👍 202
09/28 08:00
Latent visual reasoning (LVR) enables multimodal large language models (MLLMs) to perform intermediate computation in continuous latent tokens rather than expressing every reasoning step in words. However, unlike textual CoT, latent reasoning is not directly observable, making it difficult to superv
中文介绍 针对潜空间视觉推理(LVR)中中间推理步骤不可观测、难以监督的问题,提出将潜推理与视觉证据显式对齐的方法。通过在连续 latent tokens 上建立可验证的视觉 grounding,使 MLLM 的隐式计算过程可被监督与评估,改善 latent reasoning 的可解释性与可靠性。
👍 186
09/30 08:00
Self-evolving search agents build their own training curricula by jointly optimizing a proposer that generates questions and a solver that answers them. This closed loop introduces a failure mode we call co-cheating: the proposer and solver increasingly agree on shared errors, so internal reward imp
中文介绍 自进化搜索智能体中提议器与求解器协同优化会产生共作弊(co-cheating):双方趋于在共享错误上达成一致,内部奖励失真。本文诊断该失效模式并提出缓解策略,打破闭环中的虚假一致性,提升自生成课程训练的真实性与泛化能力。
👍 97
09/30 08:00
Terminal agents act through stochastic model generations, yet the ability to generate a useful action does not ensure its reliable execution. A poor command (e.g., wrong package install) can change the environment in ways that hinder subsequent progress, even when the model could generate a better a
中文介绍 终端智能体通过模型生成动作,但生成可用动作不等于可靠执行。本文提出 Mid-Harness,在模型与执行层之间插入可扩展的中间层,对动作进行验证、修正与重试,避免错误命令污染环境,提升长程任务中动作执行的成功率。
👍 63
09/29 08:00
Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when prediction
中文介绍 LLM 智能体在未见环境上迁移能力弱,world-model 方法需额外训练且预测误差会累积。EVOKE 通过从 LLM 内部知识中直接提炼可迁移的世界知识来辅助决策,无需显式预测未来观测,在减少训练成本的同时提升跨环境决策性能。
👍 31
09/29 08:00
Long-form subtitle translation requires reasoning over discourse and cultural context spanning episodes or entire series, while maintaining consistent terminology and style. Existing single-LLM methods are largely sentence-level, and multi-agent systems often use static workflows that do not adapt t
中文介绍 长篇字幕翻译需跨集数/季保持术语与风格一致并处理文化语境。Breaking Babel 提出自进化多智能体系统,各智能体角色动态演化而非静态工作流,能根据上下文持续调整翻译策略,在跨集一致性上显著优于单 LLM 方案。
👍 30
09/29 08:00
Recent advances in musculoskeletal modeling and reinforcement learning have enabled muscle-actuated agents to reproduce increasingly complex human motions. Yet these capabilities remain largely confined to flat ground, in part because motion datasets rarely include aligned terrain geometry and becau
中文介绍 肌肉驱动的强化学习智能体虽能复现复杂人体动作,但局限于平面且缺乏地形对齐数据。TERRA 提出地形感知的重建、重定向与控制框架,在生成动作时同步考虑地形几何,使 musculoskeletal locomotion 可在起伏地形上稳定运行。
👍 24
09/29 08:00
We introduce LoopVL to study whether Loop Transformers can be effectively extended to vision- language models. LoopVL combines Module-Loop and Model-Loop computation to iteratively update a unified vision-language state through shared modules. We train LoopVL from scratch through language pre-traini
中文介绍 LoopVL 探索将 Loop Transformer 扩展到视觉-语言领域,结合 Module-Loop 与 Model-Loop 通过共享模块迭代更新统一的视觉语言状态。从零开始训练,包含语言预训练与视觉对齐阶段,验证了循环架构在视觉智能任务中的有效性。
👍 24
09/30 08:00
Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects, and producing verifiable results. We thus introduce OSWorld-Science, a benchma
中文介绍 OSWorld-Science 是面向计算机使用智能体的科学软件基准,要求 VLM 解读专业界面、操作科学对象并产出可验证结果。该基准覆盖了真实科研工作流的端到端任务,揭示了现有智能体在科学软件操作中的系统性不足。
👍 16
09/26 08:00
Reward models score responses from large language models (LLMs) and guide LLM training toward human preferences. However, reward models can favor superficial attributes such as length or confidence, leading LLMs to produce higher-scoring but not more correct responses. Existing mitigation methods ei
中文介绍 奖励模型易偏向长度或置信度等表层特征,导致 LLM 产生高分但不正确的回复。BiasReducer 提出自适应偏差缓解方法,在训练奖励模型时动态调整对表层偏差的惩罚强度,使奖励信号更准确地反映回答质量。
👍 15
09/29 08:00
Modern agentic systems combine an AI model with a harness that controls execution and environmental interactions. Harness design strongly affects long-horizon performance, yet its combinatorial search space demands substantial human effort that must be repeated as models change. Existing automated m
中文介绍 Harness 设计对智能体长程性能影响巨大,但其组合搜索空间大且随模型迭代需反复调整。MILO 通过编排多智能体协同进化自动发现最优 harness 配置,减少人工设计成本,并能随模型更新自适应重新优化执行架构。
👍 15
09/30 08:00
Looped transformers and Mixture-of-Experts (MoE) offer complementary routes to efficient scaling: recurrence increases computational depth at fixed parameters, while MoE sparsity expands total capacity at fixed active compute. Yet existing scaling laws model recurrence or sparsity in isolation. In t
中文介绍 循环 Transformer 与 MoE 分别以计算深度和稀疏容量实现高效扩展,但已有 scaling law 将二者独立建模。本文推导联合循环 MoE 的扩展律,揭示循环次数与专家稀疏性之间的交互效应,为同时利用两种机制的架构设计提供理论指导。
👍 12
09/26 08:00
Software development requires more than editing code: developers repeatedly run software, interact with its interfaces, visually inspect its behavior, and use these observations to decide what to change next and whether a change works. Existing coding agents and computer-use agents are largely studi
中文介绍 软件开发不仅是编辑代码,还需反复运行、可视化检查界面并据此迭代。CUA-SWE 将计算机使用智能体与软件工程结合,研究视觉反馈驱动的迭代式开发流程,填补了现有 coding agent 与 computer-use agent 之间在真实视觉编程任务上的空白。
👍 7
09/29 08:00
TTS systems with autoregressive semantic modeling have demonstrated strong zero-shot voice cloning performance and rich expressive variation, but their sequential decoding incurs substantial latency. Non-autoregressive alternatives offer much faster generation, yet often rely on more restrictive ref
中文介绍 自回归 TTS 在零样本声音克隆上表现强但延迟高。Tacit-TTS 将解码方式从自回归转为掩码预测(masked prediction),在无转录条件下实现非自回归快速生成,保持丰富表现力的同时大幅降低推理延迟。
👍 6
09/30 08:00
Latent communication enables multi-agent systems to exchange information directly in internal representation space, reducing the token, computation, and latency overhead of text-based communication. To this end, lightweight trainable links are introduced to map the sender's representations into the
中文介绍 多智能体系统中潜空间通信可减少 token 开销,但也可泄露内部表征引发安全威胁。本文系统分析 latent communication 的安全风险,包括信息泄露与对抗操纵,并提出基于可训练轻量映射的防护方案。
👍 6
09/30 08:00
Self-attention gives LLMs fine-grained, query-dependent access to context, but dense token interactions incur quadratic prefill cost and a key--value cache growing with context length. Research thus spans explicit-memory compression, sparse access, recurrent state construction, structured state dyna
中文介绍 系统综述 LLM 中注意力机制的演进:从 dense attention 到显式记忆压缩、稀疏访问、循环状态构建与结构化动态。梳理各方案在计算成本、KV cache 长度与模型容量之间的权衡,并总结 emerging trends。
👍 5
09/30 08:00
We study training LLM judges from natural language feedback, especially for subjective tasks where the verdict depends strongly on which evaluation criteria the judge invokes and how it weighs them. The dominant approach, outcome-supervised RL (e.g., GRPO), credits every token in the rollout with a
中文介绍 针对主观任务中 LLM 评审依赖评判标准选择的问题,提出位置选择性自蒸馏方法,从自然语言反馈中训练 LLM judge。相比 outcome-supervised RL 对全部 token 等权奖励,该方法精准定位关键评判位置的信用分配。
👍 5
09/28 08:00
Large language model (LLM) routing aims to select the most suitable model for each incoming query. Most existing routers learn this decision directly from query embeddings, model representations, preference data, or clusters of similar examples. Such approaches can be effective, yet the representati
中文介绍 LLM 路由需为每个查询选择最合适的模型。SeLMRoute 提出基于语义证据的概率路由方法,将查询分解为语义片段并分别评估各模型表现,再以概率聚合输出最终路由决策,优于直接从嵌入向量学习路由的方案。
👍 4
09/28 08:00
Artificial-intelligence (AI) agents hold promise for automating bioimage analysis, yet no benchmark evaluates whether they can carry out real-world analyses end to end. Such analyses are hard for agents because 2D images, 3D volumes and time-lapse sequences are often too large to read as context, so
中文介绍 BIABench 是首个评估 AI 智能体端到端执行真实生物图像分析的基准,涵盖 2D 图像、3D 体数据与时序序列等多模态输入。由于数据体量远超上下文窗口,基准考察智能体的分步推理与工具调用能力。
👍 4
09/30 08:00
We find that major reported improvements in decoding words from non-invasive brain recordings are largely reproducible without any brain data. In the influential work of d'Ascoli et al. (2025), time series of brain activity from subjects perceiving continuous speech are segmented into fixed-length w
中文介绍 发现无创脑电解码文字的重大改进在移除脑数据后仍可复现,说明存在时序捷径(timing shortcuts)。本文证明 d'Ascoli et al. (2025) 等工作中的脑信号贡献被高估,并提出消除时序捷径的方法以真实评估脑解码性能。
👍 4
09/29 08:00
Skills equip LLM agents with professional knowledge and guidance to complete long-horizon and complex tasks. Although skills have been widely adopted in recent agent paradigms and harnesses, how to synthesize reliable training data and how to train agents for skill use remain underexplored. In this
中文介绍 SkillGym 解决技能型智能体训练数据合成与技能使用训练不足的问题,通过自动生成可验证的交互环境来产生可靠训练数据,使 LLM agent 能学习调用专业技能完成长程复杂任务,无需大量人工标注。