[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-rrsi-agent-harness-regularization":3,"topics-all":38,"news-related-68812025-96eb-4ca9-a1bc-8a82a40174dc":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"68812025-96eb-4ca9-a1bc-8a82a40174dc","Google RRSI:给 Agent 外壳自进化加正则化","Google 提出 RRSI:agent harness 递归自进化会背题过拟合,在提案与筛选两端加约束后,5 个分布外基准最高涨 4.7 分,policy token 少用 30%。","Agent 圈有个共识正在成型:同一个冻结模型,换一套由提示词、控制流、工具与记忆管理组成的「外壳」(harness),能力能差出一大截。于是越来越多人让 agent 自己改自己的外壳——迭代提出修改、跑分选优,相当于在系统层做递归自改进(RSI)。Google 团队 9 月 21 日放出的论文 RRSI 给这条路线泼了一盆冷水:这么进化出来的外壳,很可能只是在背题。\n\n## 病根:外壳进化会过拟合\n\n论文先给诊断:围绕冻结模型反复进化 harness 收益明显,但递归演化会过拟合——外壳记住了训练任务,在分布内基准上涨分巨大,一旦换到分布外(OOD)基准,涨幅会缩水甚至消失。这和深度学习里「模型背训练集」是同构的问题,只是对象从权重换成了提示词与控制流。\n\n## 双端正则化:一边限改,一边剪枝\n\nRRSI 的解法是把正则化原则嵌进进化循环的两端。提案端(proposer)使用随时间退火的预算,限制单个候选能捆绑多少条独立修改;它还以完整修改历史为条件,已被证伪的假设不会被重复提出,进化停滞时会被引导去尝试从未动过的组件。选择端(selector)配备 critic 与 pruner:critic 在评测前筛掉针对特定基准的投机逻辑,pruner 移除改动太小、太贵或已无用的部分。GitHub README 还补充了两条细则:噪声调整的地板会拦住评测方差以内的「涨分」,成本规则要求新增的推理 token 必须由实测收益支付。\n\n## 数字:OOD 涨 4.7 分,token 省 30%\n\n在覆盖编码、agentic workspace 与工程设计任务的 8 个基准上,RRSI 在进化所用的数据划分上最高涨 14.1 分,在 5 个分布外基准上最高涨 4.7 分,产出的外壳比无正则化进化少用 30% 的 policy token——更通用,还更便宜。同一套循环驱动三个实例:终端 agent(Terminal-Bench 2.1)、文档工作 agent(Harvey LAB)与工程设计 agent(EngDesign)。论文发布当天即登顶 Hugging Face 日榜。代码与项目页均已放出:仓库 google-research\u002Frrsi,项目页 regularized-rsi.com。\n\n## 所以呢\n\n这篇论文的价值不在涨分本身,而在把一个工程直觉变成可复现的方法论:harness 优化是一场搜索,搜索就需要正则化——从深度学习借来的老思想,在 agent 系统层重新生效。回头看赛道邻居:DeepSeek 开源了可插拔的 Harness 插件栈,NVIDIA SoL-Pi 用外挂 harness 砍 token,HarnessDev 干脆让 LLM 自建外壳来考。大家都默认「外壳值得被系统化优化」,RRSI 补上的一环是:优化过程本身也会过拟合,也得防。对做 agent 工程的团队,这是个直接提醒——如果你的 agent 调优只在一个固定评测集上越调越高,先怀疑它在背题,再怀疑它真的变强了。\n\n参考:论文 arXiv:2609.24972(https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.24972);代码 https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Frrsi;项目页 https:\u002F\u002Fregularized-rsi.com\u002F","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.24972","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"db3d06d8-c08a-4c33-8078-da15c28dcb8f","en","Google's RRSI Puts Regularization into Agent Harness Self-Evolution","Google's RRSI shows harness self-improvement overfits by memorizing tasks. Regularized evolution gains 4.7 OOD points with 30% fewer tokens.","A consensus is forming in the agent world: with the same frozen model, swapping the surrounding \"harness\" — the prompts, control flow, tooling, and context management — can change capability dramatically. So a growing number of teams let agents modify their own harnesses: iteratively proposing edits, evaluating on benchmarks, and keeping the winners. This is recursive self-improvement (RSI) at the agent-system level. A Google team's paper released on September 21, RRSI, pours cold water on this route: a harness evolved this way may simply be memorizing the test.\n\n## The Disease: Harness Evolution Overfits\n\nThe paper starts with a diagnosis: evolving a harness around a frozen model delivers visible gains, but recursive evolution overfits — the harness memorizes the training tasks, posts large in-distribution gains, and those gains shrink or vanish on out-of-distribution (OOD) benchmarks. It is structurally the same problem as a deep net memorizing its training set, except the object being memorized with is prompts and control flow instead of weights.\n\n## Two-Sided Regularization: Cap the Proposals, Prune the Rest\n\nRRSI embeds regularization principles into both ends of the evolution loop. On the proposal side, the proposer operates with a temporally annealed budget that caps how many independent edits one candidate may bundle; it is conditioned on the full edit history so falsified hypotheses are not redrawn, and stalled runs get redirected toward components never exercised. On the selection side, the selector is equipped with a critic and a pruner: the critic screens benchmark-specific proposals before evaluation, while the pruner removes changes that are too small, too expensive, or no longer useful. The GitHub README adds two more details: a noise-adjusted floor blocks gains within evaluation variance, and a cost rule requires any added inference tokens to be paid for by measured gain.\n\n## The Numbers: +4.7 OOD, 30% Fewer Tokens\n\nAcross eight benchmarks spanning coding, agentic workspace, and engineering design tasks, RRSI gains up to 14.1 points on the split it evolves against and up to 4.7 points on the five out-of-distribution benchmarks, while producing a harness that runs on 30% fewer policy tokens than unregularized evolution — more general and cheaper. The same loop drives three instances: a terminal agent (Terminal-Bench 2.1), a document-work agent (Harvey LAB), and an engineering-design agent (EngDesign). The paper hit #1 on Hugging Face's daily papers chart on release day. Code and project page are live: repository google-research\u002Frrsi, project page regularized-rsi.com.\n\n## So What\n\nThe value of this paper is not the score gains themselves but turning an engineering intuition into a reproducible methodology: harness optimization is a search, and every search needs regularization — an old idea borrowed back from deep learning, now reactivated at the agent-system level. Look at the neighboring work: DeepSeek open-sourced a pluggable harness plugin stack, NVIDIA's SoL-Pi cuts tokens with an external harness, and HarnessDev has LLMs build their own harnesses as an exam. Everyone assumes the harness deserves systematic optimization; what RRSI adds is the missing link — the optimization process itself can overfit, and must be defended against. For teams doing agent engineering, this is a direct warning: if your agent tuning keeps climbing on one fixed eval set, suspect memorization before you believe it got smarter.\n\nReferences: paper arXiv:2609.24972 (https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.24972); code https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Frrsi; project page https:\u002F\u002Fregularized-rsi.com\u002F","google-rrsi-agent-harness-regularization","2026-09-22T23:08:34Z","2026-09-22T23:08:37.206499Z","2026-09-22T23:08:37.206513Z",true,"agent",44,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"20568e5d-3b66-495d-8b7b-0a702f3c7877","模型在进化,训练环境却是死的:Google 开源 EnvHarness,给环境也套一层 harness","google-envharness-agent-environments","2026-08-20T10:42:06+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"0565190a-0bcd-492f-934f-0ad2ab32f485","70万参数2.8MB填一张表:Cua开源CUA-S1,单次前向替代23轮LLM","cua-s1-forms-system-one-model","2026-09-20T13:11:48+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"13378d5e-2440-496d-8c3c-7d36858e641d","不聊天的端侧基座:Needle 3 用 8-29MB 在微控制器上跑工具调用","needle-3-tiny-tool-calling-model","2026-09-19T13:09:46+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"9ba1770e-87f2-47b9-aaa1-19f4f2ba78f1","ScienceIDE:把全球科学代码变成智能体训练场","scienceide-scientific-code-agent-environments","2026-09-17T23:05:17+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"dcd8b3e1-a3c7-4614-aba4-9002219ea5f6","LibreDB Studio 0.15 发布:本地 LLM 接管数据库交互","libredb-studio-local-llm-agent","2026-09-15T00:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"089f56f3-32ff-4036-89b5-728d5f5a9359","边聊边干活:腾讯混元开源全模态交互 Agent Gander,小脑管对话、大脑管执行","hunyuan-gander-omni-interaction-agent","2026-09-09T21:07:00+00:00"]