Protecting Language Models Against Unauthorized Distillation through Trace Rewriting
• Uses trace rewriting to deter unauthorized knowledge distillation from large language models. • Introduces anti-distillation techniques that degrade training usefulness while kee