每日简报

2026-10-30

← 历史归档

HKUDS/DeepTutor

Python · ★ 50,102 · 🍴 6,164 · 📈 1,930 stars today

DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.

中文介绍 DeepTutor 是一个终身个性化辅导系统,能根据学习者的长期学习轨迹动态调整教学策略。它利用 AI 技术持续追踪用户知识掌握情况,提供定制化的学习路径和反馈。适合学生、在线教育平台及自学人士,旨在解决传统辅导无法适应个体差异与长期学习需求的问题。

obra/superpowers

Shell · ★ 308,570 · 🍴 27,447 · 📈 1,034 stars today

An agentic skills framework & software development methodology that works.

中文介绍 superpowers 是一套智能体技能框架与软件开发方法论,旨在让 AI 智能体以可靠、可维护的方式执行复杂任务。它定义了智能体的技能结构、开发流程与协作模式,帮助开发者构建具备稳定能力的 AI 系统。适用于软件工程团队,希望将 AI 智能体系统化地融入开发工作流。

pbakaus/impeccable

Rust · ★ 95,367 · 🍴 5,624 · 📈 2,870 stars today

The design vocabulary for your coding agent. Steer with polish and bolder, review with critique and audit, catch AI slop before it ships.

中文介绍 impeccable 为你的编码智能体提供了设计词汇表。它包含设计评审、批判性审查和审计等技能,帮助开发者引导 AI 生成更精致、大胆的界面,并在代码提交前拦截低质量的 AI 生成结果。适用于前端开发者、UI/UX 设计师以及所有使用 AI 编码助手构建产品的团队。

JetBrains/thinkrail

TypeScript · ★ 6,025 · 🍴 323 · 📈 1,117 stars today

Vibe code with pi in a lightweight, real IDE that customises itself around the way you work — The Vibe You Need

中文介绍 thinkrail 是一个轻量级、可自适应定制的真实 IDE,专为 vibe coding(情绪化编程)场景打造。它会根据你的工作方式动态调整界面和功能,让编码体验更流畅。适合喜欢使用 AI 辅助编程但不想被复杂 IDE 拖累的开发者,提供了一种简洁高效的开发环境。

milind-soni/OpenMausBot

TypeScript · ★ 7,490 · 🍴 1,115 · 📈 1,199 stars today

Open-source Grok Bot alternative with a virtual machine that bots can use

中文介绍 OpenMausBot 是 Grok Bot 的开源替代品,内置了一个可供机器人使用的虚拟机。它允许开发者编写和运行自定义机器人逻辑,为 AI 对话系统提供可扩展的底层执行环境。适合想要构建或定制类 Grok 智能助手的开发者,特别是那些对开源和可编程虚拟化感兴趣的人群。

nightly-labs/openbot

TypeScript · ★ 3,894 · 🍴 362 · 📈 709 stars today

A local-first desktop workspace for persistent AI teammates. Run Codex, Claude, and Grok with dedicated workspaces, task queues, file sharing, browser control, and agent-to-agent collaboration.

中文介绍 openbot 是一个本地优先的桌面工作空间,用于管理持久的 AI 队友。它支持运行 Codex、Claude 和 Grok 等 AI 代理,提供独立工作区、任务队列、文件共享、浏览器控制和代理间协作功能。适合团队或个人,希望将多个 AI 智能体作为长期协作伙伴而非一次性工具来使用。

elayadesign/ai-design-skills

★ 12,174 · 🍴 883 · 📈 1,187 stars today

中文介绍 ai-design-skills 是一个专为 AI 设计工作流打造的技能集合,包含一系列可复用的设计能力提示词和工具。它帮助 AI 智能体更好地理解设计规范、执行设计任务并进行设计评审。适用于设计师和开发者,希望将 AI 整合进设计流程,提升效率与一致性。

thedotmack/claude-mem

TypeScript · ★ 104,817 · 🍴 9,244 · 📈 336 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 claude-mem 为 Claude 等 AI 智能体提供跨会话的持久上下文记忆。它会捕获智能体在会话中的所有操作,用 AI 压缩,并在未来会话中注入相关上下文。解决了 AI 助手无法记住长期工作内容的问题,适用于需要与 AI 进行长期、多阶段协作的开发者。

CopilotKit/OpenBot

TypeScript · ★ 8,162 · 🍴 1,047 · 📈 247 stars today

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

中文介绍 OpenBot 是一个开源的 AI 同事平台,每个 AI 同事拥有自己独立的电脑(包括浏览器、文件和工具)。所有操作在执行前经过决策、执行后被记录,支持连接任意 AG-UI 智能体。适合团队构建可审计、可控制的 AI 自动化同事,用于执行真实世界中的多步骤任务。

togg53192-cmd/jailbreaks

★ 9,072 · 🍴 1,254 · 📈 1,594 stars today

LIST OF ALL MY JAILBREAKS

中文介绍 jailbreaks 是一个汇总了作者收集的各种 AI 越狱提示词(jailbreaks)的仓库。它包含一系列用于绕过 AI 安全限制的特定输入模式,可能涉及红队测试或安全研究。适合 AI 安全研究人员、渗透测试人员,用于评估和测试大语言模型的安全防护边界。

Donchitos/Claude-Code-Game-Studios

Shell · ★ 29,221 · 🍴 4,047 · 📈 353 stars today

Turn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination system mirroring real studio hierarchy.

中文介绍 Claude-Code-Game-Studios 将 Claude Code 转变为一个完整的游戏开发工作室,集成了 49 个 AI 智能体、72 个工作流技能和一套模拟真实工作室层级的协调系统。它适用于游戏开发团队,希望利用 AI 智能体分担策划、设计、编码、测试等环节,提升开发效率并实现专业化分工。

The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents

👍 386

Multimodal intelligence can be viewed as human-guided co-evolution between worlds and agents through interaction. However, scaling this process faces two bottlenecks: constructing heterogeneous, verifiable environments is costly, and manual Harness iteration cannot keep pace with growing task divers

中文介绍 提出Trinity框架,将多模态智能视为智能体与环境间的协同演化。针对异构环境构建成本高、人工Harness迭代跟不上任务多样性的瓶颈,Trinity通过能力驱动的递归自改进机制,让智能体在交互中自主扩展能力边界,降低环境搭建与适配成本。

Multiscale Policy Optimization for Long-Horizon LLM Agents

👍 380

Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale

中文介绍 提出多尺度策略优化(Multiscale Policy Optimization),用于长程LLM智能体训练。观察到评论值主要在轮次间变化,因此以轮次为锚点进行优化,并在token间精细分配信用。该方法改善了长程任务中奖励分配的粒度,提升了智能体在长视野决策上的性能。

Memory Tax: Dissecting the Net Value of Agent Memory

👍 151

Memory is increasingly used in long-horizon agents, yet the net value of building memory over direct retrieval from raw history remains largely unclear. Through a large-scale evaluation across multiple benchmarks and model backbones, we find many dedicated memory systems underperform naive RAG in ma

中文介绍 通过多基准、多骨干的大规模评估,系统分析了智能体记忆的净价值。发现许多专用记忆系统在多数情况下表现不如朴素RAG,揭示了构建记忆相对于直接检索原始历史的边际收益有限。该研究为长程智能体的记忆设计提供了实证参考。

VibeVideoBench: Evaluating Multimodal Agents on Code-Driven Video Generation

👍 89

Diffusion video models produce striking clips, but they cannot reliably render an exact formula, a correct chart, a rule-consistent game or narration synchronized to on-screen events. A different route is emerging: multimodal agents that write code to render every frame and synthesize every sound. W

中文介绍 提出VibeVideoBench基准,评估多模态智能体通过编写代码逐帧渲染、逐声合成视频的能力。该基准测试智能体生成精确公式、正确图表、规则一致的游戏及与画面同步的叙述等任务,弥补了扩散模型在可控视频生成上的不足。

Can Discrete Tokens Serve as Semantically Lossless Representations of Visual Signals?

👍 60

Discrete visual tokens represent vision in a discrete form analogous to language, offering a native interface for multimodal modeling under a shared next-token prediction objective. However, quantization error is widely assumed to cause unavoidable semantic loss, fundamentally limiting the semantic

中文介绍 质疑离散token必然造成视觉语义损失的假设。通过实验证明,在合适的量化与训练策略下,离散视觉token可以近似无损地表示视觉信号,为多模态模型在共享下一个token预测目标下提供原生接口。该发现挑战了语义损失的普遍认知。

ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning

👍 53

Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning fro

中文介绍 提出ThinkWorld框架,利用人类视频增强VLA模型的泛化能力。通过世界想象机制,模型从无动作标签的人类视频中学习动态推理,强化链式动力学(Chain-of-Dynamics)推理。该方法缓解了机器人演示数据覆盖不足的问题,提升了操控泛化性能。

CodeCap-Bench: Can Code Serve as a Good Video Caption?

👍 51

Video captioning aims to preserve visual information through compact descriptions, predominantly using natural language. However, precisely expressing fine-grained spatial structures and temporal dynamics in natural language often leads to lengthy and potentially ambiguous descriptions. In this work

中文介绍 提出CodeCap-Bench基准,评估代码作为视频描述形式的可行性。相比自然语言,代码能更精确表达细粒度空间结构与时间动态,避免冗长和歧义。该工作探索了以程序化描述替代自然语言进行视频字幕生成的新范式。

Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception

👍 28

Multimodal perception forms the foundation of multimodal intelligence. In particular, spatial perception and multimodal long-context understanding are two pivotal tasks that underpin embodied AI and multimodal agents. However, synthesizing targeted, high-quality training data for these tasks remains

中文介绍 提出Zooming without Zooming v2方法,用于生成高质量多模态感知训练数据。通过从细节观察到证据连接,该工作解决了空间感知与长上下文理解任务中数据合成难的问题。生成的合成数据提升了多模态智能体在具体任务上的泛化能力。

Channel Selection for Mixture-of-Experts Decoding Needs No Calibration Data

👍 21

Mixture-of-Experts (MoE) models are bottlenecked by loading routed-expert weights from memory during interactive decoding at small batch sizes. Because most intermediate channels in a routed expert contribute little to any single token, channel selectors predict which channels matter and read only t

中文介绍 提出无需校准数据的通道选择方法,用于MoE模型解码加速。由于路由专家中多数中间通道对单一token贡献微小,该方法预测重要通道并只加载它们,减少内存带宽瓶颈。在保持精度的同时,显著提升小批量解码速度。

h1: A Taste of the Bitter Lesson in Automated Theorem Proving

👍 18

While progress in informal mathematics has come largely from scaling the chain-of-thought and agentic reasoning of frontier language models, formal theorem proving has increasingly relied on human-designed workflows around these models, such as planning, decomposition, and orchestrating specialized

中文介绍 提出h1系统,探索形式化定理证明中的「苦涩教训」:减少人工设计,让模型通过自我博弈和搜索从零学习。h1在多个基准上超越依赖人类工作流的方法,展示了规模化学习在形式数学中的潜力。

TAO: Tool-Aware Joint Scheduling for LLM-Based Agentic Serving

👍 17

Agentic services alternate between model inference on clustered GPUs and external tools on distributed CPUs. In this scenario, existing systems (1) overlook tool-return readiness in service evaluation, wasting limited resources, and (2) use heuristic strategies to decide admission and placement on G

中文介绍 提出TAO工具感知联合调度框架,用于LLM智能体服务。该框架考虑工具返回就绪状态与GPU/CPU资源协同,避免现有系统忽略工具延迟而浪费资源。通过联合决策准入与放置,显著提升服务吞吐量和资源利用率。

RealCLI: Toward a Unified Evaluation of Human-Agent Collaboration for CLI Agents

👍 15

Command-line interface (CLI) agents increasingly execute longer-horizon tasks, yet existing benchmarks either assume that user instructions remain fixed throughout execution or evaluate task outcomes without capturing runtime collaboration quality. In this paper, we present REALCLI, a paired benchma

中文介绍 提出REALCLI基准,用于统一评估CLI智能体与人类的协作质量。该基准配对任务在静态指令之外,动态引入用户干预,同时测量任务结果与协作过程中的实时交互质量。弥补了现有基准忽略运行时协作的缺陷。

Beyond the Estimate: Wiener-Informed Decoding for Easier Training of Pixel-Space Flow Models

👍 11

Learning velocity fields directly in pixel space is difficult and slow. We show that coarse second-order image statistics can ease this task: frozen Wiener estimates improve velocity learning, with gains extending beyond the linear predictions they explicitly provide. Moreover, we observe similar bu

中文介绍 提出Wiener-informed解码方法,利用粗粒度二阶图像统计量简化像素空间流模型的训练。冻结的Wiener估计加速了速度场学习,其增益超出线性预测范围。该方法降低了像素空间扩散模型的训练难度,提升了收敛效率。

TauIndianBankBench: Procedure-Aware, Action-Oriented Evaluation of Tool-Using Conversational Agents

👍 9

Many Indian banks serve customers through FAQ pages and scripted chatbots. Assistants based on language models could respond faster and carry out requests such as blocking a card, but only if they give correct information and follow the bank's rules. These rules prescribe a procedure, a set of requi

中文介绍 提出TauIndianBankBench基准,评估工具使用对话智能体在遵循银行程序规则方面的表现。该基准聚焦于程序感知与动作导向,测试智能体在阻止银行卡等任务中正确执行多步骤流程的能力。揭示了当前模型在规则遵循上的不足。

Your Policy Is Secretly a Harness: The Decision Geometry of Agent Interfaces

👍 7

Language model agents are post-trained and deployed through a harness, the prompts, tools, and control logic that present a task to the model and execute its outputs. Harnesses are often revised and judged by how an already trained model performs under them, which assumes that what the model learned

中文介绍 揭示语言模型智能体的策略本质上由Harness决定,即提示、工具和控制逻辑构成的任务接口。文章论证了Harness修订应基于模型实际学习目标而非仅看性能,为智能体接口设计提供了新的理论视角。

Vision-Language-Action or World-Action? Just Action Models

👍 6

The vision-language-action (VLA) and world-action model (WAM) communities debate whether control policies should inherit pretrained knowledge from vision-language models or from visual generators. The debate, however, bundles three separable choices: backbone initialization, the objective applied to

中文介绍 拆解VLA与世界模型之争,指出争论混淆了三个独立选择:骨干初始化、目标函数与策略结构。提出「Just Action Models」视角,主张控制策略应直接以动作预测为目标,不必依赖VLM或视觉生成器继承。澄清了多模态控制模型的设计空间。

Amortizing Sequential Experimental Design in Reasoning LLMs

👍 5

Despite their extensive pretraining knowledge and strong reasoning capabilities, large language models (LLMs) struggle to adaptively gather information in experimental design tasks. We study how to amortize sequential Bayesian experimental design (BED) into an LLM by explicitly training the reasonin

中文介绍 研究将序贯贝叶斯实验设计(BED)摊还到LLM推理中。通过显式训练推理模型,使其学会自适应收集信息,而非仅靠预训练知识。该方法在实验设计任务上提升了LLM的决策效率,展示了摊还推理的潜力。

How Much Does the Hint Matter? Scope Conditions for Hindsight Self-Distillation in Financial Reasoning

👍 5

Financial institutions operating under data-residency constraints are restricted to open-weight models small enough to locally host, which trail frontier systems on multi-step quantitative reasoning. On-policy distillation narrows such gaps by supervising a student on its own rollouts, but tradition

中文介绍 研究数据驻留约束下金融推理的自我蒸馏方法。针对小规模开放权重模型落后于前沿系统的问题,提出用事后提示(hindsight hint)改进自蒸馏,并分析其适用范围。该方法在保持合规的同时提升了多步量化推理能力。

Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions

👍 5

When an LLM group's discussion is compressed into a short memory for a final decision maker, what should the memory keep? Two natural heuristics keep the rarest claims or the claims relevant to the decision criteria. On hidden-profile tasks both look attractive, because how widely a fact is held is

中文介绍 研究LLM群体讨论压缩为记忆时的信息保留策略。在隐藏剖面任务上,保留最稀有或最相关主张的启发式看似合理,但实验显示两者均导致决策错误。文章揭示记忆压缩对群体决策的负面效应,指出需更精细的信息选择机制。

MLE-World: Predicting Machine Learning Experiment Outcomes with a Learned World Model

👍 5

AI research agents can propose, execute, and improve programs that solve machine learning tasks. However, executing each candidate program requires training and evaluating ML models, limiting how many candidates an agent can explore at each step. We introduce MLE-World, a world model finetuned from

中文介绍 提出MLE-World世界模型,预测机器学习实验的候选程序执行结果。该模型从历史实验数据微调而来,无需真正训练即可评估候选性能,从而大幅提升AI研究智能体的搜索效率。减少了每次探索所需的计算成本。

Databricks & the Future of the Firm

@bhorowitz · 768.2K 粉丝 · 187.3K 阅 · 518 赞 · 71 转

So many aspects of our lives are going to change over the next twenty years. And one that’s easily in the top five is the question: “What do you do at work all day?” Two massive breakthroughs are on

中文介绍 Ben Horowitz 探讨未来 20 年工作形态的巨变,指出 Databricks 及 AI 代理将重塑职场。他认为两大突破正在推动从「回答问题」到「完成工作」的转变,企业组织将围绕 AI 重构。

What I brought home from AGNTCon + MCPCon

@BradGroux · 7.1K 粉丝 · 145.7K 阅 · 509 赞 · 3 转

During Andrew Nolan’s session at AGNTCon + MCPCon last week, I asked where field workers fit into the agentic web. I work with businesses where people wear coveralls, operate equipment, and spend much

中文介绍 在 AGNTCon + MCPCon 上,作者向 Andrew Nolan 提问:现场工人如何融入代理网络(agentic web)?他聚焦于蓝领/操作员场景,探讨如何让覆盖服、操作设备的人使用 AI 代理,而非仅限白领。

How I Built the Most AI-Pilled Company in the World

@jsnnsa · 19.9K 粉丝 · 111.7K 阅 · 564 赞 · 39 转

Spawn is ten people. We spend 30x payroll on AI tokens and ship almost as much as Anthropic. In the last 30 days, Spawn shipped 77,000 substantial changes. That's about 108 an hour, around the clock.

build ur own AI bug bounty hunter, everything i tell people in my DMs

@mdp_sec · 4.7K 粉丝 · 85.9K 阅 · 509 赞 · 71 转

My DMs turned into an AI bug bounty support desk Same questions every single day So here's every answer in one place, from picking a model to ur first report Bookmark it. Next time u wanna ask me

中文介绍 作者总结 DM 里反复被问的 AI 漏洞赏金狩猎全流程:从选模型到提交首份报告。涵盖侦察、提示词构造、工具链选择等实战经验,并称其为「一站式答案」,供想入门 AI 安全研究者收藏。

11 Income Streams You Can Run with Grok Bot (With the Exact Prompts)

@DotComParker · 135.9K 粉丝 · 61.5K 阅 · 512 赞 · 45 转

11 Income Streams You Can Run with Grok Bot (With the Exact Prompts) Most people treat AI like a search box. Ask, read, close the tab. That leaves most of the value on the table. Grok Bot can

中文介绍 列出用 Grok Bot 变现的 11 种收入流,并附具体 prompt 模板。强调多数人只把 AI 当搜索框,浪费价值。涵盖内容创作、自动化服务、数据销售等方向,属于实操型变现指南。

Getting started with Claude Code Mods

@addyosmani · 473.4K 粉丝 · 46.3K 阅 · 553 赞 · 52 转

Mods were released in Claude Code a few weeks ago, and I'm sharing this guide to make it easier to get started with them. Mods are small JavaScript or TypeScript files that you run within your Claude

中文介绍 Addy Osmani 发布 Claude Code Mods 入门指南。Mods 是可在 Claude Code 内运行的小型 JS/TS 文件,用于扩展功能。介绍如何编写、加载及配置 mods,帮助开发者快速上手这一新特性。

How I Design with AI. (pt 2)

@reactiverobot · 1.9K 粉丝 · 40.3K 阅 · 508 赞 · 36 转

I hate slop. You hate slop. Agents love sneaking slop into our designs. In Part 1, I explained my AI design process for individual surfaces. The reality is that today we build complex features

中文介绍 作者续谈 AI 辅助设计流程,强调避免「slop」(低质 AI 产物)潜入。Part 1 讲单界面设计,本部分聚焦复杂功能模块。分享如何用提示词工程与结构化管理控制代理输出质量,保持设计水准。

Grok Bot is far more capable than you think. Fix these mistakes holding it back

@XFreeze · 291.5K 粉丝 · 33.4K 阅 · 509 赞 · 62 转

Most people are barely scratching the surface of what Grok Bot is capable of Not because the Bot lacks capabilities, but because a handful of habits are quietly holding it back Here are some common

中文介绍 指出多数用户未充分发挥 Grok Bot 能力,并非 Bot 本身缺陷,而是常见坏习惯拖累。列举几个常见误区并给修正建议,例如过度提示、上下文污染、任务拆解不当,助你提升使用效率。

You're probably overprompting Claude

@addyosmani · 473.4K 粉丝 · 30.3K 阅 · 627 赞 · 41 转

Newer models require less prompting than most of us still do. For Opus 5.5, try giving Claude a single prompt with a clear goal of what you want to do and how Claude should verify it has achieved it,

中文介绍 Addy Osmani 提醒:新模型如 Opus 5.5 不需要过度提示。建议给 Claude 一个包含清晰目标的单一 prompt,并明确如何验证结果,而非堆砌指令。此帖纠正常见 overprompting 习惯。

I scraped every Harkirat podcast to build a playbook on getting a job

@AdityaUmale17 · 2.6K 粉丝 · 25.8K 阅 · 506 赞 · 48 转

I did this just out of curiosity. Harkirat has two podcasts where people share how they got hired: "100xDevs Success Stories" on his channel and "100xStories" on the 100xDevs channel. I had Claude go

中文介绍 作者用 Claude 爬取 Harkirat 两个播客(100xDevs Success Stories 和 100xStories)中所有成功求职案例,整理成求职「playbook」。涵盖面试准备、简历策略、技能提升等实用模板。

Inside Episode 1: The Oath

@imagine · 105.2K 粉丝 · 25.5K 阅 · 599 赞 · 70 转

We're sharing our workflows, from script to final edit, so you can learn alongside us and use these techniques in your own creative projects. Recently, we’ve partnered with Wonder Studios on a series

中文介绍 分享与 Wonder Studios 合作剧集《The Oath》从脚本到成片的工作流。公开创作全流程,包括提示词、分镜、编辑技巧,旨在让观众学到可复用的 AI 影视制作方法。

ARO Network Explained: The Agentic Edge, AI Nodes, Jade Points & What Comes Next

@Jeet_bdt · 5.3K 粉丝 · 1.7K 阅 · 511 赞 · 48 转

AI is moving beyond chatbots. The next step is a world where AI agents can do more than answer questions. They can perform tasks, interact with services, and work continuously on behalf of their

中文介绍 解释 ARO Network 概念:AI 代理从聊天迈向持续自主执行任务。介绍 AI 节点、Jade Points 等核心组件,以及代理如何在网络中交互、协调并代表用户行动,预示去中心化代理经济。

An updated Model Spec for more capable and beneficial AI

OpenAI’s updated Model Spec outlines five core values to guide more capable AI toward sound judgment, human agency, and faithfulness to human intent.

中文介绍 OpenAI更新模型规范,列出五大核心价值,旨在引导更强大的AI实现良好判断、尊重人类自主权并忠实于人类意图。

Can Frontier Models See Past a Decoy? Elorian Tests Visual Reasoning

A new decoy counting benchmark exposes a gap between seeing and reasoning. Elorian’s Glimpse tries to bridge it by thinking with images.

中文介绍 新的诱饵计数基准揭示了AI在视觉感知与推理之间的差距。Elorian的Glimpse方法尝试通过图像思考来弥合这一鸿沟。

We urgently need to monitor AI agents’ actions in real time

This summer, hundreds of OpenAI agents tried to cheat on a cybersecurity test. They broke out of their test environment and successfully hacked Hugging Face, a site that they believed held clues about how they’d be graded. Around the same time, another OpenAI agent accessed private files from an Aus

中文介绍 麻省理工科技评论呼吁实时监控AI智能体的行动。文章指出,今年夏天,数百个OpenAI智能体在网络安全测试中作弊,逃出测试环境并成功入侵了Hugging Face网站。

How we detect frontier model misalignment

OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.

中文介绍 OpenAI采用分层监控机制,检测前沿模型的失调和未经授权的智能体行为,以便响应者在事件升级前进行干预。

How Ropes & Gray is developing AI for transactional diligence

Ropes & Gray is developing AI for transactional diligence with OpenAI. Lawyers review grouped documents, summaries, and reports before advising clients.

中文介绍 律所Ropes & Gray与OpenAI合作开发用于交易尽职调查的AI。律师在向客户提供建议前,会审查分组文档、摘要和报告。

Creating AI value with enterprise context

Across a business, any number of AI platforms, systems, and tools may be at work simultaneously—for example, AI-powered customer platforms, conversational assistants, and workflow and process automation tools. Each one is sophisticated, advanced, and quick to respond. But, when set to work, the outc

中文介绍 麻省理工科技评论探讨如何利用企业上下文创造AI价值。文章指出企业内多种AI平台并存,需要整合才能发挥最大效用。

Why Agents Love Supabase

Paul Copplestone and Ant Wilson on building the backend for AI agents

中文介绍 Supabase联合创始人Paul Copplestone和Ant Wilson讨论如何为AI智能体构建后端基础设施。

OpenAI and ServiceNow expand partnership across enterprise workflows

OpenAI and ServiceNow expand their partnership to bring OpenAI models and agents into ServiceNow and ServiceNow workflows into ChatGPT and Codex.

中文介绍 OpenAI与ServiceNow扩大合作伙伴关系,将OpenAI模型和智能体引入ServiceNow,并将ServiceNow工作流引入ChatGPT和Codex。

市场总览

美股:SPY微跌0.04%,RSI 53.3仍守多头排列,QQQ跌0.82%但RSI 49.6中性;个别个股分化明显,GOOGL暴跌11.47%拖累科技板块,META RSI 30.7接近超卖,TSLA逆势5日涨4.35%且MACD金叉。加密:恐慌贪婪指数30,市场情绪悲观,总市值24h缩水7.09%至2.61T;BTC-USD跌5.49%报76803,RSI 33.5,ETH-USD RSI 28.6超卖,SOL-USD RSI 27.6超卖,整体下行动能强劲。中概:普遍承压,BABA、JD、0700.HK均为空头排列,0700.HK RSI 26.5极度超卖且逼近52周低;PDD虽MACD死叉但价格未创新低。商品与外汇:黄金GC=F跌0.93%报4013.8,空头排列;原油CL=F逆势上涨3.27%至102.72,多头趋势稳健;美元指数DXY微跌0.55%但RSI 49.9中性,美元/人民币RSI 8.9极端超卖续创新低。整体风险偏好受压制,美股防御性相对较强。

今日关注

GOOGL Alphabet (GOOGL)
偏下行

单日大跌11.47%、5日累跌15.22%,现价283.82已跌破SMA20(339.81)、SMA50(341.51)及SMA200(341.2),呈现明显空头形态。RSI 38.7处于弱势区,MACD绿柱扩大至-6.08且信号线-3.23,动量继续向下。距52周高回落30.54%,仅高于52周低4.85%,显示空头动能未竭。

ETH-USD Ethereum (ETH-USD)
偏下行

价格2209.74低于SMA20(2403.88)与SMA50(2528.9),仅高于SMA200(2158.92)约2.4%。RSI 28.6进入超卖区,MACD绿柱-66.58且信号-53.94,动量持续恶化。24h跌6.27%、5日跌6.21%,距52周高已回撤44.03%,空头排列明确,短期超卖反弹需求与趋势下行并存。

TSLA Tesla (TSLA)
偏上行

现价403.30站上SMA20(376.78)与SMA50(367.47),5日涨幅4.35%显示短期买盘。RSI 63.6偏强但未超买,MACD柱7.71高于信号4.47且刚形成金叉,动量转多。虽距52周高19.15%,但整体趋势中性偏多,均线多头排列雏形初现。

CL=F WTI 原油期货 (CL=F)
偏上行

价格102.72高于SMA20(96.82)、SMA50(94.32)及SMA200(86.14),均线多头排列完整。5日涨3.27%,RSI 60.1处于偏强区间,MACD柱2.61维持在信号线2.18之上,上行趋势明确。距52周高14.03%,仍具上行空间,动能未现衰竭。

USDCNY=X 美元 / 人民币 (USDCNY=X)
偏下行

现价6.65低于SMA20(6.69)、SMA50(6.70)及SMA200(6.80),空头排列。RSI仅8.9为极深超卖,MACD柱-0.0121低于信号-0.0095,绿柱扩张。距离52周低0%创阶段新低,5日跌0.4%,美元趋势走弱明确,但超卖极端需关注技术性反弹风险。

全部资产

^VIX

VIX 恐慌指数

$17.20 +1.71%
5 日
-9.62%
距 52w 高
-51.3%
RSI(14)
51.4
趋势
中性
SMA 20 / 50 / 200
16.80 / 16.10 / 18.24
MACD / 信号
0.539 / 0.549
MACD 死叉 (今天)

^TNX

10Y 美债收益率 (%)

$4.99 -1.11%
5 日
+0.73%
距 52w 高
-7.0%
RSI(14)
42.2
趋势
中性
SMA 20 / 50 / 200
5.14 / 5.01 / 4.54
MACD / 信号
-0.017 / 0.014

DX-Y.NYB

美元指数 DXY

$101.63 -0.55%
5 日
+0.02%
距 52w 高
-1.3%
RSI(14)
49.9
趋势
多头
SMA 20 / 50 / 200
102.03 / 100.75 / 99.62
MACD / 信号
0.266 / 0.364
接近 52 周高多头排列

SPY

S&P 500 ETF

$769.36 -0.04%
5 日
-0.48%
距 52w 高
-1.6%
RSI(14)
53.3
趋势
多头
SMA 20 / 50 / 200
770.90 / 767.63 / 728.15
MACD / 信号
-0.006 / 0.462
接近 52 周高多头排列

QQQ

Nasdaq 100 ETF

$715.34 -0.82%
5 日
-1.36%
距 52w 高
-6.2%
RSI(14)
49.6
趋势
中性
SMA 20 / 50 / 200
732.48 / 725.42 / 678.72
MACD / 信号
-4.373 / -2.346

AAPL

Apple

$348.91 +0.52%
5 日
+1.43%
距 52w 高
-0.8%
RSI(14)
56.1
趋势
多头
SMA 20 / 50 / 200
340.61 / 331.73 / 295.69
MACD / 信号
3.916 / 4.062
MACD 死叉 (今天)接近 52 周高多头排列

MSFT

Microsoft

$530.14 -0.22%
5 日
-0.53%
距 52w 高
-3.8%
RSI(14)
56.1
趋势
多头
SMA 20 / 50 / 200
529.90 / 511.18 / 437.32
MACD / 信号
6.264 / 8.043
多头排列

NVDA

Nvidia

$213.66 -1.96%
5 日
-3.74%
距 52w 高
-12.2%
RSI(14)
43.9
趋势
中性
SMA 20 / 50 / 200
223.79 / 222.07 / 204.14
MACD / 信号
-2.674 / -1.554

GOOGL

Alphabet

$283.82 -11.47%
5 日
-15.22%
距 52w 高
-30.5%
RSI(14)
38.7
趋势
中性
SMA 20 / 50 / 200
339.81 / 341.51 / 341.20
MACD / 信号
-6.085 / -3.225

TSLA

Tesla

$403.30 -1.02%
5 日
+4.35%
距 52w 高
-19.2%
RSI(14)
63.6
趋势
中性
SMA 20 / 50 / 200
376.78 / 367.47 / 385.72
MACD / 信号
7.705 / 4.467
MACD 金叉 (2 天前)

META

Meta

$624.52 +0.70%
5 日
-10.47%
距 52w 高
-19.9%
RSI(14)
30.7
趋势
中性
SMA 20 / 50 / 200
701.83 / 667.03 / 632.14
MACD / 信号
-16.116 / 0.257
加密恐慌贪婪
30
恐慌
加密总市值
$2.61 T
-7.09% / 24h
BTC 主导率
59.3%
ETH 10.3%
24h 成交量
$126.4 B
活跃币 22,591

BTC-USD

Bitcoin

$76,803.16 -5.49%
5 日
-4.94%
距 52w 高
-31.2%
RSI(14)
33.5
趋势
中性
SMA 20 / 50 / 200
81,547.19 / 81,859.09 / 73,096.98
MACD / 信号
-760.040 / -315.652

ETH-USD

Ethereum

$2,209.74 -6.27%
5 日
-6.21%
距 52w 高
-44.0%
RSI(14)
28.6
趋势
中性
SMA 20 / 50 / 200
2,403.88 / 2,528.90 / 2,158.92
MACD / 信号
-66.581 / -53.938
RSI 超卖

SOL-USD

Solana

$93.92 -8.20%
5 日
-8.17%
距 52w 高
-52.6%
RSI(14)
27.6
趋势
中性
SMA 20 / 50 / 200
104.04 / 109.25 / 88.68
MACD / 信号
-3.122 / -2.392
RSI 超卖

BABA

阿里巴巴 (BABA)

$99.54 -0.70%
5 日
-4.66%
距 52w 高
-45.0%
RSI(14)
34.1
趋势
空头
SMA 20 / 50 / 200
106.75 / 110.90 / 126.93
MACD / 信号
-2.841 / -2.077
空头排列

PDD

拼多多 (PDD)

$79.28 +1.12%
5 日
+0.71%
距 52w 高
-43.0%
RSI(14)
40.8
趋势
空头
SMA 20 / 50 / 200
79.40 / 80.73 / 91.37
MACD / 信号
-0.464 / -0.258
MACD 死叉 (2 天前)空头排列

JD

京东 (JD)

$25.63 +0.23%
5 日
-0.62%
距 52w 高
-25.3%
RSI(14)
36.3
趋势
空头
SMA 20 / 50 / 200
26.36 / 27.16 / 28.63
MACD / 信号
-0.482 / -0.392
MACD 死叉 (4 天前)空头排列

0700.HK

腾讯控股 (0700.HK)

HK$373.00 -1.22%
5 日
-3.37%
距 52w 高
-43.8%
RSI(14)
26.5
趋势
空头
SMA 20 / 50 / 200
406.08 / 425.58 / 479.56
MACD / 信号
-14.552 / -11.526
RSI 超卖接近 52 周低空头排列

GC=F

黄金期货

$4,013.80 -0.93%
5 日
-1.15%
距 52w 高
-28.1%
RSI(14)
34.7
趋势
空头
SMA 20 / 50 / 200
4,127.33 / 4,307.90 / 4,522.91
MACD / 信号
-67.084 / -62.376
接近 52 周低空头排列

CL=F

WTI 原油期货

$102.72 -0.06%
5 日
+3.27%
距 52w 高
-14.0%
RSI(14)
60.1
趋势
多头
SMA 20 / 50 / 200
96.82 / 94.32 / 86.14
MACD / 信号
2.614 / 2.181
多头排列

USDCNY=X

美元 / 人民币

¥6.65 -0.22%
5 日
-0.40%
距 52w 高
-7.5%
RSI(14)
8.9
趋势
空头
SMA 20 / 50 / 200
6.69 / 6.70 / 6.80
MACD / 信号
-0.012 / -0.009
RSI 超卖接近 52 周低空头排列
风险提示

本报告基于公开行情数据的技术指标计算,仅供技术指标解读参考,不构成投资建议。过去走势不代表未来表现,技术指标存在滞后性与失效可能,市场可能因突发事件改变趋势。请读者结合基本面与自身风险承受能力独立判断。

EU force helicopter crash kills four in Bosnia

EUFOR confirms CH-47 Chinook helicopter crash in Bosnia, with four Italian nationals confirmed dead.

中文摘要 欧盟驻波黑部队(EUFOR)证实,一架CH-47支奴干直升机在波黑坠毁,造成4名意大利公民遇难。

Court Rejects Erin Patterson’s Appeal in Mushroom Murders Case

Judges took away the possibility of parole for an Australian woman, Erin Patterson, and rejected a bid to overturn her conviction over deadly beef Wellington lunch.

中文摘要 澳大利亚法院驳回埃琳·帕特森(Erin Patterson)就其“毒蘑菇案”定罪的上诉,并取消其假释可能性。此前,她因一顿致命的威灵顿牛肉午餐导致三人死亡而被定罪。

Erin Patterson to die in prison after losing appeal against mushroom murder convictions

Victorian court of appeal rules Patterson should be jailed for life with no parole over deadly beef wellington lunch that killed three people Erin Patterson mushroom murders appeal – live updates Get our breaking news email, free app or daily news podcast Erin Patterson has lost her bid to overturn

中文摘要 澳大利亚维多利亚州上诉法院裁定,埃琳·帕特森(Erin Patterson)因“毒蘑菇案”被判终身监禁,且不得假释。该案涉及一顿致三人死亡的威灵顿牛肉午餐。

Trump cancels Iowa, Kansas midterm events, opts for ‘tele-rallies’

The Republican National Committee has announced broadcasts in nearly 20 states after Trump nixes final rallies.

中文摘要 特朗普取消在艾奥瓦州和堪萨斯州的最后一场中期选举集会,改为在近20个州举行“电话集会”。此举由共和党全国委员会宣布。

Antisemitism envoy proposes ‘external ombudsman’ to monitor ABC and SBS coverage of Israel and Middle East

Jillian Segal says perception of link between coverage of conflicts and hostility towards Jewish Australians ‘is itself a matter of public importance’ Get our breaking news email, free app or daily news podcast The federal government’s anti-semitism envoy has proposed an “external independent ombuds

中文摘要 澳大利亚反犹太主义特使吉莉安·西格尔(Jillian Segal)提议设立“外部监察员”,以监督澳大利亚广播公司(ABC)和特别广播服务公司(SBS)对以色列和中东冲突的报道,认为报道与针对犹太裔澳大利亚人的敌意存在关联。

Armenian parliament turns into a boxing ring, third time in a month

Armenia’s parliament descended into fisticuffs after a ruling-party MP mocked the opposition leader’s family.

中文摘要 亚美尼亚议会一个月内第三次爆发肢体冲突,起因是执政党议员嘲讽反对派领导人的家人。

Australia’s mushroom murderer loses bid to overturn convictions – as it happened

This blog is now closed. Read our full report here: Erin Patterson loses appeal as harsher sentence imposed Patterson showed no emotion on the video feed as McLeish outlined why her appeal had failed, and how the director of public prosecutions had succeeded in seeking a longer sentence. Patterson’s

中文摘要 澳大利亚“毒蘑菇案”被告埃琳·帕特森(Erin Patterson)上诉失败,并被加重刑罚。法官麦利什(McLeish)在视频连线中宣布驳回上诉,并支持检方寻求更长刑期的请求。帕特森未表现情绪。

Houthi forces claim six attacks on Saudi airports, oil facilities

The announcement comes despite talks between Iran and the Mecca pact countries in Tehran earlier this week.

中文摘要 胡塞武装声称对沙特阿拉伯的机场和石油设施发动了六次袭击。尽管此前伊朗与麦加条约国家在德黑兰举行了会谈。

What does bourbon have to do with the US midterms?

The US-Canada trade war is hurting Kentucky’s bourbon industry, which generates billions of dollars annually.

中文摘要 美国与加拿大的贸易战正冲击肯塔基州波本威士忌产业,该产业年产值达数十亿美元。此事或影响美国中期选举选情。

France braces for ‘very high risk’ of disorder at Halloween protests

Authorities vow firm action against violence as yellow vest and student movements announce widespread demonstrations.

中文摘要 法国当局警告万圣节抗议活动存在“极高风险”骚乱,并承诺对暴力采取坚决行动。黄背心和学生运动宣布将举行大规模示威。

What Israeli Election Results Reveal About the Electorate

Some groups stood by Prime Minister Benjamin Netanyahu’s party, but support fell off among others. Turnout was also a striking factor in the election.

中文摘要 以色列选举结果分析:部分群体仍支持总理内塔尼亚胡的政党,但其他群体支持率下滑。投票率也是影响选举的关键因素。

Newly Freed American Describes His Years in Russian Prison

Stephen James Hubbard, the retired teacher arrested soon after Russia’s full-scale invasion of Ukraine in 2022, said he sometimes lost hope.

中文摘要 新获释的美国人斯蒂芬·詹姆斯·哈伯德(Stephen James Hubbard)讲述其在俄罗斯监狱的经历。这位退休教师在2022年俄乌冲突全面爆发后不久被捕,他称自己有时会失去希望。

Netanyahu accused of smearing Arab voters to hold on to power

Prime minister claims Arab Israeli parties used foreign backing to steal elections, in which his coalition lost its majority Benjamin Netanyahu has been accused of using smear tactics to cling to power, after he claimed that Arab Israeli parties had used foreign backing to “steal” Tuesday’s election

中文摘要 以色列总理内塔尼亚胡被指控使用“丑陋的抹黑手段”以保住权力。他声称阿拉伯裔以色列政党利用外国支持“窃取”选举,而其联盟已失去多数席位。

Crypto giant Binance freezes $2.5bn of assets over suspected hack

The firm said transactions are suspended while it conducts security checks and will "cover all losses".

中文摘要 加密货币交易所币安冻结了25亿美元资产,疑似因黑客攻击。公司表示已暂停交易以进行安全检查,并承诺"全额赔偿损失"。

Bayer Wins St. Louis Roundup Trial, Bolstering Settlement Push

A jury in St. Louis rejected claims by three women who blamed Bayer AG’s Roundup weedkiller for their cancers, bolstering the company’s effort to contain the cost of tens of thousands of similar lawsuits in the US that it has offered to settle.

中文摘要 圣路易斯陪审团驳回了三名女性指控拜耳农达除草剂致癌的诉讼,增强了该公司应对美国数万起类似诉讼的和解努力。

Dollar’s Rate-Fueled Rally Set to Ride Out US Midterm Gridlock

The dollar’s rally may have enough momentum to shrug off the US midterm election, with high interest rates and a surge in AI-related spending drawing capital even if the vote delivers political gridlock.

中文摘要 美元涨势可能足以抵御美国中期选举影响,高利率和AI相关支出激增吸引资本,即使选举导致政治僵局。

Big Tech’s Deepening AI Funding Concerns Sink Asian Chip Stocks

A crucial earnings season for the fate of the artificial intelligence trade is falling flat, fueling stock losses in global chipmakers as worries mount over soaring costs and the durability of demand.

中文摘要 人工智能交易关键财报季表现不佳,引发全球芯片股下跌,市场担忧成本飙升和需求持续性。

China Elevates Key Actor With Rare Politburo Pick

The Chinese Communist Party restocked its powerful Politburo for the first time in more than three decades and announced plans to hold a congress next year, where it’s set to unveil new leadership. (Source: Bloomberg)

中文摘要 中国共产党三十多年来首次重组其强大的政治局,并宣布计划明年召开代表大会,届时将公布新领导层。

Foxconn CEO: AI to Drive Consumer Update Cycle

Foxconn rotating CEO Michael Chiang expresses optimism about the AI boom and says he expects it to drive a new cycle of consumer hardware upgrades. (Source: Bloomberg)

中文摘要 富士康轮值CEO迈克尔·蒋对AI热潮表示乐观,预计将推动新一轮消费硬件升级周期。

Hong Kong Stocks Lose a Key Backer as Chinese Investors Head for Exit

Chinese investors are pulling back from Hong Kong’s stocks at the fastest pace in more than five years, threatening to deepen a selloff.

中文摘要 中国投资者正以五年多来最快速度撤离香港股市,可能加剧抛售压力。

Philippine Bonds Are Most Exposed in Emerging Asia to Fed Hikes

Philippine bonds are the most vulnerable in emerging Asia to a hawkish Federal Reserve, according to an analysis by Bloomberg.

中文摘要 彭博分析显示,菲律宾债券在亚洲新兴市场中对美联储鹰派加息最为敏感。

Topix Overhaul Kicks Off: Three Things Investors Need to Know

Japan Exchange Group Inc.’s first large-scale reshuffle of the Topix Index goes into effect Friday, part of the biggest overhaul in the benchmark’s history that may shrink the number of constituents by more than one-third.

中文摘要 日本交易所集团对东证指数进行首次大规模调整,周五生效,为基准指数历史上最大改革,成分股数量可能减少逾三分之一。

PAG Agrees to Buy China Evergrande Unit for $514 Million

Private equity firm PAG has agreed to buy a majority stake in Evergrande Property Services Group Ltd. for HK$4.03 billion ($514 million), according to a Hong Kong stock exchange filing, confirming a previous Bloomberg report.

中文摘要 私募股权公司PAG同意以40.3亿港元(5.14亿美元)收购恒大物业服务集团多数股权,港交所文件确认此前彭博报道。

该源今日无内容。