单词学习：![[dd27d014-99f5-4f93-87e2-f5e0e22df26a.mp3]]

新闻播报：![[747c62ce-8a9b-4989-b384-29ddb05a2503.mp3]]

蚂蚁开源万亿参数大模型Ling-1T，号称多项指标居榜首

简报：

• 蚂蚁集团于10月9日发布万亿参数通用语言模型Ling-1T，并宣布开源。
• 该模型是蚂蚁百灵大模型Ling 2.0系列的首款旗舰产品，也是团队迄今推出的规模最大、能力最强的模型。
• 官方测评显示，Ling-1T在代码生成、软件开发、数学、逻辑推理等多项高难度基准测试中取得领先成绩，多项指标位居开源模型榜首。

相关链接：

• 蚂蚁发布万亿参数旗舰模型 Ling-1T并开源

Ant Group Open-sources Trillion-Parameter Large Model Ling-1T, Claims Top Spot in Multiple Metrics ✨

Brief:

• Ant Group released its trillion-parameter general language model, Ling-1T, on October 9 and announced its open-sourcing.
• The model is the first flagship product in Ant Group's Bailian Ling 2.0 large model series, and also the largest and most powerful model the team has launched to date.
• Official evaluations show that Ling-1T achieved leading results in multiple high-difficulty benchmark tests, including code generation, software development, mathematics, and logical reasoning, ranking first among open-source models in multiple metrics. 🚀🏆

Related Links:

• Ant Group Releases Trillion-Parameter Flagship Model Ling-1T and Open-sources It

OpenAI开发者日：ChatGPT升级为应用平台并发布GPT-5 Pro

简报:

• OpenAI在开发者日宣布将ChatGPT转型为集成第三方应用的平台，并发布Apps SDK，允许开发者创建直接在聊天界面中运行的交互式应用。
• 推出用于构建自主AI代理的工具包AgentKit，其核心是可视化工具Agent Builder，可帮助开发者创建能替用户执行复杂任务的代理。
• 发布顶级模型GPT-5 Pro的API，支持40万token的上下文窗口，同时推出成本更低的gpt-realtime-mini和文生视频模型Sora2的API。
• 公司披露ChatGPT周活跃用户已超过8亿，有超过400万开发者在该平台进行开发。

相关链接:

• OpenAI 开发者日重磅发布：ChatGPT 平台化，推出自主 AI 代理和顶级模型

OpenAI Developer Day: ChatGPT Upgrades to Application Platform and GPT-5 Pro Released

Brief:

• OpenAI announced at its Developer Day that ChatGPT will transform into a platform integrating third-party applications, releasing the Apps SDK to allow developers to create interactive applications that run directly within the chat interface. 🚀
• Introduced AgentKit, a toolkit for building autonomous AI agents, with its core being the visual tool Agent Builder, designed to help developers create agents capable of performing complex tasks on behalf of users.
• Released the API for the top-tier model GPT-5 Pro, supporting a 400,000 token context window. Also launched APIs for the more cost-effective gpt-realtime-mini and the text-to-video model Sora2. ✨
• The company disclosed that ChatGPT now has over 800 million weekly active users and more than 4 million developers building on the platform. 📈

Related Link:

• OpenAI Developer Day's Major Announcements: ChatGPT as a Platform, Autonomous AI Agents, and Top-Tier Models

OpenAI万亿美元算力投资曝光，联手多家巨头豪赌未来

简报：

• OpenAI已签署价值约1万亿美元的计算能力协议，将从英伟达（Nvidia）、AMD、甲骨文（Oracle）和CoreWeave等公司获得超过20千兆瓦的AI算力。
• 此次投资是在公司面临巨额亏损的背景下进行的，预计今年亏损约100亿美元，而年收入仅为120亿美元左右，但其CEO萨姆·奥特曼表示盈利并非当前的首要担忧。
• 为支撑该交易，OpenAI采用了巧妙的循环融资结构，例如英伟达计划向OpenAI投资1000亿美元，OpenAI再用这笔资金购买其芯片。
• 在与AMD价值数百亿美元的协议中，OpenAI将获得6千兆瓦的计算能力，并被授予认股权证，允许其以极低价格购买最多10%的AMD股份。
• 有硅谷投资者指出，OpenAI的资本密集程度远超谷歌或微软，并且“从一开始就没有成本控制”。

相关链接：

• AI 算力豪赌
  签署万亿美元芯片大单，烧钱规模远超谷歌微软?
• AMD 签署价值数百亿美元多年芯片协议，向 OpenAI 供应 6 吉瓦 AI 计算能力

OpenAI's Trillion-Dollar Compute Investment Revealed, Partnering with Tech Giants to Bet on the Future 🚀

Brief:

• OpenAI has reportedly signed compute capacity agreements worth approximately $1 trillion, securing over 20 gigawatts (GW) of AI compute power from companies including Nvidia, AMD, Oracle, and CoreWeave.
• This investment comes amidst the company facing significant losses, projected to be around $10 billion this year, while its annual revenue is only about$12 billion. However, CEO Sam Altman stated that profitability is not the primary concern at present.
• To support these deals, OpenAI has adopted an ingenious circular financing structure. For instance, Nvidia reportedly plans to invest $100 billion into OpenAI, which OpenAI then uses to purchase Nvidia's chips. 🔄
• In a multi-billion dollar agreement with AMD, OpenAI will acquire 6 GW of compute capacity and be granted warrants, allowing it to purchase up to 10% of AMD shares at a significantly low price.
• Silicon Valley investors have pointed out that OpenAI's capital intensity far surpasses that of Google or Microsoft, and that "there has been no cost control from the start." 🤯

Related Links:

• AI Compute Gamble: OpenAI Signs Trillion-Dollar Chip Order, Spending Far Exceeding Google and Microsoft?
• AMD Signs Multi-Billion Dollar Multi-Year Chip Deal to Supply 6 GW of AI Compute Capacity to OpenAI

谷歌携手2028洛杉矶奥运会，以云与AI技术革新赛事体验

简报：

• 2028年洛杉矶奥运会和残奥会组委会（LA28）宣布，谷歌成为其官方云服务提供商及创始合作伙伴。
• 此次合作将利用谷歌云（Google Cloud）基础设施、Gemini大模型及搜索的“AI模式”，为赛事运营和观众体验提供技术支持。
• 观众可通过官方转播方NBC环球的平台，借助AI实时获取比赛数据、赛程和选手信息。
• 合作范围还涵盖美国奥委会（Team USA），旨在打造一届数字化、高互动性的奥运盛会。

Google Teams Up with LA28 Olympics to Revolutionize Event Experience with Cloud and AI Technology 🏅

Brief:

• The Los Angeles Organizing Committee for the Olympic and Paralympic Games 2028 (LA28) announced Google as its Official Cloud Services Provider and Founding Partner.
• This partnership will leverage Google Cloud infrastructure, the Gemini large language model, and Google Search's "AI mode" to power event operations and enhance the fan experience. ☁️
• Through official broadcaster NBCUniversal's platforms, viewers will be able to access real-time competition data, schedules, and athlete information with the help of AI. 🤖
• The collaboration also extends to Team USA, aiming to create a highly digital and interactive Olympic event.

AMD下一代AI加速器MI450确认采用2nm工艺

简报：

• AMD首席执行官苏姿丰在采访中确认，其下一代AI加速器Instinct MI450将采用2nm先进制程。
• 根据进一步消息，MI450的加速计算裸晶（XCD）将使用台积电N2P工艺，而有源中介层（AID）和多媒体I/O裸晶（MID）则采用N3P工艺。

相关连接：

• AMD CEO 苏姿丰确认 Instinct MI450 加速器将采用 2nm 制程工艺
• IBM 宣布 AI 推理加速器 Spyre 本季度正式商业可用

AMD's Next-Gen AI Accelerator MI450 Confirmed to Use 2nm Process

Brief:

• AMD CEO Lisa Su confirmed in an interview that its next-generation AI accelerator, Instinct MI450, will utilize the advanced 2nm process. 🚀
• Further details indicate that MI450's Accelerator Compute Die (XCD) will leverage TSMC's N2P process, while the Active Interposer Die (AID) and Multimedia I/O Die (MID) will employ the N3P process. 💡 This is a significant step forward!

Related Links:

• AMD CEO Lisa Su confirms Instinct MI450 accelerator will use 2nm process technology
• IBM announces AI inference accelerator Spyre officially commercially available this quarter

Grok Imagine v0.9发布，支持图片一键生成音画同步视频

简报：

• xAI公司推出视频生成模型Grok Imagine v0.9，支持用户上传单张静态图片，快速生成动态视频。
• 新版本的核心亮点是原生音画同步，能够自动为视频无缝集成背景音乐、对白或唱歌元素，且音频与画面口型匹配。
• 生成的视频时长约15秒，帧率为24FPS，渲染速度快，可在15秒内完成。
• 该功能现已免费集成至Grok所有产品，包括grok.com、X平台及移动App，用户无需额外订阅即可使用。
• xAI透露，未来版本计划将视频时长延长至60秒，并探索集成量子计算以进一步缩短延迟。

相关链接：

• AI视频革命！Grok Imagine v0.9一键变身“电影导演”，上传图片秒出带唱带跳大片
• OpenAI 视频生成应用 Sora 首周 iOS 下载量超 ChatGPT，日下载峰值近 11 万

Grok Imagine v0.9 Released, Supports One-Click Generation of Lip-Sync Videos from Images 🎬

Briefing:

• xAI has launched Grok Imagine v0.9, a video generation model that allows users to upload a single static image and quickly generate a dynamic video. 🚀
• A key highlight of the new version is native audio-visual synchronization, capable of seamlessly integrating background music, dialogue, or singing elements into videos, with audio perfectly matching on-screen lip movements.
• Generated videos are approximately 15 seconds long, with a frame rate of 24FPS, and boast rapid rendering, completing within 15 seconds.
• This feature is now freely integrated across all Grok products, including grok.com, the X platform, and mobile apps, accessible to users without any additional subscription.
• xAI revealed plans to extend video duration to 60 seconds in future versions and explore the integration of quantum computing to further reduce latency. ⏱️

Related Links:

• AI Video Revolution! Grok Imagine v0.9 Transforms into a "Movie Director" with One Click, Upload an Image and Get a Singing and Dancing Blockbuster in Seconds
• OpenAI Video Generation App Sora's First-Week iOS Downloads Surpass ChatGPT, Daily Download Peak Nears 110,000