Alejo López Ávila

New framework enhances LLM safety refusal with specific boundaries

A new research paper proposes a framework called Boundary-Aware Self-Distillation to improve the safety refusal capabilities of large language models. This method focuses on defining specific refusal boundaries for diff…