Measuring Beliefs of Language Models During Chain-of-Thought Reasoning — AI Alignment Forum