单词学习：![[989ddeb9-0cd1-494c-bc25-a2b5dd5070f2.mp3]]

OpenAI斥资百亿美元与博通合作开发AI芯片

简报：据报道，OpenAI已与半导体巨头博通达成一项价值100亿美元的战略合作，将共同设计并量产自研AI芯片，以降低对英伟达的依赖。

相关链接：

• OpenAI 携手博通大手笔投资自研芯片，将向英伟达发起挑战

OpenAI Pours $10 Billion into Partnership with Broadcom for AI Chip Development 🤝💡

Brief: According to reports, OpenAI has entered into a strategic partnership worth $10 billion with semiconductor giant Broadcom to jointly design and mass-produce its own AI chips, aiming to reduce its reliance on Nvidia. 🚀

Related Link:

• OpenAI Joins Forces with Broadcom in Major Investment for Self-Developed Chips, Challenging Nvidia

OpenAI 布局招聘领域，将推出 AI 平台与技能认证项目

简报：

• OpenAI 宣布正在开发一款名为“OpenAI 招聘平台”的 AI 招聘服务，预计于2026年推出，旨在与 LinkedIn 竞争。
• 该平台将利用 AI 技术帮助企业，包括小型企业和地方政府，与具备 AI 技能的求职者进行匹配。
• OpenAI 将同步推出“AI 流利度”认证项目，计划与沃尔玛等机构合作，目标是到2030年完成对1000万美国人的认证。
• 应用程序负责人菲吉・西莫表示，此举旨在帮助人们掌握 AI 技能，以应对技术可能带来的职业冲击。

相关链接：

• OpenAI 推出 AI 招聘平台，计划与 LinkedIn 竞争
• OpenAI 明年杀入招聘市场，将帮助具备 AI 技能的人才找到工作

OpenAI Ventures into the Recruitment Sector, Launching AI Platform and Skill Certification Program

Brief:

• OpenAI has announced it is developing an AI recruitment service called the "OpenAI Hiring Platform," expected to launch in 2026, aiming to compete with LinkedIn. 🚀
• The platform will leverage AI technology to help businesses, including small enterprises and local governments, match with job seekers possessing AI skills.
• OpenAI will simultaneously launch an "AI Fluency" certification program, planning to collaborate with organizations like Walmart, with a goal to certify 10 million Americans by 2030. 🎓
• Fidji Simo, Head of Applications, stated that this initiative aims to help people acquire AI skills to navigate potential career disruptions caused by technology. 💪

Related Links:

• OpenAI Launches AI Hiring Platform, Plans to Compete with LinkedIn
• OpenAI Entering Recruitment Market Next Year, Will Help AI-Skilled Talent Find Jobs

博通AI业务驱动增长，Q3营收增22%并获百亿美元新订单

简报：

• 博通公布2025财年第三财季财报，总营收达159.52亿美元，同比增长22%。
• AI业务成为主要增长动力，该季度AI收入同比增长63%至52亿美元。
• 博通在财报会议上透露，已赢得一位新的大型AI客户，并获得了一份价值超100亿美元的AI算力机架订单。

相关链接：

• 博通 FY2025Q3 营收同比增长 22%，拿下新客户 AI 芯片大单

Broadcom AI Business Drives Growth, Q3 Revenue Up 22% and Secures Over $10 Billion in New Orders

Briefing:

• Broadcom announced its Q3 FY2025 earnings report, with total revenue reaching $15.952 billion, marking a 22% year-over-year increase. 📈
• Its AI business emerged as the primary growth driver, with AI revenue in the quarter soaring 63% year-over-year to $5.2 billion. 🚀
• Broadcom revealed during its earnings call that it has secured a new major AI customer and received an order for AI compute racks valued at over $10 billion. 💰

Related Links:

• Broadcom's FY2025 Q3 Revenue Grows 22%, Secures Major AI Chip Order from New Customer

微软推出GPT-realtime模型，支持图像输入并降价20%

简报：

• 微软于2025年9月5日在Azure AI Foundry平台正式发布其最新的语音转语音（S2S）模型GPT-realtime。
• 该模型主打自然语言处理、高音频质量和精准指令跟随，并推出了Marin和Cedar两种全新的逼真语音选项。
• 新增了创新的图像输入支持，允许用户在语音对话中加入图像进行讨论，实现多模态交互。
• 与之前的gpt-4o-realtime预览版相比，正式版价格降低了20%，按每百万代币（token）的使用量进行计算。

相关链接：

• 微软正式发布 GPT-realtime 模型，主打更逼真语音与多模态输入

Microsoft Launches GPT-realtime Model, Supports Image Input and Reduces Price by 20% 🚀

Briefing:

• Microsoft officially released its latest Speech-to-Speech (S2S) model, GPT-realtime, on September 5, 2025, on the Azure AI Foundry platform.
• The model features natural language processing, high audio quality, and precise instruction following, introducing two new realistic voice options: Marin and Cedar.
• Innovative image input support has been added, allowing users to incorporate images into voice conversations for discussion, enabling multimodal interaction. 📸
• Compared to the previous gpt-4o-realtime preview version, the official release is priced 20% lower, calculated per million tokens used. 💰

Related Links:

• Microsoft Officially Releases GPT-realtime Model, Featuring More Realistic Voices and Multimodal Input

微软CEO纳德拉分享Copilot用法：五大提示将其变为高管助理

简报：

• 微软CEO萨蒂亚·纳德拉近日分享了他个人如何将集成GPT-5的Microsoft Copilot作为核心领导力工具，以提升高管生产力。
• 纳德拉在领英上公布了五个具体的AI提示，展示了Copilot如何在他的日常工作中扮演行政助理角色，涵盖会议准备、项目管理、量化决策、时间管理和会议简报。
• 他使用的提示示例包括：要求Copilot根据历史互动总结会议要点、整合多源信息起草项目更新、以及分析数据为产品发布决策提供概率。
• 纳德拉称，Copilot已成为其日常工作流程的一部分，为所有应用程序增加了一层新的智能，正在重塑他的时间管理和战略决策方式。

相关链接：

• GPT-5变身超级行政助理，纳德拉分享在 Copilot 中的领导力用法

Microsoft CEO Satya Nadella Shares Copilot Usage: 5 Prompts to Transform it into an Executive Assistant

Brief:

• Microsoft CEO Satya Nadella recently shared how he personally uses Microsoft Copilot, integrated with GPT-5, as a core leadership tool to boost executive productivity. 🚀
• Nadella revealed five specific AI prompts on LinkedIn, demonstrating how Copilot acts as an executive assistant in his daily routine, covering meeting preparation, project management, quantitative decision-making, time management, and meeting debriefs.
• Examples of his prompts include: asking Copilot to summarize meeting key points based on historical interactions, integrating information from multiple sources to draft project updates, and analyzing data to provide probabilities for product launch decisions. 📊
• Nadella states that Copilot has become an integral part of his daily workflow, adding a new layer of intelligence across all applications and reshaping his approach to time management and strategic decision-making. ✨

Related Links:

• GPT-5 Transforms into a Super Executive Assistant: Nadella Shares Leadership Usage in Copilot

月之暗面发布Kimi K2新版，API提速并扩展上下文长度

简报：

• 月之暗面于9月5日发布Kimi K2模型最新版本（0905），提升了Agentic Coding能力和前端编程体验。
• 新版本将上下文长度从128K扩展至256K，以支持更复杂的任务。
• 全面开放高速版API，输出速度可达60-100 Token/s，大幅提升响应效率。
• 新版API完全兼容Anthropic API，并支持WebSearch Tool与全自动Context Caching功能，定价与旧版保持一致。

相关链接：

• 月之暗面发布 Kimi K2-0905：支持60-100Token/s，高速 API 全面开放
• 月之暗面 Kimi K2 宣布更新：上下文长度扩展至 256K，带来更快的 API

Moonshot AI Releases New Kimi K2 Version: Faster API and Extended Context Length 🚀

Brief:

• On September 5th, Moonshot AI released the latest version (0905) of its Kimi K2 model, enhancing Agentic Coding capabilities and front-end programming experience. ✨
• The new version extends the context length from 128K to 256K to support more complex tasks.
• The high-speed API is now fully open, offering an output speed of 60-100 Tokens/s, significantly boosting response efficiency. ⚡
• The new API is fully compatible with the Anthropic API, supports WebSearch Tool and fully automatic Context Caching, and maintains the same pricing as the old version.

Related Links:

• 月之暗面发布 Kimi K2-0905：支持60-100Token/s，高速 API 全面开放
• 月之暗面 Kimi K2 宣布更新：上下文长度扩展至 256K，带来更快的 API

谷歌AI图像编辑工具爆火，一周吸粉千万

简报：

• 谷歌AI实验项目“Nano Banana”（已集成于 Gemini 2.5 Flash Image）上线一周，为 Gemini 应用吸引了超过 1000 万新用户。
• 该功能上线后累计完成了超 2 亿次图像编辑，其受欢迎程度导致谷歌内部 TPU 资源严重过载。
• 该功能因能避免AI图像生成中常见的“恐怖谷效应”，并在多次编辑中保持人物特征一致而爆火。
• 国内AI视频平台“拍我AI”（PixVerse）已宣布接入该技术，并推出限时六天的免费创作活动。
• 目前 Gemini 免费用户每日可进行 100 次图像编辑，付费用户可进行 1000 次。

相关链接：

• AI“P 图”爆火出圈：“Nano Banana”一周为谷歌 Gemini 吸引千万新用户
• 拍我AI接入谷歌 Nano Banana，开启六天免费创作狂欢

Google AI Image Editor Goes Viral ✨, Attracting Tens of Millions in a Week

Brief:

• Google's AI experimental project "Nano Banana" (now integrated into Gemini 2.5 Flash Image) launched a week ago, attracting over 10 million new users to the Gemini app. 🚀
• Since its launch, the feature has completed over 200 million image edits, and its immense popularity has led to severe overloading of Google's internal TPU resources.
• The feature has gone viral because it effectively avoids the common "uncanny valley effect" in AI image generation and consistently maintains character features across multiple edits.
• Domestic AI video platform "PixVerse" (拍我AI) has announced its integration with this technology and launched a six-day limited-time free creation event. 📹
• Currently, Gemini free users can perform 100 image edits daily, while paid users can perform 1000 edits.

Related Links:

• AI "Photo Editing" Goes Viral: "Nano Banana" Attracts Tens of Millions of New Users to Google Gemini in a Week
• PixVerse Integrates Google's Nano Banana, Kicking Off a Six-Day Free Creation Carnival