The Straight and Narrow: Do LLMs Possess an Internal Moral Path?

Luoming Hu, Jingjie Zeng, Liang Yang +1

Enhancing the moral alignment of Large Language Models (LLMs) is a critical challenge in AI safety. Current alignment techniques often act as superficial guardrails, leaving the in…