Understanding Chain-of-Thought in LLMs through Information Theory
Jean-Francois Ton, Muhammad Faaiz Taufiq, Yang Liu
Abstract
Large Language Models (LLMs) have shown impressive performance in complex reasoning tasks through the use of Chain-of-Thought (CoT) reasoning, allowing models to break down problems into manageable sub-tasks. However, existing CoT evaluation techniques either require annotated CoT data or fall short of accurately assessing intermediate reasoning steps, leading to high rates of false positives. In this paper, we formalize CoT reasoning in LLMs through an information-theoretic lens. Specifically, our framework quantifies the `information gain' at each reasoning step, enabling the identification of failure modes in LLMs without the need for expensive annotated datasets. We demonstrate the efficacy of our approach through extensive experiments on toy arithmetic, GSM8K and PRM800k datasets, where it significantly outperforms existing outcome-based methods by providing more accurate insights into model performance on individual tasks.
BibTeX
@inproceedings{
ton2025understanding,
title={Understanding Chain-of-Thought in {LLM}s through Information Theory},
author={Jean-Francois Ton and Muhammad Faaiz Taufiq and Yang Liu},
booktitle={Forty-second International Conference on Machine Learning},
year={2025},
url={https://openreview.net/forum?id=IjOWms0hrf}
}