Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs

Amal Zouaq Full Professor & FRQS Chair in AI and Digital Health Polytechnique Montreal

September 12, 2024 April 16, 2026

Our paper Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs is accepted at NeurIPS-AFM 2024. See the publications tab for more details.