Measuring Faithfulness in Chain-of-Thought Reasoning
Does CoT reasoning really reflect the reasoning process of an LLM?
Perhaps it does...
But then again, perhaps it does not.
Recent work from Anthropic studies this question empirically.
An overview of the technical report 👇
Link to the paper: https://www.anthropic.com/index/measuring-faithfulness-in-chain-of-thought-reasoning