[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-weride-witt-atomic-physics":3,"news-related-cc43635e-9f4c-4007-b9e9-347b02f67a76":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"cc43635e-9f4c-4007-b9e9-347b02f67a76","文远知行 WITT:用\"原子级物理事实\"重写自动驾驶的数据飞轮","文远知行在 WAIC 期间发布物理 AI 认知基础大模型 WITT(World Intelligence Toward Truth),把\"理解真实世界\"拆成可被识别和验证的\"原子级物理事实\"(Atomic Physical Facts, APFs)。\n\nWITT 借鉴维特根斯坦\"世界是事实的总和\":模型不再把一段驾驶视频当整体学习,而是先识别\"自车右转、信号灯切换、行人横穿\"这类最小事实单元,再围绕这些事实做提取、推理、验证、编排。流水线被拆成事实提取、事实推理、事实验证、事实编排四件事,并配套\"6+1\"事实验证维度,给自动驾驶场景里常见的幻觉、遗漏、时序错位提供量化抓手。\n\n效率层面:相较百 B 级参数的通用大模型,WITT 可节省约 98% 的 Token 成本,单卡单日处理 1 万分钟车辆视频,数据处理效率最高提升 200 倍,平均每片段事实错误率约为通用大模型的三分之一。\n\n闭环意义在 WITT 与文远自研世界模型 GENESIS 共同构成的\"物理 AI 飞轮\":前者从真实数据中萃取、验证事实,后者据此生成高保真仿真与长尾场景。文远的护城河不只是 3000+ 辆 L4 Robotaxi,而是能把这支车队每天吐出的视频持续变成\"可被验证的事实\"——这是 L4 与 L2++ 数据能在同一套认知底座上共用的前提,也是国内同行最难抄的一段。","https:\u002F\u002Fwww.ithome.com\u002F0\u002F978\u002F055.htm","9d8c4f57-af5c-4825-9ecd-e01964415e13",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"81dc2a50-e913-4c43-89a2-f8c9ac27acfd","en","WeRide's WITT rewrites autonomy's data flywheel","During WAIC, WeRide released WITT (World Intelligence Toward Truth), a physical-AI cognitive foundation model that breaks \"understanding the real world\" down into recognizable and verifiable \"Atomic Physical Facts\" (APFs). WITT draws on Wittgenstein's \"the world is the totality of facts\": instead of treating a driving video as a whole to learn from, the model first identifies minimal fact units such as \"ego vehicle turns right, traffic light switches, pedestrian crosses\", then performs extraction, reasoning, verification, and orchestration around those facts. The pipeline is split into four tasks — fact extraction, fact reasoning, fact verification, fact orchestration — paired with a \"6+1\" fact-verification dimension, giving a quantitative grip on the hallucinations, omissions, and temporal misalignments common in autonomous-driving scenarios. On the efficiency side: compared to 100B-parameter general-purpose models, WITT saves about 98% in token cost, processes 10,000 minutes of vehicle video per single card per day, lifts data-processing efficiency by up to 200x, and keeps the average per-segment fact error rate to about one-third that of general-purpose models. The closed-loop meaning lies in the \"physical-AI flywheel\" that WITT forms together with WeRide's in-house world model GENESIS: the former extracts and verifies facts from real data, the latter generates high-fidelity simulation and long-tail scenarios on that basis. WeRide's moat is not just its 3,000+ L4 Robotaxis, but the ability to continuously turn each fleet's daily video output into \"verifiable facts\" — a prerequisite for L4 and L2++ data to share a common cognitive foundation, and the hardest piece for Chinese peers to copy.","weride-witt-atomic-physics","2026-07-17T08:00:00Z","2026-07-17T08:07:42.891479Z","2026-08-19T02:08:40.142862Z",true,"agent",87,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"a151db0c-d832-4df2-ac03-2d4e58b26e99","Kimi K3 跑通 MiniTriton:Moonshot 让 LLM 第一次从零编译出自己的 GPU 编译器","kimi-k3-minitriton-gpu-compiler","2026-07-26T14:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"7c5d7da6-2bdb-4174-b28b-d7fef5579181","商汤 SenseNova U1 Pro 把多模态 AI 卷出「长程交付」:从「好看」走向「可用」的赛道切换","sensetime-sensenova-u1-pro","2026-07-19T10:01:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"f22a15d1-07a0-4318-a5de-6cc794a54d8b","全国首个具身世界模型「我悟」拿下备案：北京人形用 Pelican-VL × WoW 打开 API 商业化通道","beijing-humanoid-wow-pelican-vl-embodied","2026-06-27T10:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"caa54bff-d57a-411c-9dae-43f1d4d46875","DeepSeek V4 重磅登场：长期记忆技术突破重塑AI能力边界","deepseek-v4-engram-ltm-long-term-memory-87pct","2026-04-22T07:05:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"f6e4aab0-7693-4c2c-bb66-c1641fc2cc3e","Ox Alpha 谜底揭晓:智谱 GLM-5.3-Flash,MIT 开源 320B MoE","ox-alpha-glm-5-3-flash-reveal","2026-08-27T13:30:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"804ab59a-a8d6-4b61-bf74-8f6f2bdae83c","智谱把 Flash 做成一件正经事:一次说清 GLM-5.3-Flash 的架构和 benchmark 真相","glm-5-3-flash-hybrid-attention-architecture","2026-08-27T08:00:00+00:00"]