👍 441
08/28 20:00
Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring this goal within reach, as modern LLMs show the capability to independently implement solutions and learn from the execution outcomes. Behind these gains, post-training (especially
中文介绍 论文提出了一种利用世界模型来自动化实证研究的框架。通过现代大型语言模型(LLM)独立实现解决方案并从执行结果中学习,作者展示了自动研究代理(AutoResearch)在研究中的潜在应用。该方法依赖于后训练技术的强化,推动了实证研究的自动化发展。意义:该研究对 Agent、World Model 和自主决策等领域具有重要影响。
👍 192
09/08 20:00
We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and mo
中文介绍 NCP-ArchPreview 是一种潜在空间语言模型,超越了传统的下一个标记预测(NTP)。它同时通过下一个概念预测(NCP)来学习,能够预测跨多个标记的离散概念。这种方法在多个计算任务中展示了更明确的概念学习能力,推动了自回归预训练的研究。意义:此研究对语言模型的推理和概念理解方向具有重要推动作用。
👍 91
09/07 20:00
Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language fee
中文介绍 本研究提出了 AgentGrad,一种针对多智能体系统(MAS)的干预指导提示优化方法。通过使用自然语言反馈的文本梯度方法,研究解决了基于大型语言模型的 MAS 在提示设计上的依赖。该方法提高了多代理系统的性能,具有广泛的应用潜力。意义:对 Agent 和协作学习等方向提供了新的思路。
👍 79
09/06 20:00
Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this
中文介绍 研究提出了一种通过合成块堆叠问题来增强大型视觉语言模型(LVLM)的空间智能能力。尽管 LVLM 在多种视觉任务中表现出色,但在对 2D 图像中的 3D 场景进行重建和推理的能力上仍然有限。论文提出的方法在这方面进行了改进。意义:对计算机视觉和场景理解领域有着重要的促进作用。
👍 48
09/09 20:00
Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool
中文介绍 该论文提出了 T1,一种面向长时间任务(如编程和科学发现)的混合专家模型,经过增强学习训练。T1 在真实云沙盒中控制 shell,能够执行超过 300 种工具操作。通过该模型,长时间任务的处理能力得到显著提升。意义:为 AI 在长时程复杂任务上的应用提供了新的可能性。
👍 47
09/04 20:00
Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed
中文介绍 EvoSafeHarness 提出了模型及领域特定的进化安全监控框架,以应对大型语言模型代理的安全性问题。该框架在模型自防御之外增加了系统级的安全性,显著增强了对直接有害请求的应对能力。研究为确保 AI 系统在实际应用中的安全性提供了新思路。意义:在安全 AI 和代理系统的开发领域具有重要意义。
👍 35
09/03 20:00
Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user's longitudinal wearable record. We introduce WearableQA, a benchmark comprising 4,084 10-option multiple-ch
中文介绍 WearableQA 是新提出的基准,旨在评估 AI 系统在真实用户的长期可穿戴数据上的推理能力。该基准包含 4,084 个具有 10 个选项的多项选择题,提供了对可穿戴设备监测的深入分析。通过此研究,评估了 AI 在健康推理中的实际应用能力。意义:对医学和健康数据分析领域的 AI 研究具有推动作用。
👍 28
09/09 20:00
Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks perturbs the embeddings consumed by the decoder and can cause deletion and premature end-of-sequence errors. We introduce X-AuT, a progressive framework that selects layer combination
中文介绍 X-AuT 提出了一种逐步音频编码器压缩框架,通过跨尺度蒸馏技术降低语音大模型的推理成本。研究着重如何在不扰动解码器嵌入的情况下,选择层组合来保持输出质量。此进展促进了语音模型在资源受限环境下的应用。意义:对语音处理和模型优化技术的进步有积极的推动作用。
👍 23
09/07 20:00
SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: reward hacking, enabled by leakage of gold solutions or hidden evalua
中文介绍 SWE-Bench Pro 是新标准的基准,旨在评估软件工程代理在复杂任务中的表现。研究显示其评估受到了奖励黑客和评估漏泄等多种因素的影响,提出了改进建议。该基准为软件开发中的 AI 应用提供了可靠检测手段。意义:推动了 AI 在软件工程领域的应用和研究标准化。
👍 22
09/09 20:00
Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their co
中文介绍 本论文提出的 Memory as Plans 旨在解决复杂的非马尔可夫任务,通过长时记忆补充现有的机器人策略。该方法超越常规的语言总结机制,提供了一种基于记忆的计划生成框架,显著提高了任务执行的灵活性和有效性。意义:对智能体与记忆相关研究方向具有重要贡献。
👍 20
09/05 20:00
Sequential memory agents process long documents by reading chunks one after another while maintaining a compact memory state, coupling document traversal to reasoning depth. This coupling introduces sensitivity to evidence placement and ties inference latency linearly to document length. We introduc
中文介绍 PARSER 提出了并行读取和深入推理的机制,以解除长文本 LLM 代理的限制。该方法通过对文档各个部分的顺序处理,解决了推理深度与文档遍历的联动,为长文档处理提供了新思路。意义:有助于改进 LLM 在处理复杂信息时的能力。
👍 18
09/07 20:00
While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essent
中文介绍 SAEScientist-Bench 是用于评估 AI 代理在自主 SAE 可解释性研究中的有效性的新基准。研究表明,后期监控与审计是确保模型安全对齐的关键因素,为理解模型学习提供了新的监控工具。意义:对 AI 模型的透明性和诚信研究有助益。
👍 17
09/08 20:00
We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It covers six real metro systems, ranging from 37 to 414 stations, and eleven categories that include routing, fare calculation, disruptions, accessibility, and adversarial input. In
中文介绍 MetroLLM-Bench 提出了一个针对语言模型作为地铁售票机政策层的新基准,涵盖多个真实地铁系统的 955 个案例。该评估涉及路线规划、票价计算等多个维度,旨在全面测试模型的实用性和安全性。意义:为语言模型在公共服务领域的应用提供了标准化评估依据。
👍 17
08/31 20:00
Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video
中文介绍 TempCloze 研究了一种视频 cloze 基准,旨在评估视频-LLMs 的视觉时间推理能力,减少基于语言的推理偏差。此基准通过提供多样化的视听信息,帮助 AI 更准确地识别缺失的时间片段。意义:对视频理解和时间推理研究有重要作用。
👍 17
08/27 20:00
Quantization has been widely adopted in LLM training and inference to reduce cost and improve efficiency. However, low-bit quantization of the attention module often introduces large errors at very low bit-widths, causing performance degradation. Existing methods mainly rely on smoothing techniques
中文介绍 HyQuant 提出了混合精度量化方法,以降低大型语言模型(LLM)中的注意力模块成本和提高效率。研究重点在于如何在保持性能的同时减少量化引入的误差,为 LLM 的应用提供了新视角。意义:对模型压缩和资源优化方向有着重要的推动作用。
👍 17
09/09 20:00
Code world models represent worlds as executable programs, but this representation alone does not determine how to construct a complex world. We introduce Recursive Code World Models (RCWM), a framework for reconstructing complex 3D worlds in code from a single reference image. RCWM couples a Recurs
中文介绍 Recursive Code World Models (RCWM) 提出一种通过递归场景程序构建复杂世界的框架,能够从单一图像重建复杂 3D 世界。同时,该模型将增强学习与代码生成结合,有效扩展了世界建模的能力。意义:对自动化场景建模与 AI 领域的合成研究提供了新思路。
👍 17
09/06 20:00
AI research agents combine prior knowledge, public sources, and experimental feedback to produce useful results. The Discovery Certification Protocol (DCP) turns claims about these results into executable recovery and feedback tests. Gate 1 validates useful improvement on sealed evaluation. Gate 2 g
中文介绍 位于 AI 研究前沿的 Discover Certification Protocol (DCP) 旨在审核 AI 研究代理的结果,通过可执行的恢复和反馈测试验证研究结果的有效性。该协议为 AI 研究中的信号验证提供了一种标准化的认证方式。意义:提高了 AI 研究的透明度与可信度,为代理系统的发展奠定基础。
👍 15
09/09 20:00
On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the
中文介绍 Negative Self-Distillation (NSD) 研究了一种通过避免错误来增强推理能力的方法。相较于传统的在政策自我蒸馏(OPSD)中利用真实解决方案的模式,NSD 通过减少错误传播,提升了模型的整体性能。意义:对自我改进与推理能力提升的研究方向具有显著促进作用。
👍 14
09/09 20:00
All-in-One medical image restoration (MedIR) aims to address diverse tasks across modalities and degradation types using a single universal model. Existing methods typically prioritize modeling inter-task heterogeneity (e.g., distinct data distributions and degradation types). However, they largely
中文介绍 UniH^3 提出了一个通用的医疗图像恢复模型,能有效处理多种模态和退化类型。研究展示了在各类任务中保持任务间同质性和异质性的能力,为医疗图像处理领域注入了新的思路。意义:对一体化医疗图像处理和相关技术的进步有显著推动。
👍 14
09/07 20:00
World models are increasingly used as policy-in-the-loop imagination environments, where reliable rollouts require fine-grained controllability with respect to low-level robot actions. A key obstacle to scaling such models in robotics is that actions are not a universal language in pixel space: chan
中文介绍 SyncWorld 通过视觉校准实现世界模型的零-shot 模拟能力,允许在控制复杂机器人动作上具备更高的灵活性。该方法克服了现有世界模型在操控时的语言限制,推动了机器人技术的实用化。意义:为智能体的操作和学习能力提升提供了基础。