REL: Working out is all you need

TL;DR

This paper introduces the Reasoning Enhancement Loop (REL), a method to improve large language models' reasoning by generating high-quality problem-solving data, inspired by human-like systematic thinking.

Contribution

It presents the REL method and a specialized dataset to enhance LLM reasoning capabilities through synthetic worked solutions.

Findings

REL improves LLM planning and reasoning performance.

High-quality reasoning data enhances model capabilities.

Synthetic data generation is effective for reasoning tasks.

Abstract

Recent developments, particularly OpenAI's O1 model, have demonstrated the remarkable potential of Large Language Models (LLMs) for complex reasoning tasks. Through analysis of O1's outputs and provided sample Chain-of-Thought (CoT) demonstrations, we observe that it approaches problem-solving in a distinctly human-like manner, systematically brainstorming ideas, testing hypotheses, verifying results, and planning comprehensive solutions. These sophisticated reasoning capabilities remain notably absent in other state-of-the-art language models. In this paper, we hypothesize that this performance gap stems from the limited availability of high-quality reasoning process data in current training sets. We demonstrate that by constructing a specialized dataset focused on explicit problem-solving workflows ("worked solutions"), we can elicit substantially improved planning capabilities from existing models. Additionally, we propose the Reasoning Enhancement Loop (REL), a method for generating synthetic worked solutions.