Understanding Chain-of-Thought in LLMs through Information Theory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ton, Jean-Francois, Taufiq, Muhammad Faaiz, Liu, Yang
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913935463022592
author Ton, Jean-Francois
Taufiq, Muhammad Faaiz
Liu, Yang
author_facet Ton, Jean-Francois
Taufiq, Muhammad Faaiz
Liu, Yang
contents 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 in 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 subtasks.
format Preprint
id arxiv_https___arxiv_org_abs_2411_11984
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Understanding Chain-of-Thought in LLMs through Information Theory
Ton, Jean-Francois
Taufiq, Muhammad Faaiz
Liu, Yang
Computation and Language
Artificial Intelligence
Machine Learning
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 in 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 subtasks.
title Understanding Chain-of-Thought in LLMs through Information Theory
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2411.11984