Yuqing Yang

New research reveals LLMs rarely admit mistakes due to internal belief states

A new arXiv paper explores the internal mechanisms that cause Large Language Models (LLMs) to admit their mistakes. Researchers found that LLMs rarely retract incorrect answers, even when they can recognize the error in…