Evolving Deeper LLM Thinking

2025/01/17 by Kuang-Huei Lee, Ian Fischer, Lee, Kuang-Huei +12 · 10 voices · 15 citations
Computer Science · Social Sciences · #Artificial Intelligence in Law #Digital Rights Management and Security #Law, AI, and Intellectual Property #cs.AI

paper · pdf · doi:10.48550/arxiv.2501.09891

openalex publication_date 2025/01/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

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.

Cited by

• Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Branching Tree Search
• Procedural Knowledge at Scale Improves Reasoning
• Instruction-Tuning Open-Weight Language Models for BPMN Model Generation
• FutureWeaver: Planning Test-Time Compute for Multi-Agent Systems with Modularized Collaboration
• On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference
• Cognitive Alpha Mining via LLM-Driven Code-Based Evolution
• MultiGA: Leveraging Multi-Source Seeding in Genetic Algorithms
• Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging
• EvoMem: Improving Multi-Agent Planning with Dual-Evolving Memory
• Adaptive Coopetition: Leveraging Coarse Verifier Signals for Resilient Multi-Agent LLM Reasoning
• Parallel Test-Time Scaling for Latent Reasoning Models
• Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models
• SAMULE: Self-Learning Agents Enhanced by Multi-level Reflection
• VerilogMonkey: Exploring Parallel Scaling for Automated Verilog Code Generation with LLMs
• CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning

Discussions

• Evolving Deeper LLM Thinking [hn, 12 points, 0 comments]
• Forget about chain-of-thoughts: "Evolving Deeper LLM Thinking", a novel method from @GoogleDeepMind called Mind Evolution to enhance how large language models (LLMs) solve complex problems. arxiv.org [bsky, 2 points, 0 comments]
• Evolving Deeper LLM Thinking [hn, 1 points, 0 comments]
• Evolving Deeper LLM Thinking The researchers explore an evolutionary search strategy for scaling inference time compute in Large Language Models. The proposed approach, Mind Evolution, uses a langua [bsky, 1 points, 0 comments]
• "In simpler terms, this approach allows AI to think more creatively and effectively, making it better at tasks like planning trips or solving complex problems in natural language. It’s a step forward [bsky, 1 points, 0 comments]
• Think about it: We're not just making AI smarter - we're teaching it to be more critical, more creative, and more effective at solving complex problems. Could this be the key to tackling even more ch [bsky, 0 points, 0 comments]
• Evolving Deeper LLM Thinking Research exploring methods to enhance large language models' reasoning capabilities and depth of thinking Read here [bsky, 0 points, 0 comments]
• Example: from arxiv.org/abs/2501.098... to gist.github.com/oaustegard/c... Example task completion: claude.ai/share/62da5a... [bsky, 0 points, 1 comments]
• A new paper from Deepmind leverages genetic algorithms (search, specifically) + LLMs to get some very impressive results in planning tasks. It outperforms o1, Best-of-N, and similar methods of by a la [bsky, 0 points, 0 comments]
• “Evolving Deeper LLM Thinking”: arxiv.org/abs/2501.09891 Use of a genetic algorithm approach with generative models for evolving better solutions to inputted tasks, where the output has a solution ev [bsky, 0 points, 1 comments]

Related

• Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoning
• Planning In Natural Language Improves LLM Search For Code Generation
• A Survey of Slow Thinking-based Reasoning LLMs using Reinforced Learning and Inference-time Scaling Law
• Evolutionary thoughts: integration of large language models and evolutionary algorithms
• Inference-Time Computations for LLM Reasoning and Planning: A Benchmark and Insights