David Dobre

A Generative Approach to LLM Harmfulness Mitigation with Red Flag Tokens

Learning Diverse Attacks on Large Language Models for Robust Red-Teaming and Safety Tuning