Evaluation-driven Scaling for Scientific Discovery

2026-04-21

AI Summary

This work addresses the challenge of efficiently scaling evaluation-driven scientific discovery loops to advance frontier research. It introduces SimpleTES, a novel framework that explicitly prioritizes scalability as a core design principle. By integrating parallel exploration, feedback-driven refinement, and local selection strategies with trajectory-level historical data, SimpleTES achieves strong cross-domain generalization. The approach synergistically combines open-source large language models (e.g., gpt-oss), verifier or simulator feedback, task-specific scoring functions, and post-training trajectory-supervised learning. Evaluated across 21 diverse scientific tasks, SimpleTES establishes new state-of-the-art results, including over 2× acceleration of the LASSO algorithm, a 24.5% reduction in quantum circuit gate overhead, and the discovery of constructions surpassing the best-known Erdős designs.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Constraint Satisfaction and Optimization: Search
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Language models are increasingly used in scientific discovery to generate hypotheses, propose candidate solutions, implement systems, and iteratively refine them. At the core of these trial-and-error loops lies evaluation: the process of obtaining feedback on candidate solutions via verifiers, simulators, or task-specific scoring functions. While prior work has highlighted the importance of evaluation, it has not explicitly formulated the problem of how evaluation-driven discovery loops can be scaled up in a principled and effective manner to push the boundaries of scientific discovery, a problem this paper seeks to address. We introduce Simple Test-time Evaluation-driven Scaling (SimpleTES), a general framework that strategically combines parallel exploration, feedback-driven refinement, and local selection, revealing substantial gains unlocked by scaling evaluation-driven discovery loops along the right dimensions. Across 21 scientific problems spanning six domains, SimpleTES discovers state-of-the-art solutions using gpt-oss models, consistently outperforming both frontier-model baselines and sophisticated optimization pipelines. Particularly, we sped up the widely used LASSO algorithm by over 2x, designed quantum circuit routing policies that reduce gate overhead by 24.5%, and discovered new Erdos minimum overlap constructions that surpass the best-known results. Beyond novel discoveries, SimpleTES produces trajectory-level histories that naturally supervise feedback-driven learning. When post-trained on successful trajectories, models not only improve efficiency on seen problems but also generalize to unseen problems, discovering solutions that base models fail to uncover. Together, our results establish effective evaluation-driven loop scaling as a central axis for advancing LLM-driven scientific discovery, and provide a simple yet practical framework for realizing these gains.

Problem

• evaluation-driven discovery
• scaling
• scientific discovery
• language models
• feedback loops

Innovation

• evaluation-driven scaling
• scientific discovery
• SimpleTES
• feedback-driven refinement
• trajectory-based learning