👍 103
10/27 08:00
Diffusion vision-language models (DVLMs) process their entire input at every denoising step. Since visual tokens make up 89-97% of this input on the benchmarks we study, most of the decoding computation is spent repeatedly processing the same image. Visual token pruning methods partially relieve thi
中文介绍 本文提出证书引导的视觉路由方法,用于高效扩散视觉语言模型(DVLMs)。该方法通过证书机制动态决定视觉令牌的处理时机,避免在每步去噪中重复处理全部视觉输入,从而显著降低计算开销,同时保持生成质量。实验表明,视觉令牌占输入89-97%,所提方法能有效剪枝冗余视觉令牌,提升推理效率。
👍 72
10/28 08:00
Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introd
中文介绍 本文提出自适应任务刷新(A-RSI)框架,用于零数据多模态自改进。现有递归自改进方法多关注生成何种经验,而忽略何时引入新经验。A-RSI通过自适应机制判断当前经验是否失效,并在适当时机刷新任务,避免无效训练,提升模型自改进效率。
👍 41
10/29 08:00
Interactive world models generate visual content conditioned on user actions, yet maintaining spatial consistency over long-horizon exploration in real time remains challenging. Existing approaches rely on either temporal context or explicit 3D memory, both of which scale poorly because memory grows
中文介绍 本文提出L3G-World,一种实时无限世界生成方法,利用潜在3D高斯记忆实现空间一致性。现有方法依赖时序上下文或显式3D记忆,随探索范围增大而扩展性差。L3G-World通过紧凑的潜在记忆表示,在长时程交互中保持空间一致性,同时实现实时生成。
👍 38
10/27 08:00
Scaling supervised training for working agents requires diverse professional tasks that yield reliable expert demonstrations. However, missing or inconsistent task information can obstruct teacher execution and weaken the resulting supervision. We propose WorkFactory, a scalable framework for constr
中文介绍 本文提出WorkFactory,一个可扩展框架,用于将专业意图转化为可执行任务,以训练工作型智能体。针对现有监督训练中任务信息缺失或不一致的问题,WorkFactory自动构建多样化的专业任务并生成可靠的专家示范,从而提升智能体在真实工作场景中的性能。
👍 30
10/23 08:00
Locating relevant evidence in long inputs is a core capability of Multimodal Large Language Models. Prior studies attribute this capability to retrieval heads, a sparse set of attention heads that transfer information from the context to the output. However, these analyses are confined to a limited
中文介绍 本文扩展了对多模态大语言模型(MLLM)检索头的分析。检索头是负责从长上下文中传递信息到输出的稀疏注意力头集合。先前研究局限于有限设置,本文通过更广泛的分析,探讨检索头在长多模态输入中的行为,揭示其能力与局限,为优化长上下文建模提供新视角。
👍 24
10/22 08:00
Large language model agents have enabled software development through natural-language interaction, but extending this paradigm to end-to-end game development remains challenging. Building a playable game requires coordinating code, scenes, assets, animations, physics, and configurations, while allo
中文介绍 本文提出VibeGame,一个利用AI原生引擎和自进化对抗智能体团队进行提示词到游戏开发的新范式。该系统通过自然语言交互协调代码、场景、资源、动画、物理和配置,并允许智能体团队在开发中自进化,以应对游戏开发的复杂性,最终生成可玩的游戏。
👍 22
10/21 08:00
A benchmark is not merely a collection of test questions; it naturally induces an empirical measure over difficulty. In this work, we develop this measure-based view of benchmark evaluation. This perspective allows us to identify and formalize two central failure modes: measure mismatch, where the b
中文介绍 本文提出基准测试的难度度量视角,将基准视为在难度上诱导的经验度量。该视角形式化了两个核心失败模式:度量失配(基准难度分布与目标分布不匹配)和度量污染(基准包含非目标难度),为基准设计提供理论指导,以提升评估的可靠性和有效性。
👍 21
10/26 08:00
Evaluating human-AI co-design requires testing whether coding agents can translate real users' multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding session
中文介绍 本文介绍Co-Design Bench,一个无参考基准,用于评估多模态协同编码。该基准包含273个任务,源自26,314个真实用户编码会话中的37,463个去重查询,测试编码智能体能否将真实用户的多模态请求转化为可工作的应用修订,为人类-AI协同设计评估提供新工具。
👍 20
10/29 08:00
One-step and few-step text-to-image generation must accurately render the details specified in the prompt despite limited opportunities for refinement. Distribution matching distillation (DMD) uses score differences to bring the generated distribution closer to a reference distribution. Following pr
中文介绍 本文提出内在分布匹配(IDM)方法,用于一步和少步文本到图像生成。针对分布匹配蒸馏(DMD)的局限,IDM利用得分差异将生成分布推向参考分布,在有限细化机会下更准确渲染提示中的细节,提升一步生成质量。
👍 19
10/28 08:00
Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeV
中文介绍 本文提出WeVisCut,一个从旁白到长视频编辑的智能体,集成整个编辑流程于一个智能体。现有编辑智能体多采用多阶段流程,由通用模型驱动,质量和成本权衡不明确。WeVisCut使用紧凑、任务自适应的策略,实现跨镜头的连贯编辑,同时优化质量与成本。
👍 18
10/28 08:00
Multi-Token Prediction (MTP) is now a common ingredient in language-model pre-training, as it improves downstream performance and enables faster inference through speculative decoding. We identify a flaw in standard MTP training, in which a residual component of the MTP gradient interferes with both
中文介绍 本文指出标准多令牌预测(MTP)训练存在缺陷,其中MTP梯度的残差分量干扰主任务训练,并影响推测解码的性能。作者提出修正方法,消除该干扰,使MTP训练更稳定,提升下游性能和解码效率,为MTP在语言模型预训练中的正确应用提供指导。
👍 18
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 本文提出REALCLI,一个配对基准,用于统一评估命令行界面(CLI)智能体与人类协作的运行时协作质量。现有基准假设用户指令固定或仅评估任务结果,REALCLI通过配对设计捕捉协作过程,为CLI智能体的人机协作评估提供更全面的衡量标准。
👍 17
10/16 08:00
Computer-use agents (CUAs) are commonly evaluated against fixed task specifications, but users may omit requirements, add constraints, or require approval before consequential actions. Completing the benchmark instruction does not guarantee serving the person who asked. We introduce PersonaCUABench,
中文介绍 本文提出PersonaCUABench,一个评估计算机使用智能体(CUAs)服务多样化用户画像的基准。针对固定任务规范忽略用户需求差异的问题,该基准引入用户画像,包括省略要求、添加约束、需批准等情形,以衡量CUAs是否真正满足用户意图,而非仅完成基准指令。
👍 14
10/29 08:00
We present Youtu-Parsing-Omni, an omni-modal parser that converts documents, images, audio, and video into structured records grounded in space and time, providing a perception layer for agents that read, locate, and cite evidence across modalities. Its design rests on one encoder for every modality
中文介绍 本文提出Youtu-Parsing-Omni,一个全能解析器,将文档、图像、音频和视频转换为基于时空的联合结构记录。该设计采用一个编码器处理所有模态,统一模式解析,为智能体提供跨模态读取、定位和引用证据的感知层,实现多模态统一解析。
👍 14
10/27 08:00
Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real websites. Capture-and-replay preserves observations, with interactio
中文介绍 本文提出WebAnything,一个从开放网络构建可扩展智能体环境的方法。现有方法依赖人工镜像或生成式构建,各有局限。WebAnything采用捕获与回放保留观察,并支持交互,无需逐站点工程,即可生成多样化、可执行、可重置的网站环境,用于网络智能体的训练与评估。
👍 13
10/28 08:00
Financial agents must form forecasts and investment judgments from decision-time evidence, with predictions evaluated against subsequent outcomes. Existing benchmarks assess financial knowledge, research workflows, or forecasting, but evaluating the full research process requires a strictly frozen i
中文介绍 本文提出投资者最后考试(Investor's Last Exam),一个冻结时间点的基准,评估AI智能体在不可见未来情况下的股权投资能力。该基准要求智能体仅基于决策时点证据进行预测和投资判断,并依据后续结果评估,严格禁止信息泄漏,以衡量全流程研究能力。
👍 13
10/15 08:00
Spoken Dialogue Models (SDMs) typically share a speech-token space for generation and synthesis, leading to long autoregressive sequences and costly inference. Lower-rate speech tokenizers shorten sequences but may introduce compression-quality trade-offs and costly replacement without breaking this
中文介绍 本文提出CoSTA,一种压缩语音令牌适配方法,用于高效口语对话模型(SDMs)。针对SDMs共享语音令牌空间导致长序列和昂贵推理的问题,CoSTA通过压缩语音令牌降低序列长度,同时保持生成质量,并避免额外训练成本,提升口语对话效率。
👍 11
10/28 08:00
Recent advances in multimodal large language models have increased interest in interaction paradigms for video understanding. The evolution from offline to streaming and proactive interaction has gradually shifted models from passive responders toward more proactive agents that can decide when to an
中文介绍 本文提出AskStream,一个流式视频理解中主动询问的框架,决定何时和问什么。该框架将多模态大语言模型从被动响应者转变为主动智能体,在不确定性下动态决定询问时机和内容,以提升流式视频理解的交互性和准确性。
👍 11
10/29 08:00
Agentic image editors decompose complex requests into steps, but when a step fails, they mostly retry with the same narrow set of tools and a judge that scores each output by eye directly. We argue that such agents are bounded by two spaces: the action space of what the agent can do, and the observa
中文介绍 本文提出EditAgent,重新思考智能体图像编辑。现有智能体在步骤失败时仅用固定工具重试,受限于动作空间和观察空间。EditAgent通过扩展这两个空间,允许智能体探索更广泛的工具和观察,提升复杂编辑请求的成功率和效率,实现更鲁棒的图像编辑。
👍 11
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 本文提出TauIndianBankBench,一个面向程序感知、动作导向的工具使用对话智能体评估基准。针对印度银行场景,该基准测试智能体是否遵循银行规则,包括程序要求和顺序约束,以正确执行如锁卡等请求,衡量智能体在真实服务场景中的合规性和有效性。