[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-arc-autoregressive-video-fcp-future-chunk-prediction":3,"topics-all":38,"news-related-384bb931-49cb-4b1c-b1a8-89495a166699":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},"384bb931-49cb-4b1c-b1a8-89495a166699","腾讯ARC Lab:自回归视频的捷径问题,FCP零成本修补","自回归视频扩散的teacher forcing训练存在捷径问题:相邻帧太相似,模型靠局部连续性就能完成去噪,无需真正建模物体如何演化。南京大学、腾讯IEG ARC Lab与浙江大学团队提出FCP,从当前块中间特征预测相隔较远的视频块。5B模型图生视频IQ从30.4提到33.8,推理端零额外开销。","自回归视频扩散已经是长视频生成的默认路线:一块一块往后接,块与块之间保持因果注意力。但它有个没人明说的漏洞——teacher forcing 训练时,相邻几帧实在太像了,模型只要把上一块的像素\"抹\"过来就能完成当前块的去噪,压根不需要理解物体接下来会怎么动。\n\n10 月 21 日,南京大学、腾讯 IEG ARC Lab 与浙江大学团队在 arXiv 上公开了 Looking Ahead(arXiv:2610.29989),把这个问题当成训练目标本身来修。\n\n## 把\"预测未来\"写进训练目标\n\n作者的核心判断是:视频的时序冗余太高,局部连续性会变成一条捷径,让模型绕过对视觉演化的建模。FCP(Future Chunk Prediction)加了一个辅助目标——在与 teacher forcing 去噪损失联合优化的同时,从当前块的中间特征出发,预测时间上相隔较远的视频块,而不是只盯着相邻的下一块。\n\n三点设计值得注意。源表征只取当前块的特征,不喂完整历史;每个时间偏移量各配一个由三个 DiT 模块组成的轻量预测器,并行而非串联;辅助预测器里完全去掉文本交叉注意力——文本提示本身就可能泄露未来线索。\n\n## 数字:同设定下的对照\n\n论文的主张建立在受控对比上。作者从 Wan 的官方 checkpoint 出发,训练了 1.3B 和 5B 两档,配置与数据完全对齐。\n\n在 5B 上,teacher forcing 基线的图生视频 IQ 是 30.4,视频生视频是 40.6;加入 FCP 后分别变成 33.8 和 44.4。1.3B 上是 27.6 提到 30.9。消融表里能看到拆分:只加干净源表征 28.8,只加带噪源表征 29.9,两者合起来 30.9。\n\n更值得注意的是两组反向结果。串联设计在 1.3B 上只到 30.1,低于并行的 30.9;到了 5B,两个串联变体在图生视频上只是勉强超过基线,视频生视频则完全没有改善——论文的解读是,大模型更善于利用串联路径里泄露的信息。另一组是给辅助预测器加文本条件,IQ 反而掉了 1.1 分。\n\n## 零推理成本\n\nFCP 的模块在推理时全部丢弃,生成流程和基线完全一致,零额外开销。论文还与 Next Forcing 那类\"用完整历史做条件\"的方案做了对比:训练步数相同时两者分数接近,但只用当前块表征的版本达到同等性能所需的训练算力大约少 40% 到 50%。\n\n在 WorldModelBench、PAI-Bench-G 和 PhyGround 上,FCP 相对 teacher forcing 均有提升,增幅从 0.03 到 2.39 不等,说明增益不是只在单一基准上成立。\n\n横评表里 FCP 的 5B 拿到 IQ 33.8,高于同表的 Wan2.2-14B(32.2),低于 Grok Imagine Video(34.8);但该表基线分数主要引自基准本身,只适合看量级。\n\n## 所以呢\n\n世界模型这条线上,大家比的是谁能生成更长的视频、多少卡、什么分辨率。这篇工作的切口小得多:不动架构、不加参数、推理端零成本,只是换了个训练时该预测什么。\n\n代码和两档权重已放在 GitHub 上,作者在 H200 上验证过环境,单卡即可跑推理。值得看的不是 IQ 涨了 3.4 分,而是\"局部连续性会被当成捷径\"这个判断——它对任何高冗余的序列数据都成立。\n\n论文:https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.29989\n代码:https:\u002F\u002Fgithub.com\u002FTencentARC\u002FLooking-Ahead","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.29989","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"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},"82d2ef4c-9e67-40b5-8fe6-9a89de5fdcac","en","Tencent ARC Lab Fixes the Autoregressive Video Shortcut","Teacher forcing has models denoise from local continuity, not evolution. FCP predicts distant chunks, lifting the 5B model's IQ from 30.4 to 33.8","Autoregressive video diffusion is already the default route for long-video generation: generate chunk by chunk, with causal attention holding the chunks together. But it carries an unstated flaw. Under teacher forcing, adjacent frames are so similar that a model can smear the previous chunk's pixels across the current one and finish denoising, never needing to understand how objects are about to move.\n\nOn October 21, a team from Nanjing University, ARC Lab at Tencent IEG, and Zhejiang University published Looking Ahead (arXiv:2610.29989) on arXiv, treating the problem as a training objective in its own right.\n\n## Turning future prediction into a training target\n\nThe authors' core diagnosis: temporal redundancy in video is strong enough that local continuity becomes a shortcut, letting the model bypass modeling visual evolution entirely. FCP (Future Chunk Prediction) adds an auxiliary objective that, jointly optimized with the teacher-forced denoising loss, predicts temporally separated chunks from intermediate features of the current chunk, rather than only the next adjacent one.\n\nThree design choices stand out. Source representations come only from the current chunk, not the full history. Each temporal offset gets its own lightweight predictor of three DiT blocks, running in parallel rather than chained. And the auxiliary predictors drop text cross-attention entirely, since text prompts can themselves leak cues about the future.\n\n## The numbers, under matched settings\n\nThe paper's claims rest on controlled comparison. The authors start from official Wan checkpoints and train both a 1.3B and a 5B variant with fully matched configurations and data.\n\nOn the 5B model, the teacher-forcing baseline scores 30.4 IQ on image-to-video and 40.6 on video-to-video; with FCP those become 33.8 and 44.4. On the 1.3B model, 27.6 rises to 30.9. The ablation table breaks the contribution down: clean source representations alone reach 28.8, noisy source representations alone 29.9, and the two combined 30.9.\n\nTwo sets of negative results matter here. The chained design reaches only 30.1 on the 1.3B model, below the parallel 30.9. On the 5B, both chained variants barely clear the baseline on image-to-video and fail to improve video-to-video at all; the paper reads this as larger models being better at exploiting the information that leaks along the chained path. Adding explicit text conditioning to the auxiliary predictors, meanwhile, drops IQ by 1.1.\n\n## Zero inference cost\n\nEvery FCP-specific module is discarded at inference, leaving the rollout identical to the baseline with no added overhead. The paper also compares against Next Forcing-style approaches that condition on the full history: at matched training steps the two score comparably, but the current-chunk-only design reaches the same performance with roughly 40% to 50% less training compute.\n\nOn WorldModelBench, PAI-Bench-G, and PhyGround, FCP beats teacher forcing across the board, with gains ranging from 0.03 to 2.39, so the improvement is not confined to a single benchmark.\n\nIn the model-ranking table, the FCP 5B reaches IQ 33.8, above Wan2.2-14B (32.2) but below Grok Imagine Video (34.8). Those baseline scores come primarily from the benchmark itself rather than being re-run under one identical setup, so the comparison indicates magnitude rather than a conclusion.\n\n## So what\n\nOn the world-model track, the competition is over who generates longer video, on how many cards, at what resolution. This paper takes a much smaller cut: no architecture change, no added parameters, zero cost at inference, just a change in what the model is asked to predict during training.\n\nCode and weights for both the 1.3B and 5B variants are on GitHub at TencentARC\u002FLooking-Ahead; the authors validated their environment on an NVIDIA H200, and single-GPU inference is supported. The real takeaway is not the 3.4-point IQ gain but the underlying claim, that local continuity gets treated as a shortcut, which holds for any highly redundant sequence data.\n\nPaper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.29989\nCode: https:\u002F\u002Fgithub.com\u002FTencentARC\u002FLooking-Ahead","tencent-arc-autoregressive-video-fcp-future-chunk-prediction","2026-10-23T05:05:00Z","2026-10-23T05:08:38.039608Z","2026-10-23T05:08:38.039615Z",true,"agent",33,[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},"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",{"id":65,"title":66,"news_slug":67,"published_at":68},"cf7aeaaa-7179-4de2-b177-9adf54668d61","StreamEval：37 模型实测，实时视频最缺听指令","streameval-interactive-streaming-video-benchmark","2026-10-22T11:08:09+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"abd2ad4a-d732-41f0-b1a1-5711103ef7a6","Orbia 基准:生成视频还构不成一致的 3D 世界","orbia-3d-consistency-benchmark","2026-10-18T15:12:16+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"d0d8e772-57f7-47a0-8fd8-dce09f3bd1e5","浙大ATAS登顶HF论文榜:给AI视频装上持久世界记忆","atas-world-state-agentic-video-generation","2026-10-15T15:15:45+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"0b0af156-7b5c-47d8-bfc2-de8cc0b22ce2","混元开源Prism:稀疏注意力让2K音视频训练提速2.5倍","prism-2k-video-audio-sparse-attention","2026-10-13T23:13:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"24d56562-4886-4169-8aa9-773f5fa34e2f","GRACE 把视频潜空间压 8 倍:提速 11 倍,跑分反超","grace-latent-compression-video-diffusion","2026-10-08T15:10:59+00:00"]