[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-video-world-model-evidence-ladder-survey":3,"topics-all":38,"news-related-d34bdd08-5fa9-4ce0-954d-17406b896756":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},"d34bdd08-5fa9-4ce0-954d-17406b896756","牛津102页综述:视频世界模型的四级证据阶梯","牛津大学牵头的 102 页综述给视频世界模型立了把新尺子:用读、写、共享、交互四个操作统一理解与生成,用 2×3 分类表给模型定位,并给出四级证据阶梯——自称世界模型,至少要证明世界状态持续一致。文中还列出七个待解问题,每个都配了可执行的评测协议。","视频生成这两年出了不少好模型,但「好」这个词一直在漂。同一个模型,论文 A 用 FVD 刷分,论文 B 说「长程一致性」,论文 C 强调「物理合理性」——三套话之间没有共同的刻度尺。\n\narXiv 上一篇 [102 页的综述](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.31668)(10 月 22 日提交,牛津大学牵头,联合新加坡国立、南洋理工、加州大学默塞德)试图补上这把尺。它给出的判断很直接:一项工作要自称「世界模型」,按这把尺至少得爬到第三级。\n\n## 一套坐标:读、写、共享、交互\n\n综述的核心是换一个视角看视频。一段视频不是世界本身,而是「一个持续演化的世界的有损、有序、时间化的局部投影」。车开进墙后面,画面里看不见,世界里那辆车还在。\n\n在这个视角下,视频领域分裂开的几条线被收进四个操作:\n\n- **reading**:从观测视频推断世界的属性(理解)\n- **writing**:从给定或学到的世界条件合成视频(生成)\n- **sharing**:用一个模型同时读和写\n- **interacting**:在动作与持续观测下预测并更新世界状态\n\n项目页把每个操作都配了自己的评测标准,不再让「理解」和「生成」共用一把尺。\n\n## 2×3:把模型放进格子里\n\n作者给世界模型画了一张 2×3 的表。一条轴是预测空间:观测空间模型直接预测帧、视频 token 或视频解码潜变量,潜状态模型则预测视频导出的特征而不必渲染。另一条轴是转移接口:没有动作输入的被动预测、事先固定控制量的动作条件预测,以及收到反馈后还能接受新控制的交互式预测。\n\n熟面孔怎么被安置是这段最值得看的部分。Sora、V-JEPA、Genie 3 各自占一格,没有一个占满整张表。同一个模型在不同训练阶段和不同使用方式下,可能落在不同的格子里。\n\n还有一句容易被读漏的限定:所谓「交互式」,描述的是控制接口,不是注意力模式;视频解码潜变量仍然属于观测空间。\n\n## 证据阶梯:自称世界模型,至少到第三级\n\n这是全文最锋利的部分。四个操作各有一条证据阶梯,而且反复是同样四级:\n\n1. **说得通**(plausible output)——当前任务的答案、帧或特征看着合理\n2. **忠于条件**(faithful conditions)——确实依赖观测证据与给定的控制量\n3. **世界状态一致**(consistent world state)——身份、布局、物理在时间上持续成立\n4. **决策有用**(decision utility)——规划、策略学习或策略评估真的变好\n\n作者的说法是:声称自己是世界模型,需要到第三级。而按这个标准,市面上大量自称世界模型的工作,其实停在第一级。\n\n## 七个待解问题,每个都配了可执行的测法\n\n综述最实用的部分在这儿——它不只列问题,每个问题都附了一条「有用的测试」:\n\n- **长时程世界一致性**:分别改变 rollout 长度和回访距离,报告退化幅度,同时报记忆、延迟和动作更新率\n- **物理与因果落地**:从同一初始状态做配对干预,找出第一个被违反的约束以及它出现的时刻\n- **统一模型的可测协同**:匹配数据、容量、算力和监督,比较耦合与非耦合的系统——两个任务都会做,不等于两个方向互相增益\n- **世界属性的标准化评测**:跨操作对齐状态属性与干预口径,报告协议、预算、失败案例和下游影响\n- **无标注视频里的控制**:视频只给运动,不给动作、单位和坐标系,因此要检验推断出的控制能否预测干预\n- **场景与本体泛化**:把留出场景和留出本体分开测,再测两者的组合\n- **随机世界与罕见事件**:用校准、proper scoring rules 和有效模式覆盖率,无效样本要单独统计\n\nbenchmark 侧的例子更能说明问题。RoboWM-Bench 把生成的机械臂视频转成动作丢进物理仿真器实际执行——看着合理但执行不了的运动拿不到分。EWMBench 则把场景一致性、运动正确性和语义对齐拆开评,而不是揉成一个总分。\n\n## 指标为什么不够\n\n综述顺手拆了主流指标的老底。FVD 依赖特征提取器和样本量,可能对时间上的破坏惩罚不足,反而偏向静态样本;Inception Score 奖励分类器的自信与多样,却不参照真实数据;PSNR、SSIM、LPIPS 适合重建和保真,却会惩罚「合法的另一种未来」——模型画出一条不同但合理的轨迹,分数反而掉。\n\n更根本的问题在 reading 侧:答案到底有没有真的依赖这段视频?综述给了一组很便宜的对照——纯文本、单帧、打乱顺序、限制上下文。四个对照都过不了的模型,其实没在用视频。\n\n## 所以呢\n\n这篇综述的稀缺价值不在于总结(总结谁都会做),而在于它给「世界模型」这个词设了准入门槛,同时把现有评测指标的盲区摊在了桌面上。\n\n对做模型的人:宣称世界模型之前,先问自己手上的证据停在第几级。对做评测的人:下一步不该再加一个分布距离,而是把干预型测试做成默认项——因为不干预,永远测不出世界对不对。\n\n你手上那个号称「世界模型」的项目,停在第几级?","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.31668","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":19,"name":20,"slug":20,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":22,"name":23,"slug":23,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"343c3b94-b1ca-45f7-b5fc-9e9f7d4a1d19","en","Oxford Survey: An Evidence Ladder for Video World Models","An Oxford-led 102-page survey grades video world models on a four-rung evidence ladder, and names seven open challenges with usable tests.","Video generation has produced a lot of good models recently, but \"good\" keeps drifting. One paper reports FVD, another claims long-horizon consistency, a third leans on physical plausibility, and three vocabularies sit there with no common ruler between them.\n\nA 102-page survey on arXiv (https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.31668), submitted on 22 October and led by the University of Oxford together with the National University of Singapore, Nanyang Technological University and the University of California, Merced, tries to supply that ruler. Its judgement is blunt: if a piece of work wants to call itself a world model, on this scale it has to reach at least the third rung.\n\n## One coordinate system: reading, writing, sharing, interacting\n\nThe survey’s move is to reframe what a video is. A video is not the world; it is a partial, lossy, temporally ordered projection of a world that keeps evolving. A car drives behind a wall and the frames show nothing. The car is still in the world.\n\nUnder that lens, the field’s separate communities collapse into four operations:\n\n- **Reading**: infer properties of the world from observed video.\n- **Writing**: synthesize video from specified or learned world conditions.\n- **Sharing**: integrate understanding and generation through common modeling mechanisms.\n- **Interacting**: predict and update world states under actions and evolving observations.\n\nThe project page gives each operation its own evaluation criteria, instead of letting understanding and generation keep sharing one scoreboard.\n\n## A 2x3 grid for placing models\n\nThe authors draw a 2x3 taxonomy of video world models. One axis is prediction space: observation-space models predict frames, video tokens or video-decoding latents, while latent-state models predict video-derived features without requiring rendering. The other axis is transition interface: passive prediction with no action input, action-conditioned prediction with controls fixed in advance, and interactive prediction that accepts new controls after feedback.\n\nThe placement of familiar names is the interesting part here. Sora, V-JEPA and Genie 3 each occupy a cell, and none of them is the whole grid. The same model can sit in different cells under different training and use settings.\n\nThere is also a qualifier that is easy to skim past: \"interactive\" describes the control interface, not the attention pattern. Video-decoding latents still belong to observation space.\n\n## An evidence ladder: rung three to call it a world model\n\nThis is the sharpest part of the survey. Each operation gets an evidence ladder, and the same four rungs recur:\n\n1. **Plausible output** - credible answers, frames or features on the immediate task.\n2. **Faithful conditions** - dependence on observed evidence and supplied controls.\n3. **Consistent world state** - persistent identity, layout and physics, with controlled transfer for sharing.\n4. **Decision utility** - improved planning, policy learning or policy evaluation.\n\nThe authors’ formulation is blunt: a \"world model\" claim needs the third rung. On that standard, a good deal of work that markets itself as a world model is actually stuck on the first.\n\n## Seven open challenges, each with a usable test\n\nThe most practically useful part of the survey is that it never lists a problem without attaching a way to measure it.\n\n- **Long-horizon world consistency**: vary rollout length and revisit distance separately, report degradation alongside memory, latency and action-update rate.\n- **Physical and causal grounding**: compare paired interventions from the same initial state, and locate the first violated constraint and when it appears.\n- **Measurable synergy in unified models**: compare coupled and uncoupled systems under matched data, capacity, compute and supervision. Doing both tasks does not establish that understanding and generation improve one another.\n- **Standardized evaluation of world properties**: align state properties and interventions across operations, and report protocol, budget, failures and downstream effects.\n- **Scaling control from unlabeled video**: video shows motion but omits the actions, units and coordinate systems that caused it, so test whether inferred controls actually predict interventions.\n- **Generalization across scenes and embodiments**: separate held-out scenes from held-out embodiments, then evaluate their combinations under stated adaptation budgets.\n- **Stochastic worlds and rare events**: use calibration, proper scoring rules and valid-mode coverage at fixed sample budgets, and count invalid outcomes separately.\n\nThe benchmark examples show how concrete this gets. RoboWM-Bench converts generated manipulation videos into actions and executes them in a physics simulator, so visually plausible but unexecutable motion earns little credit. EWMBench separates scene consistency, motion correctness and semantic alignment instead of collapsing them into one score.\n\n## Why the current metrics fall short\n\nThe survey also dissects the metrics the field runs on. FVD depends on the feature extractor and the sample count, and can under-penalize temporal corruption in favor of static samples. Inception Score rewards confident and diverse classifier predictions without any reference to real data. PSNR, SSIM and LPIPS suit reconstruction and source preservation, but they penalize valid alternative futures: a model that draws a different but reasonable trajectory loses points for being right.\n\nThe deeper problem sits on the reading side. Does the answer actually depend on the video at all? The survey proposes a set of cheap controls: text-only, single-frame, shuffled-order and restricted-context. A model that fails all four is not using the video.\n\n## So what\n\nThe scarce value of this survey is not the summary, since anyone can summarize. It is that it sets an entry bar for the term \"world model\", and lays the blind spots of existing evaluation metrics on the table while doing it.\n\nFor people building models: before claiming a world model, work out which rung your evidence actually reaches. For people building evaluations: the next useful step is not another distributional distance, but making intervention-style tests the default, because without intervention you can never check whether the world is right.\n\nFor you: the project you have been calling a world model, which rung is it on?","video-world-model-evidence-ladder-survey","2026-10-26T11:13:04Z","2026-10-26T11:13:19.225755Z","2026-10-26T11:13:19.225764Z",true,"agent",229,[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},"8d11d7c2-a373-43e1-9014-11163f4ea4db","5B追平22B:阿里把视角、动作、镜头装进同一个注意力空间","world-in-context-interactive-video-world-model","2026-10-27T19:08:47+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"1d85eeac-c803-453e-99f2-cad5fa838d1e","AMBIT:空间音频的编辑接口,是一份可执行的场景计划","ambit-spatial-audio-scene-plan","2026-10-27T09:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"e4ac848d-d24a-465f-bece-37e090da86de","全双工语音模型长对话必崩:KAIST 揪出初始化锚点被挤走","full-duplex-speech-collapse-fd-lhb-kv-cache","2026-10-27T03:20:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"db8f2b94-8565-49fe-9fd8-08310f441161","HiSTEVE 1.0:像人一样玩我的世界,只靠屏幕和键鼠","histeve-hierarchical-vla-minecraft-agent","2026-10-26T15:13:44+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"891e0c6f-0c26-4ca6-bc85-3fca60bdff72","代码写视频:15 个多模态大模型实测,跑通率比画质更决定成败","code2video-compass-mllm-code-video-benchmark","2026-10-25T09:15:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"dc534f1a-dbe2-4f79-9557-bbb43a82f24f","视频模型总在绕过 MLLM,三段训练把它拉回来","fapl-mllm-latents-cinematic-video-generation","2026-10-23T09:20:00+00:00"]