Learnijoy
ResearchAI ResearchAI Engineering & DevTools

Evolution Strategies Boost LLM Solution Coverage for Discovery

Conor F. Hayes, Elliot Meyerson, Kajetan Schweighofer, Roberto Dailey, Babak Hodjat, Risto Miikkulainen, Xin Qiu· August 14, 2026 View original

Key takeaways

  • Evolution Strategies (ES) enhance LLM solution diversity more effectively than Reinforcement Learning (RL).
  • ES optimizes directly in weight space, leading to broader output distributions.
  • Increased solution coverage from ES improves performance on discovery tasks like math benchmarks.
  • ES is a promising post-training method for LLMs in domains requiring varied outputs.

Who benefits

Scientific ResearchAI/ML EngineeringPharmaceuticalsMaterials ScienceSoftware Development

Summary

Traditional reinforcement learning (RL) narrows LLM output distributions, limiting solution diversity for discovery tasks, but a new paper shows Evolution Strategies (ES) achieve higher solution coverage and better results in math benchmarks by optimizing directly in weight space. ES provides a superior foundation for post-training LLMs in domains requiring diverse solutions.

Large Language Models are increasingly used in discovery-oriented fields like mathematics and science, where finding diverse solutions is often more valuable than just the single "best guess." While methods like pass@k allow models to explore multiple candidate solutions, standard Reinforcement Learning (RL) post-training often inadvertently narrows the model's output distribution, reducing this crucial solution coverage. This research proposes Evolution Strategies (ES) as a more effective alternative for post-training LLMs in such domains. ES is a population-based, gradient-free optimization method that directly perturbs model weights. The study demonstrates that ES consistently achieves higher pass@k scores than RL, producing a broader and more diverse range of outputs. This enhanced solution coverage, in turn, leads to improved performance on standard math benchmarks, establishing ES as a better foundation for LLM post-training in discovery problems where diverse solution generation is critical.

Why it matters

For professionals developing LLMs for scientific discovery, engineering design, or complex problem-solving, maximizing the diversity of generated solutions is key to innovation and finding optimal outcomes, making ES a valuable technique to consider.

How to implement this in your domain

  1. 1Investigate integrating Evolution Strategies into your LLM post-training pipeline for discovery-focused applications.
  2. 2Compare the pass@k performance of ES-trained models against RL-trained models on your specific problem domains.
  3. 3Experiment with different ES configurations to optimize for solution diversity and coverage.
  4. 4Apply ES to problems requiring multiple valid solutions, such as code generation, drug discovery, or material design.

Original post by Conor F. Hayes, Elliot Meyerson, Kajetan Schweighofer, Roberto Dailey, Babak Hodjat, Risto Miikkulainen, Xin Qiu

"arXiv:2608.12679v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly deployed in discovery domains such as math and science. The usual approach is to present the problem to the model and use its answer as the proposed solution. However, beyond this best g…"

View on X

Originally posted by Conor F. Hayes, Elliot Meyerson, Kajetan Schweighofer, Roberto Dailey, Babak Hodjat, Risto Miikkulainen, Xin Qiu on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI ResearchAI Investing

FlowLOB Generates Realistic, Controllable Limit Order Books Efficiently

This paper introduces FlowLOB, a conditional flow-matching generator for Limit Order Book (LOB) trajectories that offers realistic market dynamics, efficient sampling, and controllable scenario generation, outperforming existing agent-based and deep generative simulators. FlowLOB achieves high fidelity with significantly fewer computational steps than diffusion models and transfers effectively to unseen instruments.

Zhuohan Wang, Andreea Bacalum, Ollie Olby, Carmine Ventre, Namid StillmanAug 14, 2026
AI Engineering & DevToolsAI Research

Auditing Reveals Bias in Neural Combinatorial Optimization Benchmarks

This paper audits test-time budget allocation in Neural Combinatorial Optimization (NCO) solvers, revealing that reported gains from non-uniform sampling often stem from "sampling luck" rather than true allocation benefits on in-distribution data. It proposes a correction procedure and demonstrates real gains under distribution shift, emphasizing the need for rigorous evaluation.

Jinhyung BaeAug 14, 2026
AI Engineering & DevToolsAI Research

Diffusion Models Solve Mixed-Integer Optimization Problems Faster

This paper introduces Constrained Graph Diffusion (CGD), a novel learning-based approach that uses a graph-based generative diffusion model to approximately solve mixed-integer optimization problems (MIPs). CGD integrates a training-free feasibility projection operator into the diffusion process, significantly improving solution quality and feasibility while achieving substantial speedups over traditional numerical solvers.

Vincenzo Di Vito, Mehdi Taghizadeh, Deepjyoti Deka, Kaarthik Sundar, Ferdinando FiorettoAug 14, 2026