Study finds no universally best harness for automated discovery systems
LChoshen · x · 2026-07-22
A controlled study finds there is no universally best harness for automated discovery.
- Simpler harnesses often matched or beat more complex systems such as OpenEvolve.
- The best choice changed across model–problem pairs, so harnesses do not generalize reliably.
- The authors ran repeated budget-matched trials, strong baselines, and statistical hypothesis testing to handle the high-variance setting.
- Letting the system choose among harnesses online as rollouts progress can reduce the generalization problem and motivates adaptive discovery systems.
Related event: Study Finds No Universal Optimal Harness for Automated Discovery(2 posts)→
More from Research
- ICML 2026 oral paper replication scores stay middling after a stricter re-scoring — profjamesevans · 2026-07-27
- Long-running agents will need immutable event logs, this thread argues — sebpaquet · 2026-07-27
- Seed IQ navigates Doom II, prompting questions about benchmarks beyond ARC-AGI — Fit_Transition8824 · 2026-07-27
- Agentic Data Science in Practice: Agents Write Code but Answer Wrong Questions — hugobowne · 2026-07-27
- A concise canon of foundational papers in ML, systems, NLP, speech, and audio — deliprao · 2026-07-27
- TechCrunch says brain-wave signals could be the next unlock for physical AI training — TechCrunch AI · 2026-07-27