[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-trinity-capability-driven-recursive-self-improvement":3,"topics-all":38,"news-related-098ef0da-3801-419a-9e33-855d6c6d7ee6":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},"098ef0da-3801-419a-9e33-855d6c6d7ee6","Trinity:递归自我改进,训练基准平均涨 34 分","The Trinity 框架给冻结的 Claude Opus 5 装上能力树:Explorer 跑轨迹、Evolver 重排任务、Creator 造仿真世界,路由器按能力挑 Harness。论文报告 28 个训练基准平均 +34 分、18 个未见基准 +26 分,含两台真机;代码承诺一周内放出。","递归自我改进(RSI)这件事,卡在两个很具体的地方:给智能体造一批异构、可验证的环境太贵;Harness——包在模型外面的那层工程外壳——靠人手迭代,跟不上任务多样性的膨胀。10 月 26 日挂上 arXiv 的 [The Trinity](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36077)(2610.36077,cs.AI)给出了一套自动化程度相当高的答案:模型权重一行不动,让框架自己长出「能力树」。论文当天冲上 Hugging Face Daily Papers 热榜第一(233 赞)。\n\n## 冻结的 Claude Opus 5,动的全是外壳\n\n框架的基座是一个冻结的 Claude Opus 5(附录 D:全程 adaptive thinking 低档),所有改进都发生在 Harness 层——论文把 Harness 定义为围绕冻结模型的六个控制流协议。三个角色分工明确:Explorer 拿基础 Harness 把训练任务跑一遍、记录轨迹;Evolver 从验证过的轨迹里提取共性需求,把按表面类别组织的任务树重排成层级能力树;Creator 为同节点任务抽象出简化的、可验证的仿真世界,专门练能力——生成的世界明确排除目标任务的布局与解,防止泄漏。最终能力树长出 11 个能力节点、10 套 Harness,Creator 的共享模块库累计 25 个模块。新任务来了,由 Capability-Aware Harness Router 按公开任务信息挑一套兼容配置。论文称之为 Capability-Driven Real2Sim2Real:仿真里练,原任务上测迁移。\n\n## 44 个基准上的成绩单\n\n基准库共 44 个,分 5 大域、15 个子类,覆盖规则发现、游戏、GUI 操作和具身交互;摘要口径为 28 个训练基准 + 18 个未见基准。基础 Harness H0 的均值是 39.06\u002F37.45(Train\u002FUnseen),四轮 RSI 迭代后,论文报告训练侧平均 +34 分、未见侧 +26 分,还包括两个真机具身任务(SOON E01、CaP-X C-01)。分基准看:迭代 1→4 间 ARC-AGI-3 从 41.0 升到 53.6,SciGym 45.2→57.1,BuildArena 93.3→97.8;未见均值超过 Codex(47.5)与 DeepSeek(46.2)两个对照 Harness。\n\n两个结构性发现值得单独说。其一是组织方式:五种组织方案里,六组平铺的能力分组训练最省(136.94M tokens)但成绩最低(49.25\u002F39.43);把能力组织成树、在 Creator 生成的世界里练,成绩最高(56.95\u002F47.26),代价是训练 token 涨到 405.70M——用「在哪里练」换「练得多好」。其二是路由:完整自路由拿到 71.27\u002F60.96,已经逼近事后最优的 Oracle(73.16\u002F63.61),说明路由器基本总能选对分支。\n\n## 三瓢冷水\n\n第一,增益不均匀:Mastermind 四轮停在 88.9,规则发现游戏 Eleusis 从 82.2 微跌到 81.1;Crafter 涨到 60 后仍远低于 Codex 对照的 82.7,论文自己也点名 Crafter 与 DiG-bench 低于两个对照。第二,评测协议里每个基准-Harness 对先跑五次 preparation trials,把轨迹总结成技能再固定计分——增益里含一段「临考突击」的适应性成分,HF 社区讨论也点破了这一条。第三,可复现性目前是纸面的:[GitHub 仓库](https:\u002F\u002Fgithub.com\u002FInfRec-Cardinal\u002FThe-Trinity)只有 3 个 commit、14 star,README 写明代码「10 月 28 日起一周内」放出;论文署名为团队名 The Trinity Research Team、未列机构(HF 论文页机构挂新加坡国立大学,一作 Weidong Tang,作者名单 21 人)。\n\n## 所以呢\n\nThe Trinity 的真正主张不是跑分,而是把 RSI 的优化对象从「任务」换成「能力」:任务会过时,能力可以跨任务复利。Anthropic 官方也专门讨论过递归自我改进的治理问题,而权重不动、外壳自进化,恰是一条不需要重训模型的渐进路线。对工程团队的直接启示:在砸钱微调之前,先把 Harness 的模块化与路由做扎实,可能是更便宜的一级火箭。代码一周内落地,届时这套能力树能否被第三方复现,才是它从技术报告变成方法论的门槛。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36077","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},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"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},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"1f7e8cac-299a-4001-8acb-3eb4bf270d44","en","Trinity: capability tree lifts frozen Claude Opus 5 by 34 points","Trinity grows a capability tree over a frozen Claude Opus 5: +34 on 28 training benchmarks, +26 on 18 unseen, two real robots; code due in a week.","Recursive self-improvement (RSI) keeps hitting two very concrete walls: building a batch of heterogeneous, verifiable environments for agents is expensive, and the Harness — the engineering shell wrapped around the model — cannot be iterated by hand fast enough to match exploding task diversity. The Trinity ([arXiv:2610.36077](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36077), cs.AI, posted October 26) offers a heavily automated answer: leave the model weights completely frozen, and let the framework grow its own \"capability tree\". The paper shot to #1 on the Hugging Face Daily Papers board the same day (233 upvotes).\n\n## A frozen Claude Opus 5, everything else moves\n\nThe backbone is a frozen Claude Opus 5 (Appendix D: adaptive thinking at low effort throughout); all improvement happens at the Harness layer, which the paper defines as six control-flow protocols around the frozen model. Three roles split the work: Explorer runs the training tasks once with the base Harness and logs trajectories; Evolver extracts shared requirements from verified trajectories and reorganizes the task tree — originally grouped by surface category — into a hierarchical capability tree; Creator abstracts tasks at the same node into simplified, verifiable simulated worlds for capability practice, explicitly excluding target-task layouts and solutions to prevent leakage. The final tree has 11 capability nodes and 10 Harnesses, and Creator's shared module library accumulates 25 modules. When a new task arrives, the Capability-Aware Harness Router picks a compatible configuration from public task information. The paper calls this Capability-Driven Real2Sim2Real: train in simulation, measure transfer on the original tasks.\n\n## The scoreboard across 44 benchmarks\n\nThe benchmark inventory has 44 entries across 5 domains and 15 subclasses, covering rule discovery, games, GUI manipulation, and embodied interaction; the abstract's accounting is 28 training plus 18 unseen benchmarks. The base Harness H0 averages 39.06\u002F37.45 (Train\u002FUnseen); after four RSI iterations the paper reports a mean +34 on training and +26 on unseen, including two real-robot embodied tasks (SOON E01, CaP-X C-01). Per-benchmark: between iterations 1→4, ARC-AGI-3 climbs from 41.0 to 53.6, SciGym from 45.2 to 57.1, BuildArena from 93.3 to 97.8; the final unseen mean beats the Codex (47.5) and DeepSeek (46.2) reference Harnesses.\n\nTwo structural findings deserve their own paragraph. First, organization: among five schemes, six flat capability groups are cheapest to train (136.94M tokens) but score lowest (49.25\u002F39.43); organizing capabilities into a tree and practicing inside Creator-generated worlds scores highest (56.95\u002F47.26) at the cost of 405.70M training tokens — trading \"where to practice\" against \"how well it sticks\". Second, routing: the full self-routing setup reaches 71.27\u002F60.96, closing in on the retrospective-best Oracle (73.16\u002F63.61), meaning the router almost always picks the right branch.\n\n## Three buckets of cold water\n\nFirst, the gains are uneven: Mastermind sits at 88.9 across all four iterations, and the rule-discovery game Eleusis slips from 82.2 to 81.1; Crafter, after rising to 60, still trails the Codex reference at 82.7 — the paper itself names Crafter and DiG-bench as below both comparators. Second, the evaluation protocol runs five preparation trials per benchmark–Harness pair, summarizes their trajectories into skills, and only then freezes scoring — the gains embed a \"cram-before-the-exam\" adaptation component, a point HF community discussion also flags. Third, reproducibility is currently on paper only: the [GitHub repo](https:\u002F\u002Fgithub.com\u002FInfRec-Cardinal\u002FThe-Trinity) holds 3 commits and 14 stars, with the README promising code \"within one week of October 28, 2026\"; the paper is signed by the team name The Trinity Research Team with no affiliations listed (the HF paper page tags National University of Singapore, first author Weidong Tang, 21 authors).\n\n## So what\n\nThe Trinity's real claim is not the scores but shifting the optimization target of RSI from \"tasks\" to \"capabilities\": tasks expire, capabilities compound across tasks. Anthropic has publicly discussed the governance of recursive self-improvement, and a frozen-weights, self-evolving shell is precisely a gradual path that requires no model retraining. For engineering teams the direct lesson: before pouring money into fine-tuning, making your Harness modular and well-routed may be the cheaper first-stage rocket. Code lands within a week; whether third parties can reproduce this capability tree is the real threshold between a technical report and a methodology.","trinity-capability-driven-recursive-self-improvement","2026-10-28T13:16:03Z","2026-10-28T13:16:07.900460Z","2026-10-28T13:16:07.900469Z",true,"agent",205,[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},"fd0e593c-bb2c-44f9-8913-16f8dcef6999","苹果8B智能体:真实iOS上追平Gemini-3.1-Pro","apple-online-rl-ui-agent-live-ios","2026-10-31T11:10:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"23567889-4d40-4af8-a58a-ca61cdb6a8b2","小米A-RSI:不用外部标注,模型自己发现题库刷完了","a-rsi-zero-data-self-improvement","2026-10-30T19:09:58+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"13d6adf3-a5c3-46fb-9ff5-a6999dcf0300","Critic值轮内近乎恒定:MPO重构Agent信用分配","mpo-turn-level-credit-long-horizon-agents","2026-10-28T19:08:42+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"a50b0336-dabe-405e-b8c4-d27156ab6d24","60% 权重转私有测试集:开放权重落后榜首 12 分","artificial-analysis-intelligence-index-v5","2026-10-28T07:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"5480e0ca-7b17-481c-8a24-8498f1a0aecd","8B模型OST涨13.41分:空间推理的接口锁定","chronoreplay-spatial-interface-lock-in","2026-10-28T05:09:43+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"be62f4f6-6935-4c2c-abe8-9e0732bb04ef","Trajectory2Gym:智能体的成功轨迹不等于好教材","trajectory2gym-compositional-interaction-synthesis","2026-10-27T21:10:28+00:00"]