![[91fffd10-39d9-4c32-a9d7-e83da9474fc8.mp3]]

OpenAI获软银领投完成新一轮300亿美元融资，估值升至3000亿美元

简报：

• OpenAI于7月底完成新一轮由软银集团领投的300亿美元融资，总估值达到3000亿美元，成为全球估值最高的私营初创公司之一。
• 此轮资金包括软银出资225亿美元，其余由其他投资者财团提供，使得今年累计融资达到400亿美元。
• OpenAI同时与软银、甲骨文等合作，推进投资超5000亿美元的“星际之门”AI数据中心项目，部分新数据中心已在美国德克萨斯州动工。

相关链接：

• OpenAI完成300亿美元融资，估值达3000亿美元
• OpenAI融资超400亿美元，软银成最大投资方
• OpenAI与Oracle合建AI数据中心“星际之门”项目
• OpenAI总估值突破3000亿美元

OpenAI Secures New $30 Billion Funding Round Led by SoftBank, Valuation Soars to$300 Billion

Briefing:

• By the end of July, OpenAI completed a new $30 billion funding round led by SoftBank Group, bringing its total valuation to$300 billion, making it one of the world's most valuable private startups. 🚀
• This round of funding includes $22.5 billion from SoftBank, with the remainder provided by a consortium of other investors, bringing this year's cumulative funding to$40 billion. 💰
• OpenAI is also collaborating with SoftBank, Oracle, and others to advance the "Stargate" AI data center project, an initiative with investments exceeding $500 billion. Construction has already begun on some new data centers in Texas, USA. 🏗️

Related Links:

• OpenAI Completes $30 Billion Funding Round, Valued at$300 Billion
• OpenAI Raises Over $40 Billion, SoftBank Becomes Largest Investor
• OpenAI and Oracle Jointly Build "Stargate" AI Data Center Project
• OpenAI's Total Valuation Exceeds $300 Billion

马斯克被曝以超200名员工人脸数据训练xAI聊天机器人Grok

简报：

• 据《商业内幕》7月23日报道，马斯克旗下xAI公司曾要求200多名员工录制各种面部表情与对话视频，用于AI模型Grok的情绪与表情识别训练，引发员工对个人数据用途及隐私的担忧。
• 拍摄过程中，员工需与同事扮演不同角色进行自然对话，所收集视频仅用于内部训练，并需员工签署同意书，授权xAI永久访问。
• 部分员工因担忧肖像被滥用或生成虚假形象选择退出该项目。
• 此举正值xAI于7月推出视频聊天与逼真虚拟形象等新功能，相关内容及数据利用的安全性正引发外界持续关注。

相关链接：

• ithome.com 新闻1
• ithome.com 新闻2
• setn.com 报道

Elon Musk Reportedly Trained xAI Chatbot Grok Using Facial Data from Over 200 Employees

Brief:

• According to a Business Insider report on July 23, Elon Musk's xAI company reportedly asked over 200 employees to record videos capturing various facial expressions and conversations. This data was then used to train AI model Grok for emotion and expression recognition, sparking concerns among employees regarding the use of their personal data and privacy. 😟
• During the recording process, employees were required to engage in natural conversations with colleagues, playing different roles. The collected videos were reportedly for internal training purposes only, and employees had to sign consent forms granting xAI permanent access. 📄
• Some employees opted out of the project due to concerns about their likeness being misused or used to generate false images.
• This development comes as xAI launched new features in July, including video chat and realistic virtual avatars. The security of content and data utilization related to these new functionalities is continuously drawing public attention. 👀

Related Links:

• ithome.com News 1
• ithome.com News 2
• setn.com Report

阿里巴巴发布开源AI编程模型Qwen3-Coder支持超长上下文与强大Agent能力

简报：

• 2025年7月23日，阿里巴巴正式发布开源AI编程模型Qwen3-Coder，该模型采用混合专家架构、参数量高达480亿，可激活参数为35亿。
• Qwen3-Coder支持256K token原生上下文，最大可扩展到1M，能媲美国际顶尖编程模型Claude4与Claude Sonnet4。
• 该模型在7.5TB含70%代码的数据集上预训练，在WebArena、SWE-Bench等评测中获得领先成绩，展现出极强的复杂编程和多步骤任务自主规划能力。
• Qwen3-Coder现已在魔搭社区、HuggingFace等平台开源，并可通过API接入阿里的通义灵码、阿里云百炼等产品，官方还推出了命令行工具Qwen Code辅助开发。
• 该模型降低了编程门槛，让普通用户也可用简单指令完成复杂开发任务，提高多个领域效率。

相关链接：

• IT之家新闻报道
• 魔搭社区模型下载
• HuggingFace模型页面
• Qwen Code GitHub

Alibaba Releases Open-Source AI Programming Model Qwen3-Coder Supporting Ultra-Long Context and Powerful Agent Capabilities

Brief:

• On July 23, 2025, Alibaba officially released the open-source AI programming model Qwen3-Coder. This model employs a Mixture-of-Experts (MoE) architecture, boasting a massive 480 billion parameters with 35 billion active parameters. 🚀
• Qwen3-Coder supports a native context window of 256K tokens, extensible up to 1M, putting it on par with top international programming models like Claude4 and Claude Sonnet4. 🧠
• The model was pre-trained on a massive 7.5TB dataset, containing 70% code, and achieved leading results in benchmarks such as WebArena and SWE-Bench, demonstrating strong capabilities in complex programming and autonomous planning for multi-step tasks.
• Qwen3-Coder is now open-source on platforms like ModelScope and HuggingFace. It can also be integrated via API into Alibaba's Tongyi Lingma, Alibaba Cloud BaiLian, and other products. An official command-line tool, Qwen Code, has also been launched to assist development. 🛠️
• This model lowers the barrier to programming, enabling general users to complete complex development tasks with simple instructions and significantly improving efficiency across various domains.

Related Links:

• IT Home News Report
• ModelScope Model Download
• HuggingFace Model Page
• Qwen Code GitHub

阿里巴巴首款AI眼镜将发布，深度整合自家生态系统

简报：

• 据多家媒体报道，阿里巴巴将在本周推出其首款自研AI眼镜，支持语音助手、音乐播放、通话、实时翻译、会议纪要等功能，并深度整合高德地图、支付宝、淘宝等核心生态应用。
• 该眼镜采用双芯片（高通骁龙AR1与恒玄BES2800）架构，分为普通AI与带显示AI+AR两个版本，带显示版优先级更高。
• 产品由天猫精灵硬件团队和夸克AI研发团队联合打造，主导人为前华为旗舰手机主架构师宋刚。
• 行业分析认为，凭借阿里庞大业务生态和AI能力，该产品有望突破目前AI眼镜市场碎片化和功能单一的困境，助力AI眼镜普及大众市场。
• 与此同时，小米、百度、字节跳动等科技公司亦布局AI眼镜，行业即将迎来发展关键期。

相关链接：

• AI Base 新闻报道
• IT之家新闻报道

Alibaba to Launch First AI Glasses, Deeply Integrating Its Ecosystem 👓

Briefing:

• According to multiple media reports, Alibaba is set to launch its first self-developed AI glasses this week. These glasses will support functions such as voice assistant, music playback, calls, real-time translation, and meeting minutes, while deeply integrating core ecosystem applications like Amap, Alipay, and Taobao. 🚀
• The glasses feature a dual-chip architecture (Qualcomm Snapdragon AR1 and Bestechnic BES2800) and will come in two versions: a standard AI model and an AI+AR model with a display, with the latter being prioritized.
• The product is a collaborative effort between Tmall Genie's hardware team and Quark AI's R&D team, led by Song Gang, former chief architect of Huawei's flagship phones.
• Industry analysts believe that leveraging Alibaba's extensive business ecosystem and AI capabilities, this product is poised to overcome the current fragmentation and limited functionality of the AI glasses market, helping to popularize them for the mass market.
• Concurrently, tech giants like Xiaomi, Baidu, and ByteDance are also investing in AI glasses, signaling that the industry is on the cusp of a critical development phase. 💡

Related Links:

• AI Base News Report
• IT Home News Report

谷歌推出Gemini 2.5 Flash-Lite，主打高效与低价AI模型

简报：

• 谷歌宣布Gemini 2.5 Flash-Lite模型正式进入稳定版本（GA），成为速度最快、成本最低的AI模型之一，主打性能与经济性的平衡。
• 该模型原生支持高达100万token的上下文，推出每百万输入token 0.10美元、输出token 0.40美元的低价策略，相较早期预览版音频输入价格下降了40%。
• 在编码、数学、推理及多模态等基准测试项目上，2.5 Flash-Lite均优于2.0版本，开发者通过API可快速集成该模型，需在8月25日前切换至新模型别名。
• 该发布展现了谷歌在AI领域的持续创新，并为开发者和企业提供更高效、经济实用的人工智能方案。

相关链接：

• aibase新闻1
• aibase新闻2
• MSN报道
• InfoWorld报道

Google Launches Gemini 2.5 Flash-Lite: Focusing on High Efficiency and Low-Cost AI Model

Briefing:

• Google has announced that the Gemini 2.5 Flash-Lite model has officially reached General Availability (GA) 🚀, positioning itself as one of the fastest and most cost-effective AI models, emphasizing a balance between performance and affordability.
• The model natively supports up to 1 million tokens of context and introduces a low-cost strategy of $0.10 per million input tokens and$0.40 per million output tokens, with audio input prices dropping by 40% compared to the early preview version.
• In benchmark tests across coding, mathematics, reasoning, and multimodal capabilities, 2.5 Flash-Lite outperforms version 2.0. Developers can quickly integrate this model via API but need to switch to the new model alias by August 25th.
• This release demonstrates Google's continuous innovation in the AI field 💡, providing developers and businesses with more efficient and economically practical artificial intelligence solutions.

Related Links: 👇

• aibase News 1
• aibase News 2
• MSN Report
• InfoWorld Report

台积电二季度AI芯片营收超百亿美元，英伟达贡献显著

简报：

• 2024年第二季度，台积电AI相关营收达到约100亿美元，得益于与英伟达的深度合作，后者成为其AI业务最大推动力。
• 台积电期内总营收为317亿美元，同比增长61%，AI芯片需求是主要动力，高性能计算领域占营收60%。
• 当前英伟达是台积电AI芯片的主要客户，其最新一代GPU采用台积电4纳米工艺，AI业务的重要性已超过部分传统大客户。
• 台积电全年AI营收预计将继续大幅增长，公司看好2025年业绩背后的AI高性能计算需求。

相关链接：

• IT之家报道
• 三立新闻报道

TSMC's Q2 AI Chip Revenue Exceeds $10 Billion, Nvidia's Contribution Significant 🚀

Brief:

• In Q2 2024, TSMC's AI-related revenue reached approximately $10 billion, primarily driven by its deep collaboration with Nvidia, which has become the biggest contributor to its AI business.
• TSMC's total revenue for the period was $31.7 billion 💰, a 61% year-on-year increase. AI chip demand was the main impetus, with high-performance computing (HPC) accounting for 60% of revenue.
• Currently, Nvidia is TSMC's primary AI chip customer. Its latest generation GPUs utilize TSMC's 4nm process, and the importance of AI business has surpassed that of some traditional major clients.
• TSMC's full-year AI revenue is expected to continue to grow substantially, and the company is optimistic about the AI high-performance computing demand supporting its 2025 performance. 💡

Related Links:

• IT Home Report
• SETN News Report