RSIGym treats recursive self-improvement as systems engineering; Opus 5 leads at 0.4809
teortaxesTex · x · 2026-10-08
The Evolvent AI team argues recursive self-improvement (RSI) is a systems engineering problem, not just a model problem: progress hinges on the research agent's environment—what resources it can call, what it can change, and how it runs experiments.
They built RSIGym around "Everything as a Service": agents can call established research services and iterate on data, training settings, and harness code.
They also introduce RSI-Index, measuring how well frontier agents jointly improve a target model's weights and harness. Across 6 research agents and 5 benchmarks, Opus 5 leads at 0.4809. Code and experiments are being open-sourced. The thread also touches on synthetic data's performance drop and whether findings generalize beyond Qwen.
More from coding & agent
- Dev amazed as AI sets up a fresh computer's config in seconds, not minutes — Dimillian · 2026-10-08
- After a $10,811 Cloudflare bill, dev ditches delayed billing alerts for a Grok-built 3-hour usage monitor — mazzaTalk · 2026-10-08
- Voice agents that call businesses for people: caller ID, disclosure, billing and other pitfalls — Born-Mastodon443 · 2026-10-08
- Harvard study of 718 firms finds AI coding boosts individual output ~30% but firm-level shipping doesn't move — DavidLinthicum · 2026-10-08
- Dimillian shares a 4-step Codex workflow for native-looking iOS apps — Dimillian · 2026-10-08
- WSJ columnist declares 'the year of Linux' is finally here, thanks to AI agents — alexvoica · 2026-10-08