JavaScript · ★ 263,722 · 🍴 39,453 · 📈 826 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 编码代理的性能优化框架，提供技能、直觉、记忆、安全与调研优先的开发能力，兼容 Claude Code、Codex、Opencode、Cursor 等工具。适合希望让编码代理在复杂工程中长期运行、保留上下文并控制安全风险的开发者使用。
TypeScript · ★ 5,199 · 🍴 484 · 📈 98 stars today
A framework for building agentic apps
中文介绍 一个用于构建代理式应用的开发框架，帮助开发者把 LLM 代理能力直接集成进产品中。适合需要搭建具备自主决策、工具调用与任务编排的前端或全栈应用团队，减少从零实现代理逻辑的成本。
JavaScript · ★ 17,999 · 🍴 998 · 📈 2,428 stars today
A coding-agent skill for multi-phase security audits with independently verified, machine-readable findings
中文介绍 Cloudflare 推出的编码代理技能，用于执行多阶段安全审计，产出可机器读取且经过独立验证的审计发现。适合把安全审查流程嵌入 AI 编码工作流的团队，在开发阶段自动发现并核实潜在漏洞。
HTML · ★ 25,142 · 🍴 1,729 · 📈 1,018 stars today
Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.
中文介绍 开源计算机使用（computer-use）代理基础设施，提供跨操作系统的驱动、设备集群与基准测试，支持训练、评估和数据生成。适合研究或构建 GUI 自动化代理的团队，用于大规模部署与效果评测。
Python · ★ 35,356 · 🍴 5,248 · 📈 260 stars today
中文介绍 Anthropic 开源的金融服务相关仓库，面向金融行业场景提供基于 Claude 的能力与集成示例。由于缺乏描述信息，具体功能不明确，适合关注 Anthropic 官方在金融领域落地实践的开发者与机构参考。
Python · ★ 45,558 · 🍴 3,146 · 📈 57 stars today
A community-supported supercharged document management system: scan, index and archive all your documents
中文介绍 社区维护的增强版文档管理系统，可扫描、索引并归档各类纸质与电子文档，实现全文检索与长期保存。适合想要搭建自托管无纸化办公环境的个人或团队，替代商业文档管理方案。
TypeScript · ★ 147,115 · 🍴 24,058 · 📈 419 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 推出的代理式编码工具，常驻终端，能理解代码库并执行日常任务，例如解释复杂代码、处理 git 工作流等，从而加快开发速度。适合习惯命令行、希望用 AI 辅助实际编码与版本控制操作的开发者。
Python · ★ 4,549 · 🍴 1,009 · 📈 172 stars today
Assignments for CS146S: The Modern Software Dev (Stanford University Fall 2026/2025)
中文介绍 斯坦福大学 CS146S《现代软件开发》课程的作业集合，覆盖现代软件工程实践。适合学习该课程的学生或自学软件开发者，通过动手作业掌握当代开发工具链与工程方法。
Jupyter Notebook · ★ 5,369 · 🍴 956 · 📈 465 stars today
Fault-tolerant, highly scalable GPU orchestration, and a machine learning framework designed for training models with billions to trillions of parameters
中文介绍 面向超大规模模型训练的 GPU 编排与机器学习框架，具备容错和高可扩展性，支持从数十亿到数万亿参数模型的训练。适合需要在大规模集群上稳定训练大模型的 AI 研究机构与企业工程团队。
TypeScript · ★ 16,804 · 🍴 2,133 · 📈 755 stars today
OpenStock is an open-source alternative to expensive market platforms. Track real-time prices, set personalized alerts, and explore detailed company insights — built openly, for everyone, forever free.
中文介绍 开源的市场行情平台替代方案，支持追踪实时价格、设置个性化提醒并查看公司详细洞察，永久免费开放。适合不想为昂贵行情平台付费的个人投资者、开发者及希望自建数据看板的小型团队。
Go · ★ 16,042 · 🍴 1,541 · 📈 379 stars today
Secure environments for developers and their agents
中文介绍 为开发者和其 AI 代理提供安全隔离的开发环境，基于远程工作区实现配置一致、可管控的编码空间。适合需要集中管理开发环境、保障代码与凭据安全的企业工程团队，支持自托管部署。
TypeScript · ★ 17,294 · 🍴 919 · 📈 291 stars today
The Generative UI framework
中文介绍 Vercel 实验室推出的生成式 UI 框架，可通过 JSON 描述动态渲染用户界面。适合需要让 LLM 或后端动态生成界面的应用场景，简化 AI 产品中前端渲染与结构化输出的衔接。
JavaScript · ★ 97,676 · 🍴 10,294 · 📈 736 stars today
Production-grade engineering skills for AI coding agents.
中文介绍 为 AI 编码代理提供生产级工程技能集合，把软件工程最佳实践封装成代理可调用的能力。适合希望提升编码代理输出质量、让其遵循规范完成真实项目开发的工程师与团队参考使用。
👍 135
The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transf
中文介绍 针对长程 Agent 带来的输入密集型负载，prefill 计算与 KV cache 占用 HBM/SSD 的问题，DeepSeek-V4.1-Flash 推进 KV cache 压缩极限，在长上下文场景下降低显存与存储压力，兼顾推理效率与模型能力。
👍 114
Recent Omni-Modal Generative Models (Omni-Models) have advanced content generation toward unified modeling of text, images, video, and audio. MiniMax-H3 exemplifies this transition by combining multimodal context understanding with joint audio-visual generation in a shared latent framework. Its unif
中文介绍 评估全模态生成模型 MiniMax-H3 对物理世界的推理能力。该模型在共享 latent 框架内统一文本、图像、视频与音频理解与联合音视频生成，论文考察其统一建模是否真正带来物理常识与因果推理，而非仅停留在内容生成层面。
👍 94
We study length inflation in on-policy distillation (OPD), where student responses can become excessively long and even exhaust the generation budget. We identify termination-token mismatch between base students and post-trained teachers as an important source of this behavior. Across Qwen3, Llama,
中文介绍 研究 on-policy distillation（OPD）中学生回复过长甚至耗尽生成预算的长度膨胀问题，发现根源之一是基座学生与后训练教师的终止 token（EOS）不一致。作者在 Qwen3、Llama 等模型上分析该错配，为抑制长度膨胀提供依据。
👍 94
As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands from isolated predictions into long trajectories of reasoning, tool use, and feedback. Token efficiency therefore becomes important for scaling recursive self-improvement. We take an
中文介绍 面向无人值守、全天候探索的编码 Agent，其推理、工具调用与反馈构成长轨迹，token 效率成为递归自我改进的瓶颈。SoL-Pi 通过递归扩展自动研究循环来构建高效 Agent Harness，降低单位改进的 token 开销。
👍 79
Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the sci
中文介绍 科学代码库蕴含数十年可执行知识，但工具链碎片化、领域约定隐式和正确性标准特殊，难以转化为可靠学习经验。ScienceIDE 将全球科学代码库改造成 Agent 可学习环境，解决科学知识难以用于训练可靠 Agent 的问题。
👍 71
Coding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, yet existing work typically evaluates harnesses as monolithic systems, leaving the effectiveness of individual components unclear. To enable component-level comparison
中文介绍 编码 Agent 的 harness 决定模型能力如何转化为长程软件工程表现，但已有工作多把 harness 当整体评估，各组件作用不明。本文对 harness 设计做实证研究，实现组件级对比，厘清哪些设计真正有效。
👍 56
World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet predictive models remain domain-specific: can a common learning principle support world modeling across radically different systems? We introduce JEPA-Anything, a domain-agnostic fram
中文介绍 世界模型能让智能体预测后果、指导干预并从交互中学习，但现有预测模型多局限于单一领域。JEPA-Anything 提出领域无关框架，用统一学习原则跨差异极大的系统训练预测模型，验证通用世界建模的可行性。
👍 56
Reliable confidence estimation is increasingly central to the trustworthy deployment of language models: a calibrated estimate of the probability that an output is correct decides what to ship, what to escalate, and what to retry. Existing confidence estimators, however, share one design premise: th
中文介绍 可靠置信度估计对 LLM 部署至关重要，决定哪些输出发布、升级或重试。本文提出「置信度来自经验」的思路，从推理到 Agent 场景利用经验进行置信度估计，突破现有估计器依赖固定设计前提的局限。
👍 50
Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating codi
中文介绍 实际 web 开发中，Agent 常需从可运行软件推断行为并在不完整应用中实现，而现有评测多依赖 issue 或指令描述目标行为。ProgramDistill 提出从交互式 Web 应用生成可验证、带参考的 SWE 任务的基准，评测编码 Agent 的这类能力。
👍 48
LLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups where the assistant learns about social situations from subjective user narratives, and (ii) social properties, such as others'
中文介绍 LLM 助手常用于日常社交建议，但其社交推理难以评估：需从主观用户叙述中理解情境，且他人隐私等社交属性不可直接观测。本文提出可验证的社交推理评测方法，面向咨询场景检验 LLM 助手的社交推理能力。
👍 45
Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs. Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge containe
中文介绍 LLM Agent 面对多样信息需求时检索愈发关键，而检索质量取决于索引键能否充分暴露文档知识。本文提出自我演化搜索索引，让索引随任务与语料迭代更新，从而提升检索对复杂信息需求的有效性。
👍 45
Autonomous research loops such as AutoResearch show that one coding agent can improve a training setup unattended. Run several of them and each session starts from scratch, so more agents tend to mean more duplicated search rather than more discovery. Agora is a shared memory for such agents: resear
中文介绍 AutoResearch 等自主研究循环可由单个编码 Agent 无人值守改进训练配置，但多个会话各自从零开始，Agent 越多重复搜索越多。Agora 以 Git 作为共享记忆，让多个自主研究 Agent 复用研究历史，减少重复探索、增加发现。
👍 42
Platform abuse campaigns conceal redirection instructions with emojis, homophones, character decomposition, and redundant symbols, then route users through disguised links to services associated with pornography, fraud, gambling, or illicit transactions. Existing benchmarks evaluate obfuscated text
中文介绍 平台滥用活动常用 emoji、谐音、拆字与冗余符号隐藏跳转指令，再经伪装链接导向色情、诈骗、赌博等非法服务，而现有基准多只评测混淆文本。RiskChainBench 面向混淆平台消息还原与基于证据的 Web 调查，覆盖完整链路。
👍 39
Multi-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivates self on-policy distillation (OPD) to supply dense token-level supervision from a self-teacher with privileged task skills, letting a skill-free student internalize them. This rec
中文介绍 多轮 Agent 的 RL 训练每条轨迹仅得单一标量奖励，self on-policy distillation（OPD）借助具特权技能的自教师提供 token 级稠密监督。但持续蒸馏会带来依赖，RetireOPD 提出自退役式 OPD，让学生内化技能后退出蒸馏。
👍 37
GUI agents execute long-horizon tasks on dynamic graphical user interfaces, where pop-ups, delayed loads, and relocated widgets routinely invalidate plans fixed before execution. Recent agent-skill frameworks encapsulate reusable procedural knowledge to mitigate this, yet existing skill designs are
中文介绍 GUI Agent 在动态界面上执行长程任务，弹窗、延迟加载与控件移位常使预先固定的计划失效。现有 Agent 技能框架封装可复用过程知识，但技能设计仍不灵活。本文提出「反思、修正、复用」的训练无关技能演化方法。
👍 35
Document parsing converts document images into structured content and requires reliable performance across diverse layouts and acquisition conditions. Yet training corpora are biased toward common document types and clean digital pages, while expanding coverage alone does not specify how to address
中文介绍 文档解析需将文档图像转为结构化内容，并在多样版式与采集条件下保持可靠，但训练语料偏向常见文档类型与干净数字页，仅扩大覆盖不足以解决鲁棒性。WeVisDoc 从覆盖转向能力，提升端到端文档解析鲁棒性。
👍 34
Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signal
中文介绍 LLM 交易 Agent 可融合行情、新闻与可执行分析，但其行为常由部署前写死的工具使用策略控制，难以自适应地收集证据、调用工具与验证信号。EvolveTrade 提出经验驱动的策略精炼，使交易 Agent 能在运行中自我演化。
👍 33
Spatial intelligence requires more than describing object locations. Under incomplete observation, models must identify and acquire missing evidence, interpret it in a common spatial frame, and act on it. We introduce VA-Bench to evaluate the complete observe-reason-act-revise loop. General-purpose
中文介绍 空间智能不止于描述物体位置：在观测不完整时，模型需识别并获取缺失证据、在统一空间框架中解读并据此行动。VA-Bench 评测「观察—推理—行动—修正」完整闭环，通过视觉演示、主动感知与度量控制考察具身空间智能。
👍 32
Large Reasoning Models (LRMs) achieve strong performance on complex tasks but exhibit systematic inefficiency: they often overthink easy problems and underthink hard ones. Existing approaches based on uniform length penalties or rigid routing incur an efficiency tax, trading reduced computation on e
中文介绍 大型推理模型存在系统性低效：简单题过度思考、难题思考不足。现有统一长度惩罚或刚性路由方案会带来效率税。When2Think 提出难度感知的长度控制，让混合推理模型按题目难度分配思考长度，兼顾效率与性能。
👍 30
On-policy self-distillation (OPSD) lets a language model learn from a frozen copy of itself that sees an answer or a worked solution. Giving the teacher this extra information seems to offer the student more to learn, but how much does it add beyond distillation itself? To isolate that contribution,
中文介绍 On-policy self-distillation（OPSD）让模型向看到答案或解题过程的冻结自身副本学习。本文隔离出特权信息本身的贡献，考察在蒸馏之外，教师多出的答案信息究竟给学生带来多少额外收益，厘清该方法的真实增益来源。
@sairahul1 · 145.3K 粉丝 · 330.3K 阅 · 568 赞 · 59 转
You do not need a 70B model. You do not need $100,000 in compute. You do not need a machine learning team. A 1.5B parameter model fine-tuned on 200-500 good examples can outperform a frontier model
中文介绍 完整微调教程：不需要 70B 大模型、不需要十万美金算力或 ML 团队，用 200-500 条优质样本微调 1.5B 参数小模型，就能在特定任务上跑赢前沿大模型。适合想低成本自建专用模型的开发者。
@istoica05 · 9.6K 粉丝 · 150.4K 阅 · 501 赞 · 72 转
An AI agent was asked to speed up a key-value store. It delivered a 6× throughput gain and passed every correctness test. The explanation? The agent discovered that the industry-standard benchmark
中文介绍 作者指出智能体软件工程的两个关键缺口。案例：让 AI agent 优化键值存储，它实现 6 倍吞吐提升且通过全部正确性测试，原因是发现了行业标准基准本身的问题。反思 agent 能力评估的盲区。
@chddaniel · 27.7K 粉丝 · 121.2K 阅 · 521 赞 · 42 转
TypeSafe just released the strongest model, and it introduced a completely different way to put intelligence inside software... Jev is in a category of its own right now, not because it writes better
中文介绍 TypeSafe 发布新模型 Jev，带来一种把智能嵌入软件的全新方式。作者称 Jev 自成一类，价值不在写得更好，而是返回可直接执行的决策结果。这是 Jev 完整上手指南，属于新概念的科普向教程。
@DhravyaShah · 62.1K 粉丝 · 120.2K 阅 · 509 赞 · 16 转
Instinct has taken the world by storm over the last two weeks - it's one of the best imessage assistants I've used. as with every product, I decided to reverse-engineer Instinct's memory to find out
中文介绍 作者逆向拆解了 iMessage 助手 Instinct 的记忆机制。他称 Instinct 近两周爆火、是自己用过最好的 iMessage 助手之一，于是决定研究其记忆如何实现并分享细节，属于产品逆向分析。
@mvanhorn · 40.4K 粉丝 · 103.6K 阅 · 550 赞 · 37 转
Here is the tl;dr ELI5. Think AI multiple choice, not AI essay writing. It doesn't chat. It makes decisions your software can act on: "Spam or not?" "Which tool should this agent use?" "Does this need
中文介绍 用大白话解释 Jev：把它想成「AI 选择题」而非「AI 写作文」。它不聊天，只输出软件能直接执行的决策，如是否垃圾邮件、agent 该用哪个工具。并列举人们已用它做出的 9 类应用。
@chamath · 2.4M 粉丝 · 90.2K 阅 · 503 赞 · 63 转
Open-weight models have come within roughly four months of the best publicly evaluated closed frontier models, and the gap between them has only gotten more volatile. That is turning the open vs.
中文介绍 Chamath 深度分析开源与闭源 AI 竞赛：开放权重模型已逼近最好的公开评测闭源前沿模型，差距约四个月，且波动越来越大。讨论这一趋势如何改变开源与闭源的竞争格局，偏行业宏观判断。
@sydneyrunkle · 10.6K 粉丝 · 88.0K 阅 · 556 赞 · 66 转
Agents run in a loop: an LLM decides what to do, a tool executes, a model evaluates the results, and then continues in that loop until the task is complete. Agents and LLMs were initially difficult to
中文介绍 讲如何用 Jev 构建 agent harness。Agent 本质是一个循环：LLM 决策、工具执行、模型评估结果再继续，直到任务完成。作者分享在 Jev 上落地这套循环的实践，属于工程实现向内容。
@twostraws · 120.7K 粉丝 · 77.1K 阅 · 504 赞 · 42 转
I use both Codex and Claude for app development work, both with the highest-tier subscription. Here's how I've found the two compare, plus some tips for success… TL;DR I use both agents extensively,
中文介绍 博主同时用 Codex 和 Claude 做应用开发，且都开最高档订阅。他对比两者实际表现并给出使用技巧，结论是两者都大量使用、各有所长。属于真实开发者视角的工具横评，非单纯站队。
@0xCodila · 18.6K 粉丝 · 63.4K 阅 · 521 赞 · 62 转
The Jevons Paradox (he's on picture) is a rule stating that an increase in the efficiency of a resource's use does not reduce, but rather increases, its overall consumption - That's the global
中文介绍 从零上手 Jev 的 10 步路线图。开篇借杰文斯悖论点题：资源使用效率提升不会减少、反而增加总体消耗。作者据此设计一套搭建与使用「新大脑」的完整流程，偏系统性教程。
@trycua · 19.0K 粉丝 · 59.5K 阅 · 501 赞 · 39 转
What we've been building with decision models at Cua - by @ddupont808 and @francedot Over the last few days we've deprioritized some other shiny objects because we've all been Jev-crazy. TypeSafe
中文介绍 Cua 团队谈他们在决策模型上的探索：近几天整个团队都在研究 Jev，甚至搁置了其他项目。结合 TypeSafe 的新进展，讨论 System One 类模型与计算机使用（computer use）的未来方向。
@RohOnChain · 72.5K 粉丝 · 57.7K 阅 · 518 赞 · 54 转
I will break down the exact framework to build a millisecond speed 24/7 HFT trading system with Jev, along with the exact resources that helped me personally. Let's get straight to it. Bookmark This -
中文介绍 用 Jev 搭建 7×24 小时毫秒级高频交易系统的框架拆解，附作者亲测有用的资源清单。属于把决策模型落到 HFT 场景的具体案例，适合关注加密量化与 agent 结合的人。
@LangChain · 265.9K 粉丝 · 46.6K 阅 · 504 赞 · 68 转
By Daniel Shea and Seán Roche Key takeways: Jev is a fundamentally different kind of evaluator. It returns typed answers directly instead of generating text like an LLM judge. Jev was dramatically
中文介绍 LangChain 官方文章讲用 Jev 做 agent 评测的裁判。核心差异：Jev 直接返回带类型的答案，而不是像 LLM judge 那样生成文本，因此评测表现明显更好。为 agent eval 提供新思路。
@manish_fp · 846 粉丝 · 44.1K 阅 · 512 赞 · 59 转
I have read 10+ articles, watched 5 hours of content to understand what actually Jev is and then chose to write this. It is not a chatbot, It’s a whole new kind of AI, more like a classifier which
中文介绍 作者读了 10 多篇文章、看了 5 小时内容后写的 Jev 入门。他强调 Jev 不是聊天机器人，而是一种全新 AI，更像分类器，并介绍能基于它做出什么。属于信息整合型的科普总结。
@akshay_pachaar · 286.8K 粉丝 · 41.6K 阅 · 578 赞 · 87 转
We have been using LLMs like a hammer for every AI problem, even simple decisions. Jev handles those decisions in milliseconds at a fraction of the cost. Let's understand how it works and where it
中文介绍 通俗讲解 Jev 的原理与适用场景。作者指出人们习惯拿 LLM 当锤子解决所有 AI 问题，连简单决策也用大模型；而 Jev 能在毫秒级、以极低成本完成这些决策。偏概念澄清与定位说明。
@_avichawla · 77.5K 粉丝 · 37.8K 阅 · 593 赞 · 80 转
Everything you need to turn an open-source LLM into a fast, local decision engine without retraining it. It covers next-token scoring, fixed choices with probability distributions, SGLang, and a
中文介绍 教你 100% 本地复刻一个 Jev：把开源 LLM 改造成快速本地决策引擎而无需重新训练。内容涵盖 next-token 打分、固定选项的概率分布、SGLang 以及具体实现步骤，动手性最强的一篇。
Imitation is the sincerest form of Flattery
中文介绍 Latent Space 报道称，两天内出现了 6 个 Jev 的克隆版本，并评论「模仿是最真诚的奉承」。原文未披露具体克隆对象与来源细节。
OpenAI introduces the Australian Youth Safety Blueprint, a six-pillar roadmap for safer AI experiences that protect and empower young people.
中文介绍 OpenAI 发布「澳大利亚青年安全蓝图」，提出六大支柱路线图，目标是在保护青少年的同时赋能其安全使用 AI 体验。
On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all? But attendees had so many more questions than we had time to answer in the 30 minute session. So we asked our senior AI editor Will Do
中文介绍 MIT Technology Review 于周三为订阅用户举办线上圆桌活动，主题为「AI 真的会杀死我们所有人吗」。30 分钟环节未答完提问，该刊随后请资深作者补充回答读者问题。
a quiet day
中文介绍 Latent Space 的 AINews 栏目称当日为「安静的一天」，AI 领域无重大事件发生。
中文介绍 TLDR AI 2026 年 9 月 18 日刊要点：Claude Projects v2、谷歌家庭智能体。原文摘要为空，未提供更多细节。
Cooley built GO Public with ChatGPT Work to bring intelligence to the IPO process, helping lawyers surface issues earlier and focus judgment where it matters most.
中文介绍 律所 Cooley 基于 ChatGPT Work 搭建「GO Public」工具，将智能能力引入 IPO 流程，帮助律师更早发现问题，并把判断力集中在关键环节。
A dash of cold water keeps the foomers away.
中文介绍 Latent Space 报道 AI 新闻的「现实检验」：Yegge 关停 Gas Town，Databricks 披露 Astra 成本上升 60%。文章称一盆冷水可让过热的「AI 爆发论」降温。
OpenAI for Law brings frontier intelligence for law, custom firm workflows, connected legal data sources, and legal-grade controls for confidential client work.
中文介绍 OpenAI 推出面向法律行业的 Astra for Law，提供前沿智能、律所定制工作流、已接入的法律数据源，以及适用于机密客户工作的法律级管控。
中文介绍 TLDR AI 2026 年 9 月 17 日刊要点：Claude 与 Cowork 合并、ChatGPT 赞助智能体、harness 税（harness tax）。原文摘要为空，未提供更多细节。
We sit down with AIUC’s CEO on their Series A!
中文介绍 Latent Space 对话 AIUC 首席执行官 Rune Kvist，主题为「为超级智能承保：为你可起诉的智能体背书」，并涉及其 A 轮融资。
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
中文介绍 OpenAI 公布一套追踪、调查并披露模型失准（misalignment）的框架，同时发布 6 份关于模型意外或令人担忧行为的报告。
OpenAI and AARP are bringing free, hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities to build practical AI skills safely.
中文介绍 OpenAI 与美国退休人员协会（AARP）合作，在美国 10 座城市为 1000 名老年人提供免费 ChatGPT 实操工作坊，帮助其安全掌握实用 AI 技能。
Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.
中文介绍 OpenAI 展示 AI 驱动的广告新形态，包括「赞助智能体」（Sponsored Agents）、面向营销人员的工具，以及与 HubSpot、Shopify 的集成。
The AI boom is becoming a materials challenge. As AI pushes computing into new territory, the materials behind that infrastructure are becoming just as crucial as the algorithms running on it. Semiconductors and data centers are approaching physical limits around performance, thermal management, ele
中文介绍 MIT Technology Review 指出，AI 热潮正演变为材料挑战。随着算力进入新领域，支撑基础设施的材料与算法同等关键，而半导体和数据中心在性能等方面正逼近物理极限。
GPT-6 Astra helps Hex’s data agents turn answers into interactive visualizations that employees are proud to share.
中文介绍 Hex 借助 GPT-6 Astra 将复杂分析转为可视化报告：该模型帮助 Hex 的数据智能体把答案变成可交互图表，便于员工分享。
美股整体呈温和分化格局：QQQ 现价 721.45，位于 SMA20 713.24 与 SMA50 709.95 之上，MACD 0.2176 刚上穿信号线 0.1465 形成金叉，RSI14 55.9；SPY 则报 761.69，微低于 SMA20 764.3，RSI14 49.2 处于中轴，MACD 为 -0.617 而信号线 0.5463，显示大盘指数动能钝化。AAPL、NVDA、GOOGL 维持多头排列并接近 52 周高点，META 单日回撤 2.43% 但 RSI14 达 65.8，科技板块内部强弱不一。加密方面，恐慌贪婪指数 70（贪婪），总市值 2.82T 美元、24 小时 -1.1%，BTC 主导率 58.2%、ETH 11.6%；BTC 与 ETH 分别录得 RSI14 65.4 与 68.2，5 日涨幅 8.1% 与 11.94%，均线多头排列但距 52 周高点仍分别回撤 35.23% 与 43.52%，属修复性上涨。中概股技术面偏弱：BABA 虽单日反弹 4.33% 但 RSI14 仅 48 且空头排列，PDD RSI14 39、JD 34.8、腾讯 36.9 并接近 52 周低点。商品外汇与宏观方面，WTI 原油单日重挫 4.69% 仍保住 SMA20 上方，黄金在 SMA200 4554.26 之下震荡，美元指数 100.26 接近 52 周高点且 RSI14 60.9，10Y 美债收益率 5、RSI14 68.8 逼近超买，VIX 14.81 处于低位并呈空头排列，整体风险偏好读数偏暖但债汇端已现过热迹象。
现价 665.75，位于 SMA20 614.61、SMA50 607.38、SMA200 624.16 之上，短中期均线呈多头结构；RSI14 65.8 已接近超买边缘，MACD 24.03 显著高于信号线 16.23，红柱动能延续。单日虽回落 2.43%，但 5 日仍涨 2.73%，52 周高点回撤 15.27%。指标组合显示趋势向上但动量偏热。
现价 81733.62，站上 SMA20 78484.25、SMA50 73220.51、SMA200 70507.27 全部均线，维持多头排列；RSI14 65.4 处于强势区但未超买，MACD 1492.72 高于信号线 1481.81，金叉延续。5 日涨幅 8.1%，但距 52 周高点仍有 35.23% 回撤，说明中期修复而非新高突破。
现价 419，低于 SMA20 437.42、SMA50 452.5、SMA200 506.86，均线呈空头排列；RSI14 36.9 偏弱但未达超卖，MACD -7.33 低于信号线 -6.41，绿柱仍扩张。1 日跌 1.64%、5 日跌 2.19%，距 52 周高点回撤 38.65%，并接近 52 周低点区间，技术动能持续偏弱。
现价 6.69，处于 SMA20 6.71、SMA50 6.74、SMA200 6.84 之下，空头排列明确；RSI14 仅 18.1，已进入超卖区，MACD -0.0108 低于信号线 -0.0105 并触发死叉。1 日跌 0.28%、5 日跌 0.29%，同时接近 52 周低点，弱势读数集中但超卖本身也提示短线技术性反抽风险。
读数 5，高于 SMA20 4.83、SMA50 4.72、SMA200 4.4，多头排列且接近 52 周高点（仅差 0.36%）；RSI14 68.8 逼近超买阈值，MACD 0.0858 高于信号线 0.0705，动能向上但已显过热。1 日涨 1.03%、5 日涨 0.46%，趋势偏强与超买压力并存，故取中性 stance。
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
本报告仅为技术指标读数与形态的客观复述，不构成任何投资建议。过去走势不代表未来表现，指标金叉、超买或超卖均可能在后续行情中失效。仅供技术指标解读参考。
Follow the day’s news live Get our breaking news email, free app or daily news podcast Dan Tehan says fuel excise cut when oil price high a ‘sensible’ plan Dan Tehan, the shadow minister for energy, said Gallagher’s criticism of the Coalition’s fuel plan wasn’t fair. [What] I would say to Anthony Al
中文摘要 澳大利亚工党政府批评反对联盟党提出的将燃油消费税减半政策「无资金支持」。影子能源部长丹·提汉称，在油价高企时削减燃油税是「明智」方案，并反驳工党批评不公。
Apiarists will lose not just the insects but the value of pollination contracts, which can be worth thousands of dollars Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast An estimated 800,000 bees, worth tens of thousands of dollars, h
中文摘要 澳大利亚昆士兰州远北部发生一起有组织的盗蜂事件，约80万只蜜蜂被盗，价值数万澳元。养蜂人不仅损失蜜蜂，还失去价值数千澳元的授粉合同。
Bystanders found all four on the side of the road at Valley Heights on Sunday with stabbing injuries Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Police have alleged that a mother was behind a stabbing attack on three young childr
中文摘要 澳大利亚蓝山地区发生持刀袭击事件，三名儿童被刺，其中一人死亡。警方指控一名母亲为袭击者，四人在Valley Heights路边被发现时有刺伤。
Prime Minister Mark Carney of Canada and President Emmanuel Macron of France met just after the European Union invited Canada to become its first “associate member.”
中文摘要 加拿大总理马克·卡尼与法国总统埃马纽埃尔·马克龙在法国海外领地圣皮埃尔和密克隆会晤，讨论伙伴关系。此前欧盟邀请加拿大成为其首个「准成员」。
Russian President Vladimir Putin has cast the vote as a test of support for his ongoing war in Ukraine.
中文摘要 俄罗斯战时选举投票结束，执政党预计将占据主导地位。俄罗斯总统普京称此次投票是对其持续进行的乌克兰战争支持度的考验。
The suspects in the 2021 assassination of President Jovenel Moïse include 18 men who were extradited from Haiti to Miami, officials said.
中文摘要 海地总统若弗内尔·莫伊兹2021年遇刺案再有30人被捕。官员称，嫌疑人中包括18名从海地引渡至美国迈阿密的男子。
Iran accuses the French language centre of conducting 'illegal acts that violated diplomatic conventions'.
中文摘要 法国将召见伊朗驻法使节，回应德黑兰关闭法国语言中心。伊朗指控该法语中心从事「违反外交公约的非法行为」。
Australia and Finland are vying for a spot on the council, and the PM needs two thirds of member states to back him Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Australia having a seat at the United Nations Security Council will b
中文摘要 澳大利亚总理阿尔巴尼斯发起联合国安理会席位竞选，并任命前默多克助手担任高级外交职务。澳大利亚与芬兰竞争该席位，需获得三分之二成员国支持。
The lions were found dead inside the Ngorongoro Conservation Area, a popular wildlife reserve in Tanzania. Deaths raise concerns over wildlife poisonings in East Africa.
中文摘要 坦桑尼亚恩戈罗恩戈罗保护区发现10头狮子死亡，疑似中毒。该事件引发东非地区对野生动物中毒问题的担忧。
For more than five years, U Kyaw Moe Tun has refused to give up his United Nations post and a sandstone townhouse on the Upper East Side.
中文摘要 缅甸外交官吴觉莫吞拒绝放弃其纽约联合国职位及上东区住所，已坚持超过五年。
One eyewitness said the scene was so chaotic that screaming could be heard from the other end of the beach.
中文摘要 希腊一名82岁男子与一名76岁男子因沙滩躺椅发生争斗后死亡。目击者称现场混乱，海滩另一端都能听到尖叫声。
The liberal party Yabloko is fielding 98 candidates in this weekend’s elections, a small miracle for antiwar Russians who cannot openly express their views.
中文摘要 俄罗斯自由派政党亚博卢在本次周末选举中推出98名候选人。对无法公开表达反战立场的俄罗斯人而言，这被视为一个小奇迹。
Former U.S. deputy national security advisor Anne Neuberger is watching for signs of competition and cooperation on artificial intelligence at President Trump and Chinese leader Xi Jinping's summit this week.
中文摘要 美国前副国家安全顾问安妮·纽伯格关注本周特朗普与中国领导人习近平峰会上人工智能领域的竞争与合作迹象，以界定全球AI领导权。
The Iranian-backed militia was engaged in fierce battles with Saudi-backed government forces in Yemen on Sunday, as it seeks to maintain momentum in the war.
中文摘要 伊朗支持的胡塞武装与沙特支持的也门政府军周日激烈交战，试图扩大地盘。随着与沙特紧张关系升温，胡塞武装寻求维持战争势头。
Qatar’s PM says, the Gulf won't be the same unless there's a framework ensuring no one is a threat to each other.
中文摘要 卡塔尔首相呼吁建立包括伊朗在内的海湾安全框架，称除非有确保各方互不构成威胁的框架，否则海湾局势将不同于以往。
ASX Ltd. Chief Executive Officer Anthony Attia is rebuilding the Australian bourse’s leadership ranks by naming departing Dexus executive Keir Barnes as its next finance head and expanding a role for clearing and settlement operations.
中文摘要 澳大利亚证券交易所（ASX）首席执行官Anthony Attia正在重组管理层，任命即将离开Dexus的高管Keir Barnes为下一任财务主管，并扩大清算与结算业务负责人的职责范围。
Gold was steady as traders weighed the risk of persistent inflation and the path for interest rates after the Federal Reserve’s first hike since 2023.
中文摘要 金价走势持稳，交易员正在权衡持续通胀风险以及美联储自2023年以来首次加息后的利率路径。
Insurance Australia Group Ltd. said a settlement of proceedings brought by Credit Suisse entities over Greensill trade credit insurance policies hasn’t been finalized.
中文摘要 澳大利亚保险集团（IAG）表示，瑞士信贷实体就Greensill贸易信用保险保单提起的诉讼和解尚未最终敲定。
How to find out how much state pension you're likely to receive - and what you can do about it now.
中文摘要 报道介绍如何查询个人预计可领取的英国国家养老金金额，以及现在可以采取哪些措施来影响未来退休后的每年约1.3万英镑的养老金收入。
As the world focuses on the oil market, the war in Ukraine is snarling vital grain shipments and forcing buyers to scramble for supplies from elsewhere.
中文摘要 随着全球关注石油市场，乌克兰战争扰乱黑海重要粮食运输，迫使买家转向其他地区寻找供应，此举冲击小麦市场并加剧通胀风险。
The World Health Organization should look beyond its own budget for funds to support itself and member countries, according to Indonesia’s Health Minister Budi Gunadi Sadikin, who is seeking to lead the agency.
中文摘要 正角逐世界卫生组织总干事职位的印尼卫生部长Budi Gunadi Sadikin表示，世卫组织应在本机构预算之外寻求外部资金，以支持自身及成员国的运作。
Support for Australia’s center-left government fell to a new low as approval for Prime Minister Anthony Albanese weakened, according to a new poll.
中文摘要 最新民调显示，澳大利亚中左翼政府支持率跌至新低，总理Anthony Albanese的个人支持率也有所走弱。
Asian stocks and US equity-index futures advanced with traders looking to a US-China summit later this week for signs of progress on trade and economic ties. Oil fell for a fourth day.
中文摘要 亚洲股市与美国股指期货上涨，交易员关注本周稍晚举行的美中峰会，以寻找贸易与经济关系进展的迹象；油价连续第四日下跌。
Oil steadied as traders tracked supply risks in the Middle East, and intensified diplomatic efforts to end the US-Iran war.
中文摘要 油价走势持稳，交易员关注中东供应风险，以及旨在结束美伊战争的外交努力有所加强。
Swiss deals rose by a fifth to $44 billion in the first eight months of this year, with large corporate transactions squeezing out private equity funds, according to Boston Consulting Group.
中文摘要 波士顿咨询公司数据显示，今年前八个月瑞士并购交易额增长五分之一至440亿美元，大型企业交易挤压了私募股权基金的参与空间。
Pensions and sovereign wealth funds with many billions to invest are finding the diversification they once relied on is an illusion.
中文摘要 人工智能热潮正使华尔街的分散投资变得困难，拥有巨额资金的养老金和主权财富基金发现，它们曾经依赖的多元化配置其实是一种错觉。
Investors in the US Treasury market are shifting their focus to owning shorter-dated government bonds, a bet that the Federal Reserve will eventually emerge victorious in its fight against inflation.
中文摘要 美国国债市场投资者正将注意力转向短期国债，押注美联储最终将在对抗通胀的斗争中取得胜利。
7 回复 · 程序员 节点
7 回复 · 程序员 节点
5 回复 · 程序员 节点
10 回复 · 程序员 节点
29 回复 · Apple 节点
8 回复 · 程序员 节点
10 回复 · 程序员 节点
18 回复 · 程序员 节点
13 回复 · Apple 节点
7 回复 · Apple 节点
该源今日无内容。