Creating a different version of an existing AI model. One type of AI distillation is the "teacher/student" method whereby the small student model is trained to mimic the teacher model but with fewer parameters. The student model runs faster and requires less computer time.
Another method is self-distillation, which is a model that generates a new model without requiring a teacher. It uses its own outputs to retrain itself. Companies also use AI distillation to replicate the capabilities of other AI models. See
AI glossary.