单词学习：![[3c4aa7b0-b625-4618-9941-137745b4fdab.mp3]]

新闻播报：![[5e24431c-2aa4-4d58-be2d-7114f247ad36.mp3]]

阿里瓴羊发布AgentOne平台，用AI智能体助力企业成为“超级公司”

简报： 在云栖大会上，阿里巴巴旗下企业服务品牌瓴羊发布了企业级AI智能体服务平台AgentOne，该平台聚焦数据、资金、人力密集的核心场景，旨在推动企业向主动智能转型升级，助力其成为AI时代的"超级公司"。

相关连接：

• 阿里瓴羊发布AgentOne平台 推动企业向"超级公司"转型

Alibaba Lingyang Launches AgentOne Platform: AI Agents to Empower Enterprises as "Super Companies" 🚀

Brief: At the Apsara Conference, Lingyang, Alibaba's enterprise service brand, officially unveiled AgentOne, an enterprise-grade AI agent service platform. This platform focuses on core scenarios intensive in data, capital, and human resources, aiming to drive enterprises towards proactive intelligent transformation and upgrading, thereby helping them become "super companies" in the AI era. 🧠💡

Related Links:

• 阿里瓴羊发布AgentOne平台 推动企业向"超级公司"转型

生数科技发布Vidu Q2，AI视频生成可实现逼真细微表情

简报：

• 生数科技于9月25日发布新一代图生视频大模型Vidu Q2，主打“细微表情生成”。
• 该模型能精准捕捉极致的表情变化，并结合推拉运镜技术，提升了视频生成速度与语义理解能力。
• Vidu Q2提供图生视频、首尾帧视频等模式，可调时长为2-8秒，并支持电影大片或快节奏闪电出片等风格。

相关链接：

• 生数科技推出 Vidu Q2，AI细微表情生成技术发布

Shengshu Technology Launches Vidu Q2: AI Video Generation Achieves Realistic Subtle Expressions

Brief:

• On September 25th, Shengshu Technology released its new generation image-to-video large model, Vidu Q2, focusing on "subtle expression generation." 🚀
• The model can precisely capture extreme facial expression changes and, combined with push-pull camera movement techniques, has improved video generation speed and semantic understanding capabilities. ✨
• Vidu Q2 offers modes like image-to-video and start/end frame video, with an adjustable duration of 2-8 seconds, and supports styles such as blockbuster films or fast-paced rapid production. 🎬

Related link:

• Shengshu Technology Launches Vidu Q2, AI Subtle Expression Generation Technology Released

OpenAI联手甲骨文软银，在美新建五大数据中心

简报：

• OpenAI宣布将在美国新建五个数据中心，分布于德克萨斯州、新墨西哥州、俄亥俄州及中西部地区，此举将使“星际之门（Stargate）”项目的总计算能力接近7吉瓦（GW）。
• 该计划是Stargate项目的一部分，这是一个由OpenAI、甲骨文（Oracle）、日本软银及阿联酋投资公司MGX等共同发起的价值5000亿美元的AI基础设施计划。
• 在此次扩建中，甲骨文将负责开发三个新数据中心并扩建现有设施，OpenAI与软银则共同开发另外两个。
• 作为战略合作方，芯片巨头英伟达（NVIDIA）承诺向OpenAI投资高达1000亿美元，并为该项目提供GPU支持。
• Stargate项目同时也在全球扩张，并已宣布计划在挪威、阿联酋、印度和日本等国建设国际数据中心。

相关链接：

• OpenAI 在美国新建五个数据中心，推动 Stargate 项目快速发展
• OpenAI 成 AI 基建狂魔，拟启动 1 万亿美元算力扩张计划
• 輝達與OpenAI戰略合作　砸3兆打造超級AI運算中心

OpenAI, Oracle, SoftBank Partner to Construct Five New Data Centers in the US

Briefing:

• OpenAI announced the construction of five new data centers in the US, located in Texas, New Mexico, Ohio, and the Midwest region. This expansion will bring the total computing capacity of the "Stargate" project close to 7 gigawatts (GW). 📍
• This initiative is part of the Stargate project, a massive $500 billion AI infrastructure plan jointly launched by OpenAI, Oracle, Japan's SoftBank, and UAE investment firm MGX. 💰
• In this expansion, Oracle will be responsible for developing three new data centers and expanding existing facilities, while OpenAI and SoftBank will jointly develop the other two.
• As a strategic partner, chip giant NVIDIA has pledged an investment of up to $100 billion in OpenAI and will provide GPU support for the project. 🚀
• The Stargate project is also expanding globally, with plans already announced for the construction of international data centers in countries like Norway, UAE, India, and Japan.

Related Links:

• OpenAI to Build Five New Data Centers in the US, Accelerating Stargate Project Development
• OpenAI Becomes an AI Infrastructure Behemoth, Plans $1 Trillion Computing Power Expansion
• NVIDIA and OpenAI Form Strategic Partnership, Invest $3 Trillion to Build Super AI Computing Center

OpenAI多模态AI智能体GPT-Alpha曝光，基于GPT-5模型

简报：

• 科技媒体bleepingcomputer于9月25日报道，OpenAI内部正在测试一款名为“GPT-Alpha”的AI智能体，该智能体基于GPT-5模型，为高级推理与工具使用而设计。
• 此次曝光源于OpenAI一次意外的更新推送，部分用户在“Alpha Models”选项下看到了名为“Agent with Truncation”的未发布功能。
• 根据信息，“GPT-Alpha”具备多模态处理能力，可在单一会话中处理文本、代码和图像，并能联网搜索、生成编辑图片、编写调试代码以及创建和修改文档。

相关链接：

• OpenAI 最强 AI 智能体曝光：基于 GPT-5，轻松驾驭写作、编程、生图等任务

OpenAI's Multimodal AI Agent GPT-Alpha Revealed, Based on GPT-5 Model

Brief: 📰

• Tech media outlet bleepingcomputer reported on September 25 that OpenAI is internally testing an AI agent named "GPT-Alpha." This agent is based on the GPT-5 model and is designed for advanced reasoning and tool utilization.
• This revelation stems from an accidental update pushed by OpenAI, where some users spotted an unreleased feature called "Agent with Truncation" under the "Alpha Models" option.
• According to the information, "GPT-Alpha" possesses multimodal processing capabilities. It can handle text, code, and images within a single session, perform web searches, generate and edit pictures, write and debug code, and create and modify documents. 🤖💡

Related Links:

• OpenAI's Strongest AI Agent Revealed: Based on GPT-5, Easily Masters Writing, Programming, Image Generation and Other Tasks

通义千问开源模型下载量突破6亿

简报：

• 在2025云栖大会上，阿里云CTO周靖人宣布，通义千问项目已开源超过300个模型，其累计下载量已突破6亿次。

相关链接：

• 阿里云CTO谈通义千问：开源超300模型下载量破6亿
• 阿里云CTO谈通义千问：开源超300模型下载量破6亿

Tongyi Qianwen Open-Source Model Downloads Surpass 600 Million

Briefing:

• At the 2025 Apsara Conference, Alibaba Cloud CTO Zhou Jingren announced that the Tongyi Qianwen project has open-sourced over 300 models, with cumulative downloads now exceeding 600 million. 🎉🚀

Related Links:

• Alibaba Cloud CTO Discusses Tongyi Qianwen: Over 300 Models Open-Sourced, 600 Million Downloads
• Alibaba Cloud CTO Discusses Tongyi Qianwen: Over 300 Models Open-Sourced, 600 Million Downloads 📈

平头哥AI芯片性能追赶辉达，获联通亿元订单

简报：

• 阿里巴巴旗下半导体公司“平头哥”研发的最新AI芯片PPU，性能被指已逼近辉达（NVIDIA）的降级版H20芯片。
• PPU芯片采用HBM2e显存，容量达96G，并拥有与CUDA生态兼容的优势，可降低客户转换成本。
• 平头哥已获得中国联通“三江源绿电智算中心”订单，将供应超过1.6万张运算卡，涉及金额达亿元人民币。
• 该进展被视为中国应对美国芯片出口限制、推动AI算力国产替代的关键一步，并获央视《新闻联播》报道。

相关链接：

• 阿里平頭哥晶片崛起　性能逼近輝達H20獲官方億元大單

T-Head AI Chip Performance Nears NVIDIA's, Secures Hundred-Million Yuan Order from China Unicom

Brief:

• Alibaba's semiconductor company, "T-Head" (Pingtouge), has developed its latest AI chip, PPU, whose performance is reported to be approaching NVIDIA's downgraded H20 chip. 🚀
• The PPU chip utilizes HBM2e memory with a capacity of 96GB. It also offers the advantage of compatibility with the CUDA ecosystem, which can significantly reduce customer conversion costs. 💪
• T-Head has secured an order from China Unicom's "Three-River-Source Green Electricity Smart Computing Center." This deal involves supplying over 16,000 computing cards, valued at hundreds of millions of RMB. 💰
• This development is considered a crucial step for China in countering US chip export restrictions and promoting domestic substitution of AI computing power, a breakthrough also reported by CCTV's "Xinwen Lianbo."

Related Link:

• Alibaba T-Head Chip Rises, Performance Nears NVIDIA H20, Secures Official Hundred-Million Yuan Order

Meta推出ARE与Gaia2平台，以动态环境测试智能体适应力

简报：

• Meta推出了新的评估平台Agents Research Environment（ARE）和基准模型Gaia2，旨在创建与现实世界相似的动态环境来测试智能体。
• Gaia2着重评估智能体在面对不断变化的条件、截止日期、API故障及模糊指令时的适应能力，并支持评估智能体间的协作能力。
• 该平台采用异步评估方式，模拟时间流逝来测量智能体对新事件的响应能力；目前的测试显示，OpenAI的GPT-5在该基准上表现领先。

相关链接：

• 官方博客

Meta Launches ARE and Gaia2 Platforms to Test Agent Adaptability in Dynamic Environments

Briefing:

• Meta has launched new evaluation platforms, Agents Research Environment (ARE) and the benchmark model Gaia2, designed to create dynamic, real-world-like environments to test AI agents. 🚀
• Gaia2 focuses on assessing agents' adaptability when facing constantly changing conditions, deadlines, API failures, and ambiguous instructions, and also supports evaluating inter-agent collaboration. 🤝
• The platform utilizes asynchronous evaluation, simulating the passage of time to measure agents' responsiveness to new events; current tests indicate OpenAI's GPT-5 leads on this benchmark. 📈

Related Links:

• Official Blog

英伟达开源面部动画AI，助力游戏及3D应用开发

简报：

• 英伟达于9月24日宣布开源其生成式AI面部动画模型Audio2Face，旨在加速游戏和3D应用中AI智能虚拟角色的开发。
• 该技术通过分析音频特征，实时驱动虚拟角色的面部动作，生成精准的口型同步和自然表情，支持离线渲染和实时流式处理两种模式。
• 开源内容涵盖了模型、软件开发工具包(SDK)、完整的训练框架以及适用于Unreal Engine和Autodesk Maya的插件。
• 目前该技术已被《异形：侠盗入侵进化版》和《切尔诺贝利人 2：禁区》等游戏项目采用，以简化制作流程并提升角色真实感。

相关链接：

• 英伟达开源 Audio2Face 模型：AI 实时生成面部动画，多语言口型同步

NVIDIA Open-Sources Facial Animation AI to Boost Game and 3D App Development

Brief:

• NVIDIA announced on September 24th its decision to open-source its generative AI facial animation model, Audio2Face. This move aims to accelerate the development of AI-powered virtual characters in games and 3D applications. 🚀
• The technology analyzes audio features to drive virtual characters' facial movements in real-time, generating precise lip-syncing and natural expressions. It supports both offline rendering and real-time streaming modes. 🗣️
• The open-source release includes the model, a Software Development Kit (SDK), a complete training framework, and dedicated plugins for Unreal Engine and Autodesk Maya.
• Currently, the technology has been adopted by game projects like Aliens: Dark Descent and Chernobylite 2: The Exclusion Zone, aiming to streamline production workflows and boost character realism. ✨

Related Links:

• NVIDIA Open-Sources Audio2Face Model: AI Real-time Facial Animation, Multi-language Lip-sync