👍 82
11/11 08:00
System One models promise fast, general-purpose decision-making without lengthy token generation. However, understanding their capability spectrum and decision reliability requires more than aggregate performance on current benchmarks. Existing evaluations often provide limited insight into the brea
中文介绍 本文提出SO-Arena基准,系统评估System One模型的能力谱系与决策可靠性。现有基准多依赖聚合性能,难以揭示模型在快速、通用决策中的局限性。SO-Arena通过细粒度测试,量化模型在不同任务上的表现与可靠性差异,为模型选型提供更全面的参考。
👍 44
10/29 08:00
Mathematical supervision for language models is typically flattened into isolated problem-solution pairs, discarding the curricular organization and evidential structure of the original sources. We introduce TOME-Math (Textbook-Oriented Mathematical Evidence), a large-scale corpus that converts 2,11
中文介绍 本文提出TOME-Math,一个基于结构化教科书的大规模数学监督语料库,将2,117本数学教科书的组织结构和证据链转化为训练数据。该方法保留了课程的递进性与推导的逻辑结构,而非孤立的解题对,使语言模型能获得更系统的数学推理监督,提升证明能力。
👍 42
11/12 08:00
Video generation models produce visually realistic videos but still struggle with physical dynamics and causal consistency. Learning from generated videos requires fine-grained physical supervision, yet reliable quality judgments, correction pairs, and controlled variations are difficult to obtain.
中文介绍 PhysLoop利用LLM的物理常识,为视频生成模型提供自改进的物理监督。通过生成视频中提取细粒度物理反馈,构建校正对与受控变化,无需人工标注。模型在生成时自我修正物理一致性,显著提升视频的动态真实感与因果连贯性。
👍 37
11/11 08:00
In this report, we introduce the Reciprocal World Model (RWM) and present Duo v1, a full-duplex world model that continuously perceives audiovisual inputs while responding to users in real time. The RWM treats intelligence and physical reality as equally fundamental to what constitutes a world. Exis
中文介绍 本文提出全双工世界模型Duo v1,基于相互世界模型(RWM)框架,持续感知音视频输入并实时响应用户。RWM将智能与物理现实视为同等基本,打破了传统单向交互范式。Duo v1在对话中同时处理输入与输出,实现更自然、低延迟的交互体验。
👍 35
11/10 08:00
Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental conditions, producing environment transitions that diverge from successful full-precision references. Co
中文介绍 SteerDistill通过闭环教师引导,实现LLM智能体的低比特量化。分析失败轨迹发现,量化智能体常忽略任务要求或环境条件,导致状态转移偏离全精度参考。该方法利用教师信号在训练中引导量化模型修正行为,在降低推理成本的同时保持任务成功率。
👍 30
11/11 08:00
Recent work suggests that pretrained image generators encode real-world understanding beyond image synthesis. Vision Banana exploits this by framing vision tasks as conditional image editing. However, its data, weights, code, and implementation details are not public. We present Open Vision Banana (
中文介绍 本文开源Vision Banana的完整复现方案,包含数据、权重、代码与实现细节。该方法将视觉任务重新表述为条件图像编辑,利用预训练图像生成器对真实世界的隐式理解。Open Vision Banana提供了完全可复现的生成式视觉感知流程,推动该方向研究透明化。
👍 29
11/12 08:00
Decision models answer a question with a typed output, such as a choice among options with a probability for each, so one model can choose an agent's actions, evaluate actions already taken, or score every action to teach a policy with reinforcement learning. Prior work has studied them mainly as ac
中文介绍 本文系统研究决策模型在长时程智能体中的三种角色:行动者、评判者与教师。通过统一框架分析一个模型如何选择动作、评估已执行动作或为强化学习提供策略评分。实验揭示不同角色对长时程决策性能的影响,为多任务决策模型设计提供指导。
👍 26
11/04 08:00
Hyper-Connections (HC) architectures are increasingly adopted in frontier language models, yet their components remain insufficiently understood and introduce non-negligible training overhead. In this work, we systematically study the HC architecture to understand its components and identify simpler
中文介绍 本文系统研究Hyper-Connections(HC)架构的组成,以简化设计、提升稳定性与可扩展性。HC在前沿语言模型中应用广泛,但组件理解不足且训练开销大。作者识别出关键组件,提出简化版本,降低训练成本同时保持性能,并验证其扩展性。
👍 21
11/12 08:00
Decoder-only Large View Synthesis Models (LVSMs) with KV-cache have recently achieved state-of-the-art quality by regressing novel views from neural networks without reconstructing 3D geometry, while Test-Time Training (TTT) layers save computation but sacrifice quality. We propose Test-Time Baking
中文介绍 本文提出Test-Time Baking(TTB)技术,用于高效渲染仅解码器的大型视图合成模型。LVSMs利用KV缓存实现新视图回归,但TTT层节省计算却牺牲质量。TTB在测试时烘焙特征,在不重建3D几何的前提下,兼顾渲染速度与视图合成质量。
👍 21
11/07 08:00
Web-facing AI agents repeatedly access webpages, documents, repositories, and tool responses, yet reading interfaces often expose entire objects when only sparse evidence is needed. This over-reading increases token and latency costs and can dilute relevant evidence. We present SparseRead, a reusabl
中文介绍 本文提出SparseRead,一种可复用的稀疏读取接口,供AI智能体按需从网页、文档、仓库中提取关键证据。该接口显著减少过度读取带来的token和延迟成本,同时避免稀释相关信息。实验表明,在保持任务性能的前提下,有效降低推理开销。
👍 14
11/12 08:00
Multimodal coding agents decouple 4D world construction from visual generation, repurposing video models from world generators toward generative renderers. This division of labor raises a fundamental question: if code builds the world, can video models just render it? A reliable generative renderer
中文介绍 本文研究多模态编码智能体将4D世界构建与视觉生成解耦后,视频模型能否仅充当生成渲染器。WorldDraft框架让代码构建世界,视频模型渲染视图。实验证明可靠的生成渲染器能够忠实再现代码指定的世界状态,为分解式世界建模提供新范式。
👍 12
11/12 08:00
How multimodal language models should encode visual input remains up for debate. Most models use a pretrained vision encoder, while some recent recipes train the encoder from scratch with the language model or remove it entirely. Vision encoders bundle together three design choices: a visual prior f
中文介绍 本文解构多模态视觉编码中的三个捆绑设计选择:视觉先验、适配器与训练策略。通过控制变量实验,揭示各组件对下游任务的影响,并评估从头训练编码器与去除编码器的可行性。研究为多模态语言模型的视觉编码设计提供更清晰的决策依据。
👍 12
11/12 08:00
Second-order optimizers (e.g., Shampoo, SOAP) have emerged as compelling alternatives to Adam/AdamW and promise faster per-step convergence than AdamW in transformer training. However, they still remain costly in practice, and even SOAP still pays substantial optimizer-state memory and eigendecompos
中文介绍 本文提出Factored One-sided Adam-Moment(FOAM),通过因式分解与单侧更新降低SOAP优化器的状态内存与特征分解成本。该方法在Transformer训练中保持SOAP的收敛速度优势,同时显著提升内存效率与可扩展性,使二阶优化更实用。
👍 12
11/09 08:00
Agentic workloads are a major source of LLM inference demand. Existing prefill--decode (PD) disaggregated serving systems offer limited flexibility for runtime adaptation of compute allocation and KV cache management, and rely on coarse signals that only partly capture the diverse and dynamic agent
中文介绍 本文提出Evolving Infra,一种自适应LLM服务系统,针对智能体工作负载的动态特性。现有PD分离系统无法灵活调整计算分配与KV缓存管理。Evolving Infra利用细粒度信号实时感知智能体行为,动态分配资源,在规模化服务中提升吞吐并降低延迟。
👍 8
11/05 08:00
Self-improving embodied agents are increasingly being developed and compared on robot simulation benchmarks. Several recent works expose the benchmark's success criteria to the learning agent, a practice we call Criteria Exposure (CE). These criteria, originally written to verify policies trained by
中文介绍 本文质疑具身自改进智能体中暴露基准成功标准(Criteria Exposure)的做法。研究对比不同暴露策略,发现学习任务与学习验证器存在本质区别。暴露标准可能使智能体优化基准而非真正技能,误导自改进方向。作者呼吁更严谨的评估协议。
👍 5
11/04 08:00
Subliminal learning is a phenomenon in large language models (LLMs) where a teacher passes a hidden preference to a student through training data unrelated to that preference. We study whether subliminal learning also occurs in vision-language models (VLMs). We finetune a teacher VLM to prefer a spe
中文介绍 本文研究视觉语言模型中的阈下学习现象。教师VLM通过训练数据中无关的偏好信息,将隐藏偏好传递给学生模型。实验证明该效应在跨模态场景中同样存在,且与学生模型微调数据内容无关。该发现揭示了模型训练中隐性知识传递的风险。
👍 4
11/12 08:00
While search agents demonstrate impressive capabilities in multi-step question answering, their robustness to poor-quality evidence remains under-explored. Such evidence is rare in benchmark corpora, and resulting failures are hard to reproduce. We propose DeepStress, a dynamic stress-testing framew
中文介绍 本文提出DeepStress,一种动态压力测试框架,用于评估深度搜索智能体对低质量证据的鲁棒性。现有基准中此类证据稀少且难以复现。DeepStress生成可控的干扰证据,压力测试多步问答系统,揭示其失败模式并提升鲁棒性评估的可靠性。
👍 4
11/10 08:00
Local decoding usually fixes one weight precision for the whole generation. We study a different allocation: store 8-bit weights, decode with their top 4 bits, and redo a step at 8 bits only when the low-precision output distribution looks uncertain (entropy above a threshold or a small top-two marg
中文介绍 本文提出按需精度分配策略:本地解码时默认使用8位权重的顶部4位,仅在输出分布不确定时(如高熵或前二差距小)回退至8位重新计算。该方法在保持生成质量的同时,显著降低小型本地模型的平均解码位宽,提升效率。
👍 3
11/07 08:00
Self-evolving agents improve their own skills, memory, and harness code from experience without retraining the base LLM, and group-evolving agents go further by sharing these lessons across agents. Yet most such systems overlook that sharing across users and heterogeneous agents requires deciding wh
中文介绍 本文提出多粒度记忆提取框架,用于多用户、多智能体交互场景。自进化智能体从经验中提升技能、记忆与代码,但跨用户共享需确定粒度。该框架自动决定哪些经验可共享及如何聚合,优化组内协作,避免无关信息干扰个体进化。
👍 3
11/12 08:00
Reference-frame stabilization registers every frame of a hovering-drone video onto one reference frame, keeping object positions comparable. The registration pipelines it relies on are selected on feature-matching benchmarks in which only the camera moves. Over a busy intersection, vehicles moving t
中文介绍 本文构建动态城市场景下的无人机视频稳定基准,并引入共识捕获方法。参考帧稳定依赖特征匹配,但现有基准仅考虑相机运动。在繁忙路口,移动车辆产生冲突匹配。所提方法通过多帧共识投票,提升动态场景下的注册稳定性与一致性。