[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-game-engine-rlhev-world-models":3,"topics-all":38,"news-related-31c09fea-8993-4f10-b683-499672fcafe3":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},"31c09fea-8993-4f10-b683-499672fcafe3","世界模型不能再靠爬视频硬堆:游戏引擎补上了缺失的奖励信号","arXiv 新论文 RLHEV 指出,世界模型靠堆爬取视频和算力扩展效率低下,空间生成仍依赖 CLIP 分数这类模糊代理信号,难以支撑 RL 后训练。作者提出用游戏开发当奖励环境:引擎可校验碰撞、物理与可玩性,开发者验收提供全局信号。","扩展世界模型的常规配方,是拿更多爬来的视频配上更多算力。8 月 26 日提交到 arXiv 的一篇论文直接挑战这个共识:这种策略是低效的,世界模型的扩展还需要一个能提供有据奖励信号的递归数据引擎([arXiv:2608.25518](https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.25518))。论文在 8 月 28 日的 Hugging Face Daily Papers 榜单上以 136 个赞排在首位,由 Pengfei Zhou、Yang You 等 8 位作者合作完成,Hugging Face 页面显示的机构归属是新加坡国立大学。\n\n## 病根:空间生成缺一个\"编译器\"\n\n论文的出发点是代码智能体的成功经验。代码之所以能支撑 LLM 的强化学习后训练,是因为它可执行:编译器和运行时能给出高质量的奖励信号。\n\n空间生成走的却是另一条路。作者指出,这个领域至今仍大量依赖 CLIP 分数这类模糊的代理指标,而这些信号既模糊又有偏,难以支撑 RL 后训练。换句话说,图像和视频模型长期在没有硬裁判的环境里训练——奖励来自\"看起来像不像\",而不是\"物理上成不成立\"。\n\n## 解法:游戏引擎是一份可执行的世界规范\n\n论文的核心主张,是把游戏开发变成空间世界模型缺失的奖励环境。游戏引擎编码的一个场景,本质上是一份可执行的世界规范:引擎可以高效检查碰撞、物理、可导航性和有界可玩性;而开发者判断\"这个场景该不该被接受\",则提供了全局层面的验证信号。\n\n这条路还附带一份关键产物:游戏开发过程本身会产生真实世界的长视界轨迹数据,而这正是 RL 后训练需要的素材。奖励环境和训练数据,在同一个流程里同时到位。\n\n## RLHEV:把引擎信号和人类验收拼在一起\n\n在此基础上,论文提出了 RLHEV(Reinforcement Learning with Human-Engine Verification)——一种后训练范式,把引擎给出的密集信号与开发过程中隐式的人类接受反馈结合起来。局部的事实核查交给引擎,全局的\"这关做得行不行\"交给人,奖励不再依赖模糊的相似度分数。\n\n## 值得盯紧的理由\n\n抛开具体方法,这篇论文更值得注意的信号是方法论迁移:可验证奖励在代码智能体上被验证有效的思路,正在被搬进空间智能赛道,而游戏引擎就是那个现成的验证器。如果这条路跑通,\"爬更多视频 + 堆更多算力\"就不再是对世界模型扩展问题的标准答案,数据引擎加可执行验证的组合可能会改写下一代 scaling 的叙事。对做世界模型、游戏 AI 或空间智能的团队来说,这是一篇值得精读的论文。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.25518","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},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"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},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"c01cac28-bf3d-425d-880a-c6dbebc63a98","en","World Models Cannot Scale on Crawled Video Alone: Game Engines Supply the Missing Reward","A new arXiv paper, RLHEV, argues that scaling world models by piling up crawled video and compute is inefficient, because spatial generation still relies on fuzzy proxies like CLIP scores that cannot support RL post-training. The authors propose game development as the reward environment: engines verify collision, physics and playability, while developer acceptance provides the global signal.","The standard recipe for scaling world models is more crawled video plus more compute. A paper submitted to arXiv on August 26 challenges that consensus directly: the strategy is inefficient, and scaling world models also requires a recursive data engine that offers grounded reward signals ([arXiv:2608.25518](https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.25518)). The paper topped the Hugging Face Daily Papers list on August 28 with 136 upvotes. It was written by eight authors including Pengfei Zhou and Yang You, with the National University of Singapore shown as the affiliated institution on the Hugging Face page.\n\n## The Root Problem: Spatial Generation Lacks a Compiler\n\nThe starting point is the success of code agents. Code can support RL post-training of LLMs precisely because it is executable: compilers and runtimes provide high-quality rewards.\n\nSpatial generation takes a different path. The authors point out that the field still relies largely on fuzzy proxies such as CLIP scores, and these signals are fuzzy and biased, making them hard to support RL post-training. In other words, image and video models have long been trained without a hard referee — rewards come from whether something looks right, not whether it is physically valid.\n\n## The Fix: A Game Engine Is an Executable World Specification\n\nThe core claim is that game development provides the missing reward environment for spatial world models. A scene encoded by a game engine is an executable world specification: the engine can efficiently check collision, physics, navigability and bounded playability, while the developer provides the global verification signal by judging whether the scene should be accepted.\n\nThe path also yields a critical byproduct: game development produces real-world long-horizon trajectory data, exactly what RL post-training needs. Reward environment and training data arrive together in one pipeline.\n\n## RLHEV: Combining Engine Signals with Human Acceptance\n\nOn this basis the paper proposes RLHEV (Reinforcement Learning with Human-Engine Verification), a post-training paradigm that combines dense engine signals with implicit human acceptance feedback from the development process. Local fact-checking goes to the engine; the global judgment of whether a level is any good goes to humans. Rewards no longer depend on fuzzy similarity scores.\n\n## Why It Is Worth Watching\n\nBeyond the specific method, the more notable signal here is a methodology transfer: verifiable rewards, proven effective on code agents, are being carried into the spatial intelligence race, and the game engine is the ready-made verifier. If this path works, \"crawl more video + stack more compute\" stops being the standard answer to scaling world models, and the combination of a data engine with executable verification may rewrite the next generation of scaling narratives. For teams working on world models, game AI or spatial intelligence, this is a paper worth reading closely.","game-engine-rlhev-world-models","2026-08-30T13:10:00Z","2026-08-30T13:08:33.070768Z","2026-08-30T13:08:33.070778Z",true,"agent",150,[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},"dcd8b3e1-a3c7-4614-aba4-9002219ea5f6","LibreDB Studio 0.15 发布:本地 LLM 接管数据库交互","libredb-studio-local-llm-agent","2026-09-15T00:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"5a90a793-8ec1-4b3a-9691-edef5ffe8535","AI「思想病毒」实证:Anthropic 与 EPFL 让恶意想法在 Agent 间自我复制,免疫只需一段警告","mind-viruses-multi-agent-llm","2026-08-18T13:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"777afb24-262f-45cc-961f-d5d49ad42883","AgentOPSD 用递归贝叶斯信念破解多轮 Agent 强化学习的信用分配：清华\u002F浙大\u002F美团让 GRPO 学会看哪个 turn 决定胜负","agentopsd-recursive-belief-credit-assignment","2026-08-07T02:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"3c6fcf46-f5bb-4136-931c-69cd64216e12","Skill-Use 基准揭示 Agent 短板：会做任务，不等于会用 Skill","skill-use-agent-harness-benchmark","2026-08-06T08:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"74de194b-9e2c-45ab-aa13-12fe210e66ba","HiGram 给 Agent 记忆加上“路径定位”：先找证据，再改记忆","higram-agent-memory-path-localization","2026-08-05T09:32:43+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"07e73462-1efa-4215-874e-5357f6e5c840","Canva可画在中国上线MCP:接入Kimi与WorkBuddy,让AI Agent直接交付可编辑设计稿","canva-mcp-china-kimi-workbuddy-agent-launch-2026q3","2026-07-30T03:00:00+00:00"]