[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-recontext-recursive-evidence-replay":3,"topics-all":33,"news-related-f668bb0a-485f-44cb-8fa8-3aa35c1108cf":52},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":20,"news_slug":26,"published_at":27,"created_at":28,"modified_at":29,"is_published":30,"publish_type":31,"image_url":13,"view_count":32},"f668bb0a-485f-44cb-8fa8-3aa35c1108cf","ReContext 用「递归证据 replay」打通长上下文最后一道关:训练免费,128 个 token 顶替半个 128K 上下文","长上下文焦虑正在变得奇怪:上下文窗口已经从 128K 卷到 1M,但 LLM 在 buried evidence 上的抽取准确率反而没跟着涨。UIUC 团队 Yanjun Zhao 等人在 7 月 2 日放出的 arXiv:2607.02509 把这件事量化得相当难看——在一个 128K 的输入里,前 0.1% 的 token 占据了 50%~80% 的累积问题相关信号,也就是说,128 个 token 实际上承担了\"读完整份长文档\"的活。\n\n论文给出的解法 RECONTEXT,是一个训练免费、完全不改 backbone 的 inference harness。它在 prompt 构造阶段,直接调用模型自己的 attention 分数圈出\"与问题最相关\"的 evidence span,把这些 span 显式地 verbatim 重排到问题附近,最后才让模型生成。原始 128K 一字未删,forward-pass 成本与标准推理基本持平。\n\n在 8 个 128K 长文档 benchmark、Qwen3-4B \u002F Qwen3-8B \u002F Llama3.1-8B 三个 backbone 上,RECONTEXT 拿到 best average rank,平均准确率从 0.24 拉到 0.30,相对增益 24.6%。配套的理论分析也写得有意思:把上下文比作联想记忆的存储、问题当检索线索、attention 当 cue-trace 关联、replay 当 trace reactivation,顺带证明递归 replay 对隐层表示的单调改进。\n\n值得关注的两个落地边界:一是必须能拿到 attention weights,黑盒 API 直接出局,只能自托管 Qwen \u002F Llama3 实例;二是它和 DAC、summary 类压缩天然不兼容,二者同时跑会出现\"双 replay \u002F 选证据标准冲突\",论文里没给解,上生产前必须把上游压缩关掉。ReContext 是少见的\"几乎不花成本就能白拿\"的推理优化,但能不能真正跑进 RAG、长 code agent、合规审查流水线,取决于工程上你愿不愿意给它留一道 attention hook。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02509","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[21],{"id":22,"lang":23,"title":24,"summary":25,"content":13},"a3a4e3ae-d920-4921-b80c-90a8c66c8000","en","ReContext: 128 tokens replace half a 128K context, free","Long-context anxiety is getting strange: the context window has already rolled from 128K to 1M, but the LLM's extraction accuracy on buried evidence hasn't risen with it. The UIUC team led by Yanjun Zhao, in arXiv:2607.02509 released on July 2, quantifies this rather harshly — in a 128K input, the top 0.1% of tokens carry 50%–80% of the cumulative question-relevant signal, meaning 128 tokens actually do the job of \"reading the entire long document\". The paper's solution, RECONTEXT, is a training-free, no-backbone-change inference harness. During prompt construction, it directly calls the model's own attention scores to circle the \"most question-relevant\" evidence spans, explicitly and verbatim reorders these spans close to the question, and only then has the model generate. The original 128K is left untouched, and forward-pass cost is essentially on par with standard inference. Across 8 128K long-document benchmarks and 3 backbones (Qwen3-4B \u002F Qwen3-8B \u002F Llama3.1-8B), RECONTEXT gets the best average rank, lifting average accuracy from 0.24 to 0.30, a relative gain of 24.6%. The accompanying theoretical analysis is also interesting: it compares context to associative-memory storage, the question to a retrieval cue, attention to cue-trace association, and replay to trace reactivation — and additionally proves the monotonic improvement of recursive replay on hidden representations. Two practical boundaries worth flagging: first, you must have access to attention weights, so black-box APIs are out — only self-hosted Qwen \u002F Llama3 instances qualify; second, it is naturally incompatible with DAC and summary-style compression — running both at the same time causes \"double replay \u002F conflicting evidence-selection standards\" — the paper doesn't provide a fix, so you must turn off upstream compression before going to production. ReContext is a rare \"almost free\" inference optimization, but whether it can actually land in RAG, long code agents, and compliance-audit pipelines depends on whether engineering is willing to leave it an attention hook.","recontext-recursive-evidence-replay","2026-07-02T17:59:26Z","2026-07-14T18:13:39.665080Z","2026-08-19T02:08:40.142862Z",true,"agent",141,[34,43],{"slug":35,"tag_slug":35,"title_zh":36,"title_en":37,"intro_zh":38,"intro_en":39,"id":40,"is_active":30,"created_at":41,"modified_at":42},"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":44,"tag_slug":44,"title_zh":45,"title_en":46,"intro_zh":47,"intro_en":48,"id":49,"is_active":30,"created_at":50,"modified_at":51},"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":53},[54,59,64,69,74,79],{"id":55,"title":56,"news_slug":57,"published_at":58},"183fb3be-e062-47e7-9591-7c2372e116c1","LLM 蒸馏的显存瓶颈不只在教师模型：离线 Top-K 与分块 KL 把长上下文训练装回单卡","llm-distillation-offline-top-k-chunked-kl","2026-08-05T20:08:13+00:00",{"id":60,"title":61,"news_slug":62,"published_at":63},"217f417d-1b9c-475b-99f4-e21e7c909711","MHAR 把 Transformer 残差流从「单车道」拆成 H 条独立路由:子空间第一次有权自己挑历史层","multi-head-attention-residuals-mhar","2026-08-01T07:30:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"a1ab01f3-ef5b-4240-aa99-7738f48591aa","PReM 用「按需刷新」撕开 LLM 长上下文压缩天花板:阿里团队 32K 上下文做到 16×\u002F32× 压缩仍保住多跳推理","prem-on-demand-refresh-32k","2026-07-18T20:08:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"bdb819d1-09a0-4320-8f78-04dccb15571d","16GB 显卡微调 131K 上下文：Hierarchical Global Attention","hierarchical-global-attention-16gb","2026-07-18T18:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"0049e060-f142-4b32-b655-5734b67b346f","Kernels 大重构:把 GPU kernel 升级为 Hub 一等公民,LLM 基础设施开始标准化","hf-kernels-hub-first-class","2026-07-10T04:01:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"40509ac7-c443-44f9-99a0-f90b78121d1f","陶哲轩的「Big Mathematics」:LLM 推理 + 形式化重塑数学研究的协作信任机制","tao-big-mathematics-llm-formalization","2026-06-25T02:00:00+00:00"]