单词学习：![[47e36e67-0b2f-400a-a4eb-de5882870f27.mp3]]

瑞士推出国家级开源大模型Apertus，旨在打造AI公共基础设施

简报：

• 瑞士联邦理工学院(EPFL)、苏黎世联邦理工学院(ETH Zurich)及国家超级计算中心(CSCS)联合发布了瑞士首个完全开源的大语言模型Apertus。
• Apertus模型支持超过1000种语言，其中40%的训练数据为非英语，特别包含了瑞士德语等本地语言，旨在提供一个替代商业AI方案的选择。
• 模型完全开源，其架构、训练数据和流程均被公开，并采用宽松许可证，支持教育、科研及商业用途。
• Apertus依托瑞士“Alps”超级计算机训练，使用可再生能源，并提供80亿和700亿两种参数规模的版本。
• 用户可通过Hugging Face等平台获取该模型，瑞士希望借此将AI打造为透明、合规的公共基础设施，提升国家创新能力。

相关链接：

• 瑞士重磅发布自主开源AI大模型Apertus
• Apertus模型Hugging Face页面

Switzerland Unveils National Open-Source LLM Apertus, Aiming to Build AI Public Infrastructure

Briefing:

• The Swiss Federal Institute of Technology Lausanne (EPFL), ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) have jointly launched Apertus, Switzerland's first fully open-source large language model.
• The Apertus model supports over 1,000 languages, with 40% of its training data being non-English, specifically including local languages such as Swiss German. It aims to offer an alternative to commercial AI solutions. 🌍
• The model is completely open-source; its architecture, training data, and processes are all publicly available under a permissive license, supporting educational, research, and commercial applications. 🚀
• Apertus was trained on Switzerland's "Alps" supercomputer, utilizing renewable energy, and is available in two parameter sizes: 8 billion and 70 billion.
• Users can access the model via platforms like Hugging Face. Switzerland hopes to leverage this initiative to establish AI as transparent, compliant public infrastructure, thereby enhancing national innovation capabilities. ✨

Related Links:

• Switzerland Unveils Its Own Open-Source AI Model, Apertus
• Apertus Model Hugging Face Page

苹果开源视觉AI模型FastVLM与MobileCLIP2，大幅提升端侧处理速度

简报：

• 苹果公司在Hugging Face平台开源了两款视觉语言模型：FastVLM和MobileCLIP2，专为Apple Silicon等边缘设备优化。
• FastVLM模型在处理高分辨率图像时，其首词响应时间（TTFT）提升了85倍，视觉编码器体积缩小了3.4倍。
• MobileCLIP2是一款基于CLIP架构的轻量化模型，专注于在资源受限的边缘设备上实现高效的图像与文本特征对齐。
• 这两款模型均支持完全本地化处理以保护用户隐私，并已能实现在浏览器环境中对视频画面进行实时内容描述。

相关链接：

• FastVLM模型集
• 技术论文

Apple Open-Sources Vision AI Models FastVLM and MobileCLIP2, Significantly Boosting On-Device Processing Speed

Brief:

• Apple has open-sourced two vision-language models on the Hugging Face platform: FastVLM and MobileCLIP2, optimized for edge devices like Apple Silicon. ✨
• When processing high-resolution images, the FastVLM model achieves an 85x improvement in Time-to-First-Token (TTFT), and its vision encoder size is reduced by 3.4 times.
• MobileCLIP2 is a lightweight model based on the CLIP architecture, focused on achieving efficient image and text feature alignment on resource-constrained edge devices. 🚀
• Both models support fully localized processing to protect user privacy, and can already perform real-time content descriptions of video frames within a browser environment. 🔒

Related Links:

• FastVLM Model Collection
• Technical Paper

OpenAI斥资11亿美元收购Statsig，并任命其CEO为应用CTO

简报：

• 当地时间2025年9月2日，OpenAI宣布以11亿美元收购产品分析公司Statsig。
• 收购后，Statsig的创始人兼CEO维贾伊・拉吉将加入OpenAI，担任新设立的应用首席技术官（CTO）。
• Statsig的全体员工将并入OpenAI团队，但公司将继续保持独立运营，利用OpenAI的资源服务现有客户。

相关链接：

• OpenAI以11亿美元收购Statsig，组建新高管团队

OpenAI Acquires Statsig for $1.1 Billion, Appoints Its CEO as Applications CTO

Briefing:

• On September 2, 2025, local time, OpenAI announced its acquisition of product analytics company Statsig for $1.1 billion. 🚀
• Following the acquisition, Statsig's founder and CEO, Vijay Raj, will join OpenAI as the newly appointed Applications Chief Technology Officer (CTO).
• All Statsig employees will integrate into the OpenAI team, but Statsig will continue to operate independently, leveraging OpenAI's resources to serve its existing customers. 🤝

Related Links:

• OpenAI Acquires Statsig for $1.1 Billion, Forms New Executive Team ✨

ChatGPT将增设家长控制，旨在保护青少年用户

简报：

• OpenAI宣布将于下月为ChatGPT添加家长控制功能，以应对青少年用户面临的潜在风险，此举与此前发生的青少年自杀事件有关。
• 家长可通过邮件关联孩子账户，从而监控敏感提问、接收风险对话提醒，并有权决定ChatGPT如何回复孩子的问题。
• 该功能还允许家长禁用ChatGPT的记忆和聊天记录功能，以增强隐私保护。
• OpenAI同时在研发新的对话逻辑，计划将敏感话题自动转由更安全的模型（如GPT-5-Thinking或o3）进行处理。

相关链接：

• 16 岁少年与 ChatGPT 交流数月后自杀，OpenAI 称将推出家长控制功能
• ChatGPT 被指导致美国一名 16 岁少年自杀：未有效干预且起到“教学”作用

ChatGPT to Add Parental Controls, Aimed at Protecting Teenage Users

Brief:

• OpenAI announced it will add parental control features to ChatGPT next month, aiming to address potential risks faced by teenage users. This move is linked to previous incidents of adolescent suicide. 🛡️
• Parents can link their child's account via email to monitor sensitive questions, receive alerts for risky conversations, and have the power to decide how ChatGPT responds to their child's queries.
• The feature also allows parents to disable ChatGPT's memory and chat history functions for enhanced privacy protection. 🔒
• OpenAI is simultaneously developing new conversational logic, planning to automatically redirect sensitive topics to be handled by safer models (such as GPT-5-Thinking or o3). 🧠

Related Links:

• 16-Year-Old Commits Suicide After Months of Chatting with ChatGPT; OpenAI Announces Parental Controls
• ChatGPT Accused of Contributing to a 16-Year-Old's Suicide in the U.S.: Failed to Intervene Effectively and Acted as a 'Teacher'

分析师预测苹果2027年将推平价版头显Vision Air

简报：

• 分析师郭明錤报告称，苹果公司预计于2027年推出名为Vision Air的新款头显产品。
• Vision Air旨在解决Vision Pro过重与价格过高的问题，预计重量将减少超过40%，售价降低超过50%。
• 报告预测，Vision Air在2027年的出货量有望达到百万台级别。

相关链接：

• 郭明錤：苹果预计2027年推出Vision Air，预估售价降逾50%

Analysts Predict Apple to Launch Affordable Vision Air Headset in 2027

Briefing:

• Analyst Ming-Chi Kuo reports that Apple is expected to launch a new headset product named Vision Air in 2027. 🍎
• Vision Air aims to address the issues of Vision Pro being too heavy and too expensive, with an anticipated weight reduction of over 40% and a price reduction of over 50%.
• The report predicts that Vision Air's shipments could reach the million-unit level in 2027. 🚀

Related Link:

• Ming-Chi Kuo: Apple Expected to Launch Vision Air in 2027, Price Estimated to Drop Over 50% 💡

腾讯发布混元Voyager3D，业界首个原生3D重建世界模型

简报：

• 腾讯于9月2日发布了混元3D世界模型系列新成员HunyuanWorld-Voyager，官方称其为业界首个支持原生3D重建的超长漫游世界模型。
• 该模型突破了传统视频生成在空间一致性上的局限，能够生成长距离、世界一致的漫游场景，并支持将视频直接导出为3D点云格式。
• Voyager通过将场景深度预测引入视频生成过程，并采用可扩展的世界缓存机制，融合了视频生成与3D建模的优势。
• 在斯坦福大学李飞飞团队发布的世界模型基准测试WorldScore上，Voyager的综合能力位居首位。
• 该模型现已正式上线，其源代码已在GitHub和Hugging Face上免费开放，可用于虚拟现实、物理仿真、游戏开发等领域。

相关链接：

• 项目主页
• Github
• Hugging Face

Tencent Unveils Hunyuan Voyager3D, the Industry's First Native 3D Reconstruction World Model 🚀

Briefing:

• On September 2nd, Tencent released HunyuanWorld-Voyager, a new member of its Hunyuan 3D world model series, officially claiming it as the industry's first ultra-long roaming world model supporting native 3D reconstruction. ✨
• This model breaks through the limitations of traditional video generation in spatial consistency, enabling the generation of long-distance, world-consistent roaming scenes, and supports directly exporting videos in 3D point cloud format.
• Voyager integrates the advantages of video generation and 3D modeling by introducing scene depth prediction into the video generation process and employing a scalable world caching mechanism.
• On the WorldScore world model benchmark test released by Stanford University's Fei-Fei Li team, Voyager's comprehensive capabilities ranked first. 🏆
• The model is now officially live, and its source code has been freely released on GitHub and Hugging Face, available for applications in fields such as virtual reality, physical simulation, and game development.

Related Links:

• Project Homepage
• GitHub
• Hugging Face

辉达否认H100/H200芯片短缺传闻

简报：

• 针对市场传闻H100与H200芯片供应紧张甚至已售罄，辉达（Nvidia）于9月2日公开回应称传闻不实，公司拥有足夠的库存来满足所有订单。
• 辉达解释称，其云端合作伙伴能将每一颗已上线的芯片出租，但这不代表公司无法应对新的订单需求。
• 辉达同时澄清，新产品H20的销售不会影响H100、H200或下一代Blackwell芯片的供应能力。
• 尽管官方出面澄清，辉达股价在9月2日仍受传闻影响，收盘下跌1.97%。

相关链接：

• H100、H200晶片遭傳供貨吃緊？輝達火速回應了

Nvidia Denies H100/H200 Chip Shortage Rumors 🚀

Brief:

• Addressing market rumors of tight supply or even a sell-out of H100 and H200 chips, Nvidia publicly responded on September 2nd, stating the rumors are false and the company possesses sufficient inventory to fulfill all orders. 📦
• Nvidia explained that while its cloud partners can rent out every chip that goes online, this does not mean the company cannot meet new order demands.
• Nvidia also clarified that the sales of its new H20 product will not affect the supply capacity of H100, H200, or the next-generation Blackwell chips.
• Despite the official clarification, Nvidia's stock price was still impacted by the rumors on September 2nd, closing down by 1.97%. 📉

Related Link:

• Are H100 and H200 chips rumored to be in tight supply? Nvidia quickly responds!