RSI-Exam Opens for Contributions to Benchmark Recursive Self-Improvement in AI Agents
yuyinzhou_cs · x · 2026-08-28
RSI-Exam is a new benchmark designed to measure Recursive Self-Improvement (RSI) in AI agents, testing their ability to improve existing methods through autonomous, long-horizon experimentation and generalize gains to hidden data.
Key Findings:
- Spans 88 executable research tasks across 6 domains.
- Claude and GPT form the first tier, significantly outperforming others.
- Agent trajectories reveal distinctions between discovering better methods and failure modes.
Ways to Contribute:
- Research Tasks: Design challenging, executable research environments.
- Reviews: Critically evaluate task designs and metrics.
- Rollout Audits: Analyze agent trajectories to validate scoring.
Next contribution deadline: September 15, 2026.
More from Research
- DeepPhylo infers phylogenetic trees without unrealistic evolutionary assumptions — anshulkundaje · 2026-10-15
- Wander preprint explores graph foundation models via shared objective and random-walk interface — mmbronstein · 2026-10-15
- CHI Reviewer: AI Slop Is Real, but Unusually Rigorous Papers Still Getting Through — IanArawjo · 2026-10-15
- Into the Unknown: CMU benchmark finds 114 of 135 AI-written research papers fail integrity checks — arankomatsuzaki · 2026-10-15
- 8 non-negotiable principles for building AI that actually learns from experience — tokenbender · 2026-10-15
- Benchling launches Science Foundries: multiplayer agent swarms for end-to-end science workflows — nlarusstone · 2026-10-15