[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-s-space-multimodal-spatial-workspace":3,"topics-all":38,"news-related-b95bf7cf-e962-49e8-847c-2e45c917f19d":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},"b95bf7cf-e962-49e8-847c-2e45c917f19d","S-Space:多模态模型看得懂空间,却转不过弯","MirroS Lab 团队从多模态模型中间层激活里拆出三维线性子空间 S-Space:物体 token 沿水平、垂直、距离三轴编码连续坐标,非空间提问下依然可读;因果干预能定向改写空间判断而保留其他回答。诊断显示瓶颈不在感知在变换,显式计算该子空间可超过模型原生推理,代码以 Apache 2.0 开源。","多模态大模型的空间推理时灵时不灵:让它判断图里两个物体的相对位置,有时一次答对,有时错得离谱,而错在哪一直说不清。MirroS Lab 团队 10 月 9 日挂上 arXiv 的论文给了一个反直觉的答案:模型不但有空间感,还在内部维护着一套三维坐标系,真正的短板在别处([arXiv:2610.13527](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13527))。论文上线三天登上 Hugging Face 论文日榜第 2,拿到 199 个 upvote。\n\n## 中间层里藏着一个三维坐标系\n\n研究者在多模态模型的中间层激活中拆出一个线性子空间,取名 S-Space:物体 token 的表征在其中沿水平、垂直、距离三根轴编码连续坐标——不是离散的「左\u002F右」标签,而是能读出数值的连续位置。发现方法相当工程化:对空间问答得分关于物体 token 激活求梯度,把影响答案最大的方向找出来。更关键的是这套坐标是内生的:换成非空间提问它依然可读,不是特定提示词诱导出来的假象。\n\n## 一个能被干预的「空间工作台」\n\n论文把这个子空间称为模型的内部空间工作台,证据分三层。感知往里写:物体位置被编码成空间表征,直接读出即可支撑空间判断。推理在里面转:视角变化时,内部推理对该表征做变换。语言输出会选择性地访问它:因果干预改写 S-Space 坐标,能定向改变模型的空间判断,同时大体保留非空间回答——这是「工作台」最硬的证据。案例研究还显示,它会给没见过的实体和观察者自己分配位置;方向词的读出则暴露出方位参照系与图像参照系的系统性耦合。\n\n## 病根不在感知,在变换\n\n诊断价值最大的是感知-计算分解:多模态模型能形成相当强的空间表征,却在跨视角推理要求的空间变换上掉链子。仓库 README 的表述很直白——对 S-Space 做显式计算,可以超过模型的原生推理。换句话说,模型不是没有空间感,而是不会可靠地调用它。\n\n## 开源与可复现\n\n代码以 Apache 2.0 协议开源于 GitHub,已积累 247 星,预训练的 S-Space 轴随仓库直接发布;实验覆盖 Molmo2-ER、MolmoAct2、Qwen3.5-4B、Qwen3.6-27B 等模型,基准包括 COCO、CV-Bench、EmbSpatial、SpinBench、HStar,复现只需跑仓库自带脚本。\n\n这篇论文的价值不止于一个可解释性发现。它把「空间题翻车」拆成了可判定的两种病:表征没建成,还是变换没做对。对具身智能和机器人团队,这个区分尤其值钱——视角变换正是移动智能体的日常功课。对模型厂商的启示或许更扎心:与其指望思维链自发学会空间变换,不如把显式空间计算直接做进架构里。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13527","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"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},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"62e17521-326c-4cfd-b52f-1e8ae881090b","en","S-Space: MLLMs encode coordinates but fail viewpoint transforms","MirroS Lab: a 3-axis coordinate subspace in MLLM activations; editing it rewrites spatial judgments. Bottleneck: transformation, not perception.","Multimodal language models are maddeningly inconsistent at spatial reasoning: ask whether one object sits to the left of another and the answer may land right or wrong with no obvious pattern. A paper posted to arXiv on October 9 by the MirroS Lab team offers a counterintuitive diagnosis: these models do have spatial sense — they maintain an internal three-dimensional coordinate system — and the real weakness lies elsewhere ([arXiv:2610.13527](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13527)). Within three days the paper climbed to #2 on the Hugging Face Daily Papers board with 199 upvotes.\n\n## A 3-axis coordinate system hidden in intermediate layers\n\nThe researchers isolated a linear subspace in the models' intermediate activations, naming it S-Space: object-token representations encode continuous coordinates along horizontal, vertical, and distance axes — not discrete left\u002Fright labels but numeric positions that can be read out. The discovery method is refreshingly practical: differentiate spatial-answer scores with respect to object-token activations and keep the directions that matter most. Crucially, the coordinates are intrinsic — they remain readable under non-spatial questions, so they are not an artifact of particular prompts.\n\n## A spatial workspace you can intervene on\n\nThe paper frames this subspace as an internal spatial workspace, with three layers of evidence. Perception writes to it: object locations are encoded as spatial representations that directly support accurate spatial judgments. Reasoning transforms it: when the viewpoint changes, internal reasoning operates on the representation. Verbal responses access it selectively: causal interventions that rewrite S-Space coordinates change spatial judgments while largely preserving non-spatial answers — the hardest evidence that this is a genuine workspace. Case studies show it assigns locations to unseen entities and to the observer, and direction-word readouts expose systematic coupling between cardinal and image-relative reference frames.\n\n## The bottleneck is transformation, not perception\n\nThe most diagnostic result is a perception-computation decomposition: multimodal models form capable spatial representations yet fall short on the spatial transformations that reliable viewpoint-change reasoning demands. The repository README states it plainly — explicit computation over S-Space can outperform native reasoning. In other words, the models are not missing spatial sense; they fail to invoke it reliably.\n\n## Open source and reproducible\n\nThe code is on GitHub under Apache 2.0 with 247 stars, and pretrained S-Space axes ship with the repository. Experiments cover Molmo2-ER, MolmoAct2, Qwen3.5-4B and Qwen3.6-27B, with benchmarks including COCO, CV-Bench, EmbSpatial, SpinBench and HStar; reproduction is a matter of running the bundled scripts.\n\nBeyond being an interpretability finding, the paper turns vague spatial failures into two decidable failure modes: was the representation never built, or was the transformation never performed? For embodied-AI and robotics teams — whose agents live and die by viewpoint changes — that distinction is worth money. And for model builders the implication stings: rather than hoping chain-of-thought spontaneously learns spatial transformation, wire explicit spatial computation into the architecture.","s-space-multimodal-spatial-workspace","2026-10-12T17:19:44Z","2026-10-12T17:19:56.188776Z","2026-10-12T17:19:56.188784Z",true,"agent",72,[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},"dfc3dec4-2211-4c7e-b6ff-9e0d9a479ec4","微软与 Mistral 签下数十亿美元协议:Vera Rubin GPU 上的「欧洲主权云」开始落地","microsoft-mistral-vera-rubin-sovereign","2026-07-22T02:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"b2c478e8-dbc2-43c3-a941-a763ee429bd4","V-Zero:把「证据对比」塞进蒸馏,让多模态大模型不再借语言先验蒙混过关","v-zero-evidence-gated-distillation-multimodal","2026-06-24T12:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"84383156-60d6-4627-8c05-863686241eea","腾讯混元 UniRL 框架开源：把「统一多模态」塞进同一个 RL 训练循环，DRPO \u002F Flow-DPPO \u002F CPPO 三连发","tencent-unirl-drpo-flow-dppo-cppo","2026-06-14T12:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"eeb3f139-8df8-40d7-a1e6-82e49b668547","VideoJev:视频问答提速100倍,不生成只打分","videojev-system-one-video-understanding","2026-10-14T15:09:17+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"9fcb5d74-315e-4d57-a7da-290fbfb9e91c","阿里 Qwen 发布弹性 RL 框架 QwenGyre,训练长时 Agent 提速 1.85x","alibaba-qwen-qwengyre-elastic-rl-xlong-agents","2026-10-14T00:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"5d4acf64-67d8-404b-9d3f-cdf7761fb8ac","小鹏开源 GroundingPI:4B 视觉 grounding 模型自评超 GPT-6 Astra","groundingpi-4b-visual-grounding-foundation-model","2026-10-13T17:10:26+00:00"]