JavaScript · ★ 153,370 · 🍴 8,222 · 📈 1,289 stars today
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
中文介绍 给 AI 编程 Agent 注入"最懒资深工程师"思维的技能包，核心理念是"最好的代码是你从没写过的代码"，引导 Agent 少写代码、优先复用与删减。适合不想让 AI 过度生成冗余实现、希望它像老手一样克制动手的开发者，用于 Claude Code 等编码 Agent 场景。
JavaScript · ★ 75,283 · 🍴 4,516 · 📈 705 stars today
The design language that makes your AI harness better at design.
中文介绍 一套设计语言规范，用来提升 AI 编程工具在设计层面的输出质量，让 Agent 生成的界面更符合审美与一致性要求。适合用 AI 生成前端 UI、但苦于默认产出"能用不好看"的开发者，配合各类编码 Agent 使用以约束视觉风格。
JavaScript · ★ 272,223 · 🍴 40,648 · 📈 954 stars today
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
中文介绍 面向 AI Agent harness 的性能优化系统，涵盖 skills、instincts、memory、security 以及 research-first 开发范式，适配 Claude Code、Codex、Opencode、Cursor 等工具。适合希望提升 Agent 执行效率、稳定性和安全性的重度用户与团队，用于日常编码工作流调优。
TypeScript · ★ 16,808 · 🍴 813 · 📈 302 stars today
Build production-ready applications in TypeScript
中文介绍 基于 TypeScript 的生产级应用开发框架，核心是把副作用、依赖注入、错误处理和并发等纳入类型系统统一管理，让复杂业务代码更可组合、可测试。适合构建大型后端服务、需要严格类型约束和可靠错误处理的 TypeScript 工程团队。
Go · ★ 109,522 · 🍴 6,333 · 📈 505 stars today
🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talking like a caveman.
中文介绍 一个让编码 Agent"像原始人一样说话"的 skill 与代理，通过极度精简的提示语言把 token 消耗削减约 65%，从而降低成本、加快响应。适合频繁调用大模型跑编码任务、对 token 费用敏感的个人开发者与团队。
Python · ★ 89,775 · 🍴 7,898 · 📈 1,683 stars today
Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
中文介绍 给 AI Agent 装上"看全网的眼睛"，用单一 CLI 即可读取和搜索 Twitter、Reddit、YouTube、GitHub、Bilibili、小红书等平台内容，且无需 API 费用。适合需要让 Agent 抓取多平台公开信息做调研、舆情或内容汇总的场景。
TypeScript · ★ 24,675 · 🍴 6,439 · 📈 251 stars today
中文介绍 信息不足，无法确认该项目具体功能与技术细节，仅能从仓库名推断其与 T3 技术栈或代码工具相关。建议直接查看仓库 README 获取准确说明后再评估是否适用于你的开发流程。
TypeScript · ★ 95,562 · 🍴 8,459 · 📈 218 stars today
Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
中文介绍 为 Agent 提供跨会话的持久记忆：自动捕获会话中 Agent 的全部操作，用 AI 压缩处理后，在后续会话中有选择地注入相关上下文。解决 Agent 每次重启就"失忆"的问题，适合长期使用 Claude 等编码 Agent 的开发者。
TypeScript · ★ 10,558 · 🍴 1,282 · 📈 84 stars today
Agent workspace built on Cloudflare Workers for creating documents, building apps, and running agents with your company’s context and systems.
中文介绍 基于 Cloudflare Workers 构建的 Agent 工作空间，可创建文档、搭建应用并运行 Agent，同时接入公司自身的上下文与内部系统。适合希望把 AI Agent 部署在边缘、与自有业务数据打通的团队，用于内部工具与文档协作。
JavaScript · ★ 100,828 · 🍴 10,594 · 📈 305 stars today
Production-grade engineering skills for AI coding agents.
中文介绍 面向 AI 编码 Agent 的生产级工程技能集，把真实项目中的工程实践沉淀为可复用的 skill，提升 Agent 输出的代码质量与规范度。适合希望让 Claude Code、Cursor 等工具按团队工程标准工作的开发者与工程团队。
Shell · ★ 294,907 · 🍴 26,354 · 📈 578 stars today
An agentic skills framework & software development methodology that works.
中文介绍 一套可落地的 Agentic skills 框架与软件开发方法论，强调把技能组织成体系并配合明确流程来驱动 AI 完成开发任务，而非零散提示。适合想系统化使用 Agent 做工程交付的开发者与团队参考实践。
Shell · ★ 275,347 · 🍴 23,104 · 📈 750 stars today
Skills for Real Engineers. Straight from my .agents directory.
中文介绍 作者直接开源自己 .agents 目录下的 AI Agent 技能配置，面向真实工程场景，配合编码 Agent 使用。适合想参考有经验的工程师如何组织 skills、优化 AI 编码工作流的开发者，直接借鉴或复制其配置。
TypeScript · ★ 25,244 · 🍴 1,807 · 📈 256 stars today
Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.
中文介绍 面向 AI 编码 Agent 的上下文窗口优化方案：把工具输出沙箱化可减少约 98% 占用，同时持久化会话记忆，并通过 MCP 与 hooks 在 17 个平台上统一路由。适合上下文频繁爆掉的重度 Agent 用户压降 token 开销。
TypeScript · ★ 112,150 · 🍴 14,228 · 📈 408 stars today
AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI
中文介绍 一套 AI Agent 工具包，包含统一的 LLM API 封装、Agent 循环、终端 UI（TUI）以及编码 Agent 命令行工具。适合想自建 Agent 应用的开发者，用统一接口对接多家模型，快速搭建带交互界面的编码助手。
Python · ★ 45,201 · 🍴 4,898 · 📈 211 stars today
Developer-first error tracking and performance monitoring
中文介绍 开发者优先的错误追踪与性能监控平台，可实时捕获前后端异常、定位堆栈并分析性能瓶颈，帮助团队在用户报障前发现线上问题。适合各类 Web、移动和后端应用团队接入 SDK 做稳定性与性能可观测性建设。
TypeScript · ★ 149,216 · 🍴 25,359 · 📈 127 stars today
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
中文介绍 Anthropic 官方的终端 Agentic 编码工具，能理解你的代码库，通过执行日常任务、解释复杂代码和处理 Git 工作流来加快开发速度。适合习惯命令行的开发者，在本地仓库中直接让 Claude 辅助编码、重构与提交。
Python · ★ 9,397 · 🍴 2,069 · 📈 192 stars today
中文介绍 信息不足，仅能从仓库名推断这是一门关于"生产级 Agentic RAG"的课程，可能涉及把 RAG 检索与 Agent 结合并部署到生产环境。适合想系统学习 RAG 与 Agent 工程落地的开发者，具体内容需查看仓库。
Python · ★ 8,729 · 🍴 1,523 · 📈 43 stars today
中文介绍 信息不足，从仓库名可推断是美团 LongCat 团队开源的视频相关项目，可能涉及视频生成或理解模型。由于缺少描述，无法确认具体技术路线与使用场景，建议查看仓库文档了解模型能力、输入输出形式与部署要求。
TypeScript · ★ 91,599 · 🍴 9,056 · 📈 234 stars today
The open-source CapCut alternative
中文介绍 开源版的 CapCut（剪映）替代品，提供视频剪辑功能，让用户无需依赖闭源商业软件即可完成剪辑。适合注重隐私、成本或可定制性的创作者与开发者，用于日常短视频剪辑，或基于其代码二次开发自有剪辑工具。
👍 160
Streaming video LLMs must retain evidence before its relevance to future tasks is known and respond when sufficient evidence becomes available. The challenge is to form reusable factual memory without compromising real-time perception. We introduce OneStreamer, which jointly learns query-independent
中文介绍 流式视频 LLM 需在相关性未知前保留证据并及时响应，难点是兼顾可复用事实记忆与实时感知。OneStreamer 联合学习与查询无关的记忆构建与主动响应机制，统一感知、记忆与响应，在不牺牲实时性的前提下提升流式视频交互表现。
👍 153
On-policy learning has been argued to reduce catastrophic forgetting, produce sparser parameter updates, and improve generalisation. However, existing comparisons between supervised fine-tuning and reinforcement learning vary many factors simultaneously, making the contribution of rollout policy dif
中文介绍 针对「on-policy 可减少灾难性遗忘、产生更稀疏更新并提升泛化」的说法，本文系统解耦 SFT 与 RL 比较中的多重变量，单独考察 rollout 策略的贡献，厘清蒸馏动态中 on-policy 与 off-policy 学习的真实差异。
👍 57
Masked diffusion models (MDMs) generate sequences by progressively unmasking several tokens per denoising step, but their reverse process is typically factorized over positions, limiting sample quality in the few-step regime where diffusion's speed advantage over autoregressive decoding matters most
中文介绍 掩码扩散模型（MDM）逐步解掩多个 token，但其反向过程通常按位置分解，限制了少步采样质量——而少步恰恰是扩散相对自回归的速度优势所在。E-MoE 提出增强型 MoE 结构用于非因子化扩散语言模型，提升该场景下的生成质量。
👍 49
Multi-reward reinforcement learning trains large language models to satisfy multiple behavioral objectives simultaneously. Reward-wise normalization, as used in GDPO, preserves reward-specific relative information within rollout groups, but different objectives can still exhibit uneven learning prog
中文介绍 多奖励 RL 同时优化多个行为目标，GDPO 等的奖励级归一化虽保留组内相对信息，但各目标学习进度仍不均衡。本文提出密度感知的奖励聚合方法，解决稀疏奖励下多目标训练不平衡问题，使稀疏奖励真正发挥作用。
👍 44
Real-world embodied tasks, from everyday activities to professional procedures, require agents to act under physical constraints while tracking evolving object and task states. Tool use sits at the heart of such tasks, as many everyday and professional activities are tool-mediated. Understanding the
中文介绍 真实具身任务（日常与专业流程）要求智能体在物理约束下行动并追踪物体与任务状态，而工具使用是核心。EgoTools 面向第一视角视频构建以工具为中心的推理基准，考察智能体对工具中介活动的理解与状态跟踪能力。
👍 43
We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining. Our key insight is that a broad controller already holds much of t
中文介绍 研究人形机器人 loco-manipulation 的测试时进化：不重新训练，而是复用已有技能、从自身尝试中改进并保留所学。核心洞见是通用控制器已内含大量所需能力，InterEvolve 通过测试时进化奖励程序实现新任务求解。
👍 41
Decentralized multi-agent path finding (MAPF) with communication requires agents to reach individual goals without collisions under partial observability. Learnable policies trained on expert data provide an effective approach to this problem. However, when several coordinated joint actions are vali
中文介绍 部分可观测下的去中心化 MAPF 需通信协作避碰。基于专家数据的可学习策略有效，但当多个协调联合动作无效时性能受限。本文提出「去中心化 Master-Mind」，通过迭代意图去噪实现联合动作细化。
👍 37
Multi-step agents are trained on flat action streams: SFT and RLVR weight every token uniformly and ignore the sub-procedures that recur across tasks, the hierarchy that lets humans plan top-down from reusable routines. This structure sits unused, and flat training uses each scarce trajectory less f
中文介绍 多步智能体常在扁平动作流上训练，SFT 与 RLVR 对所有 token 一视同仁，忽略跨任务复现的子过程与层级结构，导致稀缺轨迹利用不足。X-Tree 将可复用经验 token 化，以层级方式提升智能体泛化效率。
👍 34
Vision language model (VLM) agents can control robots through visual feedback and action primitives, but repeated model invocations and redundant observations incur substantial token overhead. We introduce PyRUA-Lean, an interactive code-execution framework that couples feedback-driven primitive com
中文介绍 VLM 智能体可经视觉反馈与动作原语控制机器人，但反复调用模型和冗余观测带来大量 token 开销。PyRUA-Lean 提出交互式代码执行框架，耦合反馈驱动的原语组合，在成功率提高 14% 的同时减少 65% token。
👍 27
High-resolution 3D generation increasingly relies on voxel latents and multi-stage pipelines that first predict active structure and then synthesize local geometry. While effective, this design fragments continuous surfaces into many local tokens, inflates generation cost, and often weakens topologi
中文介绍 高分辨率 3D 生成多依赖体素潜变量与多阶段流程：先预测活跃结构再合成局部几何，但这会把连续表面切碎成大量局部 token，抬高成本并削弱拓扑一致性。SILSA 提出滑动窗口切片潜变量以保持拓扑。
👍 26
Furnished floor plans support real-estate visualization, interior design, and architectural workflows, yet automatic furnishing remains challenged by limited real-world data and the need to satisfy interacting geometric and functional constraints. We ask whether professional furnishing knowledge can
中文介绍 自动户型家具布置服务于房产可视化与室内设计，但受限于真实数据稀缺及几何与功能约束相互耦合。Architect-Ant 探索能否将专业布置知识引入可编辑的平面图自动家具配置流程，满足多重约束。
👍 25
Looped Transformers achieve parameter efficiency by repeatedly executing a shared block across recurrent loops. Each loop yields an intermediate representation decodable for the same next token, yet standard decoding discards earlier states. Because earlier loops embody less computation, recurrence
中文介绍 循环 Transformer 通过重复执行共享块实现参数高效，每次循环产生的中间表示都可解码同一下一 token，但标准解码丢弃早期状态。本文利用早期循环计算量更少的特点改进循环 Transformer 解码，近乎零成本提升效果。
👍 23
Group Relative Policy Optimization (GRPO) is widely used to train reasoning language models, where it computes advantages by centering and normalizing rewards across rollouts of the same prompt. For multiple rewards, GRPO sums the reward components and normalizes the total reward by its within-group
中文介绍 GRPO 广泛用于推理语言模型训练，通过组内奖励中心化与归一化计算优势；多奖励时直接求和再按总奖励组内归一化，忽略奖励间相关性。CorrGRPO 提出相关性归一化的 GRPO，改善多奖励学习中的优势估计。
👍 23
We present Multimodal Flow, a fully continuous generative model of language and vision. Most unified multimodal models either model both language and quantized images as discrete tokens or combine discrete language prediction with continuous image generation. The former introduces a visual quantizat
中文介绍 多数统一多模态模型要么把语言与量化图像都当离散 token，要么离散语言预测搭配连续图像生成，前者引入视觉量化误差。Multimodal Flow 提出在嵌入空间对语言与视觉统一做全连续流建模，避免离散化损失。
👍 21
Video generation models are increasingly being explored as world simulators for embodied planning and learning. To do so effectively, these models must not only generate visually appealing frames, but also predict how environments dynamically evolve when executing goal-directed actions. While evalua
中文介绍 视频生成模型正被探索作为具身规划与学习的世界模拟器，但需能预测目标导向动作下环境如何演化，而非仅画面美观。Ego2Act 构建第一视角视频生成中目标导向操作的评测基准，考察动作执行后的动态预测能力。
👍 21
Real-world enterprise data science and analytics workflows require reasoning across dozens of tables, performing statistical analyses, and acting on the results. Established text-to-SQL benchmarks evaluate query generation alone, and audits have found their answer keys frequently wrong. Because real
中文介绍 企业级数据科学与分析流程需跨数十张表推理、做统计分析并据结果行动，而既有 text-to-SQL 基准只考查询生成，且答案键常被审计出错误。Argo-Bench 提出面向企业级工作流的数据智能体评测基准。
👍 19
Extending a text embedding model to new modalities typically degrades text retrieval quality, and existing omni-modal embedders compensate with multi-billion parameters. We present Omni-Embed-Mini, a 0.9B-parameter model that maps text, speech, audio, images, video, and visually-rich documents into
中文介绍 文本嵌入模型扩展到新模态常导致文本检索质量下降，现有全模态嵌入器多以数十亿参数补偿。Omni-Embed-Mini 仅用 0.9B 参数，通过稠密蒸馏将文本、语音、音频、图像、视频与富视觉文档映射到统一空间而不遗忘文本能力。
👍 17
Vision-Language Models (VLMs) have shown strong multimodal reasoning capabilities, yet whether they truly capture the physical consistency underlying real-world dynamics remains unclear. Existing benchmark paradigms often suffer from fragmented evaluation, focusing on isolated cognitive stages while
中文介绍 VLM 多模态推理能力虽强，但是否真正掌握真实动态背后的物理一致性仍不明确，现有基准评估割裂、只考察孤立认知阶段。PhysVista 提出感知—推理—评估闭环的物理智能评测基准，系统检验 VLM 的物理理解。
👍 15
Preference distillation typically treats a teacher response as preferred and the student's own response as rejected. This assumes that self-generated failures are the most informative negatives and that rejects must come from a model at least as large as the student, making generation costly at scal
中文介绍 偏好蒸馏通常以教师回复为正、学生自身回复为负，假设自生成失败最有信息量且负例须来自不小于学生的模型，使规模化生成代价高昂。本文重新审视该假设，研究偏好蒸馏在不同模型规模下的缩放规律。
👍 14
On-policy self-distillation has recently emerged as an effective approach for improving language-model reasoning by supervising students with a frozen or EMA version of themselves that receives privileged information. Its application to multimodal large language models (MLLMs), however, remains larg
中文介绍 On-policy 自蒸馏通过让接收特权信息的冻结或 EMA 版本监督学生来提升语言模型推理，但在多模态大模型（MLLM）上应用仍有限。Where-OPD 提出基于合成场景的空间引导 on-policy 自蒸馏方法。
@jakeserval · 1.1K 粉丝 · 406.0K 阅 · 505 赞 · 1.1K 转
There is a lot of talk about the jobs AI takes away and very little about the ones it creates. While pessimism might be easier (and certainly gets more clicks), I’m more optimistic. From where I sit,
中文介绍 针对「AI 抢走工作」的悲观叙事，博主提出相反视角：AI 正在催生「自动化工程师」这一新岗位。核心论点是与其纠结被替代的职位，不如关注 AI 创造出的新角色，属于职业趋势观察类观点帖。
@0xROAS · 54.6K 粉丝 · 118.6K 阅 · 502 赞 · 56 转
Seedance 2.5 is one of the craziest "nano banana moments" of AI VIDEO in 2026. It's the model behind every viral video that you are seeing right now... and it's pretty simple to generate those video,
中文介绍 分享用视频生成模型 Seedance 2.5 批量做出爆款视频的方法。博主称它是 2026 年 AI 视频领域的「nano banana 时刻」，当前大量刷屏视频都出自该模型，并强调生成流程本身并不复杂，属于工具实操教程。
@0xSigil · 56.0K 粉丝 · 79.1K 阅 · 509 赞 · 72 转
6 months ago, the #1 AI was Claude Opus 4.6. Our model, Underdog 27B beats it while running on your Mac or iPhone. Most AI will run on your device, not in data centers. Intelligence that recently
中文介绍 博主发布自家模型 Underdog 27B，宣称其表现超过 6 个月前的头部模型 Claude Opus 4.6，且能在 Mac 或 iPhone 本地运行。核心主张是未来大部分 AI 会跑在终端设备而非数据中心，主打端侧智能与数据自主权。
@UberEng · 66.9K 粉丝 · 54.0K 阅 · 529 赞 · 59 转
Introduction Uber’s rapid adoption of AI agents has fundamentally changed how teams interact with code, data, and operational systems. Early ad-hoc integrations with the MCP (Model Context Protocol)
中文介绍 Uber 工程团队详解内部 MCP Gateway 平台。随着 AI agent 大规模接入代码、数据和运维系统，早期点对点 MCP 集成难以为继，文章介绍如何统一管理 MCP 连接，属于大厂基础设施层面的工程实践分享。
@VibeMarketer_ · 37.8K 粉丝 · 31.4K 阅 · 504 赞 · 59 转
i'm going to show you how to build a content machine with Opus 5.5 that turns your ideas, research and expertise into articles, posts and newsletters in your own voice. from finding your next topic to
中文介绍 一套用 Opus 5.5 搭建内容生产流水线的完整课程：把想法、调研和专业知识转化为保持个人语气的文章、社交帖和 newsletter，覆盖从选题到发布的全流程，属于 AI 内容创作的体系化教程。
@goyal__pramod · 11.4K 粉丝 · 24.7K 阅 · 504 赞 · 59 转
Now I am aware you must have been looking forward to the second part of my CUDA blog (if you haven't read it yet, check it out!), but hey! A man can have varied interests. And I believe if you are
中文介绍 博主暂别此前的 CUDA 系列博客，开写强化学习从零入门第一篇。定位是面向想系统理解 RL 原理的读者，属于基础知识科普系列的开篇，适合作为后续连载的起点。
@ActionModelAI · 60.2K 粉丝 · 6.8K 阅 · 501 赞 · 439 转
OpenAI has chosen not to release GPT-6.1 Astra after the model failed to meet its safety standards. To be clear, GPT-6.1 Astra is not GPT-6.1 Sol, the model OpenAI released at DevDay. GPT-6.1 Sol has
中文介绍 OpenAI 因未达安全标准，选择不发布 GPT-6.1 Astra，注意它与 DevDay 上发布的 GPT-6.1 Sol 并非同一模型。博主借此讨论 AI 发布权限与安全门槛问题，属于对模型发布决策的解读与评论。
中文介绍 微软在 Hugging Face 发布文章，探讨 AI 智能体声称任务已完成、但数据库状态并不一致的问题，聚焦智能体在实际执行任务时的可靠性与状态校验。
a quiet day.
中文介绍 Latent Space 的 AINews 栏目当日发布简讯，称行业消息较少，是一个相对平静的一天，未涉及具体数值或重大事件。
Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.
中文介绍 OpenAI 发布 GPT-6 系列模型实用指南，介绍初创企业如何选择 GPT-6 模型、调节推理投入、优化提示与技能、协调工具，并为生产环境准备工作流。
Enterprise AI is no longer a future ambition. It is in full operational flight. Model capabilities are advancing faster than most organizations can absorb, while the cost of performance continues to fall. Globally, AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year. F
中文介绍 MIT Tech Review 报道，自主 AI 正在重新定义企业智能。2026 年全球 AI 投资预计达 2.5 万亿美元，同比增长 44%，模型能力提升速度快于多数机构的消化能力，同时性能成本持续下降。
中文介绍 艾伦人工智能研究所（AI2）在 Hugging Face 开源 AstaBrief，这是其 Asta 工具中用于快速生成报告的模型，面向高效报告撰写场景。
After leading Meta’s Llama models, Ahmad Al-Dahle is now transforming Airbnb with AI — from how its teams develop products to how it serves guests.
中文介绍 Latent Space 报道，曾主导 Meta Llama 模型的 Ahmad Al-Dahle 现已加入 Airbnb 推动 AI 转型，涵盖团队产品开发方式以及对房客服务的改造。
On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board in what looked like a gift to its human opponent. Move 37 in game two of the five-game match looked so absurd that some commentators thought it was a…
中文介绍 MIT Tech Review 刊文指出大语言模型并不具备推理能力，并以 2016 年 3 月首尔一场围棋赛第 37 手为例，说明看似荒诞的落子曾被部分评论者误读，借此讨论对模型能力的误判。
the minimalist harness goes stable... and TypeScript!
中文介绍 Latent Space 的 AINews 报道 Pi 1.0 与 Pi Durable 发布，该极简式 harness 框架进入稳定版本，并已支持 TypeScript。
中文介绍 ServiceNow AI 在 Hugging Face 发布 AutoSynthData，用于为企业级智能体自动生成训练数据，解决企业场景下训练语料构建的问题。
We catch up with RLM first author Alex Zhang, MIT PhD, on Jev, PhD masxing, and the future of harnesses.
中文介绍 Latent Space 访谈 RLM 论文第一作者、麻省理工学院博士生 Alex Zhang，讨论其研究动机、读博经历以及 harness 框架的未来方向。
Chatham Financial uses Codex and GPT-5.6 to build technology and redesign workflows, cutting trade validation from 30 minutes to under 4.
中文介绍 Chatham Financial 使用 OpenAI 的 Codex 与 GPT-5.6 构建技术并重构工作流，将交易验证时间从 30 分钟缩短至 4 分钟以内，以扩展其资本市场专业能力。
中文介绍 TLDR AI 当日简报涵盖三个话题：决策模型、被称为「Claude 式」的科学研究方式，以及 OpenAI 安全团队的人事解雇事件。
Advanced AI may matter most for the routine work behind breakthrough ideas. Explore why execution could shape the next economy and the pace of progress.
中文介绍 OpenAI 发文《永恒的互补》指出，先进 AI 对突破性想法背后的日常执行工作可能最为关键，并探讨执行力如何影响下一阶段经济与进步速度。
Albertsons Cos. is using ChatGPT Enterprise and the OpenAI API to help teams work faster and make grocery shopping easier for millions of customers.
中文介绍 艾伯森公司（Albertsons）使用 ChatGPT Enterprise 与 OpenAI API，帮助团队提升工作效率，并为数百万顾客简化食品杂货购物体验，从内部重塑零售业务。
... but you can’t try it yet unless you are “government users and trusted cyber defenders in the Fairwind Program”
中文介绍 Latent Space 的 AINews 报道谷歌 DeepMind 推出 Gemini 4 Argon，作为对 Astra/Fable 的回应，支持 100 万 token 输出；目前仅向「Fairwind 计划」中的政府用户与受信任网络防御方开放试用。
美股三大 ETF 技术面仍偏强：SPY 报 769.64、距 52 周高仅 -1.25%，QQQ 报 749.58、RSI14 65.4 且站上 sma20 727.78，但 SPY 已出现 MACD 死叉（0.8356 对 0.8931），提示上行动量边际减弱；个股中 NVDA 报 233.95（五日 +3.95%）、MSFT 触发 MACD 金叉，而 META、AAPL、TSLA 多为死叉读数。加密方面，恐慌贪婪指数 67 处于贪婪区，总市值 2.90T 美元、24h -2.22%，BTC 主导率 58.6%、ETH 11.3%，BTC 报 84766.92、RSI14 63.7 且高于 sma20 82442.03 维持多头排列，但 MACD 1939.97 低于信号线 2093.64 呈死叉，ETH 与 SOL 同样多头排列但 SOL 亦现死叉，显示市值回落与结构分化并存。中概股整体最弱：BABA 报 105.85、RSI14 35.6，PDD 报 75.38、RSI14 33.4，JD 报 25.78、RSI14 30.4，0700.HK 报 421.2、距 52 周低仅 2.48%，四者均为空头排列叠加 MACD 死叉。商品外汇与宏观分化明显：黄金期货报 4162.3、RSI14 34.1 且呈空头排列，WTI 原油报 91.11 维持多头排列，美元人民币报 6.7 逼近 52 周低，而 10Y 美债收益率 5.28（RSI14 74.6 超买）与美元指数 101.92（RSI14 71.7 超买）双双接近 52 周高，VIX 报 15.31、空头排列但出现 MACD 金叉，风险偏好读数整体偏向高利率与强美元一侧。
10Y 收益率报 5.28，单日 +0.76%、五日 +1.79%，RSI14 达 74.6 进入超买区，MACD 0.1343 高于信号线 0.1163 维持红柱，价格 5.28 高于 sma20 5.05、sma50 4.83、sma200 4.45 呈完整多头排列，并逼近 52 周高（仅差 1.22%），动量指标读数处于全清单最极端水平。
DXY 报 101.92，五日 +0.94%，RSI14 71.7 处于超买状态，MACD 0.5794 高于信号线 0.3881，红柱扩张，价格位于 sma20 100.36 / sma50 100 / sma200 99.27 之上构成多头排列，距 52 周高仅 -0.28%，是当日宏观组中趋势与动量读数最集中的一档。
BABA 报 105.85，单日 -1.49%、五日 -3.54%，RSI14 仅 35.6 偏弱，MACD -2.3886 低于信号线 -2.1837 呈死叉绿柱，价格低于 sma20 110.07 / sma50 116.87 / sma200 130.93 为空头排列，距 52 周高 -44.17%，中概组内技术面整体承压最明显之一。
腾讯报 421.2，单日 -2.27%、五日 -3.92%，RSI14 40.6 偏弱，MACD -3.9952 低于信号线 -3.9536 呈死叉，价格跌破 sma20 433.7 / sma50 447.51 / sma200 498.88，空头排列明确，距 52 周低仅 2.48%，为清单中同时触发接近 52 周低与空头排列的标的。
GOOGL 报 343.5，单日 +1.56% 但五日 -0.12%，RSI14 恰为 50 处中性中枢，MACD -0.5463 低于信号线 -0.3972 呈死叉，价格 343.5 夹在 sma20 342.62 与 sma50 344.37 之间、略高于 sma200 338.92，趋势标签为中性，方向读数最不集中，属典型盘整结构。
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
本报告仅基于公开行情计算的技术指标读数进行客观复述，不构成任何投资建议；过去走势不代表未来表现，指标金叉、死叉、超买与超卖信号均可能失效，仅供技术指标解读参考，请读者独立判断并自行承担相关风险。
The US coastguard says it has dispatched air and surface crews to search for the plane near Nantucket.
中文摘要 一架从百慕大飞往波士顿的空中救护机失联。美国海岸警卫队表示，已派出空中和海上救援人员，在楠塔基特附近海域展开搜寻。
Connie Chan, endorsed by Nancy Pelosi to succeed her in California, says she will vote to 'end a genocide in Gaza'.
中文摘要 获美国前众议长佩洛西背书、接替其加州席位的候选人陈（Connie Chan）呼吁对以色列实施「全面武器禁运」，称将投票「结束加沙的种族灭绝」。
Man charged with nine offences after vehicle struck Knights fans attending grand final farewell, leaving 10 injured including three children Get our breaking news email, free app or daily news podcast Ten people have been injured, one critically, after a car crashed into a crowd of people in Newcast
中文摘要 澳大利亚纽卡斯尔一辆汽车冲撞参加NRL骑士队总决赛告别活动的球迷人群，造成10人受伤，含3名儿童，一名5岁女孩昏迷。一名持P牌驾照男子被控9项罪名。
President Trump says he's been having success winning races in Latin America. As Brazilian voters go to the polls, they're also deciding how much influence Trump should have in their country.
中文摘要 巴西选民在大选投票之际，也在决定美国总统特朗普对本国应有多大影响力。特朗普称其在拉美选举中连连得手，部分巴西选民将此次投票视为对特朗普的表态。
High school students say their classrooms are overcrowded, teachers are missing and buildings need repairs. The protests have spread to cities across France, and some 5,000 people have been arrested.
中文摘要 法国高中校园抗议蔓延至多座城市，学生抗议教室拥挤、教师短缺和校舍失修。目前约5000人被捕，其中多数为青少年。
Twenty-three years after President George W. Bush's campaign of "shock and awe," the U.S. military has completed its withdrawal from Iraq.
中文摘要 在时任美国总统小布什发动「震慑」行动23年后，美军已完成从伊拉克撤军。
Brazil's election enters its final day of campaigning as Lula and Flávio Bolsonaro make their last pitches to voters. Brazilians explain what worries them, and why some still can't choose.
中文摘要 巴西大选进入最后一天竞选，卢拉与弗拉维奥·博索纳罗作最后拉票。部分选民表示两人都不想选，并解释其担忧与犹豫原因。
The aircraft lost communication with flight controllers after significantly dropping in altitude, according to flight data.
中文摘要 一架载有6人的医疗飞机在美国马萨诸塞州海岸附近失踪。飞行数据显示，该机在高度大幅下降后与空管失去通信联系。
Israeli settlers attack Palestinian farmers during olive harvest
中文摘要 以色列定居者在橄榄收获季袭击巴勒斯坦农民。
The bridges, used by many commuters in the Ukrainian capital, are the latest target in a broad Russian bombing campaign against infrastructure.
中文摘要 俄罗斯转向打击基辅的桥梁，威胁乌克兰首都的重要交通命脉。这些桥梁为众多通勤者使用，是俄方针对基础设施广泛轰炸行动的最新目标。
The US president published a lawmaker's cell phone number as he called for the passage of a bill making DST permanent.
中文摘要 美国总统特朗普加大施压，要求共和党人推动通过使夏令时永久化的法案，并公开了一名议员的手机号码。
Funeral prayers were held at Gaza City’s Saint Porphyrius Greek Orthodox Church for a Palestinian mother and daughter.
中文摘要 加沙城圣波菲里乌斯希腊东正教教堂为一对在加沙空袭中丧生的巴勒斯坦母女举行葬礼祈祷。
The emperor is a symbol of Bulgarian national identity whose repatriation stoked tensions with Greece.
中文摘要 保加利亚沙皇萨穆伊尔的遗骸在约1000年后回归故土。这位君主是保加利亚民族认同的象征，其遗骸归还引发与希腊的紧张关系。
Youth-led movement demands resignation of chief election commissioner.
A photo went viral of a Palestinian boy hiding from Israeli forces on his way home from class in the occupied West Bank.
President Luiz Inácio Lula da Silva and right-wing challenger Flávio Bolsonaro remain locked in a tight race as Brazil prepares to vote in Sunday’s election, according to a pair of new polls.
中文摘要 巴西将于周日举行大选，两项最新民调显示，现任总统卢拉·伊纳西奥·卢拉·达席尔瓦与右翼挑战者弗拉维奥·博索纳罗支持率仍不相上下，选情胶着。
Borrowing is getting harder for companies.
中文摘要 派拉蒙债务下跌，令企业借款人面临融资环境收紧的困境，借贷难度上升。
The news doesn’t stop when markets close. Hosts David Gura, Christina Ruffini and Lisa Mateo bring clarity, context and a bit of humor to the weekend’s biggest headlines, LIVE from New York. Joined by The Atlantic Staff Writer Vivian Salama, “The Profiler” Online Job Scam Hunter Jay Jones, UMich Pro
中文摘要 彭博周末节目由主持人David Gura、Christina Ruffini和Lisa Mateo在纽约直播，梳理周末头条新闻并提供背景解读，嘉宾包括《大西洋月刊》撰稿人Vivian Salama及网络求职诈骗追踪者。
Flight 1073 was heading to Tel Aviv when the attack was launched in the cockpit
中文摘要 迪拜航空1073航班飞往特拉维夫期间，一名副驾驶在驾驶舱内用斧头实施「恐怖行为」，英国《金融时报》报道称。
Bloomberg Pursuits reporter Sarah Rappaport tells Bloomberg This Weekend that luxury travelers are increasingly seeking tiny hotels with as few as one to six rooms, combining the privacy of a villa with personalized high-end hotel service. Speaking with hosts David Gura and Christina Ruffini, Rappap
中文摘要 彭博Pursuits记者Sarah Rappaport表示，高端旅行者日益青睐仅有1至6间客房的微型酒店，这类酒店兼具别墅的私密性与个性化高端酒店服务。
Pointed offers a strategic twist to the news quiz format, testing not just players’ knowledge of the news but also their confidence in their answers. Join Bloomberg's Lisa Mateo, Christina Ruffini and David Gura as they play and check out the quiz for yourself at Bloomberg.com (Source: Bloomberg)
中文摘要 彭博推出新闻问答节目「Pointed」，在测试玩家新闻知识的同时考验其对答案的信心，由Lisa Mateo、Christina Ruffini和David Gura主持。
People starting to speak like AI chatbots, AI anxiety spilling into therapists' offices, NYC's AI office boom a bust for young job hunters and Tinder rolls out group dating for younger users. Join Lisa Mateo, David Gura and Christina Ruffini for a roundup of headlines you may have missed, but gotta
中文摘要 彭博汇总易被忽略的头条：人们开始像AI聊天机器人一样说话、AI焦虑蔓延至心理治疗室、纽约市AI办公热潮对年轻求职者不利，以及Tinder为年轻用户推出群体约会功能。
Washington Institute research director Dana Stroul tells Bloomberg This Weekend that the US military buildup in the Middle East, including a third aircraft carrier, is expanding President Donald Trump’s options as diplomacy with Iran remains stalled and Washington weighs its approach to the Houthis
中文摘要 华盛顿研究所研究主任Dana Stroul表示，美国在中东的军事集结，包括部署第三艘航空母舰，扩大了特朗普总统的选择空间，因与伊朗的外交仍陷僵局。
Fraud investigator Jay Jones, known online as “The Profiler,” tells Bloomberg This Weekend that AI is helping scammers create increasingly convincing fake recruiters and job listings that can lead victims to surrender money and sensitive personal information. Speaking with hosts David Gura and Chris
中文摘要 欺诈调查员Jay Jones（网名「The Profiler」）表示，AI正帮助诈骗者制造越来越逼真的虚假招聘人员和职位信息，诱使受害者交出钱财和敏感个人信息。
Atlantic Staff Writer Vivian Salama tells Bloomberg This Weekend that Egypt’s then-intelligence chief Abbas Kamel made an unusual 67-minute trip to Israel less than two weeks before the Oct. 7, 2023, attack to warn Israeli officials that intelligence indicated Hamas was mobilizing for a major assaul
中文摘要 《大西洋月刊》撰稿人Vivian Salama称，2023年10月7日袭击发生前不到两周，埃及时任情报负责人Abbas Kamel罕见地赴以色列停留67分钟，警告以方情报显示哈马斯正在动员。
Debate over artificial intelligence has largely focused on whether machines could become too powerful. Harvard philosopher Michael Sandel argues that a more immediate question is whether AI changes humans themselves: weakening authentic connection, independent thought and the ways people understand
中文摘要 哈佛哲学家Michael Sandel认为，关于AI的争论多集中于机器是否过于强大，但更紧迫的问题是AI是否改变了人类本身，削弱真实联系、独立思考及交往方式。
Chinese ecommerce company has been exploiting tax loophole that exempts small parcels from customs duty
中文摘要 中国电商公司Temu英国销售额增长逾一倍，达到1.71亿美元。该公司一直利用小额包裹免关税的税收漏洞。
6 回复 · 程序员 节点
8 回复 · 程序员 节点
5 回复 · 程序员 节点
6 回复 · Apple 节点
16 回复 · 程序员 节点
11 回复 · 程序员 节点
9 回复 · 程序员 节点
29 回复 · Apple 节点
13 回复 · Apple 节点
17 回复 · Apple 节点
该源今日无内容。