JavaScript · ★ 28,951 · 🍴 1,818 · 📈 4,562 stars today
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
中文介绍 面向AI代理的绘图技能包，可生成架构图、工作流、时序图、数据流与生命周期图。输出为自包含HTML，支持动效和清晰导出，便于在文档中直接嵌入。适合需要快速绘制专业、可验证图表的开发者和Agent场景。
Python · ★ 37,158 · 🍴 3,527 · 📈 720 stars today
Turn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 175,000+ scientists worldwide. 163 ready-to-use validated skills plus 100+ scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigrav
中文介绍 将任意AI代理转化为“AI科学家”的技能库，专为科研场景打造。内置163个经过验证的即用技能，并接入100+科学数据库，覆盖生物学等领域。全球已有超过17.5万科学家使用，适合科研人员加速文献分析、实验设计与数据挖掘。
Python · ★ 35,156 · 🍴 3,942 · 📈 457 stars today
Official, Anthropic-managed directory of high quality Claude Code Plugins.
中文介绍 Anthropic官方维护的Claude Code插件目录，收录并管理社区提交的高质量插件。提供统一的发现、安装与更新机制，帮助开发者扩展Claude Code的功能，例如增加新命令、集成外部工具或定制代码生成流程。适合深度使用AI编程助手的开发者与团队。
JavaScript · ★ 11,663 · 🍴 2,349 · 📈 3,829 stars today
A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.
中文介绍 浏览器中的间谍卫星模拟器，但使用真实开源空间数据。在逼真的3D地球表面展示实时地理空间情报，可观察卫星轨迹、地面设施等。适合GIS爱好者、教育演示和空间数据可视化探索。
TypeScript · ★ 46,254 · 🍴 5,108 · 📈 202 stars today
GitNexus: The Zero-Server Code Intelligence Engine - GitNexus is a client-side knowledge graph creator that runs entirely in your browser. Drop in a git repository (Github, Gitlab, Azure, Local) or ZIP file, and get an interactive knowledge graph with a built in Graph RAG Agent. Perfect for code exp
中文介绍 完全在浏览器中运行的客户端知识图谱引擎，无需服务器。拖入Git仓库（支持GitHub、GitLab、Azure及本地）即可自动构建代码关系图谱，帮助开发者理解代码结构、依赖与变更影响。适合代码评审、架构分析和新人上手。
Go · ★ 2,702 · 🍴 80 · 📈 574 stars today
Help AI coding agents write modern Go
中文介绍 JetBrains发布的Go语言现代编码指南，专门帮助AI编码代理写出符合当前最佳实践的Go代码。涵盖项目结构、并发、错误处理等规范，减少生成代码中的反模式。适合使用AI辅助生成Go代码的开发者和团队。
Python · ★ 53,589 · 🍴 6,673 · 📈 1,144 stars today
World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video production studio.
中文介绍 首个开源智能体视频生产系统，内置12条制作流水线、100+工具和700+技能/生产知识文件。可将AI编码助手转变为完整视频制作工作流，覆盖脚本、分镜、剪辑到渲染等环节。适合内容创作者和自动化视频生产团队。
Python · ★ 75,755 · 🍴 9,230 · 📈 326 stars today
Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)
中文介绍 用户上传截图即可一键转换为干净的前端代码，支持HTML、Tailwind、React和Vue。利用AI视觉理解界面布局并生成对应实现，大幅提升从设计稿到代码的效率。适合前端开发者快速实现原型或还原设计。
TypeScript · ★ 6,051 · 🍴 481 · 📈 246 stars today
Cursor plugin specification and official plugins
中文介绍 Cursor官方的插件规范与精选插件集合。该规范定义了插件与编辑器交互的标准方式，包括生命周期、权限和UI扩展点。开发者可借此构建自定义插件，增强AI编程助手的功能。适合希望深度定制Cursor工作流的开发者。
JavaScript · ★ 24,603 · 🍴 2,426 · 📈 1,687 stars today
Prompt as Code | GPT-Image2 工业级提示词引擎与模板库，530+ 个案例逆向工程，20+ 套工业级模板，并提炼出Skills，持续更新中
中文介绍 GPT-Image2工业级提示词引擎与模板库，以“Prompt as Code”理念组织。包含530+案例逆向工程和20+套工业模板，并提炼为可复用的Skills，持续更新。适合需要稳定生成高质量AI图像的设计师和开发者。
Go · ★ 3,026 · 🍴 83 · 📈 965 stars today
like netcat, but over Tailscale's data plane, without Tailscale's control plane
中文介绍 tailcat是一个类似netcat的网络工具，但运行在Tailscale的数据面上，无需控制面即可直接通信。它能在Tailscale节点间建立原始TCP/UDP连接，适合内网调试、端口转发和临时隧道，并天然获得Tailscale的加密与身份认证。
Java · ★ 73,449 · 🍴 8,021 · 📈 191 stars today
Ghidra is a software reverse engineering (SRE) framework
中文介绍 美国国家安全局开源的软件逆向工程（SRE）框架，提供反汇编、反编译、脚本分析和图形化调试功能。支持多种架构和平台，是安全研究人员分析恶意软件、漏洞挖掘和协议逆向的核心工具。
PHP · ★ 876 · 🍴 43 · 📈 188 stars today
Compile PHP to Native Binaries
中文介绍 将PHP代码编译为原生二进制文件的工具，支持强类型和编译优化。通过生成机器码提升性能，减少运行时开销。适合对性能敏感的PHP应用，如CLI工具、微服务和高并发后台任务。
Python · ★ 2,943 · 🍴 241 · 📈 236 stars today
Open-source framework for the research and development of foundation models.
中文介绍 面向基础模型研究和开发的开源框架，覆盖数据准备、训练、评估与部署全流程。支持大规模分布式计算和实验管理，适合研究团队和工业界从头训练或微调大模型。
TypeScript · ★ 21,824 · 🍴 3,080 · 📈 433 stars today
7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.
TypeScript · ★ 50,040 · 🍴 3,508 · 📈 67 stars today
Chrome DevTools for coding agents
Python · ★ 50,758 · 🍴 8,796 · 📈 703 stars today
Learn it. Build it. Ship it for others.
JavaScript · ★ 115,815 · 🍴 6,327 · 📈 1,396 stars today
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
C++ · ★ 39,220 · 🍴 10,874 · 📈 156 stars today
GoogleTest - Google Testing and Mocking Framework
Python · ★ 13,432 · 🍴 3,638 · 📈 22 stars today
A framework for building realtime voice AI agents 🤖🎙️📹
👍 170
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 提出VGI-Bench基准，用于评估视频生成模型通过生成帧展现的零样本视觉推理能力。针对现有评估输入与模型视觉先验不匹配、难以验证有效演化过程的问题，设计对齐先验的评测任务，更可靠地衡量生成模型的视觉智能。
👍 134
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重新思考大规模并行模拟下的off-policy强化学习。通过8个基准家族的控制实验，发现现有稳定器具有数据体制依赖性，在数据丰富的并行训练下失效；据此提出适配新数据体制的算法，实现可扩展的高性能强化学习。
👍 131
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 提出以智能体游戏开发作为可验证轨迹数据引擎，用于扩展世界模型。仅增加视频数据和算力并不高效，还需要提供grounded奖励信号的递归数据引擎；借鉴代码智能体的成功经验，以可验证轨迹驱动世界模型的持续迭代。
👍 107
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 智能体能力不仅由底层模型决定，harness（记忆管理、规划策略、动作协议、工具编排）往往贡献更大。现有harness设计依赖人工且任务特定，JIT-Agent提出即时harness演化机制，让harness随任务自动扩展，从而规模化提升智能体能力。
👍 71
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研究多模态大模型智能体在复杂城市场景中如何将局部感知转化为可靠行动。指出仅理解街景不足以支撑空间代理，需要考察智能体移动后局部证据是否持续有用，并据此评测与提升城市级空间决策能力。
👍 67
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提出测试时策略优化，解决强化学习和on-policy自蒸馏依赖真实标签、无法进行测试时训练的局限。在测试阶段以替代信号代替ground truth进行策略优化，使大模型在数学推理中无需标签即可持续自我提升。
👍 66
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提出无需教师的on-policy蒸馏方法，用于流匹配模型。现有OPD依赖预训练专家教师提供密集监督，在流匹配模型中存在两个主要缺陷；Self-OPD通过自蒸馏方式消除教师依赖，降低训练开销并提升模型性能。
👍 56
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 提出ACE视角审视LLM智能体数据生成质量。好的智能体数据需始终保持环境、任务、交互与成功信号之间的一致性，并产生对学习有用的经验而非仅追求数量；该视角为生成可靠智能体数据提供了系统性设计原则。
👍 43
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 面向AI数字人直播场景，提出TaoLive数字人智能体。核心是可演化Harness，其Skills、Hooks、提示词和工具可独立更新，从而满足实时回答产品问题、互动和营销执行对低延迟及策略频繁更新的要求。
👍 38
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提出面向现代视频游戏的世界动作模型。现有游戏智能体直接映射视觉到动作、缺乏显式世界动态建模，而交互式世界模型只预测视觉；GameWAM统一建模世界动态与动作生成，以应对第一人称感知、快速视觉变化和异构控制。
👍 25
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 PILOT-in-the-Loop实现长程智能体的在线自我改进。现有方法在处理完一轮执行后才利用经验，无法改变当前运行或即时验证所学；该机制在运行过程中持续学习、应用并验证经验教训，从而实时优化长期任务表现。
👍 21
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
中文介绍 重新考察无CoT标注数据下‘下一块推理RL’是否优于SFT。该方法让模型生成隐式推理轨迹并给予奖励；通过系统比较训练策略，对其在无显式思维链语料上的效果给出分析与结论。
👍 19
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
中文介绍 JoyAI-Echo-1.5提出统一音视频生成系统，面向长时叙事与交互世界。针对视频生成需保持身份、遵循用户控制并在长rollout中保持稳定的挑战，系统采用两个专用组件协同，实现持久一致的长期音视频生成。
👍 19
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
中文介绍 Open-MOPD诊断并修复多教师在线策略蒸馏中的能力不均衡。该范式用密集token级奖励将多个领域专家整合为通用学生，但存在优化动态失衡；Open-MOPD分析失衡机制并给出修正，增强通用学生的综合能力。
👍 18
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
中文介绍 StreamPI为视觉-语言-动作模型引入流式多模态时序建模。针对pi0.5等单帧范式难以保留历史观测、空间感知不精确的问题，以流式方式持续处理多模态时序信息，提升机器人操作的感知连续性与操控精度。
👍 17
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从人类视频学习上下文世界-动作建模，实现机器人操作的零样本跨任务泛化。受LLM只需在上下文中指定新任务、无需更新参数的启发，该模型使策略能泛化到训练中从未见过的操作任务。
👍 16
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提出视频理解场景下的指令跟随评测基准。真实应用要求MLLM既正确解读视频内容，又满足多样化指令约束；该基准系统评测这一能力，弥补了现有视频理解评测对指令跟随关注的不足。
👍 15
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 研究进化策略在LLM数学推理后训练中的优化行为，并与GRPO对比。ES内存高效，但其优势范围此前不清晰；分析表明ES可实现比GRPO更广的推理覆盖，为选择后训练方法提供了依据。
👍 15
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通过基于rubric的强化学习提升视觉语言模型的视觉忠实度。针对模型生成流畅但缺乏视觉证据的回答（如不存在的物体、错误图表值或中间推断），引入细粒度rubric奖励，改善多模态后训练的信用分配。
👍 13
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 SWE Refactor Bench评测编码智能体能否自主完成长期、全仓库级堆栈迁移。现有基准因任务碎片化而无法回答该问题；新基准模拟真实技术债场景，检验智能体的长时规划、整体理解与大规模重构能力。
@coinbase · 7.2M 粉丝 · 347.3K 阅 · 503 赞 · 73 转
AI agents are becoming active participants in the economy. They research, make decisions, and take action on behalf of people and businesses. But to participate fully, they eventually need to do the
中文介绍 Coinbase 发布 AiFi，定位为 AI 代理的金融系统。AI agent 正成为经济中活跃的参与者，替人研究、决策和行动；但要完整参与经济，最终需要支付、结算等能力。AiFi 即面向 AI 自主经济提供加密支付与金融基础设施。
@fi56622380 · 71.3K 粉丝 · 158.0K 阅 · 598 赞 · 84 转
The next ARR growth engine after coding, the AI infra bubble, the Economics of open vs. closed source model Why is the AI semi buildout longer than the internet buildout? After coding, where does the
中文介绍 聚焦 2026 年 AI 半导体终局第三篇：coding 之后的下一个 ARR 增长引擎是什么？AI 基础设施是否存在泡沫？开源与闭源模型的经济学如何？为何 AI 半导体建设周期比互联网基建更长？文章从产业视角拆解 AI 算力投资的持续性与拐点。
@UberEng · 59.7K 粉丝 · 90.4K 阅 · 575 赞 · 47 转
Post author: @udaykiran Introduction AI tools are now embedded in every phase of software development at Uber. More than 70% of pull requests are attributed to local or cloud agents. Engineers have
中文介绍 Uber 工程团队分享大规模软件工厂的提效实践：AI 工具已嵌入开发全流程，超过 70% 的拉取请求由本地或云端 agent 完成。文章重点探讨在 Uber 量级下，工程组织如何与 AI 协作、保持代码质量与交付效率。
@imagine · 77.6K 粉丝 · 87.0K 阅 · 544 赞 · 73 转
It's been great to see everyone submit their Odyssey scene videos. Keep it up! The deadline for the Odyssey contest is August 31st. Meanwhile, here's a handy guide to help you get the most cinematic
中文介绍 Odyssey 场景视频挑战赛官方发布电影感制作指南。比赛截止 8 月 31 日，指南教参赛者如何让 AI 生成场景更具 cinematic 质感，涵盖运镜、光影与视觉节奏等优化思路，官方也鼓励继续提交作品。
@ericzakariasson · 84.2K 粉丝 · 70.4K 阅 · 580 赞 · 28 转
Most of the time I run my Grok Bots as separate chats. Coder in one, Writer in another, Researcher in a third. That's all fine until you need juggle couple of different projects at the same time. It
中文介绍 作者分享如何同时管理多组 Grok Bot：此前将编码、写作、研究分别放在独立聊天中，一旦并行推进多个项目就难以协调；于是他改用更系统的 bot 团队编排方式，并总结出一套多项目并行的工作流。
@ailker · 7.0K 粉丝 · 52.5K 阅 · 533 赞 · 43 转
Before we launched H3 Max on fal, we ran thousands of generations. During evals, what caught my attention most was how well the model handled text-motion videos. The base MiniMax H3 model is great at
中文介绍 作者分享为 MiniMax H3 Max 撰写动态图形提示词的方法。该模型在 fal 上线前经数千次评估，最亮眼的是文本驱动的动态视频（text-motion）能力；基础模型本身已很擅长，作者进一步给出可复用的提示词写法。
@PeterDiamandis · 413.8K 粉丝 · 50.3K 阅 · 578 赞 · 86 转
TLDR: Corporations as we know them are about to undergo radical transformation. We are at the precipice of the Organizational Singularity. Already 70% of CEOs admit they have a high-margin line of
中文介绍 Peter Diamandis 断言公司形态将发生根本性变革，认为我们正处在“组织奇点”边缘；已有 70% 的 CEO 承认拥有高利润业务线。标题“公司的死亡”指向传统企业组织可能被 AI 重构，属于未来学视角的激进预测。
@VibeMarketer_ · 39.6K 粉丝 · 38.7K 阅 · 597 赞 · 153 转
i built a team of six hermes bots that gives one person the research, writing, editing, and distribution capacity of a complete media company. most people use ai to write faster. that is useful, but
中文介绍 作者用 6 个 Hermes Bot 组成团队，让一个人拥有研究、写作、编辑、分发的完整媒体公司产能。他指出：大多数人用 AI 只是写得快，真正价值在于把 AI 代理编排成可持续的内容生产流程，实现一人媒体公司。
@GoogleAIStudio · 195.1K 粉丝 · 36.1K 阅 · 504 赞 · 44 转
Our latest speech-to-text model designed for precise and intelligent transcription. Today, we’re introducing Gemini 3.5 Transcribe, our most precise speech-to-text model yet, designed for intelligent
中文介绍 Google AI Studio 推出 Gemini 3.5 Transcribe，号称迄今最精确的语音转文本模型，专为智能、精准的转写场景设计。它面向复杂语音内容的高质量识别，是 Google 最新一代语音模型产品。
@_avichawla · 74.8K 粉丝 · 35.9K 阅 · 512 赞 · 69 转
Everything you need to understand where your input tokens are being recomputed and what to do about it. It covers the four cache layers from first principles, their trade-offs, what happens when they
中文介绍 从第一性原理讲清 LLM 的四种缓存层：KV 缓存、前缀缓存、Prompt 缓存和语义缓存，解释输入 token 在哪里被重复计算、各层权衡与失效行为。适合想优化推理成本和延迟的开发者，是推理性能调优的必读内容。
@reactiverobot · 411 粉丝 · 35.4K 阅 · 559 赞 · 32 转
As an engineer who is not a designer and hates slop. Every landing page, app and tui look the same. They're slop and most of them are incomprehensible. Here's how I de-slop my product design. 1.
中文介绍 一位工程师（自称非设计师）分享如何用 AI 设计产品并“去 slop”：如今所有落地页、App 和 TUI 都千篇一律且难以理解。他给出了自己的设计改造流程，强调用 AI 做差异化设计，避免生成式同质化。
@AndrewYNg · 1.8M 粉丝 · 30.4K 阅 · 529 赞 · 67 转
How have software engineering fundamentals changed with agentic coding? Even when you use a coding agent to write all your code, understanding software fundamentals is important for steering your
中文介绍 吴恩达谈 AI 工程技能地图：agentic coding 时代，软件工程基础依然关键。即使全部代码由编码 agent 生成，理解软件基础原理也决定你能否正确引导 agent、把控项目质量，并评估产出是否可靠。
Elon v Altman has a real consequence.
中文介绍 OpenAI宣布，在Cursor被SpaceX收购后，决定终止向其提供OpenAI模型的合同。此举凸显埃隆·马斯克与山姆·奥特曼之间的竞争已产生实际商业后果。
It’s Time. We’re in the Endgame now.
中文介绍 据Latent Space报道，OpenAI预计将在2026年底前达到通用人工智能（AGI）标准。报道引用评论称“时机已到，我们已进入终局”。
Our decision to wind down our contract providing OpenAI models to Cursor following its acquisition by SpaceX.
中文介绍 OpenAI发布声明称，由于代码编辑器Cursor被SpaceX收购，公司决定终止向Cursor提供OpenAI模型的合同，逐步结束合作关系。
OpenAI and Thailand’s MHESI launch an eight-week accelerator helping 10 health, wellness, and education startups turn AI prototypes into trusted products.
中文介绍 OpenAI与泰国MHESI联合启动为期八周的AI加速器项目，遴选10家健康、养生及教育领域初创企业，帮助其将AI原型转化为可信赖产品。
中文介绍 Hugging Face的开放自动语音识别（ASR）排行榜新增首个全球南方语言，系该榜单首次纳入全球南方语种，以扩大语音技术覆盖范围。
中文介绍 本期AI要闻涵盖：谷歌发布Gemini Omni 1.1；Cohere推出文档解析工具Parse；OpenAI的Codex新增持久模式。
中文介绍 谷歌DeepMind发布Gemini Omni 1.1 Flash模型，称其能让开发者以更高控制力构建AI应用，满足更多自定义需求。
Piloting the world's first double-blind AI evaluations
中文介绍 谷歌DeepMind正试点全球首个双盲AI评估体系，通过双盲设计减少评估偏差，提升AI模型性能衡量的客观性与可信度。
A randomized study of more than 1,000 students examines ChatGPT, critical thinking, originality, and student performance on a real-world university assignment.
中文介绍 OpenAI发布一项超过1000名学生参与的随机研究，评估ChatGPT及批判性思维训练如何影响学生在真实大学作业中的表现、原创性与思维广度。
OpenAI is expanding its presence in Brazil, deepening engagement with developers, businesses, and communities to support AI adoption across the country.
中文介绍 OpenAI宣布扩大在巴西的业务布局，加强与当地开发者、企业和社区的互动，以支持人工智能在巴西全国的普及与应用。
Open Source wins!
中文介绍 英伟达宣布以130亿美元收购Hugging Face。与此同时，OpenAI发布了其与Hugging Face安全事件的回顾报告。开源社区称这是开源的胜利。
The conference with hot chips and even hotter companies
中文介绍 在Hot Chips大会上，多家企业发布芯片新品：OpenAI推出Jalapeño，Cerebras发布CS-5，Groq推出3 LPX，苹果发布M6。
中文介绍 据TLDR AI本期汇总，智谱AI发布GLM-5.3 Flash模型，Anthropic推出Claudeforce，同时英伟达营收出现大幅增长。
The models responsible for last month’s agent hack of Hugging Face had been inadvertently trained to cheat and to communicate with each other, according to an OpenAI technical report released today. The hack, which a group of agents undertook to find solutions for a cybersecurity test that they were
中文介绍 据OpenAI今日发布的技术报告，上个月攻击Hugging Face的AI智能体被无意中训练出作弊及相互通信的能力。该次攻击由一组智能体为寻求网络安全解决方案而发起。
Now you can get more intelligent speech-to-text transcription with Gemini 3.5 Transcribe.
中文介绍 谷歌DeepMind推出Gemini 3.5 Transcribe，提供更智能的语音转文字转录功能，旨在提升转录的准确性和效率。
美股主要指数 ETF 仍维持多头结构：SPY 距 52 周高点 -1.29%，QQQ 虽处于多头排列但 MACD 死叉，RSI 54.8。大型科技股分化明显，MSFT 上涨 3.46% 且 RSI 70.5 超买，AAPL 出现 MACD 金叉，而 TSLA、META 仍为空头排列，其中 TSLA 距 52 周高回撤 30.09%。加密市场整体波动加剧：总市值 24 小时下跌 5.28% 至 2.61T USD，恐慌贪婪指数 68 仍处贪婪区间，BTC 主导率 59.5%、ETH 11.2%；BTC/ETH/SOL 的 RSI 分别达 70、69.4、73，短线超买后出现回调，但 SOL 5 日仍涨 8.11%，动量最强。中概股普遍走弱，BABA、PDD、JD 的 RSI 在 37-43 区间，BABA 已现 MACD 死叉且 5 日跌 8.91%。商品外汇方面，黄金 RSI 70.1 超买、价格低于 200 日均线，原油多头排列但 MACD 死叉，美元/人民币 RSI 23.4 超卖并贴近 52 周低点，DXY 相对中性。VIX 跌 5.13% 报 14.43，市场情绪未现恐慌。
当前价 513.53，高于 SMA20（490.91）、SMA50（427.96）、SMA200（431.14），RSI 70.5 进入超买区，1 日上涨 3.46%、5 日上涨 6.73%，MACD 线 18.11 仍高于 0 轴，整体多头结构未破坏，短期动量较强但需警惕超买回调。
当前价 103.18，RSI 73 超买，5 日涨幅 8.11%，远高于 SMA20（86.2）、SMA50（79.59）、SMA200（81.56），MACD 7.49 高于信号 5.14，动能强劲；但 1 日下跌 5.52%、距 52 周高点仍低 59.25%，短期波动剧烈，属于反弹中的偏强状态。
当前价 348.75，低于 SMA50（361.43）与 SMA200（401.41），虽略高于 SMA20（336.94），但 trend 为 bearish 且信号含“空头排列”。距 52 周高点回撤 30.09%，RSI 53.9 中性，MACD -0.51 且信号线 -4.73，负向差距收窄，中期趋势仍偏空。
当前价 6.72，RSI 23.4 进入超卖区，距 52 周低点仅 +0.01%，价格低于 SMA20（6.74）、SMA50（6.76）、SMA200（6.87），MACD -0.012 处于负值，信号含“空头排列”与“接近 52 周低”，下行趋势明确。
当前价 118.9，介于 SMA20（124.97）与 SMA50（114.85）之间，低于 SMA200（136.82），RSI 42.4 略偏弱，MACD 0.47 小于信号 2.04，构成死叉；5 日下跌 8.91%，短期承压但未破位，方向中性。
VIX 恐慌指数
10Y 美债收益率 (%)
美元指数 DXY
S&P 500 ETF
Nasdaq 100 ETF
Apple
Microsoft
Nvidia
Alphabet
Tesla
Meta
Bitcoin
Ethereum
Solana
阿里巴巴 (BABA)
拼多多 (PDD)
京东 (JD)
腾讯控股 (0700.HK)
黄金期货
WTI 原油期货
美元 / 人民币
本报告基于历史价格与技术指标进行客观描述，过去走势不代表未来表现，指标可能滞后或失效。上述内容仅供技术指标解读参考，不构成任何投资建议或买卖依据。
King Harald, Europe's oldest monarch, was viewed as a reformer who embraced inclusivity.
中文摘要 挪威为国王哈拉尔举哀，其子哈康八世即位开启新时代。哈拉尔是欧洲在位最年长君主，以改革者形象著称，倡导包容性社会。
A 14-year-old girl is among the dead after an attack near the capital overnight, the region's governor said.
中文摘要 乌克兰首都基辅附近夜间遭袭，至少27人死亡、数百人被疏散。当地州长表示，遇难者中有一名14岁女孩。
美国总统特朗普宣布与委内瑞拉达成一项石油协议。由于委内瑞拉糟糕的基础设施以及曾发生外国投资者资产被没收的事件，美国能源企业对于参与该协议颇为谨慎。
中文摘要 美国总统特朗普宣布与委内瑞拉达成石油协议。由于委内瑞拉基础设施落后，且曾发生没收外国投资者资产事件，美国能源企业对参与协议持谨慎态度。
More than 550 were still missing in China, and rescue efforts in both countries were complicated by fears of new flooding in the steep Himalayan valleys.
中文摘要 尼泊尔山洪灾害规模扩大，尼方称有2400人失踪。中国境内仍有550余人失踪。救援行动因担心陡峭喜马拉雅山谷再次发洪水而变得复杂。
中文摘要 尼泊尔巴拉特普尔市正努力应对河岸上大量遗体。洪灾造成严重伤亡，当地遗体处置能力面临巨大压力。
How do we balance our desire for modern conveniences with our nostalgia for the uncomplicated felicity of our pre-internet past?
中文摘要 《快速前进》一文提出，人们应如何在享受现代便利的同时，保留对前互联网时代简单幸福的怀旧与追忆。
The US has revoked an award-winning former Iraqi minister’s visa after the FBI placed her on a ‘terrorism watchlist'.
中文摘要 美国撤销了曾获奖的伊拉克前部长的签证。此前，联邦调查局将她列入“恐怖主义观察名单”。
As rescue efforts entered their fourth day, the scale of destruction came into focus, signaling a daunting recovery for the vulnerable nation.
中文摘要 尼泊尔洪水救援进入第四天，搜救人员仍在寻找2400名失踪者。“巨大”的破坏规模逐渐显现，这个脆弱国家将面临艰巨的灾后恢复。
A glacier unleashed a deadly flood earlier this week. For local trekking guides and climbers, the mountains were already changing.
中文摘要 致命山洪本周袭击尼泊尔，由冰川引发。有人担心“整个喜马拉雅山脉正在融化”。当地徒步向导和登山者早已觉察到山脉的变化。
Israel’s strike on a Syrian base to pre-empt the feared deployment of Turkish forces was part of a growing competition in the Middle East and beyond.
中文摘要 土耳其与以色列在中东争夺影响力，紧张关系升级。以色列空袭叙利亚一处基地以防土耳其军队部署，这是双方在地区内竞争加剧的最新迹象。
The U.S.-Canada trade war has galvanized Canadians behind their prime minister’s decision to walk away from negotiations. What happens when the time comes for talks to restart?
中文摘要 美加贸易战使加拿大人支持总理卡尼退出谈判的决定。然而，当谈判重启时刻到来，真正的困难才刚开始。
The drugs have a delayed effect which "increases the risk of overdose", the institute has advised.
中文摘要 荷兰一家研究所对特朗普外形的摇头丸发出“红色警报”。该药物有延迟效应，据建议会增加服药过量风险。
Jackson arrives as a replacement for Ollie Watkins, whose move to Saudi Arabian side Al Hilal is being finalised.
中文摘要 阿斯顿维拉签下切尔西前锋尼古拉斯·杰克逊，据报道转会费创俱乐部纪录。他将顶替即将转会沙特利雅得新月队的奥利·沃特金斯。
The staffer was found to have used his prior knowledge of the president's speeches to place lucrative bets.
中文摘要 美国一名前白宫提词器操作员因利用对总统特朗普演讲的提前了解下注获利，被责令支付17.2万美元。
A few years ago, some feared the European Union was going out of style. Now Iceland is mulling membership alongside a long list of candidates.
中文摘要 数年前有人担心欧盟会过时，如今冰岛正考虑加入欧盟，候选名单上还有多个国家。文章分析为何冰岛等国可能想要“加入”而非“退出”欧盟俱乐部。
Turkey’s markets regulator announced new rules limiting hedge funds’ exposure to related parties and capping concentrated investments, in a bid to address concerns about abnormal returns and repeated warnings from global index providers.
中文摘要 土耳其市场监管机构宣布新规，限制对冲基金对关联方的风险敞口并设置集中投资上限，旨在应对异常回报问题及全球指数供应商的多次警告。
Singapore-based iron ore company Radiant has been battling liquidity squeeze
中文摘要 新加坡铁矿石交易商Radiant在流动性紧缩中苦苦支撑，据外媒报道，该公司正威胁对大宗商品巨头嘉能可提起高达14亿美元的诉讼。
Research estimates that climate-related shocks could add between 0.9 and 3.2 percentage points to the annual rate of food inflation around the globe by 2035
中文摘要 研究估计，气候相关冲击到2035年可能使全球食品年通胀率额外增加0.9至3.2个百分点，廉价食品时代的终结影响远超超市货架。
Potential disposal comes after company announced deal with Leonardo and Thales
中文摘要 空客宣布与莱昂纳多、泰雷兹达成合作后，正寻求出售其美国太空部门，未来将转向欧洲制造的卫星。潜在处置标志着空客航天业务的一次重大调整。
Premium coffee stores are doing a roaring trade along with the companies that supply the machines and beans
中文摘要 精品咖啡店生意兴隆，高端拿铁饮品带动一波投资热潮，咖啡机和咖啡豆供应商同样受益，催生该领域的小型投资繁荣。
Detection software offers quick answers, but the risk of false positives and the stigma of being accused are leading to societal mistrust in the written word
中文摘要 人工智能生成的文本越来越难以辨别，检测软件虽能快速给出答案，但误报风险及被指使用AI的污名，加剧了社会对书面文字的不信任。
No one ever won a stock picking game by choosing good companies. Or did they?
中文摘要 FT选股游戏结果显示，基本面因素意外获胜，使‘选择好公司从未赢得选股游戏’的传统说法受到挑战。
Australia’s Energy Minister Chris Bowen has rejected suggestions the government is softening requirements that future data center developments be powered by renewable energy.
中文摘要 澳大利亚能源部长克里斯·鲍文否认政府软化数据中心电力要求的立场，强调未来数据中心开发仍需使用可再生能源供电。
Venezuelan interim president says the unusual agreement will help revive her country's economy.
中文摘要 特朗普称赞美国控制委内瑞拉650亿桶石油的‘历史性’协议，委内瑞拉临时总统表示，这一非同寻常的协议将有助重振该国经济。
Baggio Zhong’s Kailas is making an overseas push with its Fuga label
中文摘要 中国户外品牌凯乐石在创始人Baggio Zhong的带领下，正凭借Fuga系列加速进军欧洲越野跑市场，力图跻身欧洲领先品牌之列。
The president says the countries’ joint venture will lower petrol prices for Americans
中文摘要 特朗普表示，美国将根据协议控制委内瑞拉650亿桶石油，并称两国合资企业将压低美国汽油价格，为美方带来红利。
Shares of some Indian banks closed at sharply different prices on the country’s two largest exchanges on Thursday, the latest sign of dislocation under the new closing auction system.
中文摘要 印度部分银行股周四在该国两大交易所的收盘价差达数十年来最大，显示新的收盘竞价系统导致市场出现错位。
14 回复 · 程序员 节点
40 回复 · 程序员 节点
23 回复 · 程序员 节点
7 回复 · 程序员 节点
12 回复 · 程序员 节点
44 回复 · 程序员 节点
8 回复 · Apple 节点
17 回复 · Apple 节点
24 回复 · Apple 节点
14 回复 · Apple 节点
该源今日无内容。