目录
![[06cd5d4e-d403-4367-9e28-c172f31d2024.mp3]]
ChatGPT Projects全面升级：引入深度研究与语音交互，支持多终端协作
简报：
- OpenAI在“12 Days of OpenAI”活动中宣布ChatGPT Projects重大更新，提升了项目管理和AI交互体验。
- 本次升级包括深度研究支持、语音模式集成、记忆功能优化以及移动端增强，Projects从简单文件夹升级为智能工作空间。
- 新功能使用户可结合对话、文件和网页搜索提升研究效果，通过语音在项目内实现实时协作，历史对话调用更智能，移动端支持文件上传、模型切换及视频、屏幕共享。
- Canvas功能新增PDF与Markdown导出，项目可群组管理、聊天共享，进一步强化团队协作和多终端使用场景。
- OpenAI还为企业用户提供了与HubSpot等平台的连接，推动AI在生产力和组织管理领域的深入应用。
相关链接：
ChatGPT Projects Undergo Major Upgrade: Introducing Deep Research, Voice Interaction, and Multi-Device Collaboration ✨
Brief:
- OpenAI announced significant updates to ChatGPT Projects during its "12 Days of OpenAI" event, enhancing project management and AI interaction experience. 🚀
- This upgrade includes deep research support, integrated voice mode, optimized memory functions, and mobile enhancements, transforming Projects from simple folders into intelligent workspaces. 🧠
- New features allow users to combine conversations, files, and web searches for improved research efficacy, enable real-time collaboration within projects via voice, offer smarter recall of historical conversations, and provide mobile support for file uploads, model switching, and video/screen sharing.
- The Canvas feature now supports PDF and Markdown export, while projects can be group-managed and chat-shared, further strengthening team collaboration and multi-device usage scenarios.
- OpenAI also provides enterprise users with connections to platforms like HubSpot, promoting the deeper application of AI in productivity and organizational management.
Related Links:
MiniMax Agent升级新增智能图像与多语言文档功能
简报：
- MiniMax官方宣布AI助手产品MiniMax Agent已正式上线并迎来重大升级，集成智能图像搜索、稳定图像生成、反射模式等前沿功能，提升用户在创作、数据处理和多步骤任务中的智能协作体验。
- 新版本支持中文、日文和韩文字体的Python绘图，并可一键导出内容为PDF、DOCX、PPTX格式，便捷分享和专业呈现。
- MiniMax Agent凭借多模态技术和多代理协作协议，可无缝整合文本、图像、语音处理功能，适用于个人与企业多样化生产力场景。
- 有关详细信息可访问 MiniMax Agent官网。
相关链接：
MiniMax Agent Upgrade Introduces Smart Image and Multilingual Document Features
Briefing:
- MiniMax officially announced that its AI assistant product, MiniMax Agent, has been officially launched and has received a significant upgrade. It now integrates cutting-edge features such as intelligent image search, stable image generation, and 'reflection mode,' enhancing users' intelligent collaborative experience in creation, data processing, and multi-step tasks. 🚀
- The new version supports Python plotting with Chinese, Japanese, and Korean fonts, and allows one-click export of content to PDF, DOCX, and PPTX formats, facilitating convenient sharing and professional presentation. 📊
- Leveraging multimodal technology and multi-agent collaboration protocols, MiniMax Agent seamlessly integrates text, image, and voice processing capabilities, making it suitable for diverse individual and enterprise productivity scenarios. ✨
- For more detailed information, please visit the MiniMax Agent Official Website.
Related Links:
AMD携手OpenAI发布全新AI芯片，推理性能大幅提升
简报：
- AMD与OpenAI在2025全球AI发展大会上联合推出Instinct MI400和MI350系列AI芯片，MI350系列基于CDNA4架构，推理性能较上代提升35倍，计算能力提升4倍，并拥有288GB HBM3E显存和高达8TB/s内存带宽。
- 新一代MI400系列预计搭载432GB HBM4高速显存，FP4性能达40 petaflops，支持72个GPU无缝互联，优化AI训练的低精度计算。
- MI355X GPU每美元可提供高达40%更多tokens，FP4性能达161PFLOPS，服务器部署支持风冷和液冷等多种方案。
- AMD还发布了ROCm7开源AI加速平台，推理性能提升超3.5倍，已与多家领先AI平台兼容。
- Oracle、微软、Meta、xAI等企业已宣布与AMD合作，Oracle云将率先采用Instinct MI355X方案。
相关链接：
AMD Teams Up with OpenAI to Unveil New AI Chips, Significantly Boosting Inference Performance
Briefing:
- AMD and OpenAI jointly launched the Instinct MI400 and MI350 series AI chips at the 2025 Global AI Development Conference. The MI350 series, based on the CDNA4 architecture, offers a 35x increase in inference performance and a 4x boost in compute capability compared to its predecessor. It also features 288GB of HBM3E memory and up to 8TB/s memory bandwidth. 🚀
- The new generation MI400 series is expected to feature 432GB of HBM4 high-speed memory, achieving 40 petaflops of FP4 performance. It supports seamless interconnection of 72 GPUs, optimizing low-precision computation for AI training. 💪
- The MI355X GPU delivers up to 40% more tokens per dollar, with FP4 performance reaching 161 PFLOPS. Server deployment supports various solutions, including air and liquid cooling.
- AMD also released ROCm7, an open-source AI acceleration platform, which boosts inference performance by over 3.5 times and is compatible with multiple leading AI platforms.
- Companies like Oracle, Microsoft, Meta, and xAI have announced collaborations with AMD, with Oracle Cloud being the first to adopt the Instinct MI355X solution. ✨
Related Links:
腾讯发布首个全链路工业级3D生成大模型并实现全面开源
简报：
- 腾讯在CVPR2025宣布混元3D 2.1大模型全面开源，成为首个全链路开源的工业级3D生成大模型。
- 新版本模型在几何生成质量、细节建模、PBR材质（基于物理的渲染）等方面实现显著提升，可用于游戏、电影、线上商城等领域的高质量3D角色和物品建模。
- 混元3D 2.1支持基础颜色、金属度、法线、粗糙度等多种贴图渲染，适配消费级显卡，在个人电脑上也能运行，降低开发门槛。
- 全链路开源涵盖模型权重、训练代码和数据处理流程，方便开发者定制和二次开发，并在Github等平台提供详细部署教程。
- 开源地址、详细资料和下载部署说明可访问其官方主页或Github项目页。
相关链接：
Tencent Launches First Full-Pipeline Industrial-Grade 3D Generation Large Model, Achieves Full Open Source
Briefing:
- Tencent announced at CVPR 2025 the full open-sourcing of its Hunyuan 3D 2.1 large model, making it the first full-pipeline open-source industrial-grade 3D generation large model. 🚀
- The new version of the model significantly improves geometric generation quality, detail modeling, and PBR materials (Physically Based Rendering), enabling high-quality 3D character and object modeling for games, movies, online malls, and more.
- Hunyuan 3D 2.1 supports rendering various texture maps (e.g., base color, metallic, normal, roughness), is compatible with consumer-grade graphics cards, and can even run on personal computers, significantly lowering the development barrier. 💻
- The full-pipeline open-source release covers model weights, training code, and data processing pipelines, facilitating customization and secondary development for developers. Detailed deployment tutorials are also provided on platforms like Github.
- The open-source address, detailed information, and download/deployment instructions are available on its official homepage or Github project page. ✨
Related Links:
法国Mistral AI联合英伟达进军AI基础设施，推出自主推理模型并助力欧洲技术独立
简报：
- 法国初创公司Mistral AI宣布与英伟达合作，发布Mistral Compute AI基础设施平台，旨在为欧洲企业和政府提供摆脱美国云巨头的自主解决方案。
- 公司同步推出Magistral系列推理模型，其中Magistral Small为开源版本，Magistral Medium为专有系统，两者均支持多语言推理并具备多模态能力。
- 新平台运行于18,000块英伟达Grace Blackwell芯片，数据和服务均受欧盟管辖，并承诺使用脱碳能源以降低环境影响。
- Mistral的工程创新加快了模型训练与推理速度，提升行业效率，并获得微软等超10亿美元融资支持，助力欧洲人工智能基础设施快速扩张。
相关链接：
France's Mistral AI Partners with Nvidia to Enter AI Infrastructure, Launches Sovereign Inference Models and Boosts European Tech Independence
Briefing:
- French startup Mistral AI announced a partnership with Nvidia to launch the Mistral Compute AI infrastructure platform, aiming to provide European businesses and governments with an independent solution to break free from American cloud giants. 🚀
- The company simultaneously launched its Magistral series of inference models, with Magistral Small being an open-source version and Magistral Medium a proprietary system; both support multilingual inference and possess multimodal capabilities.
- The new platform operates on 18,000 Nvidia Grace Blackwell chips. All data and services are subject to EU jurisdiction, and the company pledges to use decarbonized energy to reduce environmental impact. 🇪🇺
- Mistral's engineering innovations have accelerated model training and inference speeds, boosting industry efficiency. The company has secured over $1 billion in funding from investors including Microsoft, supporting the rapid expansion of European AI infrastructure. 💰
Related Links:
- Mistral AI Partners with Nvidia to Build "Sovereign" AI Platform
- Mistral Launches New Magistral Inference Models
如果对你有用的话，可以打赏哦
打赏
本文作者:topwind
本文链接:
版权声明:本博客所有文章除特别声明外，均采用 BY-NC-SA 许可协议。转载请注明出处！
目录