Large AI models are capable but expensive to run. Knowledge distillation trains a small model to imitate a large one, using the big model as a teacher, so the compact student captures much of the ability at a fraction of the size and cost.
This makes it practical to deploy useful models on phones and modest servers, cutting cost and energy. The smaller model rarely matches the teacher on the hardest cases, so there is a balance between size and capability, but for many real tasks a well-distilled model is more than good enough and far cheaper to run at scale.