![[40591d2f-8972-4078-a1ea-fcc4d52488cb.mp3]]

ChatGPT Projects迎来深度研究与语音模式等多项重大升级

简报：

• 2025年6月12日，OpenAI在“12Days of OpenAI”活动中宣布对ChatGPT Projects进行重大更新，推出深度研究（Deep Research）与语音模式等创新特性。
• 升级内容包括网页搜索与项目上下文结合的深度研究、语音直接交互、记忆功能优化以及移动端支持文件上传和屏幕共享。
• 新功能提升了ChatGPT Projects在个性化及团队协作场景中的应用能力，使其成为多功能AI工作中心，适用于任务管理、团队协作、学术研究等多种场景。
• OpenAI还增强了企业级集成，与HubSpot等平台无缝对接，进一步拓展了商业用途。
• 官方强调已落实数据保护合规措施，确保隐私与安全。

相关链接：

• OpenAI重磅升级ChatGPT Projects 深度研究+语音模式
• ChatGPT再进化！项目功能大升级，Canvas支持PDF导出

ChatGPT Projects Unveils Deep Research, Voice Mode, and More Major Upgrades

Brief:

• On June 12, 2025, OpenAI announced significant updates to ChatGPT Projects during its "12 Days of OpenAI" event, introducing innovative features like Deep Research and Voice Mode. 🚀
• The enhancements include Deep Research, which integrates web search with project context; direct voice interaction; optimized memory functions; and mobile support for file uploads and screen sharing.
• These new features enhance ChatGPT Projects' applicability in personalized and team collaboration scenarios, transforming it into a versatile AI work hub suitable for various use cases such as task management, team collaboration, and academic research. 💡
• OpenAI has also strengthened enterprise-level integrations, seamlessly connecting with platforms like HubSpot to further expand its commercial utility.
• OpenAI also emphasized that robust data protection compliance measures have been implemented, ensuring privacy and security. 🔒

Related Links:

• OpenAI重磅升级ChatGPT Projects 深度研究+语音模式
• ChatGPT再进化！项目功能大升级，Canvas支持PDF导出

AMD携手OpenAI推出全新AI芯片，推理性能实现35倍飞跃

简报：

• 2025全球AI发展大会上，AMD与OpenAI联合发布Instinct MI400和MI350系列AI芯片。
• MI350系列GPU基于CDNA4架构，配备288GB HBM3E内存，AI计算能力提升4倍，推理性能提升35倍。
• MI355X GPU每美元可提供多达40%的额外tokens，FP4性能达161PFLOPS，FP16性能达36.8PFLOPS。
• 新芯片支持多种冷却和大规模部署方案，并以开源AI加速平台ROCm7为开发者提供技术支持。
• 下一代MI400系列GPU预计2026年推出，将配备高达432GB HBM4高速显存，内存带宽达300GB/s。
• 多家企业包括Oracle、微软、Meta、xAI等已开始采用AMD AI芯片，Oracle将首批在云基础设施中部署Instinct MI355X解决方案。
• OpenAI CEO Sam Altman宣布将使用AMD MI300X和MI450 AI芯片。

相关链接：

• aibase.com新闻报道
• IT之家新闻1

AMD Teams Up with OpenAI to Unveil New AI Chips, Delivering a 35x Leap in Inference Performance ✨

Briefing:

• At the 2025 Global AI Development Conference, AMD and OpenAI jointly announced the Instinct MI400 and MI350 series AI chips. 🚀
• The MI350 series GPUs, based on the CDNA4 architecture, feature 288GB of HBM3E memory, boosting AI compute capabilities by 4x and inference performance by 35x.
• The MI355X GPU delivers up to 40% more tokens per dollar, with FP4 performance reaching 161 PFLOPS and FP16 performance at 36.8 PFLOPS. 💡
• The new chips support various cooling and large-scale deployment solutions, and offer developer support through the open-source AI acceleration platform, ROCm 7.
• The next-generation MI400 series GPUs are expected to launch in 2026, featuring up to 432GB of HBM4 high-speed memory and a memory bandwidth of 300GB/s.
• Several companies, including Oracle, Microsoft, Meta, and xAI, have begun adopting AMD AI chips, with Oracle being among the first to deploy Instinct MI355X solutions in their cloud infrastructure.
• OpenAI CEO Sam Altman announced that they will be using AMD MI300X and MI450 AI chips.

Related Links:

• aibase.com News Report
• IT Home News 1

谷歌新AI模型大幅提升地方气候预测精度与效率

简报：

• 谷歌研究团队推出了“动态生成下采样法”，结合物理气候建模和生成式AI，可将全球气候预测精细化至10公里分辨率，有效满足城市级农业、灾害应对等实际需求。
• 该技术核心R2D2模型，通过学习高分辨率示例，能在计算成本降低高达85%的情况下，显著提升温度、湿度、风速等变量预测准确率，相关误差较传统统计方法减少超40%。
• 此方法不仅实现更经济、快速的详细气候风险评估，还能捕捉热浪、干旱结合及野火风险等复杂天气模式，助力地方社区实现更有效应对和决策。

相关链接：

• 谷歌AI助力气候预测：突破传统模型，气象预测更精细
• 谷歌上线Weather Lab气象预测网站，AI模型精确预测台风路径

Google's New AI Model Significantly Boosts Local Climate Prediction Accuracy and Efficiency 🌎

Brief:

• Google's research team has unveiled "Resolution-Refining Diffusion Downscaling" (R2D2), which combines physical climate modeling and generative AI. This innovation allows global climate predictions to be refined down to a 10-kilometer resolution, effectively meeting practical needs like urban agriculture and disaster response.
• The core of this technology, the R2D2 model, learns from high-resolution examples. It significantly improves the prediction accuracy for variables such as temperature, humidity, and wind speed, reducing computational costs by up to 85% and errors by over 40% compared to traditional statistical methods. 📈
• This method not only enables more economical and faster detailed climate risk assessments but also captures complex weather patterns like heatwaves, combined droughts, and wildfire risks, empowering local communities with more effective responses and decision-making. 🔥

Related Links:

• Google AI assists climate prediction: Breaking through traditional models for more refined meteorological forecasts
• Google launches Weather Lab meteorological prediction website, AI model accurately predicts typhoon paths

Mistral AI与英伟达合作推出欧洲本地化AI基础设施与推理模型，推动技术自主

简报：

• 法国初创企业Mistral AI宣布与英伟达（NVIDIA）深度合作，推出Mistral Compute综合AI基础设施平台，旨在为欧洲企业和政府提供本地化解决方案，减少对美国云服务的依赖，强调数据主权与环境可持续性。
• Mistral发布Magistral系列推理模型，包括240亿参数的开源与更强大的专有版本，具备多语种逻辑推理及图像处理能力，并在速度与效率上实现工程性突破。
• 英伟达为该平台提供18,000块最新Grace Blackwell芯片，首批部署在法国数据中心，计划扩展至全欧洲，为AI本地部署和主权AI发展提供算力基础。
• 合作响应了欧洲对AI基础设施自主、数据本地存储和可持续能源的政策要求，成为欧盟推动“超级工厂”计划和区域AI产业独立的关键一步。

相关链接：

• Mistral AI携手英伟达，打造“主权”AI基础设施与顶尖推理模型
• 黃仁勳在巴黎揭示歐洲AI新藍圖
• 主權AI成新戰場！黃仁勳佈局歐洲AI基建

Mistral AI Partners with NVIDIA to Launch Localized European AI Infrastructure and Inference Models, Promoting Technological Autonomy

Briefing:

• French startup Mistral AI has announced a deep collaboration with NVIDIA to launch the Mistral Compute comprehensive AI infrastructure platform. This initiative aims to provide localized solutions for European businesses and governments, reducing reliance on US cloud services, and emphasizing data sovereignty and environmental sustainability. 🌍
• Mistral has released its Magistral series of inference models, featuring both a 24-billion-parameter open-source version and a more powerful proprietary version. These models boast multi-language logical reasoning and image processing capabilities, achieving engineering breakthroughs in speed and efficiency. ✨
• NVIDIA is supplying 18,000 of its latest Grace Blackwell chips for the platform. Initial deployments will be in French data centers, with plans for expansion across Europe, providing the computational foundation for local AI deployment and the development of sovereign AI. 🇪🇺
• This collaboration responds to Europe's policy requirements for AI infrastructure autonomy, local data storage, and sustainable energy. It marks a crucial step in the EU's push for "superfactory" initiatives and the independence of its regional AI industry.

Related Links:

• Mistral AI携手英伟达，打造“主权”AI基础设施与顶尖推理模型
• 黃仁勳在巴黎揭示歐洲AI新藍圖
• 主權AI成新戰場！黃仁勳佈局歐洲AI基建

DeepMind推出AI系统提前15天精准预报热带气旋并与美国国家飓风中心合作

简报：

• 谷歌DeepMind宣布推出一款创新AI系统，能以前所未有的精度预测热带气旋的路径与强度，提前达15天生成多种风暴演变情景。
• DeepMind与美国国家飓风中心（NHC）达成合作，这也是NHC首次将实验性AI预测纳入运营流程，预报员可实时查看并结合传统模型优化预警。
• 该AI模型在五天路径预测和风暴强度方面性能优于现有主流物理模型，同时比传统系统运算速度快约8倍，实现约一分钟生成15天预测。
• 模型训练结合全球再分析数据和近5000次气旋观测，利用最新概率建模技术“功能生成网络（FGN）”提升预报的多样性和准确性。
• DeepMind发布了“Weather Lab”互动平台，专家可评估AI模型历史和实时预报表现，为2025年大西洋飓风季提供创新技术支撑。

相关链接：

• DeepMind突破性AI 15天预报热带气旋，携手美国国家飓风中心挽救生命与经济

DeepMind Launches AI System for Accurate Tropical Cyclone Forecasts Up to 15 Days in Advance, Collaborating with the U.S. National Hurricane Center

Brief:

• Google DeepMind announced the launch of an innovative AI system capable of predicting the path and intensity of tropical cyclones with unprecedented accuracy, generating multiple storm evolution scenarios up to 15 days in advance. 🚀
• DeepMind has partnered with the U.S. National Hurricane Center (NHC), marking the NHC's first integration of experimental AI forecasts into its operational workflow. Forecasters can view these predictions in real-time and combine them with traditional models to optimize warnings. 🤝
• The AI model outperforms existing mainstream physical models in both five-day path prediction and storm intensity. It is also approximately 8 times faster than traditional systems, generating 15-day forecasts in about one minute.
• The model was trained using a combination of global reanalysis data and nearly 5,000 cyclone observations. It leverages the latest probabilistic modeling technique, "Functional Generative Networks (FGN)," to enhance forecast diversity and accuracy.
• DeepMind has launched the "Weather Lab" interactive platform where experts can evaluate the AI model's historical and real-time forecast performance, providing innovative technological support for the 2025 Atlantic hurricane season. ⛈️

Related links:

• DeepMind's Breakthrough AI Forecasts Tropical Cyclones 15 Days Out, Partnering with U.S. National Hurricane Center to Save Lives and Economies