An alignment assessment of recent cybersecurity incidents

Anthropic 最新研究：An alignment assessment of recent cybersecurity incidents…

📅 09-09📎 Anthropic Research

Emergency Department Revisit Quality Review Screening: Exploring Human Decision-Making and Artificial Intelligence Support

研究探索急诊科复诊质量审查中的人类决策过程，并评估AI辅助筛查的潜力。通过分析临床医生的判断模式，揭示人机协作在识别高风险复诊病例中的价值，为优化急诊质量改进流程提供实证依据。

📅 09-09📎 arXiv AI Papers

MOONWALK: Mediating Operations with Intent-Evidence-Action Alignment Across Junior-Supervisor Review Workflows in Animation/VFX Pre-Production

提出MOONWALK系统，用于动画/VFX前期制作的初级—主管评审流程，通过“意图—证据—行动”对齐来中介协作。它将反馈目标、依据和修改动作结构化关联，提升审查透明度、可追溯性与效率，减少理解偏差和返工。

📅 09-09📎 arXiv AI Papers

Show-Harness: Just a VLM Agent Can Play Robots

Foundation vision-language models (VLMs) exhibit broad intelligence about the world, yet translating this intelligence into robot control remains challenging. We present Show-Harness, an Embodied Harness that enables VLMs to “play” robots through a compact semantic interface linking intent to action

📅 09-09📎 arXiv AI Papers

IBIB: A Protocol for Measuring Enterprise AI Systems by Serving Route, Not Model Identifier

Enterprises deploy systems, not checkpoints. Usable capability depends jointly on weights, serving route, precision, output contract, and harness, yet all 18 audited benchmarks score advertised model identifiers. We treat this as measurement error and give a protocol that makes it reportable. It has

📅 09-09📎 arXiv AI Papers

Semigroup-JEPA: Latent Dynamics Consistency for Zero-Shot Physics Generalization

Joint-Embedding Predictive Architecture (JEPA) world models learn a compact latent representation of the world that supports prediction and planning, but their capability to learn physics and generate physically realistic dynamics remains hitherto untested. In this work, we introduce SemiGroup-JEPA

📅 09-09📎 arXiv AI Papers

JarvisGUI: Towards Cross-Device GUI Agents with Dynamic Task Composition

提出JarvisGUI，面向跨设备GUI智能体的动态任务组合框架。它可将复杂用户目标自动拆解并组合为跨设备操作序列，提升异构终端上的自动化执行与泛化能力，为跨设备智能助理提供新思路。

📅 09-09📎 arXiv AI Papers

ConvMem: Convolutional Memory for Long-Context Reasoning

提出ConvMem卷积记忆机制，用于增强大模型长上下文推理。通过卷积结构压缩与检索历史信息，缓解长序列中的遗忘与计算开销，在长文档理解、多轮推理等任务上提升性能与效率。

📅 09-09📎 arXiv AI Papers

Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMs

提出面向鲁棒大模型的层选择性遗忘方法，仅对关键层进行参数更新，避免全模型微调带来的能力损伤。该方法能精准移除目标知识，同时保留通用性能，实现更高效、可控的LLM遗忘。

📅 09-09📎 arXiv AI Papers

Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs

提出基于本体的记忆生命周期管理框架，解决大语言模型长期对话中的连贯性问题。通过结构化记忆组织与动态更新机制，实现持久上下文保持，提升模型在多轮交互中的一致性与信息延续能力。

📅 09-09📎 arXiv AI Papers

Can Foundation Models Moderate Online Content? Evaluating Instruction- vs. Example-Driven Policy Operationalization

系统评估基础模型在在线内容审核中的表现，对比指令驱动与示例驱动两种策略操作化方式。研究发现示例驱动方法在政策执行一致性和边界案例处理上更具优势，为自动化内容治理提供实践指导。

📅 09-09📎 arXiv AI Papers

4 ways Gemini makes administrative chores quick and easy

Google Gemini 产品更新：4 ways Gemini makes administrative chores quick and easy…

📎 Gemini Updates

OpenAI Does Math, Reward-Hacking, Meta Launches Personal Agent

Ben Thompson 科技战略分析：OpenAI Does Math, Reward-Hacking, Meta Launches Personal Agent…

📎 Stratechery

How GPT-5.6 Sol helps run quantum computing experiments

OpenAI 产品更新：How GPT-5.6 Sol helps run quantum computing experiments…

📅 09-08📎 OpenAI Blog

The Work Now Within Reach

OpenAI 产品更新：The Work Now Within Reach…

📅 09-08📎 OpenAI Blog

TANGO: Humanoid Navigation in Cluttered Environments with a Whole-Body Vision-Language-Action Model

本文提出TANGO，一种全身视觉-语言-动作模型，用于人形机器人在杂乱环境中的导航。结合多模态感知与全身控制，能有效规划行走路径并避开障碍。通过端到端学习提升复杂场景适应性，展现了人形机器人在真实环境中稳健移动的潜力。

📅 09-08📎 arXiv AI Papers

Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

本文引入过程图（Procedural Graphs），一种自演化的执行结构，用于增强LLM智能体的复杂任务解决能力。通过动态构建和演进任务流程，支持灵活分支与迭代优化，减少错误累积。相比固定链式推理，显著提升多步骤任务的成功率与鲁棒性。

📅 09-08📎 arXiv AI Papers

NOAH: Learning the Full Patient Journey. A Longitudinal Multimodal Time-Aware Model for Representation and Forecasting

本文提出NOAH，一种纵向多模态时间感知模型，用于学习患者完整诊疗历程。该模型整合时间维度与多种医疗数据，可有效表示病情演变并预测未来健康事件。通过捕捉长时间依赖和异质性信息，为个性化医疗和临床决策提供有力支持。

📅 09-08📎 arXiv AI Papers

A Data-Driven Framework for Identifying and Prioritizing RPA Opportunities in Healthcare Processes

本文提出一种数据驱动的框架，用于识别医疗流程中的RPA（机器人流程自动化）机会并进行优先级排序。该框架利用流程挖掘和任务特征分析，系统评估流程的自动化潜力，帮助医疗机构优先部署高价值RPA应用，从而提升运营效率、降低人工成本并减少人为差错。

📅 09-08📎 arXiv AI Papers

Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails

本文探讨工具（harness）与模型的共同进化机制，提出在线策略校正方法，帮助较弱模型在模仿学习失效的任务上实现追赶。通过迭代式反馈与策略调整，弱模型能够突破模仿瓶颈，自主改进表现，为提升低资源条件下模型能力提供了新的训练范式。

📅 09-08📎 arXiv AI Papers

ExecCritic: Learn to Test, Test to Improve for Coding Agents

论文提出ExecCritic框架，使编码代理在生成代码后学会编写并执行测试，并依据测试反馈对代码进行自我修正与优化。该方法将运行时错误信号引入迭代过程，增强代理的自我评测与纠错能力，显著提升其在复杂编程任务上的正确率和稳定性。

📅 09-08📎 arXiv AI Papers

A Generalization of Amari’s Bayesian Duality

Amari’s contributions to information geometry and machine learning are well known. Here, we revisit Amari’s work on Bayesian duality which has not received as much attention. We connect Amari’s Bayesian duality to a convex duality of Bayes’ rule. Using this connection, we present a generalization of

📅 09-08📎 arXiv AI Papers

Canonical Color as a Lens into Concept Decodability in Vision Encoders and VLMs

Visual encoders construct a representation of the image input for Vision-Language models. How much conceptual, as opposed to immediately visible, information does this representation contain? We use canonical color as a controlled test case to ask whether vision encoders make canonical-color informa

📅 09-08📎 arXiv AI Papers

DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination

Dexterous manipulation involves contact-rich and fine-grained interactions with the physical world, posing significant challenges for existing vision-language-action (VLA) models due to severe visual occlusions and complex contact dynamics. While recent works have incorporated tactile sensing into r

📅 09-08📎 arXiv AI Papers

MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents

MeClear提出针对长时程LLM智能体的记忆管理方法。利用合作博弈论归因各记忆对任务的贡献，评估其重要性；结合风险感知策略，在清理低价值记忆时避免损失关键信息，从而保持智能体长期性能并控制上下文开销。

📅 09-08📎 arXiv AI Papers

my vibe coding adventure

作者以亲身实践探索“vibe coding”——一种通过自然语言描述意图、让AI自主生成代码的全新编程范式。文章记录了在制作个人项目过程中遭遇的惊喜与挫折，剖析了这类开发方式在效率、调试及控制权上的本质变化，并反思其对未来程序员角色与软件开发流程的深远影响。

📎 Stratechery

Formalizing Fermat’s Last Theorem

本文介绍了将费马大定理进行形式化证明的研究。Anthropic研究人员利用AI辅助，将这一经典数学难题转换为机器可验证的严格逻辑表述，展示了AI在数学推理中的潜力，为形式化数学和自动化证明发展提供了关键案例。

📅 09-04📎 Anthropic Research

验证器是 RLVR 的隐形天花板，还是下一个能力突破口？

中文AI媒体：验证器是 RLVR 的隐形天花板，还是下一个能力突破口？…

📎 机器之心

从 VLA 到 WBI，具身智能的大脑在如何改变？

中文AI媒体：从 VLA 到 WBI，具身智能的大脑在如何改变？…

📎 机器之心

An Interview with OpenAI President Greg Brockman About Astra and Alignment

Ben Thompson 科技战略分析：An Interview with OpenAI President Greg Brockman About Astra and Alignment…

📎 Stratechery

2026.36: Friction and Feedback

Ben Thompson 科技战略分析：2026.36: Friction and Feedback…

📎 Stratechery

Developing Enterprise Frontier Safeguards with our customers

Anthropic携手企业客户共同研发前沿AI安全防护机制，聚焦将可信措施嵌入大模型部署全流程。文章阐述了协同构建防护体系、平衡技术创新与风险管控的经验，助力企业在真实业务场景中安全落地前沿人工智能。

📅 09-01📎 Anthropic Blog

Efficient SWE Agent Benchmarking via Trajectory-Aware Evaluation

提出轨迹感知的评估方法，用于高效基准测试SWE智能体。通过分析代理的决策轨迹，减少冗余运行并精准定位失败原因，显著提升评估效率与诊断能力，为软件工程智能体性能评估提供新思路。

📅 09-01📎 arXiv AI Papers

Adaptive Critical Token-Aware Retrieval for Repository-Level Code Generation

提出自适应关键标记感知检索策略，用于仓库级代码生成。通过动态识别与生成目标最相关的代码标记，优化检索范围与信息权重，提升长代码上下文中的生成准确性与效率，缓解无关信息干扰问题。

📅 09-01📎 arXiv AI Papers

CordisBench: Can Language Models Reason About Component Lifecycles in Dynamic Agent Harnesses?

介绍CordisBench，评估语言模型在动态智能体框架中对组件生命周期推理的能力。构建含状态转换与交互约束的任务集，发现现有模型在此类动态场景中表现薄弱，为提升语言模型的系统级理解提供基准与改进方向。

📅 09-01📎 arXiv AI Papers

The Rise of Verbal Reinforcement Learning

Natural language is emerging as a primary feedback channel for improving language agents, capable of conveying intent, preferences, and causal structure in forms interpretable by both humans and modern language models. We call this paradigm Verbal Reinforcement Learning (VRL) and offer the first uni

📅 09-01📎 arXiv AI Papers

Mechanism Design for Alignment and Control

We develop a framework for mechanism design with AI agents whose alignment (preferences) and capabilities (feasible actions and information) are unknown. We want such agents to act on our behalf so mechanisms must incentivize both honesty and obedience. A one-sided imitation structure—capabilities

📅 09-01📎 arXiv AI Papers

Designing Proactive Thought Partners for Writing

Writing involves diverse cognitive activities, from ideation to revision, and writers’ needs vary across individuals and moments. Proactive AI promises to provide the right support at the right time, yet existing proactive tools largely focus on generic textual assistance, such as autocomplete. This

📅 09-01📎 arXiv AI Papers

Scaling Near-Optimal SFT-RL Annotation Budget Allocation from Small to Large LLMs

针对大语言模型微调中监督微调与强化学习的标注预算分配问题，提出一种从小模型到大模型的可扩展近似最优分配策略。通过分析数据难度与模型能力差异，动态调整标注侧重，显著提升训练效率与性能，为大规模对齐训练提供了低成本高收益的预算规划方案。

📅 09-01📎 arXiv AI Papers

Selective Agent Guidance via Entropy: Learning Autonomous Policies from Imperfect VLM Teachers

针对不完美视觉语言模型教师指导下的智能体学习难题，提出基于熵的选择性智能体引导方法。通过计算动作分布的熵值，自动筛选高置信度教师指令，同时结合在线经验学习自主策略，有效降低错误示范影响，提升策略鲁棒性，在复杂视觉控制任务中验证了方法的有效性。

📅 09-01📎 arXiv AI Papers

From Confusion to Clarity: Confusion-Aware Retrieval and Knowledge Injection for Text Classification

针对文本分类中的混淆样本问题，提出混淆感知的检索与知识注入框架。先利用模型不确定性识别易混淆类别，再有针对性地检索外部知识并注入分类决策过程，使得模型从困惑转向清晰。实验表明该方法显著提升细粒度分类准确率，尤其在小样本和类别重叠场景下优势明显。

📅 09-01📎 arXiv AI Papers

H3-World: Turning Language Understanding into World Control

本文提出H3-World框架，旨在将语言理解能力转化为对物理世界的实际控制。通过构建层级化世界模型，实现从自然语言指令到具体动作序列的映射，并整合感知、规划与执行模块。该方法显著提升了智能体在复杂环境中的任务完成率，为具身智能与语言驱动控制提供了新范式。

📅 09-01📎 arXiv AI Papers

Mapping global methane emissions from space with deep learning

谷歌AI团队利用深度学习模型分析卫星遥感数据，绘制全球甲烷排放分布地图，自动识别油气开采等人类活动产生的泄漏源。该技术大幅提升甲烷热点检测的效率和精度，为追踪温室气体、制定气候减排政策提供了新的工具。

📅 09-01📎 Google AI Blog

Nvidia Earnings, Dollars Per Gigawatt, Open and Hugging Face

Ben Thompson 科技战略分析：Nvidia Earnings, Dollars Per Gigawatt, Open and Hugging Face…

📎 Stratechery

2026.35: Internet Hype and Real World Change

Ben Thompson 科技战略分析：2026.35: Internet Hype and Real World Change…

📎 Stratechery

Teaching Claude why

本文探讨如何通过教导Claude理解事物背后的“为什么”来提升其推理能力。研究提出新的训练方法，使模型不仅学习表面模式，还能掌握深层因果逻辑，从而更可靠地处理复杂任务，减少错误推断，增强可解释性与泛化性能。

📅 05-08📎 Anthropic Research

Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AI

Import AI 471期探讨Hugging Face作为AI模型分发中心引发的行业集中化担忧，分析太空采矿中AI自主导航与资源勘测的应用前景，并梳理五眼联盟在AI安全治理方面的最新协调立场，为读者提供产业发展的多维观察视角。

📎 Import AI

SUN: Persistent Programs For Language-Grounded Control-to-Learning-to-Real Policies

提出SUN框架，实现语言引导的控制到学习再到真实环境的持久化程序策略，提升跨场景适应性与可迁移性，为机器人操作提供统一解决方案。

📅 08-31📎 arXiv AI Papers

Auditing Anonymous AI Models: A Four-Stage Protocol for Black-Box Identity Verification

设计四阶段黑盒协议，对匿名AI模型进行身份审计与验证，在不依赖内部信息条件下实现模型归属识别，增强AI治理与溯源能力。

📅 08-31📎 arXiv AI Papers

OntoAligner-Ensemble: Voting-Based Fusion across Heterogeneous Ontology Alignment Techniques

提出OntoAligner-Ensemble方法，通过投票融合异构本体对齐技术，有效提升对齐精度与鲁棒性，解决单一技术局限，增强知识图谱集成效果。

📅 08-31📎 arXiv AI Papers

When Does Bigger Help? A Controlled Study of LLM Scale for Ontology Learning

The effect of Large Language Model (LLM) scale on ontology learning (OL) performance remains insufficiently characterized. We present a controlled evaluation of 13 models spanning dense and Mixture-of-Experts variants from the Qwen3.5 and Qwen3.6 lineages, together with proprietary GPT release varia

📅 08-31📎 arXiv AI Papers

BLOOM-WILT: Logit Tilting for Behaviour Elicitation in Automated LLM Auditing

Users of a deployed language model routinely encounter behaviours that testing almost never surfaces, since deployment puts the model through orders of magnitude more interactions than any evaluation can simulate. Automated auditors make testing cheap to scale and flexible enough to cover almost any

📅 08-31📎 arXiv AI Papers

LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering

Industrial post-training is a brownfield regime. Teams inherit a deployed checkpoint and must land targeted improvements under fixed compute and mixture budgets without regressing the rest. The maintained artifact is increasingly dataware: behavior governed by a curated post-training mixture, update

📅 08-31📎 arXiv AI Papers

Cross-Regional Grapevine Cold Hardiness Prediction via Learned Multimodal Latent Representations

本研究提出基于多模态潜在表征的跨区域葡萄藤抗寒性预测方法。通过学习不同区域间共享的潜在表示空间，有效融合气象、地理、物候等多源异构信息，显著提升模型跨区域泛化能力与预测精度，为精准农业中作物抗寒性动态评估提供了新思路。

📅 08-31📎 arXiv AI Papers

Token-Efficient Data Reasoning Agents via Adaptive Structuring of Unstructured Data

提出一种通过自适应结构化非结构化数据来提升数据推理智能体令牌效率的方法。根据任务需求动态规划数据组织方式，在保留关键信息的同时大幅压缩上下文长度，降低推理开销，使智能体在有限令牌预算下高效完成复杂数据推理任务。

📅 08-31📎 arXiv AI Papers

Reconciling Process Supervision with Outcome-Based Credit in Agentic Policy Optimization

针对智能体策略优化中过程监督与基于结果的信用分配不一致的问题，提出统一协调框架。通过将细粒度过程反馈与最终结果奖励有机结合，实现两种监督信号的互补，缓解信用分配稀疏问题，提升智能体在复杂多步任务中的策略学习效率与稳定性。

📅 08-31📎 arXiv AI Papers

Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents

Autonomous scientific research agents are increasingly applied to end-to-end scientific workflows, including literature review, data analysis, experimentation, and report generation. However, open-ended research tasks often do not clearly specify the analyses, methods, and success criteria required

📅 08-31📎 arXiv AI Papers

TimesFM-3: A zero-shot foundation model for multivariate forecasting

Google AI 发布 TimesFM-3，一款面向多元时间序列预测的零样本基础模型。该模型无需针对特定数据集微调，即可直接预测多变量间的复杂动态关系，显著提升金融、能源、运维等场景下的预测效率与泛化能力，为时序分析提供通用解决方案。

📅 08-31📎 Google AI Blog

Money, Credit, Trust, and FTX

FTX崩塌揭示了货币、信用与信任之间的脆弱关系。文章剖析了FTX如何将客户资产混淆为自有资本，模糊了货币与信用的界限，并探讨了信任在金融系统中的核心作用。作者指出，真正的信任需要透明机制支撑而非单靠声誉，为理解加密交易所风险提供了深刻洞察。

📎 The Diff

嘉立创：用 2100 万张零散的 PCB 板，拼成一门价值千亿的生意

嘉立创通过整合2100万张零散PCB板订单，以低价标准化服务积累中小客户，构建从打样到量产的全产业链平台，形成规模效应与数据壁垒，最终成就千亿级生意，展现了长尾市场聚合的巨大价值。

📎 晚点LatePost