Dialectical Alignment: Resolving the Tension of 3H and Security Threats of LLMs

Shu Yang, Jiayuan Su, Han Jiang +5

With the rise of large language models (LLMs), ensuring they embody the principles of being helpful, honest, and harmless (3H), known as Human Alignment, becomes crucial. While exi…