Distillation — Model Distillation
Definition: Training a smaller "student" model to mimic a larger "teacher" model. Distillation is how you take a flagship-quality experience and ship it at a price you can defend in a P&L.
Example
Distilling the flagship model into a 3B-param student cut inference cost 12x while keeping 94% of the quality.
When you'll hear it
Distillation shows up most often in AI strategy reviews, model evaluation discussions, and product roadmap meetings. When someone uses it, they're usually referring to model distillation — and they expect the room to already know what that means.
FAQs
What does Distillation stand for?
Distillation stands for Model Distillation.
What does Distillation mean in AI and machine-learning?
Training a smaller "student" model to mimic a larger "teacher" model. Distillation is how you take a flagship-quality experience and ship it at a price you can defend in a P&L.
Where will I hear Distillation used at work?
Distillation comes up most often in AI strategy reviews, model evaluation discussions, and product roadmap meetings. It's used as shorthand for model distillation, so people assume you already know the term.