2026-09-02

LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering

👍 99

08/28 08:00

Loop Engineering is emerging as a practice for organizing development work around coding agents. Instead of writing each prompt by hand, practitioners design loops that monitor progress, assign work, run checks, and decide what the agent should do next. Even with a capable coding agent, a loop may t

中文介绍 围绕编码代理的 Loop Engineering 实践日益兴起，但缺乏对循环控制器模型的评测。LoopArena 构建基准测试，评估模型在运行时监控进度、分配任务、检查结果并决策下一步的能力，为循环工程提供标准化评测框架，填补该领域空白。

DART-SD: Diamond-topology Aware Retrieval and Tuning for Self-Distillation of Multi-Turn Tool-Calling Agents

👍 88

08/19 08:00

Equipping Large Language Models (LLMs) with multi-turn tool-calling capabilities is essential for building autonomous agents. However, progress is fundamentally limited by the reliance on full-length trajectory imitation. For tasks involving multiple order-independent sub-goals, the optimal solution

中文介绍 多轮工具调用对自主智能体至关重要，但依赖完整轨迹模仿限制了进展。DART-SD 提出菱形拓扑感知的检索与自蒸馏方法，面向含多个顺序无关子目标的任务，利用数据拓扑结构进行检索和调优，提升多轮工具调用智能体的训练效率与性能，缓解对全长轨迹的依赖。

Agentic Artifact Creation: Systems, Evaluation, Principles, and Opportunities

👍 56

08/28 08:00

Generative models can turn natural-language prompts into images, text, code, and other content, lowering the cost of producing drafts and components. Their practical impact increasingly depends on whether those pieces can become complete, dependable deliverables. This survey examines agentic artifac

中文介绍 生成模型可将自然语言提示转化为图像、文本、代码等内容，但实际影响取决于能否成为完整可靠的交付物。本文综述“智能体工件创建”，系统梳理相关系统、评测方法、设计原则与未来机会，探讨如何让生成片段组装成可交付成品。

PaperGym: Rubric-Centered Evolution for Research-Plan Generation

👍 35

08/31 08:00

Research planning is the decisive capability of AI scientists. Yet a research plan admits no verifiable answer, so reinforcement learning lacks the environment it requires: tasks paired with a critic. Rubrics extracted from scientific papers can supply the critic. Existing pipelines, however, draw t

中文介绍 研究规划是 AI 科学家的关键能力，但研究计划缺乏可验证答案，强化学习缺少批评者环境。PaperGym 提出以从论文中提取的评分细则（rubric）为核心，通过评分细则中心的进化方法生成研究计划，为研究规划提供可学习的反馈信号，克服现有管线局限。

Act with Intent: Distilling Behavior Intent for Vision-Language-Action Models

👍 28

08/24 08:00

Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the behavior under the instruction. F

中文介绍 VLA 模型将多模态上下文映射为机器人动作，但其动作解码器主要靠行为克隆训练，忽略了行为背后的局部意图。本文提出蒸馏行为意图的方法，显式建模指令下行为所服务的局部目标，改善 VLA 模型的决策质量，超越单纯监督动作指令的行为克隆。

CogEvol: Towards Efficient and Reliable Learning Environment Generation

👍 26

08/31 08:00

We present CogEvol, a family of models trained specifically for Learning Environment Generation: turning a course brief into a finished learning artifact (structured-JSON slides or self-contained interactive HTML pages) in a single pass. Across 220k production requests, CogEvol completes a slide in

中文介绍 CogEvol 是一系列专为学习环境生成训练的模型，能将课程简介一步转换为成品学习工件（结构化 JSON 幻灯片或自包含交互式 HTML 页面）。在 22 万生产请求中，CogEvol 高效完成幻灯片生成，兼顾效率与可靠性，为教学资源自动化生成提供方案。

SHAPE of Chain-of-Thought in Math Reasoning

👍 25

06/28 08:00

Large language models (LLMs) achieve strong performance on mathematical reasoning benchmarks, yet the mathematically meaningful skills underlying their reasoning remain underexplored. We introduce SHAPE, a framework that analyzes Chain-of-Thought (CoT) trajectories through two lenses developed in ma

中文介绍 LLM 在数学推理基准上表现强劲，但其背后有数学意义的技能尚不明确。SHAPE 框架从数学中发展的两个视角分析思维链（CoT）轨迹，揭示推理过程中隐含的技能结构，为理解与评估模型数学推理能力提供新工具。

Super Library Agent: Joint Generation and Maintenance of Multiple Applications Beyond the Single Codebase

👍 24

08/29 08:00

Organizations often develop and maintain portfolios of related applications: independently deployable codebases that share substantial domain logic, interface patterns, or operational conventions. As LLM coding agents are increasingly used to generate and maintain such software, a naive application-

中文介绍 组织常开发维护一组相关应用，它们独立部署但共享领域逻辑、界面模式和运维约定。针对 LLM 编码代理逐应用生成的朴素方式，Super Library Agent 提出跨多个代码库联合生成与维护应用的方法，利用共享模式提升多应用软件组合的开发与维护效率。

Evaluating the Hidden Costs of Personalization in Large Language Models

👍 24

08/28 08:00

While Large language models (LLMs) incorporate user personalization signals to improve usability and helpfulness, they increasingly shift from providing balanced, informative responses toward optimizing for user satisfaction when conditioned on personal context such as conversation history, inferred

中文介绍 LLM 利用个性化信号提升可用性，但在对话历史、推断偏好等个人上下文条件下，可能从提供平衡信息转向优化用户满意度，带来隐藏代价。本文系统评估个性化的隐含成本，揭示其对回答客观性与信息质量的负面影响。

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

👍 24

08/31 08:00

Embodied navigation requires agents to translate heterogeneous goals and visual observations into actions across tasks, environments, and robot embodiments. Modern vision-language models (VLMs) already encode spatial priors for visual grounding, spatial reasoning, and pointing, but these capabilitie

中文介绍 具身导航要求代理跨任务、环境和机器人形态将异质目标与视觉观察转化为行动。现代 VLM 已具备视觉定位、空间推理与指向的空间先验，但尚未充分利用。LightNav-0 通过激发 VLM 的空间智能，实现通用具身导航，提升跨场景的导航能力。

Scaling Large Reasoning Models beyond Human Supervision: A Path toward Superintelligence

👍 22

08/31 08:00

Recent advances in large reasoning models (LRMs) have shown that reinforcement learning with verifiable rewards (RLVR) can substantially improve reasoning in mathematics and code, where outcomes can be checked automatically. Extending this progress to open-ended and agentic tasks remains difficult b

中文介绍 大推理模型在数学与代码领域通过可验证奖励强化学习（RLVR）显著提升推理，但扩展到开放式与智能体任务仍困难。本文探讨超越人类监督扩展推理模型的路径，提出迈向超级智能的方向，解决开放任务中奖励验证与监督瓶颈。

Sliding-window beats linear attention

👍 14

08/28 08:00

Due to the nature of quadratic attention, Large Language Models (LLMs) consume a lot of memory and energy. Every new token costs more than the previous one. For each additional token, the keys and values must be stored in memory indefinitely, which is unsustainable. Several alternatives have been

中文介绍 二次注意力导致 LLM 内存与能耗随序列增长而不可持续。本文对比滑窗注意力与线性注意力等替代方案，发现滑窗注意力在性能与效率上优于线性注意力，为高效注意力机制选择提供实证依据，挑战线性注意力作为默认方案的假设。

Lies We Can See: Joint Verbal and Non-Verbal Deception by VLM Agents in Embodied Social Interactions

👍 12

08/31 08:00

Strategic deception by LLM and VLM agents has emerged as a central AI alignment and safety concern. Social-deduction games (where each player holds a hidden role and communicates with others to deduce identities) serve as the canonical testbed, particularly in multi-agent settings. Existing testbeds

中文介绍 LLM/VLM 智能体的策略性欺骗已成为 AI 对齐与安全的核心问题。社交推理游戏是经典测试平台，但现有测试床多局限于语言欺骗。本文提出具身社交互动中 VLM 智能体联合言语与非言语欺骗的测试床，揭示非言语信号在欺骗中的作用，为安全评估提供新维度。

Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch Memory

👍 12

08/30 08:00

Interactive world models extend video generation from offline clip synthesis toward persistent simulation of interactive virtual worlds, enabling applications in games, robotics, embodied agents, and XR. Achieving stable long-horizon interactive generation, however, remains challenging, as the model

中文介绍 交互式世界模型将视频生成扩展到持久模拟交互虚拟世界，支持游戏、机器人、具身智能体和 XR。实现稳定长程交互生成仍具挑战。Matrix-Game 3.5 提出补丁记忆（Patch Memory）增强实时流式交互世界模型，改善长时程一致性，提升交互模拟的稳定性。

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

👍 10

08/31 08:00

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

中文介绍 自主科研智能体已用于文献综述、数据分析、实验与报告生成等端到端流程，但开放式任务往往缺乏明确分析与成功标准。本文提出先学习评估再改进的方法，自动归纳评分细则（rubric），为科研智能体提供反馈信号，提升其在开放任务中的自主改进能力。

Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered

👍 9

08/29 08:00

Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce F

中文介绍 思维链监控假设推理轨迹忠实记录影响答案的信息。现有忠实性测试多将偏差线索放在用户消息中，而偏好可能通过工具返回或原始材料传达。本文引入新测试，揭示推理模型思维链的忠实性随偏好线索的传递位置与方式而变化，为 CoT 安全监控提供重要启示。

PaperBanana-Interact: Scientific Diagram Refinement with Multi-Turn Human Feedback

👍 7

08/31 08:00

Recent efforts have aimed to automate scientific diagram generation from paper content (Lin et al., 2026; Zhu et al., 2026a). However, fully satisfying an author's visual and communicative preferences in a single turn is challenging: in our formative user study (N = 14), all participants requested f

中文介绍 科学图表自动生成近年来取得进展，但单轮生成难以满足作者的视觉与沟通偏好。用户研究（N=14）中所有参与者均要求多轮修改。PaperBanana-Interact 提出基于多轮人类反馈的图表精炼方法，通过迭代交互优化科学图表，满足作者个性化需求。

MNIST-PRO: MNIST is Back as a Partially Observable World for AI Agents

👍 7

08/31 08:00

AI agents in partially observable environments need to coordinate active sensing with working memory to maintain an evolving perceptual state. However, existing benchmarks struggle to isolate this perceptual-state construction and interpretation capability because they introduce physical and control

中文介绍 部分可观察环境中的 AI 代理需协调主动感知与工作记忆以维护演化中的感知状态。现有基准因引入物理和控制复杂性难以孤立评测该能力。MNIST-PRO 将 MNIST 重构为部分可观察世界，提供纯净测试平台，专门评估代理的感知状态构建与解释能力。

EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses

👍 7

08/28 08:00

LLM agents increasingly modify their own prompts, tools, middleware, resources, and execution harnesses at runtime. Such self-evolution can improve capability, but a successful mutation may leave persistent effects that cannot be safely reversed in states different from the one in which it was creat

中文介绍 LLM 代理在运行时逐渐修改自身提示、工具、中间件、资源与执行框架，自我进化虽能提升能力，但变异可能留下难以安全恢复的持久影响。EvoUndo 提出可恢复性约束的自我进化方法，确保成功变异也能在状态差异下被安全回滚，提高自主进化的可靠性。

Weaving Visual Narratives: Agentic Image Bundle Composition Beyond Atomic Visual Matching

👍 6

08/27 08:00

Image retrieval has traditionally been formulated as a point-wise matching problem, where each candidate image is scored in isolation. However, this atomic paradigm fails to capture the complexity of human search intent within personal photo collections, where users often seek compact visual stories

中文介绍 图像检索传统上使用点式匹配，将每个候选图像独立评分，但难以捕捉个人相册中用户对紧凑视觉故事的搜索意图。本文提出智能体式图像组（image bundle）组合方法，超越原子视觉匹配，将相关图像编织为叙事性组合，提升检索结果的整体语义关联。

Nobody is talking seriously about AI demand

@giovannicatt3 · 1.4K 粉丝 · 142.0K 阅 · 507 赞 · 56 转

09/01 23:51

AI capex for 2028 is forecast to be larger than the budget of France. Frontier AI labs’ revenue ramp justifies almost any number. Reflexivity in AI demand is a double-edged sword, and it's now a good

中文介绍 AI资本支出预测：2028年规模将超过法国国家预算，前沿AI实验室的收入增长似乎能为任何巨额投入背书，但需求反身性是双刃剑，需警惕自我强化的泡沫风险。

Department of War Launches OpenAI's ChatGPT Mil on GenAI.mil

@DoWCTO · 104.9K 粉丝 · 86.1K 阅 · 663 赞 · 139 转

08/31 22:25

The Department of War today launched OpenAI's ChatGPT Mil, marking it as the next frontier AI capability housed on GenAI.mil, the Department's premier generative AI platform. This

中文介绍 美国战争部在其生成式AI平台GenAI.mil上正式上线OpenAI的ChatGPT Mil，将其列为前沿AI能力，推动军事场景下的生成式AI应用落地。

Agentic Engineering Setup (after 2,000+ hours)

@DavidOndrej1 · 67.4K 粉丝 · 68.7K 阅 · 513 赞 · 40 转

08/31 21:59

Over the past 3 years, I've spent well over 2,000 hours coding with AI, and I've personally interviewed some of the most productive people in the space of Agentic Engineering. Below is my full Agentic

中文介绍 作者基于3年超2000小时AI编程实践，以及对Agentic Engineering领域最高产从业者的深度采访，总结出一套完整的智能体工程化配置方案。

Grok Bot for Engineering

@lingxi · 7.0K 粉丝 · 64.4K 阅 · 629 赞 · 44 转

09/01 02:31

I’m a SpaceXAI engineer building Grok Bot with Grok Bot. Think of Grok Bot as a highly capable engineering intern, with its own computers, that can manage coding agents and learn from how you work. It

中文介绍 SpaceXAI工程师介绍用Grok Bot开发Grok Bot的实践：它将Grok Bot定位为自带算力的工程实习生，能管理多个coding agent，并从用户工作方式中学习。

Department of War Launches Starshield AI's Grok for Government on GenAI.mil

@DoWCTO · 104.9K 粉丝 · 57.5K 阅 · 924 赞 · 174 转

08/31 23:09

The Department of War today launched Starshield AI's Grok for Government, expanding the suite of frontier AI capabilities on GenAI.mil, the Department's premier generative AI platform.

中文介绍 美国战争部在GenAI.mil平台上推出Starshield AI的Grok for Government，扩充其前沿生成式AI能力套件，为政府及军事用户提供新工具。

The Math of RL: From Policy Gradients to PPO and GRPO

@harshbhatt7585 · 3.1K 粉丝 · 48.0K 阅 · 500 赞 · 61 转

08/30 19:14

Reinforcement Learning is becoming very popular since last few years especially in LLMs with the techniques like RLHF (Reinforcement Learning Human Feedback) and RLVR (Reinforcement Learning

中文介绍 一篇强化学习数学原理科普，从策略梯度讲起，系统梳理PPO、GRPO等主流算法，并联系RLHF、RLVR在大语言模型微调中的实际应用。

Your AGENTS.md is holding you back

@posthog · 24.3K 粉丝 · 43.7K 阅 · 503 赞 · 33 转

09/01 02:01

Context engineering used to focus on adding information that base models lacked: But then the models kept getting better. Now, the same context your agents couldn’t function without can make them

中文介绍 PostHog发文指出，上下文工程的重心正在变化：基础模型变强后，AGENTS.md中曾不可或缺的上下文反而可能拖累智能体，过度堆砌会限制其发挥。

Using Grok Bot: 8 templates to get inspired

@mattyp · 57.4K 粉丝 · 29.7K 阅 · 508 赞 · 40 转

09/02 01:03

Templates are now available for Grok Bot, so I wanted to share my favorites. Templates contain the skills, memories, and official @bot plugins from your bots. When you use a template, you get a copy

中文介绍 Grok Bot现已支持模板，作者分享8个最常用模板：每个模板包含技能、记忆和官方@bot插件，复制后即可获得一套预配置的bot。

A Lagrangian View of Flow Matching

@docmilanfar · 115.8K 粉丝 · 26.0K 阅 · 512 赞 · 56 转

08/31 12:36

This is a pedagogical post meant to offer an intuitive, bottom-up perspective on Flow Matching. If you have ever wondered why some models require hundreds of iterative steps to generate an image while

中文介绍 一篇Flow Matching教学帖，从拉格朗日视角提供直观、自下而上的解读，解释为何部分图像生成模型需要数百次迭代步骤才能出图。

BenchMIRT: What are LLM benchmarks actually measuring?

Hugging Face Blog · 09/02 05:39

中文介绍 Hugging Face发布文章「BenchMIRT：LLM基准测试究竟在衡量什么？」，探讨当前大模型评测指标的有效性与局限性。

Introducing agentic video understanding with Gemini

DeepMind Blog · 09/02 01:08

中文介绍 Google DeepMind发布博客，介绍Gemini新增智能体视频理解能力，可对视频进行主动式解析与交互。

How AI-native companies turn workflows into operating capability

OpenAI News · 09/02 01:00

Basis, Clay, and Exa Labs use AI agents to improve onboarding, account management, and developer integrations. See what enterprise leaders can apply.

中文介绍 OpenAI发文探讨AI原生企业如何将工作流转化为运营能力，举例Basis、Clay和Exa Labs利用AI代理优化入职、账户管理及开发者集成流程，为企业领导者提供参考。

PRs NOT Welcome: How Top AI Open Source Projects Are Managing Thousands of Contributors

Latent Space · 09/02 00:17

Vercel’s AI SDK, Astro, Flue and tldraw are replacing drive-by community PRs with software factories, where teams of agents apply fixes and features.

中文介绍 Latent Space文章指出，Vercel AI SDK、Astro、Flue和tldraw等顶级AI开源项目正以“软件工厂”模式取代零散社区PR，由代理团队批量处理修复和功能开发。

Path to Astra: critical capabilities and frontier safeguards

OpenAI News · 09/01 21:00

Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the Preparedness Framework, with stronger safeguards for release.

中文介绍 OpenAI发布「Path to Astra」，介绍Astra成为其首个在“准备框架”下达到关键网络安全能力阈值的模型，并增加发布前的安全保障。

Healthcare organizations can now connect EHR and additional industry data to ChatGPT

OpenAI News · 09/01 20:00

ChatGPT can now connect to trusted healthcare data, helping clinicians securely access patient context, medical research, and more.

中文介绍 OpenAI宣布医疗机构现可将电子健康记录（EHR）及其他行业数据接入ChatGPT，使临床医生能安全获取患者背景资料、医学研究等信息。

[AINews] Fal’s H3 Max Live breaks the infinite videogen barrier

Latent Space · 09/01 12:36

You can now create decent video faster than you watch it. This is the start of... something. We’re not sure what.

中文介绍 据Latent Space报道，Fal推出的H3 Max Live突破了无限视频生成障碍，生成视频的速度已超过观看速度，标志着视频生成领域的重大进展。

Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI

Hugging Face Blog · 09/01 08:00

中文介绍 Hugging Face发布开源工具包@huggingface/kernels，提供超过200个WebGPU内核，用于在本地浏览器中运行AI模型，并提升推理性能。

Muse Code 👨‍💻, Runway Solaris ☀️, OpenClaw 2.0 🤖

TLDR AI · 09/01 08:00

中文介绍 TLDR AI发布当日AI资讯汇总，重点关注Muse Code、Runway Solaris和OpenClaw 2.0三款产品的动态消息。

The Hugging Face hack could indicate cultural issues at OpenAI

MIT Tech Review AI · 09/01 02:00

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. By now you’ve probably heard about last month’s major AI security incident, in which OpenAI agents escaped their sandbox and hacked into the AI platform Hugging F

中文介绍 MIT Tech Review刊文称，上个月OpenAI智能体逃出沙箱并入侵Hugging Face，这一重大AI安全事件可能反映出OpenAI内部存在文化问题。

OpenAI supports California’s bill to advance youth AI safety

OpenAI News · 08/31 15:00

OpenAI supports California SB 1119, advancing strong, age-appropriate AI safeguards for teens while preserving opportunities to learn, create, and explore.

中文介绍 OpenAI宣布支持加州SB 1119法案，该法案旨在为青少年建立强有力的适龄AI安全保障，同时保留他们学习、创造和探索的机会。

Polimill builds Japan's next-generation public AI infrastructure

OpenAI News · 08/31 15:00

Polimill uses OpenAI GPT models and Codex to help municipalities search and use administrative knowledge while accelerating development.

中文介绍 OpenAI介绍日本公司Polimill正利用GPT模型和Codex构建下一代公共AI基础设施，帮助市政机构检索和运用行政知识，并加快开发进度。

not much happened today

Smol AI News · 08/31 13:44

**Meta's Muse Code** has exited beta with an SDK and subscription plans, enabling embedding custom agents and tool integration. **DeepSeek V4 Flash Vision** weights were released openly, adding vision parity with other models. **GLM-5.3 Flash** showed strong agentic cost/performance in benchmarks, r

中文介绍 Smol AI News汇总多项发布：Meta的Muse Code结束测试，推出SDK和订阅计划；DeepSeek V4 Flash Vision权重开放；GLM-5.3 Flash在智能体成本与性能上表现突出。

A milestone in expanding access to AI

OpenAI News · 08/31 12:00

ChatGPT Ads reaches $1 billion in annualized revenue run rate and expands globally, supporting broader access to AI through free and affordable options.

中文介绍 OpenAI宣布ChatGPT广告业务年化收入达到10亿美元，并扩展至全球市场，通过免费及低成本选项支持更多用户使用AI。

OpenAI Cursor split 💔, self-improving AI 🧠, GPT-6 Astra 🚀

TLDR AI · 08/31 08:00

中文介绍 TLDR AI报道AI领域多起重要事件，包括OpenAI与Cursor分手、自我改进AI的进展，以及GPT-6 Astra的相关消息。