单词学习：![[28f21cd5-e263-4de7-b935-7460321b4476.mp3]]

新闻播报：![[71f5c66a-bad1-439e-9e8f-24dd57c69b5e.mp3]]

OpenAI联合甲骨文斥资150亿美元建AI算力园区

简报：

• 2025年10月23日，OpenAI、甲骨文（Oracle）与Vantage Data Centers宣布在美国威斯consin州启动“Lighthouse”大型数据中心园区建设，该项目是全球AI基建“星际之门”（Stargate）计划的一部分。
• 项目总投资150亿美元，计划建设四个数据中心，提供约1吉瓦的算力，预计2028年竣工。
• 该园区将实现100%零排放能源供能，建设期间将创造约4000个工作岗位，完工后提供超1000个长期职位。
• Lighthouse项目是OpenAI与甲骨文等公司合作推进的“星际之门”计划的延续，该计划拟在美国新建五个AI数据中心，总投资超过4000亿美元。

相关链接：

• OpenAI、Oracle 再加码 AI 基建:150亿美元 Lighthouse 园区启动建设

OpenAI and Oracle Partner on dJyH6kDkfC5RlM16.answerText5 Billion AI Computing Campus

Briefing:

• On October 23, 2025, OpenAI, Oracle, and Vantage Data Centers announced the groundbreaking of the "Lighthouse" hyperscale data center campus in Wisconsin, USA. This project is a key component of the global AI infrastructure "Stargate" initiative. 🚀
• The project involves a total investment of dJyH6kDkfC5RlM16.answerText5 billion and plans for four data centers, providing approximately 1 gigawatt of computing power. Completion is anticipated by 2028.
• The campus will be powered by 100% zero-emission energy. It is projected to create around 4,000 jobs during the construction phase and over 1,000 long-term positions upon completion. 🌱
• The Lighthouse project is a continuation of the "Stargate" initiative, advanced by OpenAI, Oracle, and other companies. This broader plan aims to establish five new AI data centers across the U.S. with a total investment exceeding $400 billion. 💰

Related Links:

• OpenAI, Oracle Double Down on AI Infrastructure: dJyH6kDkfC5RlM16.answerText5 Billion Lighthouse Campus Launched

微软升级Windows 11 Copilot，新增语音唤醒、屏幕分析及AI代理功能

简报：

• 微软发布Windows 11重大更新，通过全新的Copilot和Agentic体验，旨在将每台PC转变为AI PC。
• 已向所有用户推出的新功能包括：支持“Hey, Copilot”语音唤醒的Copilot Voice，以及能分析屏幕内容并提供协助的Copilot Vision。
• 面向Windows测试人员的预览功能包括：可直接操作本地文件的Copilot Actions、集成OneDrive及Google服务的连接器、以及名为Manus的通用型AI代理，可执行创建网站等复杂任务。
• 微软还推出了专为游戏玩家打造的Gaming Copilot，并与华硕ROG Xbox Ally等设备合作，提供游戏推荐与操作协助。
• 微软强调所有新功能将以安全为核心，用户始终拥有控制权，相关功能默认关闭。

相关链接：

• Windows 11测试版为Copilot应用带来AI文件搜索功能
• Copilot AI可以看见你的整个Windows 11屏幕
• 微软Gaming Copilot AI承诺成为您在Windows 11游戏中的伙伴

Microsoft Upgrades Windows 11 Copilot with Voice Activation, Screen Analysis, and AI Agent Features

Briefing:

• Microsoft has released a major Windows 11 update, aiming to transform every PC into an AI PC with new Copilot and agentic experiences. 🚀
• New features rolled out to all users include Copilot Voice, supporting "Hey, Copilot" voice activation, and Copilot Vision, which can analyze screen content and offer assistance.
• Preview features for Windows Insiders include Copilot Actions, capable of directly operating local files; connectors integrating OneDrive and Google services; and a general-purpose AI agent named Manus, designed to perform complex tasks like website creation.
• Microsoft also introduced Gaming Copilot, specifically designed for gamers, partnering with devices like ASUS ROG Xbox Ally to provide game recommendations and operational assistance. 🎮
• Microsoft emphasizes that all new features prioritize security, with users always maintaining control, and relevant functionalities are off by default.

Related Links:

• Windows 11 test brings AI file search to the Copilot app
• Copilot AI can see your entire display on Windows 11, Mac should copy it
• Microsoft Gaming Copilot AI promises to be your Windows 11 gaming pal

字节跳动发布Seed 3D 1.0，可从单张图像生成仿真级3D模型

简报：字节跳动Seed团队于10月23日发布3D生成大模型Seed3D 1.0，该模型基于Diffusion Transformer架构，能够从单张图像端到端地生成包含精细几何、真实纹理和PBR材质的高质量仿真级3D模型，旨在为具身智能发展提供世界模拟器支持。

相关链接：

• 项目主页
• 在线体验入口

ByteDance Unveils Seed 3D 1.0: Generating Simulation-Grade 3D Models from Single Images 🚀

Brief: ByteDance's Seed team released its large 3D generation model, Seed3D 1.0, on October 23rd. Based on a Diffusion Transformer architecture, the model is capable of end-to-end generation of high-quality, simulation-grade 3D models—complete with intricate geometry, realistic textures, and PBR materials—from just a single image. Its goal is to provide world simulator support for the advancement of embodied AI. ✨

Related Links:

• Project Homepage
• Online Experience Portal

相关链接:

• https://www.aibase.com/zh/news/22212

谷歌发布Willow量子芯片，首度实现可验证量子优势

简报：

• 谷歌于10月22日宣布，其最新量子芯片Willow实现了全球首个可验证的量子优势，相关成果已发表于《自然》期刊。
• 在运行“量子回声”（Quantum Echoes）算法时，Willow芯片用2.1小时完成了全球最快超级计算机Frontier需要3.2年才能完成的任务，速度快约13,000倍。
• 该研究由新晋诺贝尔物理学奖得主Michel Devoret领衔，其关键技术基于量子调控与超导量子电路。
• 谷歌CEO桑达尔·皮查伊表示，这一里程碑是“量子计算走向实际应用的重要一步”。

相关链接：

• 谷歌再登《自然》!Willow量子芯片实现可验证量子优势，快13，000倍击败超级计算机

Google Releases Willow Quantum Chip, Achieving Verifiable Quantum Advantage for the First Time

Brief:

• Google announced on October 22 that its latest quantum chip, Willow, has achieved the world's first verifiable quantum advantage. The relevant results have been published in the journal Nature.
• When running the "Quantum Echoes" algorithm, the Willow chip completed a task in 2.1 hours that would take the world's fastest supercomputer, Frontier, 3.2 years to complete, making it about 13,000 times faster.
• The research was led by Michel Devoret, the newly-awarded Nobel laureate in physics, and its key technology is based on quantum control and superconducting quantum circuits.
• Google CEO Sundar Pichai said that this milestone is "an important step towards the practical application of quantum computing."

Related Links:

• Google Again Published in Nature! Willow Quantum Chip Achieves Verifiable Quantum Advantage, 13,000 Times Faster Than Supercomputer

腾讯开源混元世界模型1.1，实现秒级3D场景重建

简报：

• 腾讯于10月22日正式发布并开源混元世界模型1.1（WorldMirror），在多视图和视频输入支持、单卡部署以及生成速度等方面进行了重大升级
• 该模型能够在数秒内从视频或图片中生成专业级3D场景，采用纯前馈架构，对于8-32视图输入仅需1秒即可完成推理
• 混元世界模型1.1采用多模态先验引导机制，支持相机位姿、相机内参和深度图等多种信息注入，实现点云、深度图、相机参数、表面法线和新视角合成等多种3D几何预测
• 模型已在GitHub上开源，普通用户可通过HuggingFace Space在线体验，上传多视图图像或视频实时预览生成的3D场景
• 其前身混元世界模型1.0于今年7月发布，是业界首个兼容传统CG管线的开源可漫游世界生成模型

相关链接：

• 项目主页
• Github项目地址
• Hugging Face模型地址

Tencent Open-Sources Hunyuan World Model 1.1, Enabling Second-Level 3D Scene Reconstruction

Briefing:

• Tencent officially released and open-sourced Hunyuan World Model 1.1 (WorldMirror) on October 22nd, bringing significant upgrades in multi-view and video input support, single-GPU deployment, and generation speed. 🚀
• The model can generate professional-grade 3D scenes from videos or images in mere seconds. Utilizing a pure feedforward architecture, it completes inference for 8-32 view inputs in just 1 second.
• Hunyuan World Model 1.1 employs a multi-modal prior guidance mechanism, supporting the injection of various information such as camera pose, camera intrinsics, and depth maps. This enables diverse 3D geometric predictions, including point clouds, depth maps, camera parameters, surface normals, and novel view synthesis.
• The model has been open-sourced on GitHub. General users can experience it online via HuggingFace Space, uploading multi-view images or videos to preview the generated 3D scenes in real-time. 🌐
• Its predecessor, Hunyuan World Model 1.0, was released in July this year, and was the industry's first open-source navigable world generation model compatible with traditional CG pipelines. ✨

Related Links:

• Project Homepage
• GitHub Project
• Hugging Face Model