Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs

TL;DR

This paper identifies the underthinking phenomenon in o1-like large language models, where frequent thought switching hampers reasoning depth and accuracy, and proposes a decoding strategy to mitigate this issue, improving performance without fine-tuning.

Contribution

It introduces a novel metric to quantify underthinking and proposes TIP, a decoding method that reduces premature thought switching to enhance reasoning depth in LLMs.

Findings

Thought switching correlates with incorrect answers.

TIP improves accuracy on challenging datasets.

The approach does not require model fine-tuning.

Abstract

Large language models (LLMs) such as OpenAI's o1 have demonstrated remarkable abilities in complex reasoning tasks by scaling test-time compute and exhibiting human-like deep thinking. However, we identify a phenomenon we term underthinking, where o1-like LLMs frequently switch between different reasoning thoughts without sufficiently exploring promising paths to reach a correct solution. This behavior leads to inadequate depth of reasoning and decreased performance, particularly on challenging mathematical problems. To systematically analyze this issue, we conduct experiments on three challenging test sets and two representative open-source o1-like models, revealing that frequent thought switching correlates with incorrect responses. We introduce a novel metric to quantify underthinking by measuring token efficiency in incorrect answers. To address underthinking, we propose a decoding strategy with thought switching penalty TIP that discourages premature transitions between thoughts, encouraging deeper exploration of each reasoning path. Experimental results demonstrate that our approach improves accuracy across challenging datasets without requiring model fine-tuning. Our findings contribute to understanding reasoning inefficiencies in o1-like LLMs and offer a practical solution to enhance their problem-solving capabilities.