Argus: A General-Purpose Agentic Runtime for Long-Horizon Reasoning

Long-horizon reasoning requires an agentic runtime that can persist when evidence supports its current approach and pivot when measurements reveal failure, hidden constraints, or a misspecified objective. We present Argus, a persistent, self-evolving runtime in which Manager, Planner, Engineer, and

📅 08-05📎 arXiv AI Papers

OctoLong: Mid-Training On Cross-Repository Code Contexts Enhances Long-Context Modeling

Context lengths of language models (LMs) have dramatically increased, driven by the demands for in-context learning, self-improvement, and long-horizon agentic workflows. Existing long-context corpora, however, are dominated by books, academic articles, and code repositories, which are finite resour

📅 08-05📎 arXiv AI Papers

Teaching Nemotron Greek: Mining a Corpus, Adapting Retrieval, and Grounding Generation for Modern Greek across Specialist Domains

Modern Greek is absent from NVIDIA’s Nemotron retrieval models and from major multilingual retrieval benchmarks, despite being important for retrieval-augmented generation (RAG) in legal, energy, financial, and medical applications. We present an end-to-end adaptation of the Nemotron retrieval stack

📅 08-05📎 arXiv AI Papers

OPD-V: Visual On-Policy Self-Distillation with Modality Balance

提出OPD-V方法，通过视觉在策略自蒸馏与模态平衡机制提升强化学习性能。该方法在策略学习过程中利用自蒸馏约束视觉表征，并动态平衡视觉与状态模态的信息贡献，有效缓解视觉过拟合，提高样本效率与泛化能力，在多个连续控制基准上验证了其优越性。

📅 08-05📎 arXiv AI Papers

SSTQ:Privacy-Preserving Vector Quantization via Subsampled Stochastic TurboQuant

提出SSTQ框架，面向隐私保护场景下的向量量化任务。通过子采样随机TurboQuant机制，在压缩高维向量的同时注入受控随机性，防止数据重建攻击，兼顾量化精度与隐私预算。实验表明SSTQ在保持较低失真率的同时显著增强安全性，适用于联邦学习等隐私敏感场景。

📅 08-05📎 arXiv AI Papers

Chained Recursive Language Models for Multi-Iteration Reasoning

提出链式递归语言模型框架，通过将模型输出循环反馈至输入实现多轮迭代推理。不同于单次前向推理，该方法允许模型逐步修正思维链、累积推理深度，在复杂数学与逻辑任务上显著提升准确率，同时保持参数规模不变，为推理时扩展计算提供了新范式。

📅 08-05📎 arXiv AI Papers

Robust and Efficient Motion Reasoning for Privacy-Aware Classroom Incident Recognition

针对隐私感知课堂事件识别中的运动推理鲁棒性与效率问题，提出一种兼顾准确性与隐私保护的方法。该方法在受限信息条件下高效提取运动特征，提升异常事件检测能力，为课堂安全监控提供可靠且保护隐私的智能识别方案。

📅 08-05📎 arXiv AI Papers

Representational separation between unitary and channel quantum generative models via shared classical randomness at shallow depth

通过共享经典随机性，研究浅层量子生成模型中幺正与信道两种表示之间的分离特性。理论分析表明，在浅层深度下两种模型表达能力存在本质差异，为量子生成模型的架构选择与经典模拟界限提供了新的理论依据。

📅 08-05📎 arXiv AI Papers

CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs

CoPlan提出一种可信协同智能接口，利用基于角色的可争议论证图实现护理计划制定。该方法支持多方参与和透明推理，增强决策可解释性与可质疑性，从而提高护理规划中的信任度和协作效率。

📅 08-05📎 arXiv AI Papers

ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment

本文提出ABSeeker，一种通过答案回溯的信用分配方法，用于训练长时程搜索代理。该方法有效应对稀疏奖励下的信用分配难题，显著提升复杂推理与多步决策任务中的搜索效率和成功率，为训练高性能搜索智能体提供了新的技术路径。

📅 08-05📎 arXiv AI Papers

Google Earnings, The Frontier Case, Amazon Earnings

本文分析谷歌与亚马逊最新财报，揭示云业务分化态势与AI投资回报差异。前沿案例部分深入探讨大模型竞赛中算力成本与战略选择的博弈，对比两家科技巨头的商业模式演进，为观察AI时代平台经济提供清晰框架。

📎 Stratechery