antidistillation sampling

New framework proposed for evaluating LLM distillation defense robustness

A new research paper proposes a framework for understanding and evaluating defenses against distillation attacks on black-box large language models. The paper introduces a threat model that considers an attacker's query…