RSIGym: A Flexible Environment for Recursive Self-Improvement

Key points

• Recursive self-improvement requires carrying accepted changes into later improvement cycles, while studying agent-proposed changes also requires substantial research infrastructure.1
• RSIGym exposes training, inference, rollout, evaluation, and sandbox execution through reusable services, with shared budget and permission controls supporting Data, Harness, and Joint improvement tracks.1
• We define RSI-Index as the mean fraction of the remaining performance gap closed across five benchmarks covering software engineering, terminal interaction, mathematics, scientific reasoning, and skill-based tasks.1
• We open-source the full RSIGym codebase and results to support reproducibility and further research.1

Sources (1)

• [1]RSIGym: A Flexible Environment for Recursive Self-Improvement
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 04:05 PM

  We introduce RSIGym, an agent-native research environment based on Everything as a Service (EaaS).

  Recursive self-improvement requires carrying accepted changes into later improvement cycles, while studying agent-proposed changes also requires substantial research infrastructure.

Extractive summary: sentences quoted from the sources.