Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models

The development of Reasoning Large Language Models (RLLMs) has significantly improved multi-step reasoning capabilities, but it has also made hallucination problems more frequent and harder to eliminate.