What is this training technique called?

Knowledge distillation trains a smaller, more efficient "student" model to mimic the outputs and behavior of a larger, more capable "teacher" model, effectively transferring much of the teacher's learned skill into a compact form. The student learns from the teacher's responses across many examples, absorbing patterns of reasoning and style that would be far more expensive to learn from scratch, letting companies deploy fast, cheap tools that still perform impressively well despite their smaller size.

Encryption is a security technique that scrambles data so only authorized parties can read it, an entirely different concept from transferring skills between models. Overclocking means running hardware faster than its standard settings to boost performance, a hardware trick unrelated to how a model is trained.

This is a major reason many fast, inexpensive AI chatbots available today still perform impressively well on everyday tasks, since they effectively inherit distilled knowledge from far larger, costlier models that would be impractical to run directly for every user request.