单词学习：![[5938bc7b-88ea-4f76-8484-f0a962d9b421.mp3]]

新闻播报：![[d7e39ecd-e24a-4f7f-a41a-c7298a3f9afb.mp3]]

月之暗面完成数亿美元新融资，Kimi k2模型获国际认可

简报：

• 月之暗面（Moonshot AI）即将完成新一轮数亿美元融资，距上一轮约3亿美元融资（2024年8月）不久
• 公司今年7月凭借全新升级的Kimi k2模型强势回归，扭转了因DeepSeek爆发而沉寂的局面
• 美国知名投资人Chamath Palihapitiya在播客《All In》中表示，其公司已将主要工作切换到Kimi k2
• Vercel首席执行官Guillermo Rauch发文称，Kimi k2在内部智能体测试中性能超越了GPT-5和Claude 4.5
• Kimi k2的卓越表现使月之暗面重新成为资本和技术圈焦点，站上全球AI舞台中央

相关链接：

• Kimi k2性能被赞超越GPT-5，月之暗面再获数亿美元重磅融资

Moonshot AI Completes Hundreds of Millions in New Funding Round, Kimi k2 Model Gains International Recognition 🚀

Brief:

• Moonshot AI is set to complete a new funding round worth hundreds of millions of dollars, following shortly after its previous ~$300 million financing round in August 2024
• The company made a strong comeback in July this year with its newly upgraded Kimi k2 model, reversing the quiet period caused by DeepSeek's emergence
• Prominent U.S. investor Chamath Palihapitiya stated on the "All In" podcast that his company has switched its primary operations to Kimi k2 💡
• Vercel CEO Guillermo Rauch posted that Kimi k2 outperformed GPT-5 and Claude 4.5 in internal agent testing
• Kimi k2's exceptional performance has propelled Moonshot AI back into the spotlight of capital and tech circles, positioning it at the center of the global AI stage 🌟

Related Link:

• Kimi k2 Performance Praised as Surpassing GPT-5, Moonshot AI Secures Major Funding of Hundreds of Millions

字节跳动发布Seed 3D 1.0：单图生成仿真级3D模型

简报：

• 字节跳动Seed团队推出3D生成大模型Seed 3D 1.0，可从单张图像端到端生成高质量仿真级3D模型，包含精细几何、真实纹理和PBR材质
• 该模型采用Diffusion Transformer架构，1.5B参数规模超越业界3B参数模型，在几何质量、材质纹理、视觉清晰度等维度获得较好评分
• Seed 3D 1.0支持分步生成策略构建完整3D场景，可无缝导入Isaac Sim等仿真引擎，为具身智能和机器人训练提供多样化操作场景
• 团队构建了完整的三阶段数据处理管线，将海量异构原始3D数据转化为高质量训练集
• 未来计划引入多模态大语言模型提升3D生成质量，推动模型在世界模拟器中的大规模应用

相关链接：

• 项目主页
• 体验入口

ByteDance Releases Seed 3D 1.0: Single-Image to Simulation-Grade 3D Model Generation 🚀

Brief:

• ByteDance's Seed team has launched Seed 3D 1.0, a 3D generation foundation model that performs end-to-end generation of high-quality simulation-grade 3D models from a single image, featuring fine-grained geometry, realistic textures, and PBR materials 🎨
• The model employs a Diffusion Transformer architecture with 1.5B parameters, surpassing industry 3B parameter models in performance, achieving superior scores across dimensions including geometric quality, material textures, and visual clarity
• Seed 3D 1.0 supports step-by-step generation strategies to construct complete 3D scenes, seamlessly integrating with simulation engines like Isaac Sim to provide diverse operational scenarios for embodied intelligence and robot training 🤖
• The team has built a comprehensive three-stage data processing pipeline that transforms massive heterogeneous raw 3D data into high-quality training datasets
• Future plans include introducing multimodal large language models to enhance 3D generation quality and drive large-scale application of the model in world simulators

Related Links:

• Project Homepage
• Experience Portal

OpenAI收购Sky团队，ChatGPT将深度整合macOS系统

简报：

• OpenAI宣布收购Mac平台AI语言界面应用Sky背后的团队Software Applications Incorporated（SAI），整个Sky团队将转投OpenAI
• 收购目标是让ChatGPT更自然地融入Mac日常工作流程，Sky应用具备理解上下文、适应用户操作并直接与其他应用协同工作的能力
• 被收购的SAI团队曾开发知名的Workflow应用，该应用后被苹果收购并更名为现在的Shortcuts（快捷指令）
• 此次收购交易已由独立董事会委员会审核通过，OpenAI首席执行官Sam Altman此前曾通过投资公司持有Sky的被动股份
• 此举与OpenAI推动ChatGPT驱动的自动化广泛战略高度契合

相关链接：

• 直击Mac桌面!OpenAI收购Sky团队，ChatGPT即将深度嵌入macOS工作流

OpenAI Acquires Sky Team, ChatGPT to Deeply Integrate with macOS 🚀

Brief:

• OpenAI announces acquisition of Software Applications Incorporated (SAI), the team behind Sky, a Mac platform AI language interface application. The entire Sky team will join OpenAI 💼
• The acquisition aims to enable ChatGPT to integrate more naturally into daily Mac workflows. The Sky app features context understanding capabilities, adapts to user operations, and works directly with other applications
• The acquired SAI team previously developed the renowned Workflow app, which was later acquired by Apple and rebranded as the current Shortcuts app
• The acquisition has been reviewed and approved by an independent board committee. OpenAI CEO Sam Altman previously held passive shares in Sky through an investment company 📱
• This move aligns closely with OpenAI's broader strategy to promote ChatGPT-driven automation ✨

Related Links:

• Direct Access to Mac Desktop! OpenAI Acquires Sky Team, ChatGPT Set to Deeply Embed into macOS Workflow

OpenAI新浏览器Atlas存在即时注入漏洞，安全主管公开示警

简报：

• OpenAI安全主管Dane Stuckey公开警告，全新智能浏览器ChatGPT Atlas面临"即时注入"(Prompt Injection)攻击威胁，短期内难以彻底根除
• 攻击者可通过在网页、邮件中嵌入恶意指令，诱导AI执行非预期操作，可能导致购买行为被干扰、电子邮件内容或登录凭证等私人数据被窃取
• OpenAI已部署两项防御措施：一是"注销模式"阻断AI访问用户数据，二是"监视模式"要求用户在敏感网站关键操作中进行人工确认
• 尽管OpenAI已开展大规模安全测试并引入多重防护机制，但安全主管坦言"即时注入"仍是棘手的开放性问题
• 团队正加速研发更多防护功能与快速响应系统，以便在面对潜在攻击时第一时间干预和修复

相关链接：

• OpenAI 内部警示
  Atlas 面临"即时注入"安全威胁
• OpenAI's ChatGPT Atlas Browser Has a Big Problem

OpenAI's New Atlas Browser Faces Prompt Injection Vulnerability, Security Chief Issues Public Warning ⚠️

Brief:

• OpenAI's Security Chief Dane Stuckey has publicly warned that the new intelligent browser ChatGPT Atlas faces "Prompt Injection" attack threats that will be difficult to completely eliminate in the short term
• Attackers can embed malicious instructions in web pages and emails to trick the AI into executing unintended operations, potentially leading to interference with purchasing behavior and theft of private data such as email content or login credentials 🔓
• OpenAI has deployed two defensive measures: first, a "logged-out mode" that blocks AI access to user data, and second, a "monitoring mode" that requires manual user confirmation for critical operations on sensitive websites
• Although OpenAI has conducted large-scale security testing and introduced multiple protection mechanisms, the security chief admits that "prompt injection" remains a thorny open-ended problem
• The team is accelerating the development of additional protective features and rapid response systems to enable immediate intervention and fixes when facing potential attacks 🛡️

Related Links:

• OpenAI Internal Warning: ChatGPT Atlas Faces "Prompt Injection" Security Threat
• OpenAI's ChatGPT Atlas Browser Has a Big Problem

Google向公众免费开放内部AI培训资源，推出"Google Skills"学习平台

简报：

• Google于2025年10月23日推出"Google Skills"学习平台，首次系统性地向全球公众开放内部AI知识和实训资源
• 平台整合了DeepMind AI研究团队、Google Cloud、Gemini AI模型开发团队以及Google Education的精华内容
• 对Google Cloud用户完全免费，非Cloud用户每月仅需$29订阅，相比大学同类课程5万至6万美元学费实现"零门槛"学习
• 提供超过700个真实云环境实操实验室、内置Gemini Code Assist编程助手，强调"动手学习"的实战训练模式
• 完成课程可获得官方数字证书和技能徽章，被超过82%招聘方认可，并与150多家合作企业建立直通就业渠道

相关链接：

• Google重磅推出"Google Skills"平台，内部AI知识向公众免费开放

Google Opens Internal AI Training Resources to the Public for Free, Launches "Google Skills" Learning Platform 🚀

News Brief:

• Google launched the "Google Skills" learning platform on October 23, 2025, systematically opening its internal AI knowledge and hands-on training resources to the global public for the first time 📚
• The platform integrates premium content from DeepMind AI research team, Google Cloud, Gemini AI model development team, and Google Education
• Completely free for Google Cloud users, with only a $29 monthly subscription for non-Cloud users—achieving "zero-barrier" learning compared to$50,000-$60,000 tuition fees for similar university courses
• Offers over 700 hands-on labs in real cloud environments with built-in Gemini Code Assist programming assistant, emphasizing a "learning by doing" practical training approach
• Upon course completion, participants receive official digital certificates and skill badges recognized by over 82% of recruiters, with direct employment channels established through partnerships with more than 150 companies 💼

Related Link:

• Google Unveils "Google Skills" Platform, Opening Internal AI Knowledge to the Public for Free

微软Copilot秋季重大更新：支持32人协作群聊，引入长期记忆与浏览器AI模式

简报：

• 微软推出Copilot秋季重大更新，新增群聊功能最多支持32人实时协作，AI可自动总结对话、提供选项并分配任务
• 引入长期记忆功能，用户可保存跑步目标、生日等关键信息，Copilot将在未来对话中自动调用这些数据
• 服务整合能力扩展至Outlook、Gmail、Google Drive和日历，用户可通过自然语言搜索文档和邮件，需主动批准访问权限
• Microsoft Edge浏览器推出"Copilot模式"，可分析打开的标签页、总结比较内容、填写表格或提供预订建议
• 正式推出AI形象吉祥物"Mico"，被视为经典助手Clippy的精神继承者，能对语音做出反应并显示情绪
• 新增"实时学习"辅导模式和语音控制功能，所有功能现已在美国上线，将在未来几周推广至其他国家

相关链接：

• 微软 Copilot 新增群聊、记忆与 Edge AI 模式：最多可让 32 人实时协作
• 微软推出全新AI形象 Mico，Clippy 回归"变身"人工智能好友

Microsoft Copilot Fall Major Update: Supports 32-Person Collaborative Group Chat, Introduces Long-Term Memory & Browser AI Mode

Brief:

• Microsoft launches major fall update for Copilot, adding group chat functionality that supports up to 32 people in real-time collaboration, with AI automatically summarizing conversations, providing options, and assigning tasks 💬
• Introduces long-term memory feature allowing users to save key information such as running goals and birthdays, which Copilot will automatically recall in future conversations
• Service integration capabilities expand to Outlook, Gmail, Google Drive, and Calendar, enabling users to search documents and emails through natural language, with access requiring user approval
• Microsoft Edge browser launches "Copilot Mode," capable of analyzing open tabs, summarizing and comparing content, filling forms, or providing booking suggestions
• Officially unveils AI mascot character "Mico," regarded as the spiritual successor to classic assistant Clippy, which can react to voice commands and display emotions 🤖
• Adds "Learn Live" tutoring mode and voice control functionality, with all features now available in the United States and rolling out to other countries in the coming weeks 🌍

Related Links:

• Microsoft Copilot Adds Group Chat, Memory & Edge AI Mode: Supports Real-Time Collaboration for Up to 32 People
• Microsoft Introduces New AI Character Mico, Clippy Returns "Transformed" as AI Companion

阿里Qwen Deep Research升级：一键生成多模态研究内容

简报：

• 阿里巴巴Qwen团队对Deep Research工具进行重大升级，可在网页版Qwen Chat中激活使用
• 新版支持生成带引文的综合研究报告，并可一键创建交互式网页与多说话者播客，仅需1-2次点击完成
• 由Qwen自研模型体系提供支持，包括Qwen3-Coder(代码生成)、Qwen-Image(图像渲染)与Qwen3-TTS(语音合成)
• 演示显示模型可主动发现数据差异、计算复合增长率，并生成图文并茂的可视化报告
• 该工具为Qwen专有托管服务，用户无需配置基础设施即可直接使用，实现"文字、视觉、听觉"三位一体的成果展示

相关链接：

• 阿里 Qwen 推出 Deep Research:一键生成报告、网页与播客

Alibaba Qwen Deep Research Upgrade: One-Click Generation of Multimodal Research Content 🚀

Brief:

• Alibaba's Qwen team has launched a major upgrade to its Deep Research tool, now accessible via the web-based Qwen Chat platform
• The new version supports generating comprehensive research reports with citations and can create interactive webpages and multi-speaker podcasts with just 1-2 clicks 📊
• Powered by Qwen's proprietary model ecosystem, including Qwen3-Coder (code generation), Qwen-Image (image rendering), and Qwen3-TTS (speech synthesis)
• Demonstrations show the model can autonomously identify data discrepancies, calculate compound growth rates, and generate visually rich reports with charts and graphics
• The tool operates as a Qwen-exclusive hosted service, allowing users to access it directly without infrastructure configuration, delivering a "text, visual, and audio" integrated research experience 🎯

Related Link:

• Alibaba Qwen Launches Deep Research: One-Click Generation of Reports, Webpages, and Podcasts