Daily Papers · 2026-09-16
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：2026.09.15 | 智能体超级智能黎明；实时可编辑空间视频生成
1. Atria Dawn：智能体超级智能的黎明
- 英文标题：Atria Dawn: The Dawn of Agentic Superintelligence
- arXiv：2609.15818v1
💡 一句话解读：这篇研究想解决的是：AI智能体如何参与自身后续版本的研发，并改变人类研究者的角色。它推出一个面向科研与工程的基础智能体模型Atria Dawn Preview，用可验证经验流程训练，在16项基准中五项得分最高；同时通过769条人机协作记录发现，人类更多负责判断方向与把关，智能体负责执行与修改。
随着AI智能体成为其后续版本开发的参与者，它们重塑了智能的生产方式以及人类研究者的角色。我们推出Atria Dawn Preview，一个面向科学研究与工程工作流的基础智能体语言模型，目标是在真实世界中拓展智能体生产力的边界。该模型通过可验证经验流程（Verifiable Experience Pipeline）训练，将工具介导的交互与可执行环境及外部验证结果相连接。在涵盖真实世界研究、工程与数字工作的16项基准上，Atria Dawn Preview与前沿智能体相比具有竞争力，并在其中五项上取得已报告的最高分。除独立性能外，我们还将该模型背后的真实研发过程作为人机协作案例进行研究，分析了来自56名参与者的769条任务记录以及智能体日志。当被要求在可比条件下评估已完成任务时，参与者认为约三分之一的AI辅助完成任务在无AI情况下不可行。更引人注目的是，智能体频繁提出方法并实施修改，而人类保留大部分最终决策，并通过判断与反馈引导探索。这些观察表明，协作正从任务级执行转向项目级伙伴关系，人类精力集中于什么值得追求以及证据应如何引导研究。因此，迈向更自主的AI研究，必须同时提升发现能力与有意义的人类监督能力，保留人类对持续开发的风险与方向的问责权威。
📄 English Abstract
As AI agents become participants in the development of their successors, they reshape both the production of intelligence and the role of human researchers. We introduce Atria Dawn Preview, a foundation agentic language model designed for scientific research and engineering workflows, with the goal of expanding the frontier of agent productivity in the real world. This model is trained via a Verifiable Experience Pipeline that connects tool-mediated interactions to executable environments and externally verified outcomes. Across 16 benchmarks spanning real-world research, engineering, and digital work, Atria Dawn Preview is competitive with frontier agents and achieves the highest reported score on five of them. Beyond standalone performance, we examine the real research-and-development process behind this model as a case study of human–AI collaboration, analyzing 769 task records from 56 participants together with agent logs. When asked to evaluate completed tasks under comparable conditions, participants rated about one-third of completed AI-assisted tasks as infeasible without AI. More strikingly, agents frequently propose methods and implement revisions, while humans retain most final decisions and guide exploration through judgment and feedback. These observations indicate a shift from task-level execution to project-level partnership, with human effort concentrating on what is worth pursuing and how evidence should guide research. Progress toward more autonomous AI research must therefore advance both the capacity for discovery and the capacity for meaningful human oversight, preserving accountable human authority over the risks and direction of continued development.
2. Vidu S2：实时交互、可编辑与空间视频生成
- 英文标题：Vidu S2: Real-Time Interactive, Editable, and Spatial Video Generation
- arXiv：2609.11638v1
💡 一句话解读：这篇研究想解决的是：现有视频生成模型无法实时生成可交互的数字人，也无法实时编辑视频流。为此提出 Vidu S2，包含实时数字人模型 Vidu S2-Avatar 和实时视频编辑模型 Vidu S2-Editing，并探索了实时空间视频生成，实验表明其优于所有基线。
我们提出了 Vidu S2，它包含 Vidu S2-Avatar（一个实时交互式数字角色模型）和 Vidu S2-Editing（一个实时视频编辑模型）。此外，我们还探索了 Vidu S2-Avatar 和 Vidu S2-Editing 实现实时空间视频生成的可行性。与 Vidu S1 相比，Vidu S2-Avatar 支持实时 720p 视频生成、可随时更新的动态参考生成，以及更强的指令跟随能力（例如跳舞）。Vidu S2-Editing 支持实时编辑视频流，包括风格渲染、服装替换、角色替换和背景替换。实验表明，Vidu S2 优于所有基线方法。可玩的在线演示见 https://vidu.com/vidu-stream。
📄 English Abstract
We present Vidu S2, which comprises Vidu S2-Avatar, a real-time interactive digital-character model, and Vidu S2-Editing, a real-time video editing model. Moreover, we explore the feasibility of real-time spatial video generation for both Vidu S2-Avatar and Vidu S2-Editing. Compared with Vidu S1, Vidu S2-Avatar supports real-time 720p video generation, generation with dynamic references that can be updated at any moment, and stronger instruction following, such as dancing. Vidu S2-Editing supports editing a video stream in real time, including style rendering, clothing replacement, character replacement, and background replacement. Experiments show that Vidu S2 outperforms all baselines. A playable online demo is available at https://vidu.com/vidu-stream.
3. ZGCM-1：一个完全开放且极其高效、面向数学与智能体搜索的基础模型
- 英文标题：ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
- arXiv：2609.13356v1
💡 一句话解读：小模型记不住整个互联网，怎么办？这篇论文训练了一个70亿参数的小模型ZGCM-1，让它学会“边想边查工具”，用外部工具弥补自身容量不足。团队还设计了高效训练流程和AI自动管理研发，使它在数学推理和智能体搜索上能媲美大它几十倍的模型，并全部开源。
在本工作中，我们提出了ZGCM-1，一个完全开放的70亿参数稠密基础模型，从零开始训练，在数据、系统和算法效率上达到极致。ZGCM-1基于一个核心前提：紧凑模型无法被动记忆开放网络，但可以通过将审慎的内部思考与主动的外部工具使用相结合，来克服参数容量限制。为在256K上下文下支持这一范式，我们开发了一套端到端、高效率的开放训练配方：架构与系统协同设计：交错的门控滑动窗口与全注意力，以及稳定的FP8 Muon优化器；渐进式课程与MDP中期训练：上下文从16K、64K扩展到256K，并将交互轨迹重构为马尔可夫决策过程。此外，我们建立了一种AI原生的研发工作流，其中智能体集群自主管理集群运维、数据整理和快速诊断评估。大量评估表明，ZGCM-1-7B在通用基准上在7B模型家族中具有竞争力。在若干具有挑战性的数学推理和智能体搜索套件上，它仍能与大几个数量级的前沿模型（如Qwen3-235B-A22B和GLM-5.1）竞争。我们还表明，我们的预训练设计在16K预训练时间到损失上带来了约4.2倍的效率提升。在整个开发生命周期中，我们提炼出八条可操作的经验发现——涵盖架构扩展、SFT质量剪枝、长上下文泛化和智能体协同训练动态。为促进社区研究，我们开源了预训练、中期训练和后训练阶段的模型权重、中间检查点、训练代码、各阶段数据及数据配方，以及W&B日志。
📄 English Abstract
In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency. ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate internal thinking with active external tool use. To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gated sliding-window and full attention, and a stable FP8 Muon optimizer; Progressive Curriculum & MDP Mid-Training: context scaling across 16K, 64K, and 256K, and the reformulation of interaction traces into Markov Decision Processes. Furthermore, we establish an AI-native R&D workflow where agent swarms autonomously manage cluster operations, data curation, and rapid diagnostic evaluation. Extensive evaluations show that ZGCM-1-7B is competitive across 7B model family on general benchmarks. On several challenging mathematical reasoning and agentic search suites, it remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1. We also show that our pre-training design offers a ~4.2x efficiency improvement in 16K pre-training time-to-loss. Across the full development lifecycle, we distill eight actionable empirical findings-spanning architectural scaling, SFT quality pruning, long-context generalization, and agentic co-training dynamics. To facilitate community research, we open-source model weights from the pre-training, mid-training, and post-training stages, intermediate checkpoints, training code, per-stage data and data recipes, and W&B logs.
4. Dream-RSI：通过演化世界实现递归自我改进
- 英文标题：Dream-RSI: Recursive Self-Improvement through Evolving Worlds
- arXiv：2609.14858v1
💡 一句话解读：这篇论文要解决AI自主探索时策略难以适应复杂搜索空间的问题。它提出Dream-RSI，用历史探索记录构建一个“梦境模拟器”，让AI在模拟中低成本试错并改进探索策略，再用于实际探索，形成自我改进循环。
递归自我改进对自主AI智能体日益重要，其进展取决于在复杂领域中能否发现高价值解决方案。这一过程的驱动力是有效探索，但管理和改进探索策略仍是主要瓶颈。当前系统面临一个根本困境：固定策略无法适应搜索空间的扩展，而在线策略优化需要在延迟且昂贵的反馈下、在长时程展开中导航巨大的元搜索空间。我们提出了Dream-RSI，一个可扩展且能递归自我改进的探索框架。一个轻量级编排层使探索变得显式且可编程，同时保持底层编码智能体不变。我们的关键洞察是，积累的发现历史可以作为已实现搜索空间上的回放模拟器。通过从历史发现树构建的回放模拟器中进行“ dreaming”，Dream-RSI能够获得即时、低成本的离策略反馈，以评估和优化探索策略，而无需重复进行昂贵的在线评估。改进后的策略随后被重新部署到在线环境中以驱动进一步发现，在自我改进循环中持续扩展模拟器池。在算法工程、数学优化和GPU内核工程中，Dream-RSI在若干设置下实现了有竞争力或更好的发现质量，同时大幅降低了发现成本。
📄 English Abstract
Recursive self-improvement is becoming increasingly vital for autonomous AI agents, where progress hinges on discovering high-value solutions across complex domains. The driver of this process is effective exploration, however, managing and improving exploration strategies remains a major bottleneck. Current systems face a fundamental dilemma: fixed strategies fail to adapt as search spaces scale, while online policy optimization requires navigating vast meta-search spaces under delayed and expensive feedback over long-horizon rollouts. We introduce \textsc{Dream-RSI}, a framework for scalable and recursively self-improving exploration. A lightweight orchestration layer makes exploration explicit and programmable while leaving the underlying coding agent unchanged. Our key insight is that accumulated discovery history can serve as a replay simulator over the realized search space. By performing dreaming in the replay simulator constructed from historical discovery trees, \textsc{Dream-RSI} secures immediate, low-cost off-policy feedback to evaluate and refine exploration policies without invoking repetitive, expensive online evaluations. The improved policy is subsequently redeployed online to drive further discovery, continuously expanding the simulator pool in a self-improving loop. Across algorithm engineering, mathematical optimization, and GPU kernel engineering, \textsc{Dream-RSI} achieves competitive or improved discovery quality while substantially reducing discovery cost in several settings.
5. PhysBrain 1.5：从视觉语言模型到物理基础模型
- 英文标题：PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models
- arXiv：2609.14973v1
💡 一句话解读：这篇研究想让一个AI模型同时具备三种能力：看懂物理环境、决定怎么动手、预测动手后会发生什么。做法是从一个通用的看图说话模型出发，把语言、机械臂动作和画面变化都变成同一种“词”来一起训练，先看大量人类操作视频学，再用真人示范、机器人轨迹和模拟数据微调。
我们提出 PhysBrain 1.5，一个用于理解物理环境、生成动作并预测未来状态的统一模型。受“观察—交互—环境变化”这一物理闭环的启发，我们将这些能力纳入同一个学习框架。我们从通用视觉语言模型（vision-language model）出发，将语言回复、末端执行器运动（end-effector motion）和稠密视觉目标编码为离散序列，并通过自回归下一词元预测（autoregressive next-token prediction）联合优化。预训练阶段的具身监督完全来自人类交互视频，以任务为中心的片段将语义与空间上下文同恢复出的运动及后续观察配对。随后，我们在人类演示、机器人轨迹和模拟经验的混合数据上通过监督微调对模型进行适配。在 28 个具身理解基准上，我们的 8B 模型平均得分 72.5，创下新的开源最优结果，并与 GPT-6-Astra、Gemini 3.6 Flash 等领先闭源模型表现相当；它在 14 个基准上取得最佳开源成绩，同时保留通用多模态能力。除这些理解评测外，定性示例显示该模型能够生成末端执行器轨迹，并通过空间对齐的 RGB、深度和机器人掩码输出预测未来场景。
📄 English Abstract
We present PhysBrain 1.5, a unified model for understanding physical environments, generating actions, and predicting future states. Motivated by the physical loop of observation, interaction, and environmental change, we bring these capabilities into a common learning framework. Starting from a general vision–language model, we encode language responses, end-effector motion, and dense visual targets as discrete sequences and jointly optimize them with autoregressive next-token prediction. Pre-training draws its embodied supervision entirely from human interaction videos, using task-centered episodes to pair semantic and spatial context with recovered motion and subsequent observations. We then adapt the model through supervised fine-tuning on a mixture of human demonstrations, robot trajectories, and simulated experience. Across 28 embodied understanding benchmarks, our 8B model achieves an average score of 72.5, setting a new open-source state of the art and performing on par with leading proprietary models such as GPT-6-Astra and Gemini 3.6 Flash. It achieves the best open-source results on 14 benchmarks while retaining general multimodal capabilities. Beyond these understanding evaluations, qualitative examples show the model’s ability to produce end-effector trajectories and predict future scenes through spatially aligned RGB, depth, and robot-mask outputs.
6. 分组值注意力：通过按需键重建实现高效KV缓存
- 英文标题：Grouped Value Attention: Efficient KV Caching via On-Demand Key Reconstruction
- arXiv：2609.13285v2
💡 一句话解读：Transformer 生成文本时，缓存历史信息（KV 缓存）会随长度增长，成为内存和速度瓶颈。GQA 虽能减少缓存，但仍要存键和值。本文提出 GVA：只存分组的值，用可学习的线性映射重建键，推理时映射可吸收进查询，无需实际生成键，并用小通道保留位置信息。缓存减少约 45-47%，精度接近 GQA。
KV 缓存是 Transformer 解码的主要瓶颈：其内存占用和缓存读取流量随序列长度增长。分组查询注意力（GQA）通过共享键值头降低此开销，但每一步仍同时存储键和值。我们提出分组值注意力（GVA），它存储分组的值，并用一个学习到的线性映射重建内容键。在推理时，该映射可被吸收到查询中，从而在预期的解码路径中无需实际生成内容键。一个小的共享解耦 RoPE 通道通过单独缓存的位置键保留位置信息。在所研究的配置中，这种表示相对于匹配的 GQA 将持久缓存标量减少约 45-47%。在 3.5 亿参数规模、300 亿 FineWeb-Edu token 下，16 维位置变体在五项任务上达到 44.18 的平均准确率，而 GQA 为 44.36，MLA 为 43.88。这些结果表明，在更紧凑的缓存表示下，基准准确率接近 GQA。为了将这种紧凑表示转化为更快的自回归推理，我们开发了自定义解码内核，目前正在评估其端到端推理性能，并计划很快开源发布。
📄 English Abstract
The KV cache is a primary bottleneck for Transformer decoding: its memory footprint and cache-read traffic grow with sequence length. Grouped-query attention (GQA) reduces this cost by sharing key-value heads, but still stores both a key and a value at every step. We introduce Grouped Value Attention (GVA), which stores grouped values and reconstructs content keys with a learned linear map. At inference, the map can be absorbed into the query, eliminating the need to materialize content keys in the intended decode path. A small shared decoupled RoPE channel retains positional information through a separately cached positional key. For the configurations studied, this representation reduces persistent cache scalars by approximately 45-47% relative to matched GQA. At the 350M-parameter scale with 30B FineWeb-Edu tokens, the 16-dimensional positional variant reaches 44.18 average accuracy across five tasks, compared with 44.36 for GQA and 43.88 for MLA. These results demonstrate near-GQA benchmark accuracy with a more compact cache representation. To translate this compact representation into faster autoregressive inference, we have developed custom decoding kernels and are currently evaluating their end-to-end inference performance with an open-source release planned soon.
7. LynnReal-Omni：面向智能体视觉工作流的原生多模态视频生成
- 英文标题：LynnReal-Omni: Native multi-modal Video Generation for Agentic Visual Workflows
- arXiv：2609.15863v1
💡 一句话解读：视频生成模型随机性强、难精确控制，长视频还容易跑偏。该研究把「智能体视觉创作」提供的明确参考（如3D场景、游戏画面）与扩散模型结合，提出统一的多模态视频生成框架，支持多种输入与任务，并推出加速版实现实时生成。
视频扩散模型具有随机性且难以控制：精确内容往往需要反复采样且无法保证成功，长时程场景在外观、交互和时间连贯性上会出现漂移。智能体视觉创作提供了显式参考、可编辑的3D场景或可执行的游戏状态以实现稳定控制，但其本身并不能保证高物体或角色保真度。将两者结合可实现稳定、高质量的生成。为实现这一结合，我们提出了LynnReal-Omni，一个原生多模态视频生成框架，构建于一个320亿参数的共享多模态扩散Transformer（diffusion transformer）之上，统一了文本到视频、图像条件生成、参考引导生成、结构控制、编辑、退化视频修复和长视频生成。它接受异构视觉输入，包括外观参考、可编辑3D渲染和游戏录制，使智能体能够在统一模型中组合视觉条件。我们还训练了一个专用的270亿参数Flash共享多模态扩散Transformer用于实时渲染。我们构建了系统化的数据流水线，用于视频清洗、主体关联、多模态标注和对齐控制构建，产出了包含多镜头音视频片段的精选语料库，并引入了MSAVP，一种包含100个提示、20项指标的评价设计，将指令遵循、生成合理性、视觉质量、时间行为和音频协调性分开评估。LynnReal-Omni-Flash通过模型和解码加速（包括轻量级VAE解码器）进一步降低推理成本；在单张H100上，22帧540p视频的热生成和解码在LynnReal-Omni上耗时843毫秒，在Flash上耗时377毫秒。这些结果为实时流式视频生成提供了基础，使LynnReal-Omni成为智能体视觉创作的一个统一、可控且高效的基础。
📄 English Abstract
Video diffusion models are stochastic and hard to control: precise content often requires repeated sampling without guaranteed success, and long-horizon scenes drift in appearance, interactions, and temporal coherence. Agentic visual creation provides explicit references, editable 3D scenes, or executable game states for stable control, but does not by itself guarantee high object or character fidelity. Combining the two can enable stable, high-quality generation. To realize this combination, we present LynnReal-Omni, a native multimodal video generation framework built on a 32B shared multimodal diffusion transformer that unifies text-to-video, image-conditioned generation, reference-guided generation, structural control, editing, degraded video restoration, and long-video generation. It accepts heterogeneous visual inputs, including appearance references, editable 3D renders, and game recordings, allowing agents to compose visual conditions within a unified model. We also train a dedicated 27B Flash shared multimodal diffusion transformer for real-time rendering. We build a systematic data pipeline for video cleaning, subject association, multimodal annotation, and aligned control construction, yielding a curated corpus of multi-shot audiovisual segments, and introduce MSAVP, a 100-prompt, 20-metric evaluation design that separates instruction following, generating plausibility, visual quality, temporal behavior, and audio coordination. LynnReal-Omni-Flash further reduces inference cost through model and decoding acceleration, including a lightweight VAE decoder; on one H100, warm generation and decoding of a 22-frame 540p video take 843 ms with LynnReal-Omni and 377 ms with Flash. These results provide a foundation for real-time streaming video generation, making LynnReal-Omni a unified, controllable, and efficient basis for agentic visual creation.
8. RSIAgent：新环境中面向递归自我改进的自主探索
- 英文标题：RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environments
- arXiv：2609.15364v1
💡 一句话解读：这篇研究要解决的问题是：数字助手进入陌生软件环境时，常因不熟悉界面和操作后果而频频出错。它的办法是让多个AI角色分工合作——一个负责出题、一个负责动手尝试、一个负责检查结果，边探索边把「什么操作在什么条件下会带来什么结果」记成笔记，先广撒网再深挖细节，之后直接翻笔记干活，不用重新训练模型。
数字智能体常常需要适应新环境，而这些环境的界面、工具和失败模式并未被预训练模型充分掌握。我们提出 RSIAgent，一个无需训练的多智能体框架，通过自主记忆构建实现递归自我改进。RSIAgent 协调课程智能体、执行智能体和验证智能体，持续探索环境、验证结果并保留环境特定知识，包括动作、条件与后果之间可复用的因果（causal）关系。它进一步采用「先广后深」的探索策略，将用于发现多样环境结构的并行广度递归自我探索，与用于挖掘困难案例、隐藏约束、边界条件及此前未知因果依赖的聚焦深度自我探索相结合。所得记忆被冻结后可直接复用于下游任务，无需更新模型参数。在 OSWorld-v2 和 Agent’s Last Exam 上的实验表明，RSIAgent 大幅提升了强开源模型，使 Kimi-K3 和 GLM-5.3 能够超越包括 GPT-6 在内的前沿闭源模型。
📄 English Abstract
Digital agents must often adapt to new environments whose interfaces, tools, and failure modes are not fully captured by pretrained models. We introduce \textbf{RSIAgent}, a training-free multi-agent framework for recursive self-improvement through autonomous memory construction. RSIAgent coordinates curriculum, actor, and verifier agents to continually explore the environment, validate outcomes, and retain environment-specific knowledge, including reusable causal relationships between actions, conditions, and consequences. It further adopts a \textbf{broad-then-deep} exploration strategy, combining parallel broad recursive self-exploration for discovering diverse environment structures with focused deep self-exploration for uncovering hard cases, hidden constraints, boundary conditions, and previously unknown causal dependencies. The resulting memory is frozen and can be directly reused for downstream tasks without updating model parameters. Experiments on OSWorld-v2 and Agent’s Last Exam show that RSIAgent substantially improves strong open-source models, enabling Kimi-K3 and GLM-5.3 to outperform frontier closed-source models including GPT-6.
9. 发现基础模型：迈向开放式发现智能
- 英文标题：Discovery Foundation Models: Toward Open-Ended Discovery Intelligence
- arXiv：2609.15973v1
💡 一句话解读：当前AI只会解决人类给定的问题，不会自己发现新问题、新知识。本文提出“发现智能”，让AI像科学家一样自主提出假设、做实验、修正理论，并构建了可训练、可评估的通用系统，还展示了在真实药物发现中的闭环应用。
基础模型已从对现有知识的学习和推理，逐步发展到通过行动、工具使用和结果反馈进行学习。我们认为下一个前沿是进一步的转变：从解决和作用于人类指定的问题，转向参与新问题、新表示、新解释和新知识的创造过程。我们将这种能力称为发现智能（Discovery Intelligence），并将发现基础模型（Discovery Foundation Models, DFMs）定义为面向开放式发现的通用模型系统。DFM 在一个可修正的研究状态上运行，支持七项耦合能力：问题发现、问题形式化、表示构建、假设形成、干预、基于证据的修正以及持续发现改进。我们通过 Zetema 实例化该框架，它耦合了显式研究状态动态、验证与实验门控、外部 grounding 以及跨任务发现技能演化。我们进一步通过 GALILEO 将该框架落地，这是一个真实的治疗发现系统，其中干实验室推理、机器人和手工湿实验室实验、外部生物证据以及迭代假设与设计修正形成了一个闭环物理发现回路。然后，我们制定了一种统一的能力形成和以过程为中心的评估方法，使发现行为能够在最终答案性能之外被训练、改进和度量。这些组件共同将发现确立为基础模型系统的一种可学习、可执行和可评估的能力。我们将这一转变视为智能扩展的更广泛进展：从对现有知识的学习，到从行动结果中学习，最终到参与新知识发现所依赖的结构的构建、测试和修正。代码：https://github.com/Gen-Verse/DFM-Plans
📄 English Abstract
Foundation models have progressed from learning and reasoning over existing knowledge, to increasingly learning through action, tool use, and outcome feedback. We argue that the next frontier is a further transition: from solving and acting within problems specified by humans to participating in the process by which new problems, representations, explanations, and knowledge are created. We refer to this capability as Discovery Intelligence. We formulate Discovery Foundation Models (DFMs) as general-purpose model systems for open-ended discovery. A DFM operates over a revisable research state and supports seven coupled capabilities spanning problem discovery, formulation, representation construction, hypothesis formation, intervention, evidence-grounded revision, and continual discovery improvement. We instantiate this framework with Zetema, which couples explicit research-state dynamics, verification and experimental gating, external grounding, and cross-task Discovery Skill evolution. We further ground the framework with GALILEO, a real therapeutic-discovery system in which Dry-Lab reasoning, robotic and hands-on Wet-Lab experimentation, external biological evidence, and iterative hypothesis and design revision form a closed physical discovery loop. We then formulate a unified approach to capability formation and process-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance. Together, these components establish discovery as a learnable, executable, and evaluable capability of foundation-model systems. We view this shift as a broader progression in intelligence scaling: from learning over existing knowledge, to learning from action outcomes, and ultimately to participating in the construction, testing, and revision of the structures through which new knowledge is discovered. Code: https://github.com/Gen-Verse/DFM-Plans
10. BVB：在 Blender 中通过程序化重建对智能体视频理解进行基准测试
- 英文标题：BVB: Benchmarking Agentic Video Understanding via Programmatic Reconstruction in Blender
- arXiv：2609.15478v1
💡 一句话解读：现有视频理解测试主要靠问答，但真正理解视频的智能体应能按视频内容重新做出一段动画。为此，研究者提出 BVB 基准：让智能体在 Blender 里用代码重建真实视频，并从“事实保留”和“视觉相似”两方面打分。结果显示，模型画面能做得像，但视频里的关键事实往往记不全。
多模态智能体无需依赖扩散模型，就能通过编程在 Blender 等软件中生成复杂视频。然而，视频理解基准仍主要通过问答来评估模型。如果智能体真正理解一段视频，它就能以程序化方式重建该视频。我们提出 BVB（Blender-VideoBench），一个通过要求智能体将真实世界视频重建为 Blender 动画场景来测试这一能力的基准。为确保公平比较，每个智能体都在相同沙箱中、在共享成本限制下，通过一个轻量级框架 Mini-BVB 编写重建程序。该基准从其动画相机渲染每个重建结果，并从两个维度进行评估：（1）双 VQA 衡量重建保留了多少时空事实；（2）潜在相似度（Latent Similarity）衡量重建与源视频在感知上的接近程度。我们的总分采用平方根均值，偏向均衡表现。我们评估了来自 10 个模型家族的 51 种配置，并分析了语义保留、感知相似度、推理努力和成本。最佳模型达到 88.6 的潜在相似度，但仅保留 53.7% 的源视频正确时空答案。额外推理能提升视觉相似度，却无法弥合事实准确性上的差距。在一项有 15 名评分者和五种配置的盲测中，潜在相似度与人类偏好高度相关。这些结果表明，程序化重建是检验智能体视频理解的一种可行测试，而语义保留仍是主要挑战。
📄 English Abstract
Multimodal agents can create complex videos in software such as Blender by coding without relying on diffusion models. Yet video understanding benchmarks still evaluate models mainly through question answering. If an agent truly understands a video, it can reconstruct it programmatically. We introduce BVB, Blender-VideoBench, a benchmark that tests this ability by asking agents to reconstruct real-world videos as animated Blender scenes. To ensure fair comparison, each agent programs the reconstruction through a lightweight harness, Mini-BVB, in an identical sandbox under a shared cost limit. The benchmark renders each reconstruction from its animated camera and evaluates it on two axes: (1) Dual VQA measures how many spatiotemporal facts the reconstruction preserves. (2) Latent Similarity measures how closely the reconstruction matches the source video perceptually. Our overall score, a square-root mean, favors balanced performance. We evaluate 51 configurations from 10 model families and analyze semantic retention, perceptual similarity, reasoning effort, and cost. The best model reaches 88.6 Latent Similarity but retains only 53.7% of the source-correct spatiotemporal answers. Additional reasoning improves visual similarity but does not close this gap in factual accuracy. In a blind study with 15 raters and five configurations, Latent Similarity correlates strongly with human preference. These results show that programmatic reconstruction is a viable test of agentic video understanding, and that semantic retention remains the main challenge.
11. 无损投机解码究竟有多无损？数值精度在 Orthrus 中的作用
- 英文标题：How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus
- arXiv：2609.15504v1
💡 一句话解读：这篇论文要检验一种加速语言模型生成的方法（Orthrus）是否真的能无损地产生和原模型完全一样的输出。作者在不同数值精度下测试后发现，常用的低精度（BF16）下只有不到一半的提示能完全匹配，而高精度（FP32）下则全部匹配，说明这种无损性其实依赖于计算精度。
Orthrus 是一种混合自回归-扩散架构，它通过并行生成多个词元（token）来加速自回归语言模型推理，同时使用一个冻结的自回归骨干网络。其核心主张是，一种模型内共识机制能够实现无损投机解码（lossless speculative decoding），产生与自回归模型相同的输出序列。我们独立复现了 Orthrus，并在不同数值精度下检验了这一主张。在 BF16 推理下，对于来自 12 个领域的 1190 个提示，作者提供的检查点仅有 45% 的情况出现精确轨迹匹配，而我们独立训练的模型为 43%。精确匹配的概率还与参考模型的响应条件困惑度（response-conditional perplexity）强相关。尽管存在这种轨迹偏差，Orthrus 在下游 lm-eval-harness 基准测试上并未表现出系统性退化。相比之下，使用 FP32 重复轨迹评估时，所有评估的提示均实现了精确轨迹匹配。这些结果表明，Orthrus 的实际无损性取决于数值精度，并且精确轨迹等价性应与下游任务性能分开评估。
📄 English Abstract
Orthrus is a hybrid autoregressive-diffusion architecture that accelerates autoregressive language-model inference by generating multiple tokens in parallel while using a frozen autoregressive backbone. Its central claim is that an intra-model consensus mechanism enables lossless speculative decoding, producing the same output sequence as the autoregressive model. We independently reproduce Orthrus and examine this claim under different numerical precisions. Under BF16 inference, exact trajectory matching occurs in only 45% of cases for the authors’ checkpoint and 43% for our independently trained model across 1,190 prompts from 12 domains. The probability of exact matching is also strongly associated with the response-conditional perplexity of the reference model. Despite this trajectory divergence, Orthrus does not show systematic degradation on downstream lm-eval-harness benchmarks. In contrast, repeating the trajectory evaluation with FP32 yields exact trajectory matching on all evaluated prompts. These results show that the practical losslessness of Orthrus depends on numerical precision and that exact trajectory equivalence should be evaluated separately from downstream task performance.
12. AlayaVista：从全景状态到透视视频的流式世界建模
- 英文标题：AlayaVista: Streaming World Modeling from Panoramic States to Perspective Video
- arXiv：2609.14462v1
💡 一句话解读：现有交互式视频世界模型在相机移动时，要么视野窄、容易丢失画面外内容，要么生成慢、画质差。AlayaVista 把“全景世界演变”和“当前视角画面生成”分开处理：先用一张普通照片扩展出360度全景，再让全景随相机移动而演变，最后从中渲染出你当前看到的清晰视频，从而兼顾大范围场景和低延迟高画质。
交互式视频世界模型需要在相机运动下保持宽广的场景上下文，同时以低延迟生成高保真观测。现有方法面临表示上的权衡：透视模型基于局部视图操作，必须在长时间推演中保持屏幕外内容；而更广的空间覆盖通常通过合成全球视频或构建显式三维表示来获得。受视觉感知中全局上下文与选择性局部敏锐度互补作用的启发，我们提出 AlayaVista，一种相机可控的流式视频世界模型，将全景世界演化与透视观测合成解耦。给定单张透视图像，AlayaVista 使用预训练的全景扩展模型构建 360 度场景先验，然后将场景演化为以相机为条件的全景潜在状态。一个潜在视口渲染器将该状态映射为所请求的透视视频潜在表示，而一个透视细化器则恢复细节、抑制伪影并进行超分辨率。为支持高效流式处理，我们将全景生成器适配为分块自回归生成，并将全景生成和透视细化都蒸馏为少步过程。为提供该设计所需的监督，我们构建了 MUGEN，一个大规模真实世界全景视频数据集，包含 1,318 小时、分辨率至少为 4K 的视频，以及丰富的语义和几何标注。
📄 English Abstract
Interactive video world models must maintain broad scene context under camera motion while producing high-fidelity observations with low latency. Existing approaches face a representation trade-off: perspective models operate on local views and must preserve off-screen content over long rollouts, whereas broader spatial coverage is typically obtained by synthesizing full-sphere videos or constructing explicit 3D representations. Motivated by the complementary roles of global context and selective local acuity in visual perception, we present AlayaVista, a camera-controllable streaming video world model that decouples panoramic world evolution from perspective observation synthesis. Given a single perspective image, AlayaVista constructs a 360-degree scene prior using a pretrained panorama expansion model and then evolves the scene as a camera-conditioned panoramic latent state. A latent viewport renderer maps this state to the requested perspective video latents, while a perspective refiner restores details, suppresses artifacts, and performs super-resolution. To support efficient streaming, we adapt the panoramic generator to chunk-autoregressive generation and distill both panoramic generation and perspective refinement into few-step processes. To provide the supervision required by this design, we construct MUGEN, a large-scale real-world panoramic video dataset containing 1,318 hours of videos at resolutions of at least 4K, together with rich semantic and geometric annotations.
13. Kaininja：将原生3D生成器扩展至部件级
- 英文标题：KaiNinja: Extending Native 3D Generators to the Part Level
- arXiv：2609.15659v2
💡 一句话解读：现有3D生成器只能把一张图片变成一整块融合的模型，而编辑、绑定骨骼、仿真等后续工作都需要按部件拆分的模型。作者提出“双体积”表示法，解决了两个部件接触面无法在单个体积里表达的问题，并据此构建KaiNinja，让生成器直接输出部件级模型，无需分割网络，速度和质量不降反升。
原生3D生成器能将单张图像转换为单个网格。TRELLIS.2及其同类方法可生成高保真、非水密且带材质的几何体，但输出是一个融合的整体对象，而编辑、绑定和仿真等下游工作则针对部件级资产。一个朴素的想法是在TRELLIS.2生成的融合网格上运行3D分割网络，但此类流程速度慢且受限于分割精度。我们希望找到一种简单方法，将现有原生3D生成器扩展到部件级。但我们面临一个关键问题：O-Voxel网格在每个体素中只存储一层表面，因此无论分辨率多高，单个体积都无法表示两个部件相接触的界面。我们引入双体积表示来解决这一问题，并提出KaiNinja——基于TRELLIS.2的O-Voxel表示的双体积形式所构建的部件级扩展。KaiNinja在保持TRELLIS.2生成速度和质量的同时将其扩展到部件级，且流程中无需掩码或分割器。其训练数据来自多种来源，包括CAD模型和由LLM驱动的智能体创作的资产；据我们所知，这是首个在智能体创作的部件数据上训练的3D生成模型。出乎意料的是，我们还发现，与在同一数据集上微调的相同骨干网络相比，整体对象的保真度有所提升。与不同范式的部件生成流程相比，它将整体对象倒角距离降低了40%，并将严格部件F分数提高了16%。
📄 English Abstract
Native 3D generators turn one image into a single mesh. TRELLIS.2 and its peers deliver high-fidelity non-watertight geometry with materials, but the output is one fused object, while downstream work such as editing, rigging and simulation operates on part-level assets. A naive idea is to run a 3D segmentation network on the fused mesh that TRELLIS.2 generates, but such pipelines are slow and bounded by the accuracy of the segmentation. We want a simple way to extend an existing native 3D generator to the part level. But we face a critical problem: the O-Voxel grid stores one sheet of surface per voxel, so a single volume cannot represent the interface where two parts touch, at any resolution. We introduce a dual-volume representation to solve this problem and put forward KaiNinja, a part-level extension of TRELLIS.2 built on a dual-volume form of its O-Voxel representation. KaiNinja keeps the generation speed and quality of TRELLIS.2 while extending it to the part level, with no mask or segmenter in the pipeline. Its training data come from sources of many kinds, including CAD models and assets authored by an LLM-driven agent; to our knowledge it is the first 3D generative model trained on agent-authored part data. Surprisingly, we also find that whole-object fidelity improves over the same backbone fine-tuned on the same dataset. Against part generation pipelines of different paradigms, it lowers whole-object Chamfer distance by 40% and raises strict part F-score by 16%.
14. 全模态流式思维
- 英文标题：Omni-Streaming Thinking
- arXiv：2609.15128v1
💡 一句话解读：这篇研究要解决的是：流式全模态模型在视频和音频还没播完时，就急着根据画面下结论，之后即使声音推翻了它，模型仍会继续沿用旧结论。作者提出“全模态流式思维”（OST），把结论先标为“待验证”，分别保存画面和声音证据，到验证时间点再核对，发现矛盾就削弱旧结论并更新状态，最后再决定是否作答。
流式全模态模型必须根据目前观察到的视频片段和同步音频，决定回答什么以及何时回答。视觉线索常常在一句话或一个声音事件尚未结束时就支持某种解释；如果这种解释作为事实进入记忆，那么即使之后音频与之矛盾，后续推理仍可能继续沿用该解释。我们把这种失败称为“过早的跨模态承诺”。我们提出“全模态流式思维”（Omni-Streaming Thinking，OST），它生成结构化输出，包括目前已观察到的证据、对未来证据的预测，以及基于这些证据的主张。每条主张最初被标记为待定，并关联到一个未来的验证区间。音频和视觉证据被分开存储，OST 在验证区间结束时，用指定模态的证据来核验该主张。当检测到矛盾证据时，一个反驳过程会降低该主张及其依赖状态的影响，然后利用新证据引导状态更新。一个“回答门”决定回答关键主张是否满足给出回应的条件。在冻结的 Qwen3-Omni-30B-A3B-Instruct 主干模型上做轻量适配后，OST 在五个流式与音视频基准上平均相对超过最强开源基线 10% 以上。我们还提出了 OST-DiagBench，它固定视频、编辑音频，以测试一致、缺失、矛盾、共存和字幕-语音冲突等情况。OST 达到 d-prime = 2.95，而开源基线最高仅为 1.38，同时减少了视觉诱发的听觉幻觉。
📄 English Abstract
Streaming omni-modal models must decide what and when to answer from the video chunks and synchronized audio observed so far. Visual cues often support an interpretation before an utterance or sound event is complete. If that interpretation enters memory as a fact, later reasoning can keep relaying it even after audio contradicts it. We call this failure premature cross-modal commitment. We propose Omni-Streaming Thinking (OST), which generates structured outputs that include evidence observed so far, forecasts of future evidence, and claims based on this evidence. Each claim is initially marked as pending and linked to a future verification interval. Audio and visual evidence are stored separately, and OST checks a claim against the evidence from the specified modality at the end of the verification interval. When contradictory evidence is detected, a refutation process reduces the influence of the claim and its dependent states, and then guides a state update using the new evidence. An answer gate decides whether the answer-critical claims meet the conditions for giving a response. Using a frozen Qwen3-Omni-30B-A3B-Instruct backbone with lightweight adaptation, OST outperforms the strongest open baselines on five streaming and audio-visual benchmarks by more than 10% relative on average. We also introduce OST-DiagBench, which holds video fixed and edits audio to test agreement, absence, contradiction, coexistence, and subtitle-speech conflict. OST reaches d-prime = 2.95, compared with at most 1.38 for open baselines, while reducing vision-induced auditory hallucinations.
15. LLaDA-UI：将块级扩散引入视觉语言 GUI 智能体
- 英文标题：LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents
- arXiv：2609.13287v1
💡 一句话解读：这篇研究要解决的问题是：扩散语言模型解码快，但能否用来做实时看屏幕、点按钮的图形界面智能体还不清楚。作者提出 LLaDA-UI，先让视觉编码器与扩散语言模型对齐，再用界面操作数据训练，在多个测试中表现优于同类模型。
扩散大语言模型（dLLMs）通过块级并行、任意顺序生成实现高解码效率，使其对延迟敏感的应用颇具吸引力。图形用户界面（GUI）智能体是这一范式的天然试验场，因为它们必须实时反复感知屏幕状态并输出结构化、空间定位的动作。然而，dLLMs 能否在保持并行解码优势的同时扩展为能力强的多模态 GUI 智能体，仍是一个未解问题。我们提出 LLaDA-UI，一个 167 亿参数、基于混合专家（MoE）的块级扩散视觉语言 GUI 智能体。LLaDA-UI 采用两阶段训练流程：通用多模态预训练将原生分辨率视觉编码器与 LLaDA2.0-mini-base 扩散语言主干对齐，随后在多样化的移动端、桌面端、网页及定位数据上进行 GUI 智能体监督微调。在广泛采用的定位基准和跨多个平台的导航基准上，LLaDA-UI 大幅优于 Qwen2.5-VL-7B，并在所报告的六个 GUI 基准中的四个上超过 Qwen3-VL-8B。这些结果确立了块级扩散作为多模态 GUI 智能体实用生成范式的地位。
📄 English Abstract
Diffusion large language models (dLLMs) achieve high decoding efficiency through block-parallel, arbitrary-order generation, making them attractive for latency-sensitive applications. GUI agents represent a natural testbed for this paradigm, as they must repeatedly perceive screen states and emit structured, spatially grounded actions in real time. However, whether dLLMs can be extended into capable multimodal GUI agents while preserving their parallel decoding advantage remains an open question. We present LLaDA-UI, a 16.7B-parameter MoE-based, block-wise diffusion vision-language GUI agent. LLaDA-UI follows a two-stage training pipeline: general multimodal pre-training aligns a native-resolution vision encoder with the LLaDA2.0-mini-base diffusion language backbone, followed by GUI-agent supervised fine-tuning on diverse mobile, desktop, web, and grounding data. Across widely adopted grounding benchmarks and navigation benchmarks spanning multiple platforms, LLaDA-UI substantially outperforms Qwen2.5-VL-7B and surpasses Qwen3-VL-8B on four of six reported GUI benchmarks. These results establish block-wise diffusion as a practical generative paradigm for multimodal GUI agents.