![[5dd0946a-d289-4865-a0fd-fb9e8aa70031.mp3]]

OpenAI寻求5000亿美元估值，新模型GPT-5却反响平平

简报：

• 人工智能公司OpenAI正计划进行一轮60亿美元的股权出售，目标估值高达5000亿美元，软银、Dragoneer投资集团和Thrive资本等正在参与谈判。
• 与资本市场的热度相反，OpenAI于本月最新发布的GPT-5模型遭遇了用户的冷遇。
• 用户反馈称，新版本的写作质量和个性化特征均未达到预期标准，甚至不如之前的版本。
• OpenAI首席执行官山姆·奥特曼承认，尽管GPT-5展现了“普遍智能”的潜力，但其关键的“持续学习”能力仍处于早期发育阶段。
• 在此次股权出售前，OpenAI在过去12个月里已获得微软、软银等投资者超过400亿美元的投资。

相关链接：

• OpenAI豪赌5000亿美元估值神话：60亿股权大甩卖背后，GPT-5却遭用户冷遇
• OpenAI 计划通过股权出售成为全球最有价值的私营公司，估值达5000亿美元

OpenAI Seeks $500 Billion Valuation, While New GPT-5 Model Receives Lukewarm Reception

Brief:

• Artificial intelligence company OpenAI is planning a $6 billion equity sale, targeting a valuation as high as$500 billion. SoftBank, Dragoneer Investment Group, and Thrive Capital are reportedly involved in negotiations. 💰
• In stark contrast to the fervor in capital markets, OpenAI's latest GPT-5 model, released this month, has been met with a cold shoulder from users. 🥶
• Users report that the new version's writing quality and personalization features have not met expected standards, with some even finding it inferior to previous versions.
• OpenAI CEO Sam Altman admitted that while GPT-5 shows potential for "general intelligence," its critical "continuous learning" capability is still in its early developmental stages.
• Prior to this equity sale, OpenAI has already secured over $40 billion in investments from Microsoft, SoftBank, and other investors over the past 12 months. 🚀

Related Links:

• OpenAI豪赌5000亿美元估值神话：60亿股权大甩卖背后，GPT-5却遭用户冷遇
• OpenAI 计划通过股权出售成为全球最有价值的私营公司，估值达5000亿美元

为AI编程立规矩：OpenAI联盟发布AGENTS.md标准

简报：

• 由OpenAI Codex、Amp、Google Jules、Cursor、RooCode和Factory组成的联盟，为AI编程助手推出了名为AGENTS.md的新文件标准。
• 该标准为AI代理提供构建步骤、测试指令、项目架构等结构化关键信息，使其能像人类开发者一样迅速掌握项目并开始工作。
• 它的出现旨在为行业提供统一的沟通语言，让开发者一次配置就能适配多种AI编程工具，提升协作效率。
• 目前，该标准已获得Cursor、Amp、Jules、Factory、RooCode和Codex等主流AI编程工具的支持。

相关链接：

• https://www.aibase.com/zh/news/20665

Setting the Standard for AI Programming: OpenAI Coalition Unveils AGENTS.md

Briefing:

• A coalition comprising OpenAI Codex, Amp, Google Jules, Cursor, RooCode, and Factory has launched a new file standard named AGENTS.md for AI programming assistants. 🚀
• The standard provides AI agents with structured key information such as build steps, testing instructions, and project architecture, enabling them to quickly grasp projects and begin work, much like human developers.
• Its emergence aims to provide the industry with a unified communication language, allowing developers to configure once and adapt to various AI programming tools, thereby enhancing collaboration efficiency. 💡
• Currently, the standard has received support from mainstream AI programming tools including Cursor, Amp, Jules, Factory, RooCode, and Codex. ✨

Related Link:

• https://www.aibase.com/zh/news/20665

Meta重组AI业务，成立超级智能实验室引发内部调整

为提升人工智能产品开发效率并应对市场竞争，Meta CEO扎克伯格正重组其人工智能部门，计划将其拆分为四个分别专注于研究、新一代“超级智能”技术、产品开发和基础设施的小组，而此举也可能引发裁员、高管离职和内部紧张局势。

相关链接：

• ​扎克伯格计划重组 Meta 人工智能部门，裁员与新团队引发内部变动
• Meta 公司再次重组 AI 团队，成立超级智能实验室

Meta Restructures AI Business, Forms Super Intelligence Lab 🚀, Triggering Internal Adjustments 🤔

To enhance the efficiency of AI product development and address market competition, Meta CEO Mark Zuckerberg is restructuring its artificial intelligence department. The plan involves splitting it into four groups, each focusing on research, next-generation "super intelligence" technology, product development, and infrastructure. This move could also lead to layoffs, executive departures, and internal tensions. 😬

Related Links:

• Zuckerberg Plans to Restructure Meta's AI Division, Layoffs and New Teams Cause Internal Changes
• Meta Again Restructures AI Team, Establishes Super Intelligence Lab

英特尔与软银达成20亿美元投资协议

简报： 美国芯片制造商英特尔（Intel）于8月19日深夜发表声明，证实已与日本软银集团（SoftBank）签署最终证券买卖协议，软银将投资20亿美元，以每股23美元的价格购买英特尔的普通股。

相关链接：

• 快訊／英特爾發聲明證實與軟銀合作！強強聯手　20億美元入股大利多

Intel and SoftBank Reach $2 Billion Investment Agreement

Brief: U.S. chipmaker Intel issued a statement late on August 19, confirming it has signed a definitive securities purchase agreement with Japan's SoftBank Group. Under the agreement, SoftBank will invest $2 billion to purchase Intel's common stock at a price of$23 per share. 🤝💰📈

Related Link:

• Flash News / Intel Issues Statement Confirming Partnership with SoftBank! Strong Alliance, $2 Billion Investment a Major Benefit

企业微信5.0发布，主打三大AI功能

简报：

• 8月20日，企业微信发布5.0版本，全面融合AI能力，推出智能搜索、智能总结、智能机器人三大核心功能。
• AI能力被深度融入智能表格、AI写邮件等办公模块，旨在打造一体化协作体验。
• 官方数据显示，企业微信已连接超过1400万家真实企业与组织，每日服务微信用户超7.5亿。
• 新版本同时提供可将数据与服务部署于自有服务器的私有化部署选项，并为出海企业提供海外版WeCom。

相关链接：

• 企业微信5.0重磅官宣：推出三大全新AI能力，连接真实企业超1400万
• 企业微信邮箱重磅更新：AI一键帮写邮件
• AI全副武装！企业微信5.0引入智能搜索与总结，告别繁琐工作流

WeChat Work 5.0 Released, Featuring Three Major AI Functions

Briefing:

• On August 20th, WeChat Work released version 5.0, fully integrating AI capabilities and introducing three core functions: Smart Search, Smart Summary, and Smart Assistant.
• AI capabilities are deeply integrated into office modules such as Smart Spreadsheets and AI Email Writing, aiming to create a unified collaborative experience. 🤝
• Official data indicates that WeChat Work has connected over 14 million real enterprises and organizations, serving more than 750 million WeChat users daily. 📈
• The new version also offers a private deployment option, allowing data and services to be deployed on an enterprise's own servers, and provides an international version, WeCom, for businesses expanding overseas. 🌍

Related Links:

• WeChat Work 5.0 Officially Announced: Introducing Three New AI Capabilities, Connecting Over 14 Million Real Enterprises
• WeChat Work Email Gets Major Update: AI One-Click Email Writing
• Fully AI-Equipped! WeChat Work 5.0 Introduces Smart Search and Summary, Bidding Farewell to Tedious Workflows

DeepSeek发布V3.1大模型，开源685B参数版本并支持128K上下文

简报：

• DeepSeek在Hugging Face上开源了其新模型V3.1-Base，该模型参数约为685B，并将上下文长度拓展至128K。
• DeepSeek线上模型版本已同步升级至V3.1，用户可通过官方网页、App及小程序进行测试，API接口调用方式保持不变。
• 备受关注的DeepSeek R2模型目前仍没有确定的发布时间。

相关链接：

• DeepSeek 开源新模型 V3.1：约 685B 参数，上下文长度拓展至 128K
• DeepSeek 通知线上模型版本升级至 V3.1，上下文长度拓展至 128k

DeepSeek Releases V3.1 Large Language Model, Open-Sources 685B Parameter Version with 128K Context Support 🚀

Briefing:

• DeepSeek has open-sourced its new V3.1-Base model on Hugging Face, featuring approximately 685B parameters and an extended context length of up to 128K.
• DeepSeek's online model version has been simultaneously upgraded to V3.1. Users can test it via the official website, app, and mini-program, while API call methods remain unchanged.
• The highly anticipated DeepSeek R2 model still does not have a confirmed release date. ⏳

Related Links:

• DeepSeek Open-Sources New Model V3.1: Approx. 685B Parameters, Context Length Extended to 128K
• DeepSeek Announces Online Model Version Upgrade to V3.1, Context Length Extended to 128K ✨