RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
Google Research's RRSI regularizes recursive agent-harness evolution, lifting out-of-distribution scores while cutting policy tokens 30%.
RRSI constrains recursive self-improvement of LLM agent harnesses—prompts, control flow, tools, memory, and context—so evolution does not overfit training tasks. A temporally annealed edit budget and history-based exploration limit proposals, while a critic and pruner drop benchmark-specific, tiny, costly, or obsolete changes. Across eight coding, workspace, and engineering benchmarks, RRSI gains up to 14.1 points in-distribution and 4.7 on five out-of-distribution sets, using 30% fewer policy tokens than unregularized evolution. Code is released under Google Research.
- Proposer uses an annealed edit budget and favors unexplored evolution paths.
- Critic and pruner reject narrow, tiny, expensive, or stale harness edits.
- Gains up to 14.1 in-distribution and 4.7 out-of-distribution points on eight benchmarks.
- Resulting harness uses 30% fewer policy tokens than unregularized evolution.
- Tags
- rrsiagent-harnessrecursive-self-improvementregularizationllm-agentsgoogle-researchout-of-distribution
- Organizations
- Google Research
From the original article
1 min read at huggingface.co
An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level.
By Peng Xia, Rujun Han, Zifeng Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.