![[3c9506bb-b4d7-4c41-8abf-7df0897eb479.mp3]]

OpenAI发布免费GPT-5，微软生态同步全面集成

简报：

• OpenAI正式发布GPT-5系列模型，并向全球用户免费开放。其核心是集成了文本、图像和语音的多模态推理架构，能根据任务复杂度自动切换模式。
• GPT-5被称为“博士生水平的AI”，在推理、编程和创意写作方面实现重大突破，在AIME2025评测中表现优于竞争对手。
• 微软宣布旗下Copilot、Microsoft 365 Copilot、Azure AI Foundry及GitHub Copilot等核心产品已全面深度整合GPT-5。
• 开发者可通过Azure AI Foundry直接调用新模型，付费GitHub Copilot用户也可使用GPT-5提升代码生成和逻辑优化效率。
• 模型引入了“安全完成”机制以智能过滤不当内容，并提供更多个性化设置，支持与Gmail、谷歌日历等外部服务集成。

相关链接：

• Introducing GPT-5
• OpenAI发布GPT-5 微软旗下生态宣布已全面接入
• OpenAI发布GPT-5模型，多模态推理能力大幅提升并全面免费开放

OpenAI Launches Free GPT-5, Microsoft Ecosystem Fully Integrated 🚀

Briefing:

• OpenAI has officially released the GPT-5 series models, making them freely available to users worldwide. At its core is a multimodal inference architecture integrating text, images, and audio, capable of automatically switching modes based on task complexity. ✨
• GPT-5 is hailed as 'PhD-level AI,' achieving significant breakthroughs in reasoning, programming, and creative writing, and has outperformed competitors in the AIME 2025 evaluation.
• Microsoft announced that its core products, including Copilot, Microsoft 365 Copilot, Azure AI Foundry, and GitHub Copilot, have been fully and deeply integrated with GPT-5.
• Developers can directly call the new models via Azure AI Foundry, and paid GitHub Copilot users can also leverage GPT-5 to enhance code generation and logic optimization efficiency.
• The model introduces a 'safe completion' mechanism to intelligently filter inappropriate content, offers more personalized settings, and supports integration with external services like Gmail and Google Calendar. 🛡️

Related Links:

• Introducing GPT-5
• OpenAI发布GPT-5 微软旗下生态宣布已全面接入
• OpenAI发布GPT-5模型，多模态推理能力大幅提升并全面免费开放

特斯拉战略转向：解散Dojo团队，拥抱英伟达

简报：

• 据悉，特斯拉已解散其Dojo超级计算机团队，此举标志着公司在AI算力上的一次重大策略调整。
• 特斯拉将放弃其自研芯片的路线，转而与芯片巨头英伟达进行更深入的合作。
• 未来的自动驾驶和人工智能模型训练，特斯拉将主要依赖采购的英伟达GPU集群来提供算力支持。

相关链接：

• 特斯拉解散 Dojo 超级计算机团队，放弃自研芯片转向英伟达合作
• 特斯拉 Dojo 团队解散的背后：成本、性能与马斯克的战略转向

Tesla's Strategic Turn: Dojo Team Disbanded, Nvidia Embraced! ⚡

Brief:

• Reportedly, Tesla has disbanded its Dojo supercomputer team, marking a major strategic shift in the company's approach to AI computing power. 🚀
• Tesla will abandon its in-house chip development path, opting instead for deeper collaboration with chip giant Nvidia.
• For future autonomous driving and AI model training, Tesla will primarily rely on purchased Nvidia GPU clusters to provide computing power. 🤖

Related Links:

• Tesla disbands Dojo supercomputer team, abandons self-developed chips, turns to Nvidia for cooperation
• Behind the disbandment of Tesla's Dojo team: Cost, performance, and Musk's strategic shift

黄仁勋登顶《财富》全球商界人物榜首

简报：

• 美国《财富》杂志公布最新全球商界百大影响力人物榜单，英伟达（NVIDIA）CEO黄仁勋超越特斯拉CEO马斯克，位居榜首。
• 《财富》指出，在AI晶片需求激增的推动下，英伟达市值已突破4兆美元，巩固了其在生成式AI产业的领先地位。
• 台湾台积电董事长魏哲家在此次榜单中名列第40位，超微（AMD）CEO苏姿丰也同样上榜。
• 榜单前十名的其他成员还包括微软CEO纳德拉、Meta创办人扎克伯格和OpenAI CEO奥特曼。

相关链接：

• https://www.36kr.com/newsflashes/3416492819795332
• 輝達正面對決！駁斥中國晶片安全指控　稱遠端操控根本不可能
• 台灣囡仔好威！全球百大人物揭曉　黃仁勳超越馬斯克奪冠、這2人也上榜

Jensen Huang Tops Fortune's Global Business Persons List

Brief:

• American magazine Fortune has released its latest list of the Top 100 Most Influential Global Business Persons, with NVIDIA CEO Jensen Huang surpassing Tesla CEO Elon Musk to claim the top spot. 🏆
• Fortune notes that fueled by soaring demand for AI chips, NVIDIA's market capitalization has surpassed $4 trillion, solidifying its dominant position in the generative AI industry. 🚀
• TSMC Chairman C.C. Wei of Taiwan ranked 40th on the list, and AMD CEO Lisa Su also made the cut. 🌟
• Other notable figures in the top ten include Microsoft CEO Satya Nadella, Meta founder Mark Zuckerberg, and OpenAI CEO Sam Altman.

Related Links:

• https://www.36kr.com/newsflashes/3416492819795332
• 輝達正面對決！駁斥中國晶片安全指控　稱遠端操控根本不可能
• 台灣囡仔好威！全球百大人物揭曉　黃仁勳超越馬斯克奪冠、這2人也上榜

小米新一代多模态大模型开源，多项基准测试创纪录

简报：

• 小米大模型团队宣布开源最新一代多模态大模型Xiaomi MiMo-VL-7B-2508，包含RL（强化学习）与SFT（监督微调）两个版本。
• 新模型在学科推理（MMMU）、文档理解（ChartQA）、图形界面定位（ScreenSpot-v2）及视频理解（VideoMME）四项核心能力上刷新了纪录。
• 其中MMMU基准测试得分首次突破70分，ChartQA得分升至94.4，内部VLM Arena评分从1093.9跃升至1131.2。
• 模型引入了“/no_think”指令，用户可自由切换展示推理链条的“思考”模式与直接生成答案的“非思考”模式。

相关链接：

• 小米开源最新多模态大模型 Xiaomi MiMo-VL-7B-2508
• MiMo-VL-7B-RL-2508 开源地址
• MiMo-VL-7B-SFT-2508 开源地址

Xiaomi's New Generation Multimodal Large Model Open-Sourced, Setting Records in Multiple Benchmarks

Briefing:

• Xiaomi's large model team announced the open-sourcing of its latest generation multimodal large model, Xiaomi MiMo-VL-7B-2508, which includes both RL (Reinforcement Learning) and SFT (Supervised Fine-tuning) versions. 🚀
• The new model has set new records in four core capabilities: academic reasoning (MMMU), document understanding (ChartQA), graphical interface localization (ScreenSpot-v2), and video understanding (VideoMME).
• Notably, the MMMU benchmark score surpassed 70 points for the first time, the ChartQA score rose to 94.4, and the internal VLM Arena score jumped from 1093.9 to 1131.2. 📈
• The model introduces a "/no_think" command, allowing users to freely switch between a "thinking" mode that displays the reasoning chain and a "non-thinking" mode that directly generates answers. 💡

Related Links:

• Xiaomi Open-Sources Latest Multimodal Large Model Xiaomi MiMo-VL-7B-2508
• MiMo-VL-7B-RL-2508 Open-Source Address
• MiMo-VL-7B-SFT-2508 Open-Source Address

华为将发AI推理新技术，旨在减少对HBM依赖并优化大模型

简报：据报道，华为计划于8月12日在金融AI推理应用论坛上发布一项AI推理领域的突破性技术，该成果或能降低中国AI推理对HBM（高带宽内存）的依赖，并提升国内大模型推理性能。

相关链接：

• https://www.aibase.com/zh/news/20342
• https://www.ithome.com/0/874/282.htm
• https://www.aibase.com/zh/news/20357

Huawei to Unveil New AI Inference Technology Aiming to Reduce HBM Reliance and Optimize Large Models

Briefing: According to reports, Huawei is slated to unveil a breakthrough AI inference technology at the Financial AI Inference Application Forum on August 12th. This innovation could potentially lessen China's AI inference reliance on HBM (High Bandwidth Memory) and significantly boost the performance of domestic large models. 🚀💡📈

Related Links:

• https://www.aibase.com/zh/news/20342
• https://www.ithome.com/0/874/282.htm
• https://www.aibase.com/zh/news/20357