Evolving Deeper LLM Thinking

TL;DR

This paper introduces Mind Evolution, an evolutionary search strategy that leverages language models to improve inference efficiency and effectiveness in large language models, outperforming traditional methods in planning tasks.

Contribution

The paper presents a novel evolutionary approach for scaling inference in LLMs that does not require formal problem specification, demonstrating superior performance in planning benchmarks.

Findings

Mind Evolution outperforms Best-of-N and Sequential Revision strategies.

Achieves over 98% problem-solving success in benchmarks.

Operates effectively without formal solvers.

Abstract

We explore an evolutionary search strategy for scaling inference time compute in Large Language Models. The proposed approach, Mind Evolution, uses a language model to generate, recombine and refine candidate responses. The proposed approach avoids the need to formalize the underlying inference problem whenever a solution evaluator is available. Controlling for inference cost, we find that Mind Evolution significantly outperforms other inference strategies such as Best-of-N and Sequential Revision in natural language planning tasks. In the TravelPlanner and Natural Plan benchmarks, Mind Evolution solves more than 98% of the problem instances using Gemini 1.5 Pro without the use of a formal solver.