👍 80
09/24 08:00
While Large Language Models (LLMs) rely on highly non-linear components, in this work we demonstrate that they exhibit fundamental linearity: when inputs from distinct text streams are linearly combined, the model outputs a superposition of the individual next-token distributions. We term this the S
中文介绍 提出 SLOP(线性叠加现象),表明 LLM 在多条独立文本流的输入做线性组合时,其输出是各单流下一 token 分布的线性叠加。该发现揭示高度非线性 Transformer 内部存在基底层级的线性结构,为线性探测和模型可解释性提供了新证据。
👍 45
09/22 08:00
Prefill and decode reward different approaches to quantization: low-precision arithmetic accelerates prompt processing, while compact weights reduce memory traffic during generation. We propose "disaggregated quantization" (DQ), which specializes computation formats, weights and storage placement to
中文介绍 提出「分离式量化」DQ,针对 LLM 推理中 prefill 与 decode 阶段量化目标不同的问题,分别为两者定制计算精度、权重格式与存储布局。prefill 用低精度加速 prompt 处理,decode 用紧凑权重降低访存开销,从而在同硬件上同时提升吞吐与生成速度。
👍 21
09/24 08:00
Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure
中文介绍 公开可复现的 GLM-4.5-Air-Base(106B-A12B)后训练配方 Rufus-Air,按串行流水线组织 8 个阶段:SFT、Reasoning RL、Coding RL、Instruction-Following RL、General/Coding/Search Agent 与 RLHF。论文完整记录数据、奖励设计与基础设施,便于社区复现。
👍 20
09/25 08:00
Scaling language models to long contexts is limited by the quadratic cost of self-attention. Block sparse attention offers an efficient alternative, but selecting the retained blocks remains a bottleneck. Conventional block selection requires scoring all query-block pairs and therefore remains quadr
中文介绍 提出具有对数线性复杂度的块稀疏注意力。传统块选择需对所有 query-block 对打分,仍为二次复杂度;本文设计可在对数线性时间内选出保留块的方案,从而在长上下文场景下显著降低 self-attention 成本,同时保持精度。
👍 13
09/23 08:00
Recently, Large Language Models (LLMs) have been increasingly able to solve advanced mathematical problems, including many that have been open for decades. This opens the door to expansion of mathematical knowledge at unprecedented scale. Yet, while LLMs may be able to conjecture and prove more and
中文介绍 探索让 LLM 主动发现有趣数学问题的能力。在 LLM 已能求解大量长期开放难题的背景下,本文设计方法引导模型生成有价值的新猜想与证明方向,为以自动化方式扩展数学知识、发现新定理开辟路径。
👍 11
09/24 08:00
Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discrete decisions are tightly coupled to geometric, kinematic, and dynamic constraints. Generalized TAMP addresses this difficulty by exploiting regularities across problem insta
中文介绍 提出面向广义任务与运动规划(TAMP)问题的 coding agent。通过利用问题实例之间的共性,让代码生成式 agent 从自然语言任务描述直接构造解算程序,将离散决策与几何、运动学、动力学约束耦合求解,提升泛化性。
👍 11
09/24 08:00
We introduce PUBG Ally, an embodied agent for PUBG: BATTLEGROUNDS that can reason, act autonomously, and play alongside players as a voice-enabled teammate. Building such a teammate requires combining two difficult capabilities: it must perceive and respond to a constantly changing game world under
中文介绍 提出 PUBG Ally,一个可在 PUBG:BATTLEGROUNDS 中作为语音队友参与的具身 agent。它需同时具备在快速变化的游戏世界中进行感知推理与自主操作的能力,并通过语音与玩家实时协作,是复杂多模态具身 AI 的一项落地示例。
👍 10
09/24 08:00
Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained wit
中文介绍 提出 Jev,一个基于强化学习训练的校准决策模型,可作为 AI 对齐失败检测器的零样本方案。相比逐条解码的生成式评审或每次仅给一个固定标签的分类器,Jev 能在单次前向中输出带校准概率的多准则评分。
👍 9
09/24 08:00
Deep search requires LLM agents to decompose complex queries, search for evidence, and synthesize grounded answers, yet existing ReAct-style agents suffer from two limitations: role coupling, where one policy must handle planning, evidence use, and synthesis; and context accumulation, where growing
中文介绍 提出 IterSynth,重新设计深度搜索 agent 的迭代合成范式。针对 ReAct 风格 agent 中规划、证据使用与合成耦合并同上下文不断累积的两类缺陷,本文通过角色解耦的迭代合成解耦职责、控制上下文增长,提升合成答案的接地性与稳定性。
👍 8
09/24 08:00
Jev is a fast, low-cost decision model that answers natural-language questions with choices, binary judgments, and scores. As its public ecosystem grows rapidly, it remains unclear how Jev is used across applications and how public attention relates to project distribution. To answer these questions
中文介绍 对 Jev 这一低成本决策模型的公开生态进行数据驱动分析,涵盖其功能边界、典型应用场景以及社区关注度与项目分发之间的关系。基于公开数据刻画 Jev 的实际使用情况与生态演化,为理解此类专用小模型的落地提供实证依据。
👍 8
09/24 08:00
In this paper, we propose RGBD20K, a novel dataset for facilitating the development of more robust and general RGB-D semantic segmentation by encompassing abundant categories and high-quality annotations. RGBD20K possesses several attractive properties: (1) Expanded Semantic Space. In particular, it
中文介绍 提出 RGBD20K,一个面向 RGB-D 语义分割的大规模基准数据集,覆盖更广泛的类别空间并配备高质量标注。其扩展的语义空间与丰富的实例多样性旨在推动更鲁棒、更具泛化性的 RGB-D 语义分割算法的发展。
👍 8
09/19 08:00
Modern Transformer design and compression both reduce to allocating capacity under a budget. The standard scalars for these decisions, #Params and #FLOPs, capture size and compute but not architectural structure: two architectures with identical parameter budgets but different depth-width, head, or
中文介绍 提出「神经谱容量」作为架构容量度量。传统 Params 与 FLOPs 只反映规模与算力,无法区分深度-宽度、head 等结构差异;该方法仅从网络结构规范出发度量并指导架构与压缩设计,为 Transformer 架构搜索与量化提供更细粒度指标。
👍 7
09/23 08:00
Time series agents answer analytical questions by calling external tools, and which tools they carry is decided by people before the agent runs. However, we identify two failures in this setup. Human-Agent Tool Misalignment: a library of 21 expert-curated tools helps on some tasks and hurts on other
中文介绍 提出 TimeEvo,让时间序列 agent 在失败驱动下自演化其工具集合。作者指出专家预置的 21 个工具存在人机工具错配,且部分任务会被多余工具拖累;本文通过运行时的失败反馈自适应增删工具,提升分析准确性与效率。
👍 7
09/25 08:00
Existing multi-agent benchmarks primarily test in competitive settings, short-horizon interactions under 20 steps, or simply aggregate individual performance, failing to isolate and highlight genuine collaboration capabilities of LLM-based agents. We introduce AgentWorld, a benchmark of 100 human-an
中文介绍 提出 AgentWorld 基准,包含 100 个人工设计的长时程协作场景,用于评测多 LLM agent 的真实协作能力。相较现有偏竞争、短步数或仅聚合个体表现的基准,AgentWorld 更强调长地平线、强耦合的协作任务。
👍 7
09/24 08:00
Tool-calling agents produce heterogeneous outputs, interleaving structured tool invocations with user-facing natural language summaries. This output heterogeneity presents a structural failure mode in standard on-policy Reinforcement Learning (RL): algorithms like GRPO indiscriminately broadcast a h
中文介绍 提出 SLCA-GRPO,解决工具调用 RL 中的跨段信用错配问题。工具调用 agent 的输出混合结构化调用与面向用户的自然语言总结,标准 GRPO 会将统一奖励无差别广播到所有段;本文按段分配奖励,避免信用污染,提升策略学习质量。
👍 7
08/29 08:00
Modern AI agents routinely cross trust boundaries: they ingest untrusted content, combine it with privileged instructions, persist intermediate beliefs in long-term memory, and invoke privileged tools. This creates an attack surface in which malicious payloads can enter through model inputs and caus
中文介绍 提出 AgentKernel,一个「信任原生」的 agentic 操作系统。针对现代 AI agent 频繁跨越信任边界、恶意载荷可从模型输入渗透至记忆与工具调用的问题,AgentKernel 在系统层内建隔离与权限控制,缩小攻击面。
👍 6
09/24 08:00
Long-horizon reasoning remains a central challenge for large language models (LLMs) under sparse-reward regimes. We argue that this brittleness arises from two biases induced by complex reasoning spaces: an exploration bias, where models are drawn toward locally plausible but structurally unstable b
中文介绍 提出 SAGE,通过拓扑引导缓解长程推理在稀疏奖励下的偏差。作者指出复杂推理空间中模型存在探索偏差与不稳定结构偏好两类脆弱性,SAGE 引入拓扑先验引导搜索路径,提升长程推理的稳定性与成功率。
👍 5
09/23 08:00
World-action models (WAMs) transfer visual and motion priors from pretrained video generators to robot control by jointly modeling visual dynamics and actions. Existing WAMs, however, predict dense future frames during training, repeatedly modeling largely unchanged content and coupling action-condi
中文介绍 提出 DeltaWAM,用于双臂操作的世界动作模型。针对现有 WAM 在训练时反复预测大量静态未来帧、将动作条件与视觉建模耦合过紧的问题,DeltaWAM 仅建模视觉与动作的增量变化,提升训练效率与动作控制精度。
👍 4
09/24 08:00
Banking assistants must use account-specific information to answer requests and, in many cases, take actions through tools. Evaluating only the final response misses important errors. An assistant may ask for information it already has, rely on stale context, select the wrong account, or write an in
中文介绍 提出 IndicBankBench,面向印度零售银行场景的 LLM 助手安全与可靠性评测基准。除最终响应外,还评测助手在工具调用、账户选择、上下文时效性等中间环节的正确性,覆盖印度本地化银行业务与合规需求。
👍 4
09/20 08:00
Adapting large language models to individual users remains challenging due to the tension between fine-grained personalization and scalable deployment. We present CARD, a hierarchical framework that achieves effective personalization through progressive refinement. CARD first clusters users accordin
中文介绍 提出 CARD,一个用于个性化文本生成的层次化框架。CARD 先按用户群体聚类,再在奖励引导的解码中做渐进细化,从而在保留可规模化部署的同时实现细粒度个性化,缓解个性化与可扩展性之间的张力。