Google Research 开源 RRSI：让 AI 智能体改进自身 harness 而不过拟合
Google Research Open-Sources RRSI: AI Agents That Improve Their Own Harness Without Overfitting - MarkTechPost
Google Cloud AI Research 联合 UNC-Chapel Hill、斯坦福和圣路易斯华盛顿大学发布 RRSI（Regularized Recursive Self-Improvement），让 LLM 智能体重写自己的 harness，包括提示词、工具、记忆、控制流和子智能体，模型权重不变。RRSI 对改进循环本身施加约束，以缓解基准特定拟合、追逐噪声和复杂度累积三种失败模式，全部 6 个留出集均有提升：以 Gemini 3.5 Flash 为策略时，Terminal-Bench 2.1 从 64.6 升至 78.7，SWE-bench Verified 从 76.8 升至 79.0。代码以 Apache 2.0 开源，需 Python 3.10+，支持任意 LiteLLM 模型字符串，默认使用 Vertex AI 上的 Claude Opus 4.8。
推荐理由
AI 评分 75
RRSI 把正则化思路搬到智能体自我改进循环上，读者可了解它如何在 6 个留出集上避免过拟合。
主题
正文
当前显示原文，中文全文译文尚未就绪。
Google Cloud AI Research , with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released RRSI (Regularized Recursive Self-Improvement) . It lets an LLM agent rewrite its own harness: prompts, tools, memory, control flow and sub-agents. Model weights never change. RRSI constrains the improvement loop itself, so gains hold on benchmarks the agent never optimized against.
Deployable? Yes, as a research framework. The code is Apache 2.0 , needs Python 3.10+, and accepts any LiteLLM model string. Defaults assume Claude Opus 4.8 on Vertex AI.
Harness evolution loops propose edits, score them on a fixed evolve set and keep the winner. The same tasks are reused every round, so the loop can memorize them. The RRSI research names 3 failure modes: benchmark-specific fitting, noise chasing and complexity accumulation. Each one widens the gap between evolve-set scores and real transfer.
RRSI keeps every harness component editable. It regularizes how the search moves instead.
The research team frame these as analogies to classic regularizers. The edit budget maps to L0, pruning to Lasso (L1) and the cost rule to Ridge (L2).
All 6 held-out splits improved. With Gemini 3.5 Flash as the policy, Terminal-Bench 2.1 rose from 64.6 to 78.7. SWE-bench Verified rose from 76.8 to 79.0.
The harness is also lighter. On the agentic workspace instance, RRSI uses 2.42M policy tokens per trial. Unregularized evolution uses 3.80M. The abstract reports this as 30% fewer; the project page says 36%.
Scores come from Table 1 of the RRSI research paper. All methods share the same starting harness, policy, evolve split and candidate budget.
*Per the RRSI research team. OOD average is the mean of JobBench, GDPval and APEX-Agents, computed from Table 1 .
Meta-Harness leads on the Harvey LAB evolve split. RRSI has the smallest evolve gain but the only OOD average more than 1 point above H0.
git clone https://github.com/google-research/rrsi.git && cd rrsi pip install -e ".[dev]" python3 rrsi.py --domain coding baseline python3 rrsi.py --domain coding run
Each round drafts 2 candidates in separate git worktrees, screens them, evaluates both and fast-forwards the branch to the winner. The coding instance also needs Docker and harbor . New domains plug in through a single adapter module.