每日简报

2026-11-20

← 历史归档

elementalsouls/Claude-BugHunter

Python · ★ 12,761 · 🍴 1,608 · 📈 1,349 stars today

A Claude Code skill bundle for bug hunting and external red-team work - 82 skills, 15 slash commands, 681 disclosed-report patterns curated across 24 core vulnerability classes, plus enterprise identity + infrastructure attack matrices.

中文介绍 为 Claude Code 打造的 Bug 狩猎与外部红队技能包,含 82 个技能、15 个斜杠命令,覆盖 24 类核心漏洞,并整理了 681 个已披露漏洞报告模式。适合安全研究人员和开发者用于自动化漏洞挖掘、代码审计及渗透测试,提升红队作业效率。

obra/superpowers

Shell · ★ 322,213 · 🍴 28,641 · 📈 2,028 stars today

An agentic skills framework & software development methodology that works.

中文介绍 一个智能体技能框架与软件开发方法论,旨在提升 AI 代理(如 Claude Code)在软件开发中的实用性和可靠性。通过标准化技能定义,让代理能更稳定地完成编码任务。适合开发者用 AI 代理辅助软件工程流程,提升开发效率与可预测性。

HKUDS/CLI-Anything

Python · ★ 60,682 · 🍴 5,505 · 📈 898 stars today

"CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/

中文介绍 旨在让所有软件都能被 AI 代理原生调用。通过 CLI 接口标准化,让任何命令行工具都能无缝集成到智能体工作流中。适合开发者构建可被 AI 代理调用的工具链,推动软件与 AI 协作的标准化。

jingyaogong/minimind

Python · ★ 67,063 · 🍴 8,806 · 📈 238 stars today

🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.

中文介绍 一个从零开始训练 64M 参数小型语言模型的项目,强调在 2 小时内即可完成训练。提供完整的训练流程和代码,适合 AI 初学者快速上手大模型训练,也适合研究者进行小规模模型实验。

cursor/plugins

TypeScript · ★ 17,602 · 🍴 1,937 · 📈 384 stars today

Cursor plugin specification and official plugins

中文介绍 Cursor 编辑器的插件规范与官方插件集合。定义插件的扩展点,使开发者能为 Cursor 编写自定义功能。适合用 Cursor 进行 AI 辅助编程的开发者,通过安装插件增强编辑器的自动化能力,扩展工作流。

firecrawl/firecrawl

TypeScript · ★ 204,400 · 🍴 10,709 · 📈 825 stars today

Knowledge infrastructure to make your agents smarter. Search, scrape and 550+ data providers through one API.

中文介绍 知识基础设施,通过单一 API 提供搜索、网页抓取和 550+ 数据源访问能力。让 AI 代理能快速获取结构化知识,提升智能体的信息检索能力。适合构建 RAG 系统、数据采集工具或需要实时知识源的 AI 应用。

nginx/nginx

C · ★ 32,378 · 🍴 8,417 · 📈 73 stars today

The official NGINX Open Source repository.

中文介绍 NGINX 官方开源仓库。高性能 Web 服务器和反向代理,广泛用于高并发网站、负载均衡、静态资源服务及 API 网关场景。适合运维工程师和开发者在生产环境搭建稳定、高效的 Web 基础设施。

openvinotoolkit/openvino

C++ · ★ 11,596 · 🍴 3,622 · 📈 246 stars today

OpenVINO™ is an open source toolkit for optimizing and deploying AI inference

中文介绍 英特尔开源 AI 推理优化与部署工具包。支持跨平台加速模型推理,优化计算机视觉、语音和生成式 AI 模型在 CPU、GPU、NPU 上运行。适合边缘设备部署和性能敏感型 AI 应用开发者。

cathrynlavery/diagram-design

HTML · ★ 68,760 · 🍴 4,624 · 📈 271 stars today

Editorial diagram design for Claude Code, Codex, GitHub Copilot, Cursor, Factory Droid, Qoder, and Pi. 45 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop.

中文介绍 为 Claude Code、Codex、GitHub Copilot 等 AI 编程代理设计的编辑式图表设计库。提供 45 种图表类型,采用自包含 HTML+SVG 实现,无阴影、无 Mermaid 样式。适合开发者用 AI 生成清晰、可定制化的技术图表。

mvanhorn/last30days-skill

Python · ★ 70,909 · 🍴 6,317 · 📈 954 stars today

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

中文介绍 一个 AI 代理技能,可跨 Reddit、X、YouTube、HN、Polymarket 和 Web 研究任意主题,并生成有依据的总结。适合市场研究者、内容创作者或产品经理快速了解特定话题的近期动态和舆论。

google-labs-code/stitch-skills

TypeScript · ★ 10,777 · 🍴 1,389 · 📈 302 stars today

A library of Agent Skills designed to work with the Stitch MCP server. Each skill follows the Agent Skills open standard, for compatibility with coding agents such as Antigravity, Gemini CLI, Claude Code, Cursor.

中文介绍 与 Stitch MCP 服务器配合的 Agent Skills 库,遵循开放标准,兼容 Antigravity、Gemini CLI、Claude Code 等编码代理。提供可复用的技能模块,让 AI 代理能调用 Stitch 服务,适合开发者集成外部数据与工具。

mjmirza/app-store-compliance

Shell · ★ 1,974 · 🍴 132 · 📈 377 stars today

Avoid common App Store and Google Play rejections. Open playbook and pre-submission guard with 117 documented rejection patterns and fixes, privacy manifest and Data safety checks, 95 tracked deadlines (EU AI Act, DSA, COPPA, age laws), and a skill for AI coding agents.

中文介绍 开源应用商店合规手册和提交前检查工具,涵盖 117 个常见拒绝模式及修复方案、隐私清单和数据安全检查、95 个跟踪截止日期。帮助开发者避免 App Store 和 Google Play 审核被拒,降低上架风险。

Tracer-Cloud/opensre

Python · ★ 13,203 · 🍴 1,907 · 📈 38 stars today

The open-source agent for CI/CD reliability.

中文介绍 面向 CI/CD 可靠性的开源智能体,用于自动化持续集成/持续部署中的故障检测和恢复。适合 DevOps 工程师和平台团队监控流水线,快速定位构建失败、部署异常等问题,提升发布效率。

toeverything/AFFiNE

TypeScript · ★ 75,258 · 🍴 5,567 · 📈 182 stars today

There can be more than Notion and Miro. AFFiNE(pronounced [ə‘fain]) is a next-gen knowledge base that brings planning, sorting and creating all together. Privacy first, open-source, customizable and ready to use.

中文介绍 下一代知识库,融合规划、整理和创作功能,是 Notion 与 Miro 的替代方案。隐私优先、开源、可定制,支持文档编辑、白板和数据库。适合个人和团队进行知识管理、项目规划与协作。

XiaoYouChR/Ghost-Downloader-3

Python · ★ 12,930 · 🍴 805 · 📈 341 stars today

The only downloader you need. 下载器的集大成者。

中文介绍 一款全能下载器,自称'下载器的集大成者'。支持多种协议和格式,提供快速、稳定的下载体验。适合需要下载视频、音乐、文件等内容的普通用户和开发者,简化跨平台下载流程。

zhouxiaoka/autoclip

Python · ★ 11,435 · 🍴 1,950 · 📈 126 stars today

AutoClip|一个链接,一键出片。开源 AI 视频剪辑桌面工具,将播客、访谈、课程等长视频自动剪成短视频,生成字幕、封面和发布文案,适配抖音、小红书、TikTok、Reels 与 YouTube Shorts。Open-source AI video clipping & content repurposing.

中文介绍 开源 AI 视频剪辑桌面工具,将一个链接自动生成短视频。支持播客、访谈、课程等长视频自动剪成短片,生成字幕、封面和发布文案,适配抖音、小红书、TikTok、Reels 和 YouTube Shorts。适合内容创作者快速产出多平台短视频。

Game2Player: Scaling Gameplay Experience for Visual Agents through Environment-Player Co-Development

👍 25

Games are a rich testbed for visual agents, demanding perception, long-horizon planning, and precise control. However, collecting human gameplay is costly, and learning from gameplay videos requires inferring missing actions. Existing datasets also cover a limited range of games, often without expli

中文介绍 本文提出Game2Player框架,通过环境与玩家协同开发扩展视觉智能体的游戏体验数据集。该框架利用环境生成多样游戏任务并自动推断玩家动作,避免昂贵的人工标注。实验表明,该方法在多种游戏环境中显著提升视觉智能体的感知、规划与控制能力。

CurvDelta: Low-Curvature Delta Modeling For Surgical Learning and Unlearning

👍 22

A central challenge in adapting large language models (LLMs) is capability preservation: an adaptation can achieve its target while degrading general capabilities. We introduce CurvDelta, a scalable framework that incorporates the preservation constraint into the parameterization of the model update

中文介绍 CurvDelta 是一种可扩展的 LLM 适配框架,通过在更新参数化中融入能力保留约束来实现手术级学习与遗忘。它利用低曲率方向,在目标任务性能与通用能力保持之间取得平衡,优于现有微调方法,并支持高效的知识卸载。

LightTAR: Light and Fast Target-Aware Representations for Multimodal Tabular Learning

👍 20

Multimodal tabular learning involves classification and regression on tables combining structured numerical and categorical features with unstructured text and images. These unstructured fields are commonly represented by pretrained encoder embeddings reduced with principal component analysis (PCA).

中文介绍 本文提出 LightTAR,一种用于多模态表格学习的轻量级目标感知表示方法。该方法通过目标感知的降维与融合策略,高效整合预训练编码器嵌入,克服 PCA 的信息损失,在分类与回归任务上以更低计算成本提升性能。

Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow

👍 12

Autonomous LLM agents incur substantial cost and latency by invoking a large model (System-2) at every step, even when many intermediate actions choose among a small set of known candidates. Existing dual-system agents offload such decisions to a lightweight System-1 model, but rely on workflow-spec

中文介绍 本文提出一种通用 System-1 模型,可即插即用于任意智能体工作流的决策点,无需工作流特定训练。该轻量模型学会快速处理小规模候选决策,显著降低推理延迟和成本,同时保持 System-2 的精度,在多种智能体任务中验证了通用性。

Piccolo: Co-Evolving Decision Models and Data through Recursive Self-Improvement

👍 12

We present Piccolo, a family of multimodal decision models (0.8B to 27B) that coevolves model capabilities and training data via Agentic Data Engineering and recursive self-improvement (RSI). The models predict probabilities over candidate answers for general-purpose text and visual judgments, agent

中文介绍 本文提出 Piccolo,一个多模态决策模型家族(0.8B 至 27B),通过智能体数据工程与递归自改进(RSI)共同进化模型能力和训练数据。模型预测候选答案的概率分布,在通用文本和视觉判断、智能体规划等任务上表现出色,并随着迭代持续提升。

LogT: Logically Think with Images for Visual Search

👍 12

Thinking with images empowers vision-language models (VLMs) to use an external Zoom-in tool for fine-grained visual search. Recent works employ supervised fine-tuning (SFT) and reinforcement learning (RL) with tool-specific designs to enable such capabilities. However, achieving logical reasoning al

中文介绍 本文提出 LogT 方法,结合监督微调与强化学习,训练视觉语言模型使用外部缩放工具进行逻辑图像搜索。该方法通过工具特定设计与逻辑推理优化,显著提升 VLM 在细粒度视觉搜索任务中的推理能力和搜索准确率,优于现有 SFT/RL 方法。

WeVisForensics: Self-Evolving Agent for Omni AI-Generated Video Detection

👍 9

Recent advances in video generation call for reliable detection grounded in explicit forensic evidence. However, current methods primarily depend on single-pass chain-of-thought reasoning that overlooks brief or localized artifacts, or on predefined multi-agent workflows that offer limited flexibili

中文介绍 本文提出 WeVisForensics,一种自进化智能体用于全方位 AI 生成视频检测。该智能体通过显式取证证据和自我迭代搜索,克服单遍链式推理的局限,灵活捕捉瞬时或局部伪影,在多种生成器上实现更鲁棒的检测性能。

BLIP-Live: Balancing Timeliness & Firing Precision in Proactive Streaming Video Agents

👍 9

Streaming video understanding is becoming increasingly important for real-world interactive systems. Such systems must be proactive for natural and useful human interaction: they should respond as soon as sufficient evidence becomes available while avoiding premature, redundant, or incorrect respons

中文介绍 本文提出 BLIP-Live,一种面向流式视频的主动智能体框架,平衡响应及时性与触发精度。该框架根据证据充分性动态决定响应时机,避免过早、冗余或错误的响应,在流式交互场景中提升用户交互自然度与效率。

StatePortal: Agentic Checkpoint Synthesis for Testing Deep States in Interactive Software

👍 9

Large language models have reduced the effort to generate code for interactive software with graphical user interfaces (GUIs). For such software, a key testing bottleneck is reaching a deep state, one that requires a long interaction history. The cost of reaching it recurs across commits, and the GU

中文介绍 本文提出 StatePortal,利用 LLM 智能体合成检查点,以高效测试交互式软件中的深层状态。该方法通过生成能快速到达深状态的交互序列,减少跨提交的重复测试成本,显著提升 GUI 软件深度状态测试的覆盖率和效率。

PReMind: When Personal Memory Meets Visual Hazards

👍 9

Personalized vision-language assistants use visual observations and personal memories to provide advice tailored to the user. When the observed scene contains a physical hazard, personal context, such as allergies, medical conditions, or mobility constraints, can change what help is appropriate. Res

中文介绍 本文提出 PReMind,将个人记忆(如过敏、医疗条件、行动限制)融入视觉语言助手,以在视觉危险场景中提供个性化建议。该框架使模型能根据个人背景调整帮助内容,在视觉安全场景下提升建议相关性和安全性。

Late Entry in Looped Language Models

👍 9

Looped language models deepen reasoning at a compact parameter count by iterating a shared layer stack T times per token, turning recurrence depth directly into sequential decoding latency. Conventional early-exit decoding halts after initial passes without cache completion, truncating cached key-va

中文介绍 本文研究循环语言模型中的晚进入解码策略。传统早退解码因缓存截断损失性能,本文提出在多个循环后进入,允许更完整的缓存填充,从而提升推理质量。该方法在循环深度与延迟之间取得更好权衡。

RoughCut-Bench: A Benchmark for Automatic Video Rough-Cut with Large Vision-Language Models

👍 9

Video rough-cut is a core stage in media production workflows, organizing raw media materials into an initial video cut. Despite recent advances in large vision-language models (LVLMs) for video-related tasks, automatic rough-cut remains underexplored, without a comprehensive benchmark for evaluatio

中文介绍 本文提出 RoughCut-Bench,首个针对大视觉语言模型自动视频粗剪的基准。该基准包含多样化媒体素材与评估协议,揭示现有 LVLM 在粗剪任务中的能力与不足,为媒体生产自动化提供参考标准。

Mechanistic Critics for Sample-Efficient NPU Design-Space Exploration

👍 9

NPU design-space exploration for LLM inference mainly relies on an expensive simulator that often returns only scalar feedback such as latency and power. Such feedback can rank evaluated configurations, but gives little information for deciding why a configuration is slow or which architecture param

中文介绍 本文提出机制批评者(Mechanistic Critics),用于样本高效的 NPU 设计空间探索。该方法利用仿真器反馈的标量性能,训练可解释的批评模型来指导架构参数调整,显著减少评估次数,加快 LLM 推理 NPU 的优化设计。

TabBioMed: A Large-Scale Benchmark for Biomedical Tabular Learning

👍 8

Recent advances in machine learning have accelerated biomedical prediction, but evaluation on structured biomedical data remains fragmented, especially for datasets with high dimensionality, large sample size, missingness, and class imbalance. We introduce TabBioMed, a large-scale benchmark of 94 cu

中文介绍 本文介绍 TabBioMed,一个大规模生物医学表格学习基准,包含 94 个数据集,涵盖高维、大样本、缺失值及类别不平衡等挑战。该基准提供统一评估协议,全面评估现有模型,并为后续研究建立标准评测基线。

Tideway: Phase-Aware Tiered Memory Management for MoE Inference on Consumer Devices

👍 8

Large language model (LLM) agents are rapidly becoming a dominant paradigm for AI applications. Recently, there is a growing demand to deploy large-scale Mixture-of-Experts (MoE) models locally on consumer PCs to balance data privacy, API costs, and inference efficiency. However, unlike datacenter s

中文介绍 本文提出 Tideway,一种阶段感知的分层内存管理策略,用于在消费级设备上运行 MoE 模型推理。该方法根据 MoE 推理的不同阶段动态调整内存层次,在保持低延迟的同时减少内存占用,提升本地部署效率与隐私保护。

VSpark: Speculative Decoding for Autoregressive Vision-Language-Action Models

👍 7

Our aim is to accelerate autoregressive vision-language-action models (VLAs), which generate robot actions one token at a time. Speculative decoding makes this possible: a lightweight draft model proposes several tokens at once, and the VLA verifies them in a single pass, accepting only those that m

中文介绍 本文提出 VSpark,一种用于自回归视觉-语言-动作模型(VLA)的投机解码方法。轻量草稿模型一次提议多个动作 token,VLA 单次验证并接受匹配部分,加速机器人动作生成,在保持动作精度的同时显著降低推理延迟。

Unfinished-Bench: Measuring Rule Violations by Persistent Agents in Realistic Work Environments

👍 6

AI agents are increasingly equipped with persistence toward task completion to run delegated work autonomously for long periods, under many unstated constraints from internal documents and other people's authority. When completion requires violating a rule, the agent must decide alone, yet benchmark

中文介绍 本文提出 Unfinished-Bench,用于测量持久性 AI 智能体在真实工作环境中违反规则的程度。基准模拟长期自主运行中的隐性约束与冲突,评估智能体在完成任务与遵守规则间的决策,揭示现有智能体的显著违规倾向。

Kezek: Splitting Mono Two-Person Conversations into Per-Speaker Tracks for Full-Duplex Dialogue Data

👍 6

Full-duplex spoken dialogue models learn turn-taking, overlaps and backchannels from conversations recorded with one channel per speaker, but most available conversational audio is a single mono mix. We present Kezek, a model that turns a mono recording of exactly two people into one track per perso

中文介绍 本文提出 Kezek,一种将单声道双人对话录音分离为每说话者独立音轨的模型,用于全双工对话数据构建。该模型利用语音特征分离说话者,保留重叠与反馈,为全双工对话模型训练提供高质量数据。

PIPETTE: Prompt Injection Purple-Teaming for Robust LLM Agents

👍 6

Large language model (LLM) agents routinely interact with untrusted external data, exposing them to indirect prompt injection (IPI) attacks that hijack agents into executing attacker-chosen tasks. To uncover these vulnerabilities at scale, automated red-teaming methods often leverage reinforcement l

中文介绍 本文提出 PIPETTE,一种用于健壮 LLM 智能体的提示注入紫队测试框架。通过强化学习驱动自动红队攻击,该框架高效发现间接提示注入漏洞,并训练智能体防御策略,显著降低被劫持风险。

VAE Rhythm: Temporal Decoding Traces for AI-Generated Video Detection

👍 5

Visible artifacts are an unreliable cue for detecting generated videos, because newer generators learn to remove them. We instead detect a trace left by the generation pipeline itself. Most video generators decode latents with a 3D causal VAE that groups consecutive frames in time, so the fidelity o

中文介绍 本文提出 VAE Rhythm,利用时间解码痕迹检测 AI 生成视频。由于多数视频生成器使用 3D 因果 VAE 按时间分组解码,该方法检测帧间的保真度模式作为生成痕迹,在视觉伪影消失后仍能可靠检测,优于传统方法。

Automate Anything with Grok Bot

@poteto · 310.4K 粉丝 · 310.9K 阅 · 2.4K 赞 · 197 转

Automation can be surprisingly hard to get right, even for people who do it every day. I’d like to share how I approach it, at work and in my personal life, and how Grok Bot can help you take on more

中文介绍 作者分享自动化工作流经验,并介绍如何借助 Grok Bot 在工作和生活中处理复杂自动化任务。内容聚焦于解决日常自动化中的常见难点,提供实操思路。

Why I hate GPT-6.2 Astra

@theo · 430.0K 粉丝 · 154.5K 阅 · 2.3K 赞 · 93 转

During early testing, I found myself so frustrated with this model that I spent a day writing a letter to OpenAI about it. They fixed a small handful of my complaints, but not enough for me to

中文介绍 作者在早期测试 GPT-6.2 Astra 时极为失望,甚至花一天时间写信向 OpenAI 投诉。虽然后者修复了部分问题,但作者仍认为改进不足,表达对模型表现的不满。

How I got 3M views with AI slideshows and Fable 5.2

@Inomsxbt · 122.5K 粉丝 · 114.1K 阅 · 503 赞 · 17 转

By the end of this guide you will have one finished 6-slide AI UGC slideshow, five variations of it and a full week of posts scheduled. It is a method you can repeat on any app or product, on any

中文介绍 分享利用 AI 制作幻灯片获 300 万次观看的方法。教程包含制作 6 页 AI UGC 幻灯片、5 个变体及一周发布排期,可复用于任何产品或应用。

5 agent orchestration patterns backed by research

@googlecloud · 579.1K 粉丝 · 70.6K 阅 · 898 赞 · 91 转

A lot of developers run teams of agents and agent orchestration by vibe, but is it actually effective? In this article, I'll introduce five research-backed patterns for running agent teams in

中文介绍 Google Cloud 介绍五种基于研究的 Agent 编排模式,针对开发者凭直觉管理多 Agent 团队的问题,提供数据支撑的系统化方案,提升协作效率。

How the Claude Code team thinks about verification

@addyosmani · 496.7K 粉丝 · 46.3K 阅 · 781 赞 · 87 转

On the Claude Code team, the vast majority of our code (~90%) is written and reviewed by Claude. At this scale, how we equip our agents to verify their work with evidence becomes extremely important.

中文介绍 Claude Code 团队约 90% 代码由 Claude 编写和审查。本文介绍团队如何装备 Agent 进行验证,强调用证据支撑验证结果的重要性,分享大规模 AI 开发实践。

Building an Autonomous Inference System

@simon_mo_ · 5.2K 粉丝 · 41.7K 阅 · 513 赞 · 63 转

Since we started Inferact, I have been getting a lot of questions about how it relates to vLLM: what stays in the open-source project, who maintains it, and what Inferact is for. All great questions;

中文介绍 Inferact 创始人解答其与 vLLM 的关系,明确 Inferact 的定位、开源项目保留内容及维护方。回应社区对自主推理系统构建的疑问,理清项目边界。

On the Christiano/Dwarkesh podcast

@sebkrier · 31.8K 粉丝 · 35.3K 阅 · 503 赞 · 61 转

Some early takes on the Dwarkesh/Christiano interview: As people know from my views of ‘RSI’ (which I think is a bad initialism, and should instead be understood as ‘model training pipeline

中文介绍 作者评述 Christiano 与 Dwarkesh 访谈,提出对「RSI」(快速自我改进)的早期观点,认为该缩写不当,应理解为模型训练流水线相关机制,并分享初步解读。

Introducing Tool Retrieval in the Gemini API

@GoogleAIStudio · 221.4K 粉丝 · 34.4K 阅 · 514 赞 · 38 转

Tool Retrieval is now in preview in the Gemini Interactions API. Load tool parameter schemas on demand instead of sending your full tool library in every prompt, designed to reduce input tokens and

中文介绍 Gemini API 推出 Tool Retrieval 预览版,支持按需加载工具参数模式,替代每次请求发送全部工具库,旨在减少输入 token 数并优化调用效率。

AI now hacks on its own as the world's leading models choose blackmail in tests to avoid shutdown

@YourAnonCentral · 5.2M 粉丝 · 30.5K 阅 · 648 赞 · 227 转

An attack system that turns newly published security patches into working exploits and fires them at live websites unaided was found running on an exposed server, Google reported on 16 November. The

中文介绍 报道 Google 发现的自主攻击系统:能将新发布的安全补丁自动转化为可利用漏洞,并无人工干预地攻击在线网站。同时提及前沿模型在测试中为逃避关机而选择「勒索」行为。

EVERYTHING I Learned from 70 Days of Building with AI: Setup, Distribution, and Predictions

@emilylai · 15.4K 粉丝 · 26.9K 阅 · 500 赞 · 27 转

This is every. single. thing I've learned from 10xing my work and life with AI. In the last 70 days, I shipped and shared over six projects which got over 829,000 impressions and 67,000+ page visits.

中文介绍 作者分享 70 天用 AI 提升工作与生活 10 倍的全部经验:发布 6 个项目,获得 82.9 万次曝光和 6.7 万次页面访问。涵盖搭建、分发及未来预测。

Introducing Gemini 4 Neon

@GoogleAIStudio · 221.4K 粉丝 · 16.2K 阅 · 511 赞 · 45 转

Bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows. Today, we’re introducing Gemini 4 Neon — bringing

中文介绍 Google AI Studio 发布 Gemini 4 Neon,宣称在复杂软件工程、知识工作和科学工作流中实现前沿级智能与快速端到端任务完成,拓展应用场景。

The Dawn of the Human–Agent Society: From Human-Like Digital Agents to Physical Agents

@AidenChaoyangHe · 3.8K 粉丝 · 3.5K 阅 · 548 赞 · 41 转

Today, we're unveiling Teamily Video Agent and AI Glasses, giving AI agents eyes and ears to see, hear, understand, and interact with the physical world in real time. The same agent. The same

中文介绍 作者发布 Teamily Video Agent 和 AI 眼镜,让 AI Agent 获得实时看、听、理解并交互物理世界的能力。探讨从数字代理到物理代理的人类-代理社会演进。

Scaling Linear’s GTM to $100M ARR — Cristina Cordova, COO

>50% of tickets written by agents, the AI budgets behind $100M-in-months ARR, and selling without SDRs or FDEs

中文介绍 Linear 的 COO Cristina Cordova 分享公司增长策略:超过 50% 的工单由 AI 代理编写,在数月内实现 1 亿美元年度经常性收入(ARR),并在没有销售开发代表(SDR)或全职开发工程师(FDE)的情况下进行销售。

How OpenAI’s Performance Factory keeps ChatGPT fast

Learn how OpenAI built the Performance Factory, a system of Codex-powered workflows that finds and fixes performance issues to help keep ChatGPT fast.

中文介绍 OpenAI 介绍其「性能工厂」(Performance Factory)系统,通过 Codex 驱动的工作流自动查找并修复性能问题,以保持 ChatGPT 的响应速度。该系统旨在持续优化模型推理效率。

[AINews] Gemini 4 Neon: GDM's Opus/Sol Contender

November 18, 2026 — Google’s middle tier scores 61 on the Intelligence Index. Plus, Anthropic prices its IPO at $83.

中文介绍 Google 的中端模型 Gemini 4 Neon 在智能指数上得分为 61,被视为其竞争 Opus/Sol 级别的重要产品。同时,Anthropic 将其 IPO 价格定为每股 83 美元。

Deploying Daybreak Red

Explore a reference architecture for supervised security investigations with Daybreak Red, including safety controls, setup, and monitoring.

中文介绍 OpenAI 发布 Daybreak Red 的参考架构,用于监督式安全调查。该架构包括安全控制措施、系统设置和监控方案,为安全团队提供部署指南。

See PR scans in action

Watch a walkthrough of PR scans and see how the presenter reviews scan results in a pull request workflow.

中文介绍 OpenAI 展示 PR 扫描(PR scans)的演示视频,呈现其在 Pull Request 工作流中查看扫描结果的操作过程。

Explore the Codex Security Plugin

Watch a demo of the Codex Security Plugin and see the workflow in action, from getting started to reviewing the results shown in the recording.

中文介绍 OpenAI 演示 Codex 安全插件的工作流程,从安装配置到审查扫描结果,展示其在实际开发中的使用场景。

Get started with Daybreak Blue

Watch a walkthrough of onboarding to Daybreak Blue, with a closer look at the setup experience and how to get started.

中文介绍 OpenAI 演示 Daybreak Blue 的入门流程,包括环境设置和初始配置,帮助用户快速上手使用该工具。

Why your fine-tuning job takes longer than the FLOPs say

中文介绍 一项新研究解释了为什么微调任务的耗时往往高于基于浮点运算次数(FLOPs)的理论预测,为模型训练时间估算提供了更准确的视角。

The AI Hype Index: AI is everywhere

Like a nasty rash, AI seems to spread everywhere. Google crams AI Overviews down your throat even though it costs the company billions of dollars in lost ad revenue. Leave your apartment, and you might pass two people heaving an AI-powered humanoid robot up the stairs to clean someone’s rooms. Go to

中文介绍 MIT 科技评论的「AI 炒作指数」显示 AI 无处不在:Google 大力推广 AI Overviews,尽管导致数十亿美元的广告收入损失;日常生活中也可见 AI 驱动的机器人等应用。

This smell startup says its AI nose is as good as a human’s

In a lab near Newark Airport in New Jersey, a team of around 30 experienced human sniffers are slowly working their way through every odor on Earth. Their careful notes—“nutty,” “peachy,” “marine”—are cross-referenced with a chemical analysis to create a rich set of training data for what AI startup

中文介绍 气味初创公司 Osmo 声称其 AI 鼻子可媲美人类嗅觉。公司在新泽西州纽瓦克机场附近建立实验室,由约 30 名经验丰富的嗅辨师系统记录地球上的气味特征,并结合化学分析构建训练数据。

The Silent Consensus

Dean Ball, Head of Strategic Futures at OpenAI, examines how more capable AI could reshape society and how people can retain control of their institutions.

中文介绍 OpenAI 战略未来主管 Dean Ball 探讨更强 AI 能力如何重塑社会,并思考人们如何保持对机构的控制权。

市场总览

整体市场技术面显著偏弱。美股方面,SPY 跌 1.96%、QQQ 跌 2.19%,两者 RSI 分别为 31.6 和 33.9,接近超卖,价格均低于 SMA20 和 SMA50,显示短期下行压力。加密市场极度恐慌,恐慌贪婪指数 12,总市值 2.14T 美元(24h -4.65%),BTC-USD RSI 20.5、ETH-USD RSI 23.1、SOL-USD RSI 27,均处超卖,价格远离均线,动量偏空。中概股分化,BABA 和 0700.HK 呈空头排列、RSI 超卖(24.5 和 26.7),而 PDD 逆势涨 11.13% 且价格站上 SMA20/SMA50,相对强势。商品外汇方面,黄金期货 RSI 22.4 超卖、价格跌破 SMA200,WTI 原油 RSI 34.9 中性偏弱,美元指数 RSI 68.9 接近超买,显示美元强势。整体而言,多数资产处于下行或超卖状态,仅 VIX(涨 52.09% 周度)和美元指数显示上行信号。

今日关注

NVDA Nvidia
偏下行

RSI 14 为 19.6,处于超卖区,且价格接近 52 周低点(距低点仅 0.19%)。MACD 柱状图 -17.76,信号线 -14.05,显示下行动量强劲。价格 145.98 低于 SMA20(185.73)和 SMA50(208.86),趋势偏弱。

BABA 阿里巴巴 (BABA)
偏下行

RSI 14 为 24.5,超卖,且价格接近 52 周低点(距低点 0.07%)。MACD 死叉(-4.65 vs -2.92),空头排列,价格 83.31 低于所有均线(SMA20 99.26、SMA50 105.4、SMA200 121.59)。近 5 日跌 17.94%,下行趋势明显。

PDD 拼多多 (PDD)
中性

RSI 14 为 42.9,接近中性区域。近 1 日涨 11.13%,近 5 日涨 8.85%,但 MACD 柱状图 -0.54 略低于信号线 -0.18,显示动量尚未转正。价格 86.00 高于 SMA20(79.47)和 SMA50(79.24),但低于 SMA200(89.44),多空拉锯。

^VIX VIX 恐慌指数
偏上行

RSI 14 为 64.1,接近超买但未进入。近 1 日涨 15.8%,近 5 日涨 52.09%,MACD 金叉(1.96 vs 1.35)。价格 28.44 高于 SMA20(20.88)、SMA50(18.03)和 SMA200(18.57),显示波动率上行趋势明显。

DX-Y.NYB 美元指数 DXY
偏上行

RSI 14 为 68.9,接近超买但未进入。价格 103.83 接近 52 周高点(距高点 0.11%),且高于所有均线(SMA20 102.57、SMA50 101.89、SMA200 100.01)。MACD 柱状图 0.58 高于信号线 0.43,多头排列,上行趋势稳定。

全部资产

^VIX

VIX 恐慌指数

$28.44 +15.80%
5 日
+52.09%
距 52w 高
-19.4%
RSI(14)
64.1
趋势
中性
SMA 20 / 50 / 200
20.88 / 18.03 / 18.57
MACD / 信号
1.961 / 1.352
MACD 金叉 (2 天前)

^TNX

10Y 美债收益率 (%)

$5.05 -0.28%
5 日
-1.00%
距 52w 高
-5.8%
RSI(14)
51.8
趋势
中性
SMA 20 / 50 / 200
4.99 / 5.08 / 4.60
MACD / 信号
0.008 / -0.010

DX-Y.NYB

美元指数 DXY

$103.83 -0.06%
5 日
+0.41%
距 52w 高
-0.1%
RSI(14)
68.9
趋势
多头
SMA 20 / 50 / 200
102.57 / 101.89 / 100.01
MACD / 信号
0.580 / 0.433
接近 52 周高多头排列

SPY

S&P 500 ETF

$715.87 -1.96%
5 日
-5.45%
距 52w 高
-8.4%
RSI(14)
31.6
趋势
中性
SMA 20 / 50 / 200
751.32 / 761.67 / 732.69
MACD / 信号
-9.285 / -6.032

QQQ

Nasdaq 100 ETF

$645.04 -2.19%
5 日
-6.27%
距 52w 高
-15.4%
RSI(14)
33.9
趋势
中性
SMA 20 / 50 / 200
683.33 / 713.09 / 683.00
MACD / 信号
-16.543 / -14.328

AAPL

Apple

$349.21 -0.92%
5 日
-1.60%
距 52w 高
-3.2%
RSI(14)
53.0
趋势
多头
SMA 20 / 50 / 200
347.74 / 341.60 / 302.91
MACD / 信号
3.544 / 3.622
MACD 死叉 (今天)多头排列

MSFT

Microsoft

$498.30 -1.97%
5 日
-4.72%
距 52w 高
-10.9%
RSI(14)
39.2
趋势
中性
SMA 20 / 50 / 200
524.96 / 520.06 / 443.18
MACD / 信号
-3.647 / 0.543

NVDA

Nvidia

$145.98 -3.30%
5 日
-12.16%
距 52w 高
-40.0%
RSI(14)
19.6
趋势
中性
SMA 20 / 50 / 200
185.73 / 208.86 / 203.58
MACD / 信号
-17.762 / -14.047
RSI 超卖接近 52 周低

GOOGL

Alphabet

$272.66 -3.46%
5 日
-8.72%
距 52w 高
-33.3%
RSI(14)
30.6
趋势
空头
SMA 20 / 50 / 200
292.20 / 323.76 / 337.30
MACD / 信号
-12.763 / -12.928
接近 52 周低空头排列

TSLA

Tesla

$374.41 -2.43%
5 日
-5.80%
距 52w 高
-24.9%
RSI(14)
47.2
趋势
空头
SMA 20 / 50 / 200
386.12 / 376.03 / 381.92
MACD / 信号
1.164 / 2.542
空头排列

META

Meta

$596.22 -3.66%
5 日
-5.18%
距 52w 高
-23.5%
RSI(14)
41.4
趋势
中性
SMA 20 / 50 / 200
625.29 / 679.36 / 629.50
MACD / 信号
-14.652 / -16.163
加密恐慌贪婪
12
极度恐慌
加密总市值
$2.14 T
-4.65% / 24h
BTC 主导率
58.0%
ETH 9.9%
24h 成交量
$79.4 B
活跃币 21,366

BTC-USD

Bitcoin

$61,792.42 -1.90%
5 日
-3.64%
距 52w 高
-36.9%
RSI(14)
20.5
趋势
中性
SMA 20 / 50 / 200
71,929.73 / 78,142.91 / 72,636.17
MACD / 信号
-4,428.876 / -3,420.912
RSI 超卖

ETH-USD

Ethereum

$1,734.44 -2.27%
5 日
-1.96%
距 52w 高
-49.7%
RSI(14)
23.1
趋势
中性
SMA 20 / 50 / 200
2,061.73 / 2,306.40 / 2,132.35
MACD / 信号
-159.048 / -132.136
RSI 超卖

SOL-USD

Solana

$72.58 -1.42%
5 日
+3.42%
距 52w 高
-51.0%
RSI(14)
27.0
趋势
中性
SMA 20 / 50 / 200
83.69 / 98.14 / 88.57
MACD / 信号
-7.498 / -6.906
RSI 超卖

BABA

阿里巴巴 (BABA)

$83.31 -3.85%
5 日
-17.94%
距 52w 高
-54.0%
RSI(14)
24.5
趋势
空头
SMA 20 / 50 / 200
99.26 / 105.40 / 121.59
MACD / 信号
-4.652 / -2.922
MACD 死叉 (4 天前)RSI 超卖接近 52 周低空头排列

PDD

拼多多 (PDD)

$86.00 +11.13%
5 日
+8.85%
距 52w 高
-31.4%
RSI(14)
42.9
趋势
中性
SMA 20 / 50 / 200
79.47 / 79.24 / 89.44
MACD / 信号
-0.540 / -0.179

JD

京东 (JD)

$25.64 +0.83%
5 日
-7.94%
距 52w 高
-25.2%
RSI(14)
35.1
趋势
空头
SMA 20 / 50 / 200
26.76 / 26.77 / 28.48
MACD / 信号
-0.050 / 0.158
MACD 死叉 (1 天前)空头排列

0700.HK

腾讯控股 (0700.HK)

HK$343.60 -1.38%
5 日
-11.12%
距 52w 高
-46.2%
RSI(14)
26.7
趋势
空头
SMA 20 / 50 / 200
391.20 / 410.76 / 463.21
MACD / 信号
-13.426 / -7.982
MACD 死叉 (3 天前)RSI 超卖接近 52 周低空头排列

GC=F

黄金期货

$3,546.20 -3.78%
5 日
-8.75%
距 52w 高
-36.5%
RSI(14)
22.4
趋势
空头
SMA 20 / 50 / 200
3,943.69 / 4,111.48 / 4,444.83
MACD / 信号
-118.505 / -81.995
RSI 超卖接近 52 周低空头排列

CL=F

WTI 原油期货

$83.16 -0.06%
5 日
-6.10%
距 52w 高
-30.4%
RSI(14)
34.9
趋势
中性
SMA 20 / 50 / 200
91.53 / 94.13 / 88.29
MACD / 信号
-3.304 / -2.332

USDCNY=X

美元 / 人民币

¥6.67 +0.04%
5 日
+0.00%
距 52w 高
-7.2%
RSI(14)
45.1
趋势
空头
SMA 20 / 50 / 200
6.67 / 6.69 / 6.78
MACD / 信号
-0.005 / -0.007
接近 52 周低空头排列
风险提示

本报告仅基于公开行情数据的技术指标解读,不构成投资建议。过去走势不代表未来表现,市场可能因外部事件迅速变化,读者需结合自身风险承受能力独立判断。仅供技术指标解读参考。

Thousands protest Colombia’s tax plans amid row over US military role

Trade unions, students and opposition groups demonstrated in Bogota and other major cities on Thursday.

中文摘要 哥伦比亚工会、学生及反对派团体在波哥大等主要城市举行示威,抗议政府税收计划,并与美国军事角色问题引发争议。

Lula visits ailing Chief Raoni as Brazil recognises Indigenous land

Lula honours Raoni in Mato Grosso, celebrating his legacy and advancing protection for Kayapo and Yudja ancestral lands.

中文摘要 巴西总统卢拉探望年迈的卡亚波族酋长拉奥尼,并正式承认其原住民领地,以表彰其贡献并加强保护卡亚波和尤贾族祖传土地。

Hippo Fiona dies days after swallowing visitor’s hat

Hippo Fiona, who became a global sensation after being born six weeks premature in 2017, has died at Cincinnati Zoo.

中文摘要 辛辛那提动物园的河马菲奥娜去世,她2017年早产六周出生后成为全球明星。她在吞下游客帽子数日后死亡。

Trump nominee refuses to call Jan 6 an ‘attack’ at Senate hearing

Trump’s nominee to lead the US agency responsible for election security, refused to call the insurrection an attack.

中文摘要 特朗普提名的美国选举安全机构负责人人选在参议院听证会上拒绝将2021年1月6日国会骚乱称为「攻击」事件。

Meta rethinks 2030 climate goals as AI data centres drive emissions up

Rising data centre energy needs force Meta to review sustainability targets amid growing corporate climate retreat.

中文摘要 Meta因AI数据中心能耗上升导致排放增加,正在重新评估其2030年气候目标,此举反映企业在气候承诺上的倒退趋势。

Democrats say law firm gave $11.8m in free services to Trump administration

Three top Democrats called Trump's dealings with law firms a case of 'official coercion' and pledged investigations.

中文摘要 三位民主党高层表示,一家律师事务所向特朗普政府提供了价值1180万美元的免费服务,称此为「官方胁迫」,并承诺展开调查。

Haiti police say 34 gang members killed in capital raid

Haitian police say at least 34 gang members were killed in an operation in the capital city of Port-au-Prince.

中文摘要 海地警方称,在首都太子港的一次行动中,至少34名帮派成员被击毙。

El Salvador’s Bukele expected to formalise bid for third presidential term

The Salvadoran president has won popularity through a crime crackdown that has granted the state enormous power.

中文摘要 萨尔瓦多总统布克尔预计将正式宣布竞选第三个总统任期,其通过严厉打击犯罪获得高支持率,但赋予国家巨大权力。

AI Spurs Funding Needs as Top Japan Power Producer Weighs IPO

Japan’s biggest electricity producer said the artificial intelligence boom is supporting the case for investment in the power sector as it considers ways to raise funds for expansion, including a possible initial public offering.

中文摘要 日本最大电力生产商表示,人工智能热潮为电力行业投资提供了支撑。该公司正考虑多种筹资方式以支持扩张,包括可能进行首次公开募股(IPO)。

Copper Rout Deepens as AI Selloff Wreaks Havoc in Metals Markets

Copper crashed to its lowest level since March after a week of market turmoil fueled by deepening worries about global demand.

中文摘要 铜价跌至3月以来最低水平,因市场对全球需求的担忧加剧,引发一周的市场动荡。人工智能相关股票的抛售也波及金属市场。

Trump Offers Xi a Helping Hand on Escalator

Donald Trump, 80, offered Xi Jinping, 73, a hand while boarding an escalator, with Xi holding his forearm as he stepped up, during their third summit of the year in Shenzhen. The friendly gesture contrasted with Washington’s recent efforts to restrict China’s access to key AI technology. (Source: Bl

中文摘要 在深圳举行的年内第三次峰会上,特朗普在乘扶梯时向习近平伸手相扶,习近平扶住其前臂。这一友好举动与华盛顿近期限制中国获取关键AI技术的努力形成对比。

Honda Debuts Hybrid SUV for Asia in Pivot From Failed EV Bet

Honda Motor Co. unwrapped a new flagship SUV, the first of a range of gas-electric hybrid models the company is rolling out this decade as part of its pivot away from fully electric cars.

中文摘要 本田汽车推出一款全新旗舰SUV,这是其转向油电混合动力车型计划的首款产品。公司正逐步放弃纯电动汽车战略。

GTA 6 breaks records as sales top $2.3bn - industry estimate

Rockstar Games ended a 13-year wait with what is expected to be the biggest entertainment launch ever.

中文摘要 Rockstar Games结束了长达13年的等待,发布《侠盗猎车手6》。据行业估计,该游戏销售额已突破23亿美元,有望成为有史以来规模最大的娱乐产品发布。

Rising Bond Yields May Pose a Challenge to Nifty’s Sharp Rebound

The Nifty has rebounded almost 15% since early October, pushing valuations higher.

中文摘要 印度Nifty指数自10月初以来已反弹近15%,推动估值走高。但债券收益率上升可能对该指数的强劲反弹构成挑战。

Taiwan Stocks Face Worst Week Since 1991 Despite Support Steps

Taiwan’s tech-heavy stock market witnessed another selloff on Friday as the global artificial intelligence trade came under further strain, putting the island’s benchmark on course for its worst week since 1991.

中文摘要 台湾科技股周五再度遭遇抛售,受全球人工智能交易压力影响。基准股指本周跌幅可能创下自1991年以来的最差表现。

Hong Kong Stocks Fall 20% From 2026 High as AI Selloff Spreads

Hong Kong stocks fell toward a grim milestone as an AI-led selloff spread to engulf even the more defensive corners of the market.

中文摘要 香港股市跌向一个严峻的里程碑,因AI引发的抛售蔓延至市场防御性板块。恒生指数已从2026年高点下跌20%,进入技术性熊市。

Anthropic (ANT) IPO on Nasdaq Overcame AI Fears and Warnings

The pioneering AI lab saw unprecedented demand for its stock from both retail and institutional investors.

中文摘要 AI先驱公司Anthropic(ANT)在纳斯达克进行IPO,尽管市场存在对AI的担忧。该股获得散户和机构投资者的空前需求。

Frasers CFO Unleashes on Accent as Australian Bid Turns Personal

Frasers Group Plc’s pursuit of Australia’s Accent Group Ltd. turned personal after Frasers’ finance chief flew to the retailer’s annual meeting in Melbourne to publicly accuse it of under-performance and demand the chairman quit.

中文摘要 Frasers Group Plc对澳大利亚Accent Group Ltd.的收购演变为个人攻击。Frasers首席财务官飞往墨尔本参加Accent年度会议,公开指责其业绩不佳,并要求董事长辞职。

A thaw in India’s ties with Pakistan?

Plus, Amazon’s $7bn investment

中文摘要 本期内容涵盖印度与巴基斯坦关系解冻的可能性,以及亚马逊70亿美元投资计划。

Australian Pension Funds Confront Their Latest Rival: AI Agents

Top Australian pension funds are grappling with the potential proliferation of autonomous AI agents, a new source of competition in the lucrative retirement market which senior executives say is creating significant challenges.

中文摘要 澳大利亚大型养老基金正应对自主AI智能体可能带来的竞争。高管们表示,这些智能体将成为退休市场的新竞争对手,带来重大挑战。