AI Distillation

AI distillation is a training technique in which a smaller "student" model is taught to imitate a larger "teacher" model by learning from its outputs, so it can perform similarly at a fraction of the cost. For governance teams, the concern is that distillation can be used to copy a proprietary model's capabilities by querying it at scale, which may breach a provider's terms of service, raise intellectual property questions, and bypass safety guardrails built into the original system. Companies should therefore review both whether their own models are being distilled by third parties and whether their vendors' models were trained using distilled outputs.

1 item