English

Rational Metareasoning for Large Language Models

Computation and Language 2025-06-25 v3 Artificial Intelligence Machine Learning

Abstract

Being prompted to engage in reasoning has emerged as a core technique for using large language models (LLMs), deploying additional inference-time compute to improve task performance. However, as LLMs increase in both size and adoption, inference costs are correspondingly becoming increasingly burdensome. How, then, might we optimize reasoning's cost-performance tradeoff? This work introduces a novel approach based on computational models of metareasoning used in cognitive science, training LLMs to selectively use intermediate reasoning steps only when necessary. We first develop a reward function that incorporates the Value of Computation by penalizing unnecessary reasoning, then use this reward function with Expert Iteration to train the LLM. Compared to few-shot chain-of-thought prompting and STaR, our method significantly reduces inference costs (20-37\% fewer tokens generated across three models) while maintaining task performance across diverse datasets.

Keywords

Cite

@article{arxiv.2410.05563,
  title  = {Rational Metareasoning for Large Language Models},
  author = {C. Nicolò De Sabbata and Theodore R. Sumers and Badr AlKhamissi and Antoine Bosselut and Thomas L. Griffiths},
  journal= {arXiv preprint arXiv:2410.05563},
  year   = {2025}
}
R2 v1 2026-06-28T19:12:15.737Z