Automated Discovery Has No Universally Superior Harness
2026-07-20 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors studied different ways to run automated discovery systems, which are used to find new things using large language models. They found that no single setup works best for every problem, and simpler methods sometimes do better than more complex ones. Instead of picking one fixed method, the authors suggest adjusting the approach based on early progress during the search. They also created a way to start many methods, stop the weaker ones early, and focus computing power on the better ones, which improves results. The authors provide all their testing data to help others evaluate future methods.
autonomous discovery systemsevolutionary searchOpenEvolveTTT-Discoverlarge language models (LLMs)budget allocationstatistical analysishyperparameter tuningadaptive allocationonline adaptation
Authors
Akshat Gupta, Jermaine Lei, Alexander Lu, Gopala Anumanchipalli, Leshem Choshen
Abstract
Autonomous discovery systems such as OpenEvolve and TTT-Discover are often used as general-purpose harnesses. However, in practice these are composite systems combining several design choices about archives, parent selection, exploration, and budget allocation into a single recipe. Because discovery runs are expensive and inherently stochastic, existing harnesses are often compared using too few independent trials to distinguish key methodological improvements from run-to-run variance. We systematically decompose OpenEvolve-style evolutionary search and the TTT-Discover search harness into its constituent components and systematically evaluate 30 budget-matched harnesses across 12 model-problem pairs using more than 3.1 million LLM rollouts and repeated-trial statistical analysis. Our results show that discovery harnesses have a generalization problem: No fixed harness is reliably superior across the evaluated model-problem pairs, and variants of OpenEvolve generally underperform simpler alternatives. Thus, harness choice is better viewed as a hyperparameter rather than as a universal recipe, and should be tailored to the specific problem and underlying model. We also find that early discovery progress predicts final performance, and use this property to present a budget-matched adaptive-allocation experiment that starts multiple harnesses, prunes weak partial runs, and reallocates compute to stronger survivors, outperforming both commitment to a randomly sampled fixed harness and a non-adaptive harness ensemble. Together, these results motivate shifting from fixed harness selection to online adaptation guided by early performance. We release all run pools including baseline null distributions for every model-problem pair as reusable statistical infrastructure against for future harness proposals.