👍 170
08/25 20:00
Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames. Yet reliable evaluation remains challenging: benchmarks should adopt inputs aligned with the visual priors of current video models, require valid evolving processes r
中文介绍 VGI-Bench 提出了一个新的评估框架,旨在探测视频生成模型的视觉智能。通过设计与当前模型视觉先验一致的输入,该方法解决了评估可靠性问题。实验表明,VGI-Bench 能有效衡量模型的生成能力和零-shot 视觉推理能力,为视频生成领域的研究提供了重要的基准和工具。意义:该研究促进了在视频生成和视觉理解领域的评估方法发展。
👍 134
08/24 20:00
Massively parallel simulation changes the data regime in which off-policy reinforcement learning (RL) is trained, challenging stabilizers designed for data-limited replay. Through controlled experiments across eight benchmark families, we show that these stabilizers are data-regime-dependent: parame
中文介绍 WarpSAC 提出了一种新的可扩展的非策略强化学习框架,重点在于重新思考探索与利用之间的权衡。通过在八个基准系列上的实验,结果显示,该方法在数据丰富的环境下,优化了训练过程。WarpSAC 的关键贡献在于为非策略 RL 提供了一种更稳定的训练方法,有助于在数据丰富的仿真环境中实现更高效的学习。意义:该研究推动了可扩展 RL 方法在复杂任务中的应用。
👍 132
08/25 20:00
A common strategy for scaling world models is to train on more crawled video with more compute. We argue that this strategy is inefficient: scaling world models also requires a recursive data engine that offers grounded reward signals. The success of code agents illustrates why this matters. As code
中文介绍 本论文提出了一种递归数据引擎的策略,以支持世界模型的扩展。该方法强调以代码代理为例,解决通过更多爬取视频和计算资源进行简单扩展的低效问题。实验结果表明,具有 grounded 奖励信号的数据引擎能显著提升模型训练的效率。意义:此研究推动了世界模型和数据生成的深度整合,为 AI 代理能力提升提供了新思路。
👍 107
08/25 20:00
Agent capability is not determined by the model alone. The agent harness, encompassing memory management, planning strategy, action protocol, and tool/skill orchestration, can dominate the contribution of the underlying foundation model. Yet harness design remains manual, task-specific, and fundamen
中文介绍 JIT-Agent 关注于通过即时演化的代理框架来提升人工智能系统的智能水平,强调设计代理的拉力和任务的灵活性。提出的强化学习方法体现了内存管理、规划策略等多方面的融合。实验结果显示,优化的代理设计能够显著增强模型的综合性能。意义:研究为代理智能的提升提供了新视角,并推动了 RLHF 和代理设计领域的发展。
👍 71
08/26 20:00
Multimodal large language models (MLLMs) can interpret a street view, but urban agency depends on whether such local evidence remains useful after the agent starts to move. In this paper, we investigate how far current MLLM agents can turn local urban perception into reliable action in a complicated
中文介绍 UrbanGround 研究了多模态大语言模型如何在城市环境中进行有效的空间代理。文章提出了一种方法来评估当前模型在复杂城市环境中从局部感知转化为可靠行动的能力,并展示了使用历史感知的有效策略。研究结果强调了当前语言模型在动态环境中的挑战与应用潜力。意义:该研究为智能城市和环境交互中的 AI 代理设计提供了重要参考。
👍 68
08/26 20:00
Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth wit
中文介绍 TTPO 引入了一种新的测试时策略优化框架,以克服现有后期训练方法对真实标签的依赖。通过置换真实标签为相关信号,TTPO 在测试时实现了有效的模型自我改进。实验结果表明,该方法能显著提高模型在无标签条件下的适应能力,为后期训练研究提供了新的视角。意义:该研究对 RL 和自我监督学习等领域产生深远的影响。
👍 66
08/26 20:00
On-policy distillation (OPD), which leverages a pre-trained, specialized teacher model to provide dense supervisory signals, has achieved significant success in Large Language Models (LLMs) and has recently been adapted to flow matching models. However, this paradigm suffers from two major issues: F
中文介绍 Self-OPD 针对没有教师模型的环境提出了一种在政策蒸馏中利用流动匹配模型的新策略。通过实验证明,该方法在大型语言模型任务中取得了一定程度的成功,且解决了传统 OPD 模型中的两个主要问题。此研究为无需教师的政策优化提供了有效方案,有助于提升模型自适应能力。意义:推动了自我监督学习和政策优化的研究进展。
👍 58
08/26 20:00
LLM agents increasingly rely on generated interaction data to learn how to interact with external environments. Agentic data generation must maintain consistency among environments, tasks, interactions, and success signals while producing experience that is useful rather than merely abundant. Existi
中文介绍 本研究聚焦于 LLM 代理的数据生成方式,探讨了如何确保生成的数据在不同环境、任务和成功信号间的一致性。提出的 ACE 方法使得生成的数据不仅丰富且实用。实验结果表明,一致性有助于提升 LLM 的效果和互动能力,推动智能代理和人机交互的研究。意义:该研究为智能代理的训练和应用方向提供了重要的理论依据。
👍 43
08/21 20:00
AI-powered digital avatar streamers must answer product questions, engage viewers, and execute marketing strategies in real time, demanding low latency, frequent strategy updates, and accurate yet effective responses. Evolvable Harnesses, whose Skills, Hooks, prompts, and tools can be updated indepe
中文介绍 本论文探讨了 AI 驱动的数字化虚拟主播如何以更高效的方式进行实时互动与观众参与。提出的 Evolvable Harnesses 框架支持高清视频流中的独立更新和策略调整。实验显示,设计的代理能在快速变化的环境中保持高性能。意义:为数字内容创作和实时互动领域的高效代理设计奠定了基础。
👍 39
08/24 20:00
Modern video games combine first-person perception, rapid visual changes, persistent world state, and heterogeneous native controls. Existing game agents map visual and task context directly to actions but lack explicit world dynamics modeling, whereas interactive game world models predict visual fu
中文介绍 GameWAM 提出了一个针对现代视频游戏的世界行动模型,专注于动态世界状态与用户控制的有效整合。通过相较于先前代理缺乏世界动力学建模的挑战,结果表明,GameWAM 能显著提升代理在复杂环境中的决策能力。意义:推动了游戏智能体的发展,为游戏设计和 AI 测试提供了新的思路。
👍 26
08/26 20:00
Long-horizon agent runs generate experience that can improve both the current run and future work. Most self-improvement methods process this experience only after execution ends, so they cannot redirect the active run or immediately apply and validate lessons learned from it. We argue that self-imp
中文介绍 StreamPI 提出了一种新的多模态时间建模框架,强调在视觉-语言-行动模型中的应用。实验结果表明,该方法能有效改善模型对历史观察的记忆能力,提升空间认知的准确性。StreamPI 在处理动态信息时的强大能力为机器人操作任务的成功提供了支持。意义:推动了多模态学习和机器人技术领域的研究进展。
👍 21
08/23 20:00
Recent work proposes next-chunk reasoning RL for leveraging no-CoT data---corpora such as worked solutions and textbook derivations that contain reasoning-rich content but lack explicit chain-of-thought annotations. The method trains a model to generate implicit reasoning traces and rewards them by
👍 19
08/23 20:00
Video generation is progressing beyond isolated clips toward long-form narratives and interactive worlds, requiring models to preserve identities, follow user controls, and remain stable over extended rollouts. We present JoyAI-Echo-1.5, a unified audio-visual generation system with two purpose-buil
👍 19
08/18 20:00
Multi-teacher on-policy distillation (M-OPD) has emerged as a promising paradigm for consolidating domain-specialized reinforcement learning (RL) experts into a single generalist student via dense, token-level reward supervision. Despite its practical success, the optimization dynamics governing mul
👍 18
08/25 20:00
Vision-Language-Action (VLA) models have demonstrated effectiveness in robot manipulation, yet state-of-the-art models such as pi0.5 operate under a single-frame paradigm, limiting their ability to retain past observations and develop precise spatial perception. In this paper, we propose StreamPI, a
👍 17
08/25 20:00
Zero-shot cross-task generalization, where a policy must execute manipulation tasks never seen during training, remains a central challenge in robot learning. In large language models, a novel task can be performed simply by specifying it in the context, without any parameter update. This form of in
中文介绍 Zero-WAM 描述了一种新的在上下文中建模世界-行动的框架,该方法特别关注机器人任务中的零-shot 跨任务泛化能力。实验结果显示,该模型能够在没有额外参数调整的情况下成功执行未见任务,突出其强大的适应能力。意义:此研究在机器人学习和迁移学习领域具有重要应用前景与指导价值。
👍 16
08/25 20:00
Multimodal Large Language Models (MLLMs) have shown strong performance in video understanding. However, their ability to follow instructions in this domain remains under-explored. Real-world video understanding requires models not only to interpret video content correctly, but also to satisfy divers
中文介绍 Video-IFBench 针对多模态大语言模型在视频理解场景下的表现进行了评估,探讨了其指令跟随能力。研究表明,模型在解读视频内容时,有必要满足不同的指令需求,从而提升其在实际应用中的有效性。研究结果为构建更智能的视频理解系统奠定了基础。意义:推动了视频理解和多模态学习的研究进展。
👍 15
08/26 20:00
Evolution Strategies (ES) have recently emerged as a memory-efficient post-training paradigm for LLM reasoning. However, the optimization behavior of ES remains understudied, making it hard to define its advantage scope compared to mainstream post-training paradigms (e.g., Group Relative Policy Opti
中文介绍 本研究探讨了演化策略在 LLM 理由推理中的应用,揭示了其较主流后期训练方法(如 GRPO)的更广泛的推理覆盖范围。通过分析优化动态,研究结果为理解 LLM 优化算法的有效性提供了新的视角,强化了对演化策略的认识。意义:对后期优化算法的研究和应用有深远影响。
👍 15
08/25 20:00
Vision-language models can produce fluent answers that are insufficiently grounded in the visual evidence: a single unsupported object, chart value, or intermediate inference can undermine an otherwise plausible response. We argue that this is a credit-assignment failure in multimodal post-training.
中文介绍 V-Rubrics 提出了通过基于评分标准的方法进行可视化可信性评估,解决了多模态语言模型在视觉证据支持不足问题。论文论证了这种信用分配失败的根源,并提出改进策略。研究为多模态后期训练和模型解释性提供了新的思路。意义:有助于提高多模态学习模型的实用性和可解释性。
👍 13
08/23 20:00
Modern software systems accumulate technical debt over decades of development, which makes migration expensive and largely manual. As coding agents become increasingly capable at bug fixing, can they autonomously perform such migrations? Existing benchmarks cannot answer this question because they e
中文介绍 本研究探讨了现代软件系统中,编码代理在长期、全仓库堆栈迁移中的能力。由于现有基准无法解决技术债务问题,该论文提出了一种新的评估方式,以衡量编码代理的自主迁移能力。研究结果将为提高编码代理在软件迁移任务中的表现提供指导。意义:该研究将在智能化软件开发与维护领域产生深远的影响。