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

Open in new window