👍 386
10/26 08:00
Multimodal intelligence can be viewed as human-guided co-evolution between worlds and agents through interaction. However, scaling this process faces two bottlenecks: constructing heterogeneous, verifiable environments is costly, and manual Harness iteration cannot keep pace with growing task divers
中文介绍 提出Trinity框架,将多模态智能视为智能体与环境间的协同演化。针对异构环境构建成本高、人工Harness迭代跟不上任务多样性的瓶颈,Trinity通过能力驱动的递归自改进机制,让智能体在交互中自主扩展能力边界,降低环境搭建与适配成本。
👍 380
10/26 08:00
Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale
中文介绍 提出多尺度策略优化(Multiscale Policy Optimization),用于长程LLM智能体训练。观察到评论值主要在轮次间变化,因此以轮次为锚点进行优化,并在token间精细分配信用。该方法改善了长程任务中奖励分配的粒度,提升了智能体在长视野决策上的性能。
👍 151
10/25 08:00
Memory is increasingly used in long-horizon agents, yet the net value of building memory over direct retrieval from raw history remains largely unclear. Through a large-scale evaluation across multiple benchmarks and model backbones, we find many dedicated memory systems underperform naive RAG in ma
中文介绍 通过多基准、多骨干的大规模评估,系统分析了智能体记忆的净价值。发现许多专用记忆系统在多数情况下表现不如朴素RAG,揭示了构建记忆相对于直接检索原始历史的边际收益有限。该研究为长程智能体的记忆设计提供了实证参考。
👍 89
10/27 08:00
Diffusion video models produce striking clips, but they cannot reliably render an exact formula, a correct chart, a rule-consistent game or narration synchronized to on-screen events. A different route is emerging: multimodal agents that write code to render every frame and synthesize every sound. W
中文介绍 提出VibeVideoBench基准,评估多模态智能体通过编写代码逐帧渲染、逐声合成视频的能力。该基准测试智能体生成精确公式、正确图表、规则一致的游戏及与画面同步的叙述等任务,弥补了扩散模型在可控视频生成上的不足。
👍 60
10/27 08:00
Discrete visual tokens represent vision in a discrete form analogous to language, offering a native interface for multimodal modeling under a shared next-token prediction objective. However, quantization error is widely assumed to cause unavoidable semantic loss, fundamentally limiting the semantic
中文介绍 质疑离散token必然造成视觉语义损失的假设。通过实验证明,在合适的量化与训练策略下,离散视觉token可以近似无损地表示视觉信号,为多模态模型在共享下一个token预测目标下提供原生接口。该发现挑战了语义损失的普遍认知。
👍 53
10/27 08:00
Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning fro
中文介绍 提出ThinkWorld框架,利用人类视频增强VLA模型的泛化能力。通过世界想象机制,模型从无动作标签的人类视频中学习动态推理,强化链式动力学(Chain-of-Dynamics)推理。该方法缓解了机器人演示数据覆盖不足的问题,提升了操控泛化性能。
👍 51
10/27 08:00
Video captioning aims to preserve visual information through compact descriptions, predominantly using natural language. However, precisely expressing fine-grained spatial structures and temporal dynamics in natural language often leads to lengthy and potentially ambiguous descriptions. In this work
中文介绍 提出CodeCap-Bench基准,评估代码作为视频描述形式的可行性。相比自然语言,代码能更精确表达细粒度空间结构与时间动态,避免冗长和歧义。该工作探索了以程序化描述替代自然语言进行视频字幕生成的新范式。
👍 28
10/27 08:00
Multimodal perception forms the foundation of multimodal intelligence. In particular, spatial perception and multimodal long-context understanding are two pivotal tasks that underpin embodied AI and multimodal agents. However, synthesizing targeted, high-quality training data for these tasks remains
中文介绍 提出Zooming without Zooming v2方法,用于生成高质量多模态感知训练数据。通过从细节观察到证据连接,该工作解决了空间感知与长上下文理解任务中数据合成难的问题。生成的合成数据提升了多模态智能体在具体任务上的泛化能力。
👍 21
10/27 08:00
Mixture-of-Experts (MoE) models are bottlenecked by loading routed-expert weights from memory during interactive decoding at small batch sizes. Because most intermediate channels in a routed expert contribute little to any single token, channel selectors predict which channels matter and read only t
中文介绍 提出无需校准数据的通道选择方法,用于MoE模型解码加速。由于路由专家中多数中间通道对单一token贡献微小,该方法预测重要通道并只加载它们,减少内存带宽瓶颈。在保持精度的同时,显著提升小批量解码速度。
👍 18
10/25 08:00
While progress in informal mathematics has come largely from scaling the chain-of-thought and agentic reasoning of frontier language models, formal theorem proving has increasingly relied on human-designed workflows around these models, such as planning, decomposition, and orchestrating specialized
中文介绍 提出h1系统,探索形式化定理证明中的「苦涩教训」:减少人工设计,让模型通过自我博弈和搜索从零学习。h1在多个基准上超越依赖人类工作流的方法,展示了规模化学习在形式数学中的潜力。
👍 17
10/27 08:00
Agentic services alternate between model inference on clustered GPUs and external tools on distributed CPUs. In this scenario, existing systems (1) overlook tool-return readiness in service evaluation, wasting limited resources, and (2) use heuristic strategies to decide admission and placement on G
中文介绍 提出TAO工具感知联合调度框架,用于LLM智能体服务。该框架考虑工具返回就绪状态与GPU/CPU资源协同,避免现有系统忽略工具延迟而浪费资源。通过联合决策准入与放置,显著提升服务吞吐量和资源利用率。
👍 15
10/15 08:00
Command-line interface (CLI) agents increasingly execute longer-horizon tasks, yet existing benchmarks either assume that user instructions remain fixed throughout execution or evaluate task outcomes without capturing runtime collaboration quality. In this paper, we present REALCLI, a paired benchma
中文介绍 提出REALCLI基准,用于统一评估CLI智能体与人类的协作质量。该基准配对任务在静态指令之外,动态引入用户干预,同时测量任务结果与协作过程中的实时交互质量。弥补了现有基准忽略运行时协作的缺陷。
👍 11
10/26 08:00
Learning velocity fields directly in pixel space is difficult and slow. We show that coarse second-order image statistics can ease this task: frozen Wiener estimates improve velocity learning, with gains extending beyond the linear predictions they explicitly provide. Moreover, we observe similar bu
中文介绍 提出Wiener-informed解码方法,利用粗粒度二阶图像统计量简化像素空间流模型的训练。冻结的Wiener估计加速了速度场学习,其增益超出线性预测范围。该方法降低了像素空间扩散模型的训练难度,提升了收敛效率。
👍 9
09/30 08:00
Many Indian banks serve customers through FAQ pages and scripted chatbots. Assistants based on language models could respond faster and carry out requests such as blocking a card, but only if they give correct information and follow the bank's rules. These rules prescribe a procedure, a set of requi
中文介绍 提出TauIndianBankBench基准,评估工具使用对话智能体在遵循银行程序规则方面的表现。该基准聚焦于程序感知与动作导向,测试智能体在阻止银行卡等任务中正确执行多步骤流程的能力。揭示了当前模型在规则遵循上的不足。
👍 7
10/27 08:00
Language model agents are post-trained and deployed through a harness, the prompts, tools, and control logic that present a task to the model and execute its outputs. Harnesses are often revised and judged by how an already trained model performs under them, which assumes that what the model learned
中文介绍 揭示语言模型智能体的策略本质上由Harness决定,即提示、工具和控制逻辑构成的任务接口。文章论证了Harness修订应基于模型实际学习目标而非仅看性能,为智能体接口设计提供了新的理论视角。
👍 6
10/27 08:00
The vision-language-action (VLA) and world-action model (WAM) communities debate whether control policies should inherit pretrained knowledge from vision-language models or from visual generators. The debate, however, bundles three separable choices: backbone initialization, the objective applied to
中文介绍 拆解VLA与世界模型之争,指出争论混淆了三个独立选择:骨干初始化、目标函数与策略结构。提出「Just Action Models」视角,主张控制策略应直接以动作预测为目标,不必依赖VLM或视觉生成器继承。澄清了多模态控制模型的设计空间。
👍 5
10/14 08:00
Despite their extensive pretraining knowledge and strong reasoning capabilities, large language models (LLMs) struggle to adaptively gather information in experimental design tasks. We study how to amortize sequential Bayesian experimental design (BED) into an LLM by explicitly training the reasonin
中文介绍 研究将序贯贝叶斯实验设计(BED)摊还到LLM推理中。通过显式训练推理模型,使其学会自适应收集信息,而非仅靠预训练知识。该方法在实验设计任务上提升了LLM的决策效率,展示了摊还推理的潜力。
👍 5
10/26 08:00
Financial institutions operating under data-residency constraints are restricted to open-weight models small enough to locally host, which trail frontier systems on multi-step quantitative reasoning. On-policy distillation narrows such gaps by supervising a student on its own rollouts, but tradition
中文介绍 研究数据驻留约束下金融推理的自我蒸馏方法。针对小规模开放权重模型落后于前沿系统的问题,提出用事后提示(hindsight hint)改进自蒸馏,并分析其适用范围。该方法在保持合规的同时提升了多步量化推理能力。
👍 5
10/27 08:00
When an LLM group's discussion is compressed into a short memory for a final decision maker, what should the memory keep? Two natural heuristics keep the rarest claims or the claims relevant to the decision criteria. On hidden-profile tasks both look attractive, because how widely a fact is held is
中文介绍 研究LLM群体讨论压缩为记忆时的信息保留策略。在隐藏剖面任务上,保留最稀有或最相关主张的启发式看似合理,但实验显示两者均导致决策错误。文章揭示记忆压缩对群体决策的负面效应,指出需更精细的信息选择机制。
👍 5
10/27 08:00
AI research agents can propose, execute, and improve programs that solve machine learning tasks. However, executing each candidate program requires training and evaluating ML models, limiting how many candidates an agent can explore at each step. We introduce MLE-World, a world model finetuned from
中文介绍 提出MLE-World世界模型,预测机器学习实验的候选程序执行结果。该模型从历史实验数据微调而来,无需真正训练即可评估候选性能,从而大幅提升AI研究智能体的搜索效率。减少了每次探索所需的计算成本。