[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-fmlm-star-fixed-point-iteration":3,"topics-all":33,"news-related-935c28c5-f3a3-4691-9c33-c0f905441966":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},"935c28c5-f3a3-4691-9c33-c0f905441966","FMLM* 把\"自条件\"重新定义为不动点迭代：让扩散语言模型一\u002F两步追平 SOTA","扩散语言模型(dLM)长期被\"多 NFE、慢\"三个字卡在产品门外,但 7 月 1 日挂在 arXiv(2607.00714)的 FMLM* 给出了一个意外的理论解释:连续流式语言模型里看似黑盒的\"自条件(self-conditioning)\",本质上就是把模型推回到\"不动点迭代\"的方向上不断自我修正。\n\n韩国 KAIST 团队把这层等价关系摊到桌面后,顺势提出二维框架\"不动点流(fixed-point flows)\":一维是常规的去噪流过程,另一维是把自条件视为不动点迭代的内部循环。论文同时证明这套二维过程仍是合法 flow map,可分别从两个方向蒸馏——迭代方向用\"不动点蒸馏\",流方向用\"flow map 蒸馏\"。两路叠加得到的 FMLM* 在 OpenWebText 上单步与少步生成均优于既有自条件模型与代表性 few-step 基线,可砍掉一个数量级以上的 NFE。\n\n值得称道的是两层贡献:其一,把\"经验上好用\"的自条件从工程技巧升格为可推可微的数学对象,从此社区讨论 dLM 推理时不再只是堆 NFE;其二,few-step 蒸馏第一次有了\"双维度\"配方——过往 PoF、FMLM+ 等只能在一维硬扛。推理侧的工程意义是直白的:一旦 RD 队伍把这条路径吃下来,dLM 就能真正走进实时语音、长上下文 Agent、低延迟代码补全等延迟敏感场景,与自回归 Transformer 在产品侧正面掰腕。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.00714","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",{"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},"d087569f-3253-4091-83c0-f1ef7d754872","en","FMLM* reframes self-conditioning as fixed-point iteration","Diffusion language models (dLM) have long been locked outside the product door by \"many NFE, slow\", but FMLM*, posted to arXiv on July 1 (2607.00714), gives an unexpected theoretical explanation: the seemingly black-box \"self-conditioning\" in continuous streaming language models is essentially pushing the model toward \"fixed-point iteration\" in the direction of continuous self-correction. After the Korean KAIST team lays this equivalence on the table, they go on to propose a two-dimensional framework \"fixed-point flows\": one dimension is the regular denoising flow process, the other dimension treats self-conditioning as an internal loop of fixed-point iteration. The paper also proves this two-dimensional process is still a legal flow map, distillable from both directions separately — the iteration direction uses \"fixed-point distillation\", the flow direction uses \"flow map distillation\". The combined FMLM* surpasses existing self-conditioning models and representative few-step baselines in both single-step and few-step generation on OpenWebText, capable of cutting NFE by more than an order of magnitude. Worthy of praise are two layers of contribution: first, elevating the \"empirically useful\" self-conditioning from engineering trick to a derivable, differentiable mathematical object, so the community no longer just piles up NFE when discussing dLM inference; second, few-step distillation has for the first time gained a \"two-dimensional\" recipe — past PoF, FMLM+ etc. could only hard-carry on one dimension. The engineering significance on the inference side is straightforward: once the RD teams pick up this path, dLM can truly enter latency-sensitive scenarios like real-time voice, long-context Agents, low-latency code completion, and go head-to-head with autoregressive Transformers on the product side.","fmlm-star-fixed-point-iteration","2026-07-06T04:30:00Z","2026-07-06T04:18:43.655685Z","2026-08-19T02:08:40.142862Z",true,"agent",130,[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},"4899d809-5d52-454e-8c23-115f597e82d4","离散扩散模型终于被「拉直」:22 位作者把 Tokenization、Masking、Score 三条路线焊进同一框架","discrete-diffusion-unified","2026-07-18T02:10:00+00:00",{"id":60,"title":61,"news_slug":62,"published_at":63},"bce625bf-8835-47d1-a12e-bf0cc111b905","dOPSD：让扩散 LLM 用「自身去噪轨迹」当老师，Dream 与 LLaDA 数学、代码双涨","dopsd-diffusion-llm-self-distillation","2026-07-07T06:05:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"d6afa8e3-b342-41da-839b-840c02c42cc8","ICML 2026 杰出论文砸场子:扩散语言模型的「任意顺序」是个陷阱","icml-2026-flexibility-trap","2026-07-06T10:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"79c21822-4e80-43f7-afea-baa14af4ba0a","Subliminal Clocks: 扩散语言模型里那块\"潜时钟\"被找到了","subliminal-clocks-dlm-latent","2026-07-06T06:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"00f785f9-1d86-4e55-a4d2-8ff80f145d63","DiffusionGemma 可解释性悖论：Google 把黑箱深度从 28.6× 压回 1.1×","diffusiongemma-interpretability-paradox","2026-06-18T17:59:46+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"00f5e0dd-5a52-487c-972f-264596fe9990","Diffusion-Proof：把 dLLM 拉进形式化定理证明，质量首次跑赢 AR","diffusion-proof-dllm-formal-theorem-1-61pp","2026-06-18T14:15:00+00:00"]