping notes on site and answering faster. On the supplier's test set, mostly adult general medicine notes, the small model scores close to the large one. On pediatric or transplant notes the gap can be much wider, since a student matches its teacher most closely on material like its distillation data. Run both on a sample from your own specialties before choosing, and ask which of the teacher's known failure modes the student was tested for.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →