Evolving Deeper LLM Thinking logo

Evolving Deeper LLM Thinking

Free

Evolutionary search strategy for scaling LLM inference time compute

FreeFree tier
Type
Open Source

About Evolving Deeper LLM Thinking

Evolving Deeper LLM Thinking introduces Mind Evolution, an evolutionary search strategy designed to scale inference time compute in Large Language Models. The approach leverages a language model to iteratively generate, recombine, and refine candidate responses, avoiding the need to formalize the underlying inference problem as long as a solution evaluator is available. In natural language planning tasks such as TravelPlanner and Natural Plan benchmarks, Mind Evolution outperforms Best-of-N and Sequential Revision strategies, solving over 98% of problem instances using Gemini 1.5 Pro without a formal solver. The method represents a novel way to enhance LLM reasoning through evolutionary algorithms.

Key Features

Evolutionary search strategy using a language model to generate, recombine, and refine candidate responses
Scales inference time compute without formalizing the underlying inference problem
Requires only a solution evaluator, no formal solver needed
Outperforms Best-of-N and Sequential Revision in natural language planning tasks
Achieves over 98% success rate on TravelPlanner and Natural Plan benchmarks with Gemini 1.5 Pro

Pros & Cons

Pros
  • Significantly outperforms other inference scaling strategies like Best-of-N and Sequential Revision
  • Avoids the complexity of formalizing the problem logic
  • Leverages the language model's own generative capabilities for refinement
  • High success rate (over 98%) on challenging planning benchmarks
Cons
  • Requires a problem-specific solution evaluator to guide the search
  • May be computationally expensive due to iterative generation and refinement
  • Effectiveness is demonstrated mainly on planning tasks; generalizability to other domains needs further study

Best For

Natural language planning tasks (e.g., TravelPlanner benchmark)General LLM reasoning and decision-making problems where a solution evaluator is available

FAQ

What is Mind Evolution?
Mind Evolution is an evolutionary search strategy for scaling inference time compute in Large Language Models. It uses the LLM to generate, recombine, and refine candidate responses guided by a solution evaluator.
How does Mind Evolution compare to other inference strategies?
Controlling for inference cost, Mind Evolution significantly outperforms Best-of-N and Sequential Revision in natural language planning tasks, achieving over 98% success rate on TravelPlanner and Natural Plan benchmarks using Gemini 1.5 Pro.
Does Mind Evolution require a formal solver?
No, Mind Evolution avoids the need to formalize the underlying inference problem as long as a solution evaluator is available.