单词学习：![[2026051160836175.mp3]]

新闻播报：![[2026051151239923.mp3]]

淘宝深度整合通义千问AI，开启对话式智能购物新模式

简报：

• 阿里巴巴将通义千问AI与淘宝深度整合，接入淘宝和天猫超40亿款商品，通过对话式交互替代传统关键词搜索。
• 用户在通义千问应用中可通过与AI智能体对话，直接完成商品浏览、比价、下单，并获取物流调度到售后服务的全方位支持。
• 淘宝APP内同步上线由通义千问赋能的AI购物助手，可根据用户偏好智能推荐商品，并提供虚拟试穿和30天价格走势追踪等实用功能。
• 该模式将人工智能技术深入嵌入真实交易全流程，凸显了国内电商与海外平台在AI应用策略上的发展差异。

相关链接：

• 阿里巴巴推出通义千问 AI，革新淘宝购物体验

Taobao Deeply Integrates Tongyi Qianwen AI, Ushering In a New Era of Conversational Smart Shopping 🛍️🤖

Brief:

• Alibaba has deeply integrated Tongyi Qianwen AI with Taobao, connecting it to more than 4 billion products across Taobao and Tmall, and replacing traditional keyword searches with conversational interactions.
• Within the Tongyi Qianwen app, users can directly browse products, compare prices, and place orders through conversations with AI agents, while also receiving comprehensive support ranging from logistics coordination to after-sales service.
• An AI shopping assistant powered by Tongyi Qianwen has also been launched inside the Taobao app, where it can intelligently recommend products based on user preferences and offer practical features such as virtual try-ons and 30-day price trend tracking.
• This model embeds artificial intelligence deeply into the entire real-world transaction process, highlighting the differences in AI application strategies between domestic e-commerce platforms and their overseas counterparts. 🚀

Related Link:

• Alibaba Launches Tongyi Qianwen AI, Transforming the Taobao Shopping Experience

联发科台湾首建研发资料中心，全面推进AI发展战略

简报：

• 联发科在台湾地区正式设立了首座研发资料中心。
• 该设施的建立旨在进一步加速联发科在人工智能（AI）领域的战略布局与发展。

相关链接：

• 联发科在台设立首座研发资料中心，加速 AI 布局

MediaTek Establishes Its First R&D Data Center in Taiwan, Fully Advancing Its AI Development Strategy 🚀

Brief:

• MediaTek has officially established its first R&D data center in Taiwan.
• The facility is intended to further accelerate MediaTek’s strategic expansion and development in the field of artificial intelligence (AI) 🤖.

Related Link:

• MediaTek Establishes Its First R&D Data Center in Taiwan to Accelerate AI Expansion

谷歌Gemini API文件搜索迎大升级：多模态RAG支持原生图文解析与页面级溯源

简报：

• 谷歌宣布对Gemini API的文件搜索功能进行重大升级，为开发者提供更完善的多模态检索增强生成（RAG）能力。
• 新版功能基于Gemini Embedding2模型构建，开发者无需搭建复杂的向量数据库，即可在API内部完成从数据上传到信息检索的完整RAG工作流。
• 系统具备统一的多模态嵌入能力，可原生识别和处理PDF、文档及图片等视觉信息，解决了传统RAG系统无法有效读取图表、设计图等非文本内容的痛点。
• 新增自定义元数据过滤功能，开发者可通过添加部门、时间、分类等标签条件进行预设过滤，使AI输出的回答更加聚焦。
• 支持页面级引用溯源，AI生成答案时会明确标注信息来源的具体文档页码，大幅提升内容核实效率与透明度。
• 目前该增强版文件搜索功能已向全球开发者开放，支持通过Google AI Studio或Google Cloud平台接入体验。

相关链接：

• Google 升级 Gemini API 文件搜索：多模态 RAG 能力实现全方位跨越

Google Gemini API File Search Gets a Major Upgrade: Multimodal RAG Now Supports Native Image-Text Parsing and Page-Level Source Tracing 🚀

Brief:

• Google has announced a major upgrade to the file search capabilities of the Gemini API, providing developers with more comprehensive multimodal retrieval-augmented generation (RAG) features.
• Built on the Gemini Embedding2 model, the new functionality allows developers to complete the entire RAG workflow—from data upload to information retrieval—directly within the API, without the need to build a complex vector database.
• The system offers unified multimodal embedding capabilities and can natively recognize and process visual information in PDFs, documents, and images, addressing a key limitation of traditional RAG systems that struggle to effectively interpret non-text content such as charts and design diagrams. 🖼️
• A new custom metadata filtering feature allows developers to preset filtering conditions by adding tags such as department, time, and category, enabling AI-generated responses to be more focused.
• Page-level citation tracing is now supported, allowing AI-generated answers to clearly indicate the exact document page where the information originated, greatly improving verification efficiency and transparency. 🔎
• The enhanced file search feature is now available to developers worldwide and can be accessed through Google AI Studio or the Google Cloud platform.

Related link:

• Google Upgrades Gemini API File Search: Multimodal RAG Capabilities Achieve a Comprehensive Leap

单词学习：![[2026050666508954.mp3]]

新闻播报：![[2026050648223011.mp3]]

智源联合多所医院推出业内首个心脏磁共振多模态智能体

简报：

• 北京智源人工智能研究院联合北京安贞医院、河南医药大学第一附属医院，正式发布业内首个心脏磁共振多模态智能体BAAI Cardiac Agent。
• 该系统基于Agent-Expert架构，能够动态协调多个“专家”子模型，对心脏核磁共振影像完成端到端自动化分析。
• 智能体实现了一站式“结构分割与分析-功能定量评估-疾病诊断与分类-智能化报告”的全流程诊疗闭环，可自动输出符合临床规范的标准化报告。

相关链接：

• 智源发布业内首个心脏磁共振多模态诊断智能体BAAI Cardiac Agent

BAAI and Multiple Hospitals Launch the Industry’s First Multimodal Cardiac MRI Agent ❤️🩺🤖

Brief:

• The Beijing Academy of Artificial Intelligence, in collaboration with Beijing Anzhen Hospital and the First Affiliated Hospital of Henan Medical University, has officially released the industry’s first multimodal cardiac MRI agent, BAAI Cardiac Agent.
• Built on an Agent-Expert architecture, the system can dynamically coordinate multiple “expert” sub-models to perform end-to-end automated analysis of cardiac MRI images.
• The agent enables a one-stop, full-process clinical closed loop covering “structural segmentation and analysis - functional quantitative assessment - disease diagnosis and classification - intelligent reporting,” and can automatically generate standardized reports that comply with clinical norms.

Related Link:

• BAAI Releases the Industry’s First Multimodal Diagnostic Agent for Cardiac MRI, BAAI Cardiac Agent

台积电高层就加码对美投资事宜作出回应

简报：

• 由于提供的参考资料为空，无法提取台积电高层关于是否“再加码对美投资”的具体表态与详细决策内容。

相关链接：

• 再加码对美投资？台积电高层这样说

TSMC Executives Respond to the Possibility of Increasing Investment in the U.S. 🇺🇸🏭

Briefing:

• As the provided reference materials are empty, it is not possible to extract TSMC executives’ specific statements or detailed decision-making content regarding whether the company will “further increase its investment in the U.S.”

Related Link:

• Increase investment in the U.S. again? Here’s what TSMC executives said

AI需求助推三星电子跻身“万亿美元俱乐部”

简报：

• 受人工智能相关股票需求蓬勃发展的推动，三星电子市值成功突破1万亿美元。
• 三星电子成为继台积电之后，第二家跻身万亿美元市值俱乐部的东亚企业。

相关链接：

• 三星电子借AI热潮市值突破1万亿美元

AI Demand Propels Samsung Electronics into the “Trillion-Dollar Club” 🚀

Briefing:

• Driven by booming demand for AI-related stocks, Samsung Electronics has successfully surpassed a market capitalization of dJyH6kDkfC5RlM16.answerText trillion.
• Samsung Electronics has become the second East Asian company to join the trillion-dollar market cap club, following TSMC. 📈

Related Link:

• Samsung Electronics Surpasses dJyH6kDkfC5RlM16.answerText Trillion in Market Value Amid the AI Boom

新闻播报：![[2026050571812800.mp3]]

OpenAI加速开发首款AI Agent手机，最快于2027年上半年量产并搭载定制天玑芯片

简报：

• 天风国际分析师郭明錤透露，OpenAI可能正加速首款AI Agent手机开发，目标最快于2027年上半年量产。
• 新机预计采用基于天玑9600的定制版本处理器，有望由联发科独家获得订单，并于2026年下半年由台积电N2P工艺生产。
• 该机关键规格包含双NPU架构（AI算力分层）、LPDDR6 + UFS 5.0（缓解内存瓶颈）以及pKVM + inline hashing（提升安全性）等。
• 若新机开发顺利，预计2027与2028年共计出货约3000万部。
• 郭明錤指出，AI Agent将从根本上重新定义手机，用户不再仅是打开App，而是通过手机直接执行任务满足需求。

相关链接：

• 郭明锜：OpenAI 首款 AI Agent 手机目标最快于 2027 年上半年量产
• 郭明錤：OpenAI 联手高通 & 联发科开发手机芯片，自研手机项目预计 2028 年量产

OpenAI Accelerates Development of Its First AI Agent Smartphone, with Mass Production as Early as H1 2027 and a Custom Dimensity Chip 📱🤖

Brief:

• TF International analyst Ming-Chi Kuo revealed that OpenAI may be accelerating development of its first AI Agent smartphone, with mass production targeted as early as the first half of 2027.
• The new device is expected to use a customized processor based on the Dimensity 9600. MediaTek is likely to secure the order exclusively, with production scheduled for the second half of 2026 using TSMC’s N2P process.
• Key specifications are expected to include a dual-NPU architecture (for layered AI computing power), LPDDR6 + UFS 5.0 (to ease memory bottlenecks), and pKVM + inline hashing (to enhance security) 🔐.
• If development proceeds smoothly, total shipments are projected to reach around 30 million units across 2027 and 2028.
• Kuo noted that AI Agents will fundamentally redefine smartphones: instead of simply opening apps, users will rely on their phones to directly execute tasks and fulfill their needs.

Related links:

• Ming-Chi Kuo: OpenAI’s first AI Agent smartphone targets mass production as early as H1 2027
• Ming-Chi Kuo: OpenAI teams up with Qualcomm & MediaTek to develop smartphone chips; self-developed phone project expected to enter mass production in 2028

英特尔重组领导层：高通前高管执掌客户端计算与AI，正式任命新CTO

简报：

• 英特尔宣布高通前执行副总裁 Alex Katouzian 将于5月加盟，出任执行副总裁兼客户端计算与物理AI事业部总经理。
• 公司正式任命 Pushkar Ranade 为首席技术官，其此前担任代理首席技术官一职。
• 新任首席技术官将负责推进公司技术战略，重点推动量子计算、神经拟态计算、光子学及新型材料等前沿领域。
• Katouzian 与 Ranade 两位高管未来将直接向英特尔首席执行官陈立武汇报工作。

相关链接：

• 高通前高管将加入英特尔，领导客户端计算与物理AI事业部
• 英特尔宣布两项重要领导层任命

Intel Reshuffles Leadership: Former Qualcomm Executive to Lead Client Computing and AI, New CTO Officially Appointed 🚀

Brief:

• Intel announced that former Qualcomm Executive Vice President Alex Katouzian will join the company in May as Executive Vice President and General Manager of the Client Computing and Physical AI Business.
• The company has officially appointed Pushkar Ranade as Chief Technology Officer, following his previous role as interim CTO.
• The new CTO will be responsible for advancing the company’s technology strategy, with a focus on cutting-edge areas including quantum computing, neuromorphic computing, photonics, and new materials.
• Both Katouzian and Ranade will report directly to Intel CEO Lip-Bu Tan in their new roles. 🤖

Related Links:

• Former Qualcomm Executive to Join Intel to Lead Client Computing and Physical AI Business
• Intel Announces Two Major Leadership Appointments

Meta完成对物理AI初创企业ARI的收购，团队将并入超级智能实验室

简报：

• Meta于本月1日完成对物理AI模型初创企业Assured Robot Intelligence（ARI）的收购，具体交易财务条款未披露。
• ARI创立仅一年，其核心使命是实现物理通用人工智能（AGI），致力于研发能直接从人类经验中学习的类人形态通用物理智能体。
• Meta表示，ARI团队将带来在模型设计、机器人控制及自主学习前沿技术方面的深厚专业知识，并将应用于全身人型机器人控制领域。
• ARI联合创始人Wang Xiaolong确认，团队将加入Meta Superintelligence Labs，共同推动个人超级智能进入物理世界。

相关链接：

• 成立仅 1 年，物理 AI 初创企业 ARI 被 Meta 收购

Meta Completes Acquisition of Physical AI Startup ARI, Team to Join Superintelligence Labs 🤖

Brief:

• Meta completed its acquisition of physical AI model startup Assured Robot Intelligence (ARI) on the 1st of this month. The financial terms of the deal were not disclosed.
• Founded just one year ago, ARI’s core mission is to achieve physical artificial general intelligence (AGI), focusing on developing humanoid general-purpose physical agents capable of learning directly from human experience.
• Meta said the ARI team brings deep expertise in model design, robot control, and cutting-edge autonomous learning technologies, which will be applied to full-body humanoid robot control.
• ARI co-founder Wang Xiaolong confirmed that the team will join Meta Superintelligence Labs to help bring personal superintelligence into the physical world 🚀

Related link:

• Just 1 Year After Its Founding, Physical AI Startup ARI Acquired by Meta

黄仁勋力挺人工智能：是工业规模的就业生成器而非“失业预兆”

简报：

• 英伟达首席执行官黄仁勋于2026年5月4日晚公开反驳“人工智能末日论”，称AI是工业规模的就业机会生成器。
• 黄仁勋认为AI是美国实现“再工业化”的最佳机遇，新型工业工厂在生产AI基础设施硬件时必然需要大量工人。
• 他强调自动化某个特定任务并不等同于取代整份工作，外界误解了工作任务与工作目的之间的关系。
• 黄仁勋对“AI将统治人类”等科幻式担忧表示批评，担心过度恐惧会导致人们不敢参与和利用这项技术。
• 他重申关键在于让所有人掌握AI工具，因为人们更可能被“会使用AI的人”取代，而非被AI本身取代。

相关链接：

• 英伟达黄仁勋称人工智能创造就业机会，反驳“AI 末日论”
• 黄仁勋：人们更可能被会使用 AI 的人取代

Jensen Huang Backs Artificial Intelligence: A Generator of Jobs at Industrial Scale, Not a “Harbinger of Unemployment” 🤖🏭

Briefing:

• On the evening of May 4, 2026, Nvidia CEO Jensen Huang publicly pushed back against “AI doomerism,” saying that AI is a generator of job opportunities at industrial scale.
• Huang said AI represents the best opportunity for the United States to achieve “reindustrialization,” and that new industrial factories producing AI infrastructure hardware will inevitably require a large workforce.
• He emphasized that automating a specific task does not equal replacing an entire job, and that many people misunderstand the relationship between job tasks and the purpose of a job.
• Huang criticized science-fiction-style fears such as “AI will dominate humanity,” warning that excessive fear could discourage people from engaging with and making use of the technology.
• He reiterated that the key is to ensure everyone learns to use AI tools, because people are more likely to be replaced by “those who know how to use AI” than by AI itself. 📈

Related links:

• Nvidia’s Jensen Huang Says Artificial Intelligence Creates Jobs, Rebuts “AI Doomerism”
• Jensen Huang: People Are More Likely to Be Replaced by Those Who Know How to Use AI

苹果为缓解芯片产能瓶颈，正与英特尔及三星洽谈代工合作

简报：

• 苹果公司正与英特尔、三星电子进行初步洽谈，计划将其作为台积电之外的设备主处理器备选供应商，目前尚未下达任何订单。
• 受近期AI数据中心建设热潮及Mac需求超预期导致的芯片短缺影响，苹果考虑增加供应商以缓解产能瓶颈、避免增长受限。
• 苹果目前对采用非台积电技术仍存有顾虑。
• 苹果、英特尔、三星及台积电的发言人均对该消息拒绝置评。

相关链接：

• 苹果考虑让英特尔、三星代工设备处理器

Apple in Talks With Intel and Samsung on Foundry Partnerships to Ease Chip Capacity Bottlenecks

Briefing:

• Apple is holding preliminary talks with Intel and Samsung Electronics about potentially adding them as alternative suppliers of main processors for its devices alongside TSMC, though no orders have been placed so far.
• Amid recent chip shortages driven by the boom in AI data center construction and stronger-than-expected Mac demand, Apple is considering expanding its supplier base to ease capacity constraints and avoid limiting growth 📈.
• Apple still has reservations about adopting manufacturing technology other than TSMC’s.
• Spokespeople for Apple, Intel, Samsung, and TSMC all declined to comment on the report 🤐.

Related link:

• Apple Considers Having Intel and Samsung Manufacture Device Processors

单词学习：![[2026050399174248.mp3]]

新闻播报：![[2026050316627210.mp3]]

ChatGPT Images2.0上线：印度首周下载量达500万领跑全球，CEO重申AI不取代人类

简报：

• OpenAI最新推出图像生成工具ChatGPT Images2.0，强化了印地语等非拉丁文本的渲染能力，并引入“思考”功能以支持图像变体生成与修复。
• 全球首发反响呈现地域差异，全球首周下载量环比增长11%，而以印度为首的新兴市场（含巴基斯坦、越南等）推广期间周增幅高达79%。
• 印度成为该功能首周最大的用户群体所在地，下载量约500万次，显著高于美国的200万次，用户广泛将其用于肖像、头像等“自我表达”创作。
• 5月2日，OpenAI CEO萨姆·奥尔特曼在社交平台表示，公司的目标是打造增强和提升人类能力的工具，而不是创造取代人类的实体，并反驳了“AI取代工作”的长远悲观论调。

相关链接：

• OpenAI 发布 ChatGPT Images2.0，印度市场贡献首周最大用户增量
• OpenAI CEO 萨姆 · 奥尔特曼：我们不会创造取代人类的实体

ChatGPT Images 2.0 Launches: India Leads the World with 5 Million Downloads in the First Week, as CEO Reaffirms AI Will Not Replace Humans 🚀

Brief:

• OpenAI has launched its latest image-generation tool, ChatGPT Images 2.0, with enhanced rendering capabilities for non-Latin text such as Hindi, and a new “thinking” feature to support image variant generation and inpainting. 🎨
• The global debut has shown clear regional differences in reception. Worldwide, first-week downloads rose 11% week-over-week, while emerging markets led by India—including Pakistan and Vietnam—saw weekly growth during the promotion period reach as high as 79%.
• India became the largest user base for the feature in its first week, with approximately 5 million downloads, significantly surpassing the United States at 2 million. Users have widely adopted it for “self-expression” creations such as portraits and avatars.
• On May 2, OpenAI CEO Sam Altman stated on social media that the company’s goal is to build tools that augment and elevate human capabilities, rather than create entities that replace humans, pushing back against long-term pessimistic claims that “AI will take away jobs.” 🤖

Related Links:

• OpenAI Releases ChatGPT Images 2.0, with India Contributing the Largest First-Week User Growth
• OpenAI CEO Sam Altman: We Will Not Create Entities That Replace Humans

奥尔特曼回应 AI 取代论：OpenAI 致力于开发赋能人类的工具

简报：

• OpenAI 首席执行官萨姆·奥尔特曼在 X 平台明确表示，公司的目标是打造增强和提升人类能力的工具，而非创造取代人类的实体。
• 针对当前各行业因 AI 引发的裁员潮及焦虑情绪，奥尔特曼认为长期看这种悲观主义是错误的，未来人们将能找到更有意义的工作。
• 奥尔特曼描述了一个美好的未来愿景，称在 AI 的辅助下，即使人们不想努力工作，依然可以享受“繁荣而精彩的生活”。
• 与此同时，OpenAI 发布了 ChatGPT Images 2.0 图像生成工具，通过强化非拉丁文本渲染及“思考”功能，在印度市场首周下载量突破 500 万次。
• 行业数据显示，虽然全球 AI 用户增长趋于温和，但在印度、越南等新兴市场，多模态 AI 工具的本地化应用正成为新的增长引擎。

相关链接：

• OpenAI CEO 萨姆 · 奥尔特曼：我们不会创造取代人类的实体
• OpenAI 发布 ChatGPT Images 2.0，印度市场贡献首周最大用户增量

Altman Responds to AI Replacement Claims: OpenAI Is Committed to Building Tools That Empower Humans 🤖✨

Brief:

• OpenAI CEO Sam Altman stated clearly on X that the company’s goal is to build tools that augment and enhance human capabilities, rather than create entities that replace people.
• In response to the current wave of AI-related layoffs and growing anxiety across industries, Altman argued that such pessimism is wrong in the long run, and that people will be able to find more meaningful work in the future.
• Altman outlined an optimistic vision of the future, saying that with AI’s assistance, people could still enjoy a “prosperous and wonderful life” even if they do not want to work hard.
• At the same time, OpenAI released its ChatGPT Images 2.0 image-generation tool. By strengthening non-Latin text rendering and “reasoning” capabilities, the product surpassed 5 million downloads in its first week in the Indian market 📈.
• Industry data shows that while global AI user growth is becoming more moderate, localized applications of multimodal AI tools in emerging markets such as India and Vietnam are becoming a new engine of growth.

Related links:

• OpenAI CEO Sam Altman: We Will Not Create Entities That Replace Humans
• OpenAI Launches ChatGPT Images 2.0, with India Contributing the Largest First-Week User Increase

四巨头今年AI资本开支上调至7250亿美元，微软暗示软件收费模式迎变

简报：

• 谷歌、亚马逊、微软及Meta 2026年计划在AI领域共投入7250亿美元，较去年大幅增长77%，仅今年一季度四家巨头已累计支出1300亿美元。
• 各大厂商AI投入呈翻倍式增长：亚马逊计划投入超2000亿美元，微软与Meta开支预计均达1900亿美元，Alphabet投入突破1450亿美元。
• 尽管面临内存芯片及核心组件价格上涨带来的成本压力，谷歌与微软均表示现有计算能力仍远未满足爆发式的市场需求。
• 微软CEO萨蒂亚·纳德拉暗示，软件收费模式正从固定的“按席位计费”转向“席位费+使用费”的混合模式，AI算力成本将向B端终端客户传导。

相关链接：

• 四大科技巨头2026年 AI 资本开支上调至7250亿美元，同比激增77%

Big Four Raise AI Capital Expenditure to $725 Billion This Year, as Microsoft Signals a Shift in Software Pricing Models 🤖💰

Briefing:

• Google, Amazon, Microsoft, and Meta plan to invest a combined $725 billion in AI in 2026, up 77% from last year. In the first quarter of this year alone, the four giants had already spent a total of dJyH6kDkfC5RlM16.answerText30 billion.
• AI spending by the major tech firms is rising at a near-doubling pace: Amazon plans to invest more than $200 billion, while both Microsoft and Meta are each expected to spend dJyH6kDkfC5RlM16.answerText90 billion, and Alphabet’s investment will exceed dJyH6kDkfC5RlM16.answerText45 billion.
• Despite cost pressures from rising prices for memory chips and other core components, both Google and Microsoft said their existing computing capacity still falls far short of meeting surging market demand.
• Microsoft CEO Satya Nadella indicated that software pricing models are shifting from a fixed “per-seat subscription” structure to a hybrid model of “seat fee + usage-based fee,” with AI computing costs likely to be passed on to enterprise customers 📈.

Related link:

• The Four Tech Giants Raise 2026 AI Capital Expenditure to $725 Billion, Up 77% Year-on-Year

Meta 收购人形机器人初创公司 ARI，加速布局“具身智能”与物理 AGI

简报：

• Meta 宣布正式收购由前英伟达研究员王晓龙及纽约大学教授勒雷尔·平托创立的人形机器人初创公司 ARI。
• ARI 核心团队将并入 Meta 超级智能实验室（Superintelligence Labs），其在自学习模型和全身人形机器人控制技术方面的积淀将助力 Meta 实现硬件与模型协同的人形机器人战略。
• 此项收购旨在强化 Meta 在复杂动态环境下机器人理解与预测人类行为的能力，支持通用人工智能（AGI）在物理世界通过交互进行训练。
• 泄露的备忘录显示 Meta 内部已启动针对消费者市场的 AI 机器人研究计划，旨在通过“具身智能”突破数字数据训练的局限。
• 随着亚马逊此前收购同赛道公司 Fauna Robotics，人形机器人领域的巨头竞争与资本动作正在显著加速。

相关链接：

• 剑指 AGI 物理训练，Meta 吞并 ARI 补齐全身人形机器人控制版图

Meta Acquires Humanoid Robotics Startup ARI, Accelerating Its Push into “Embodied Intelligence” and Physical AGI 🤖

Brief:

• Meta has officially announced the acquisition of ARI, a humanoid robotics startup founded by former NVIDIA researcher Xiaolong Wang and New York University professor Lerrel Pinto.
• ARI’s core team will be integrated into Meta’s Superintelligence Labs, where its expertise in self-learning models and full-body humanoid robot control technologies will support Meta’s strategy of combining hardware and models for humanoid robotics.
• The acquisition is intended to strengthen Meta’s capabilities in enabling robots to understand and predict human behavior in complex, dynamic environments, while supporting the training of artificial general intelligence (AGI) in the physical world through interaction.
• A leaked memo shows that Meta has already launched an internal AI robotics research initiative aimed at the consumer market, with the goal of using “embodied intelligence” to overcome the limitations of training solely on digital data.
• Following Amazon’s earlier acquisition of Fauna Robotics in the same sector, competition among major players and capital activity in the humanoid robotics field are accelerating significantly 📈.

Related link:

• Targeting Physical Training for AGI, Meta Acquires ARI to Strengthen Its Full-Body Humanoid Robot Control Capabilities 🚀

专家警告AI替代初级员工将透支未来，科技巨头对AI与就业影响态度不一

简报：

• 麻省理工学院研究科学家安德鲁·麦卡菲警告，企业用AI取代Z世代初级岗位将破坏未来管理者和核心人才的培养通道，并牺牲未来学习机会。
• 德勤研究数据显示，Z世代是所有世代中最熟悉AI的群体，约76%的Z世代使用独立AI工具。
• 亚马逊AWS CEO否认AI会让程序员失业，表示公司对软件开发者的需求正在加速增长，今年计划招聘11000名软件开发工程师实习生。
• Meta CEO扎克伯格在全员大会上澄清，公司近期裁员是因为增加AI资本开支，需要适当缩小规模，并非因为员工被AI替代。
• Meta预计在今年5月20日裁掉约10%的员工，并筹划下半年继续裁员；同时因追踪员工操作以训练AI及裁员解释不充分，引发了部分员工的内部抗议。

相关链接：

• 麻省理工 AI 专家警告：用自动化替代 Z 世代入门员工是“透支未来”
• 亚马逊 AWS CEO：AI 并未让程序员丢饭碗，今年计划招聘 11000 名实习生
• Meta CEO 扎克伯格谈裁员：是因为资本支出、并非“员工被 AI 替代”

Experts Warn That Replacing Junior Employees with AI Could Mortgage the Future, as Tech Giants Differ on AI’s Impact on Jobs 🤖📉

Briefing:

• Andrew McAfee, a research scientist at MIT, warned that companies replacing entry-level Gen Z roles with AI would undermine the pipeline for developing future managers and core talent, while also sacrificing future learning opportunities.
• Research from Deloitte shows that Gen Z is the most AI-familiar generation of all, with about 76% of Gen Z using standalone AI tools.
• Amazon AWS CEO denied that AI will make programmers unemployed, saying the company’s demand for software developers is accelerating, and that it plans to hire 11,000 software development engineer interns this year.
• Meta CEO Mark Zuckerberg clarified at an all-hands meeting that the company’s recent layoffs were due to increased capital expenditures on AI, which required an appropriate reduction in size, rather than employees being replaced by AI.
• Meta is expected to lay off about 10% of its workforce on May 20 this year and is planning further layoffs in the second half of the year; at the same time, internal protests have emerged among some employees over the company’s tracking of employee activity to train AI and what they see as insufficient explanations for the layoffs. ⚠️

Related links:

• MIT AI Expert Warns: Replacing Entry-Level Gen Z Employees with Automation Is “Mortgaging the Future”
• Amazon AWS CEO: AI Has Not Cost Programmers Their Jobs, Company Plans to Hire 11,000 Interns This Year
• Meta CEO Zuckerberg on Layoffs: Driven by Capital Expenditures, Not Because “Employees Were Replaced by AI”

亚马逊AWS CEO否认AI取代程序员，计划招聘1.1万名实习生并警惕用工结构断层

简报：

• 亚马逊AWS首席执行官Matt Garman明确表示AI并未让程序员丢饭碗，公司对软件开发人员的需求正加速增长，今年计划招聘1.1万名软件开发工程师实习生。
• Matt Garman指出，虽然未来编写Java代码等传统技能重要性下降，但软件工程师仍需掌握构建应用、解决客户问题等复合技能及核心技术知识。
• 麻省理工学院专家安德鲁·麦卡菲对此发出警示，认为企业若用自动化替代Z世代入门员工是“透支未来”，不仅破坏了培养未来管理者的“学徒阶梯”，也会失去对AI工具最熟练的人才来源。
• 与亚马逊增加招聘形成对比，Meta CEO扎克伯格表示因增加AI资本开支需要缩小公司规模，计划于5月20日裁员约10%，并表示后续裁员并非由员工被AI替代导致，而是资金分配调整。
• Meta内部正推行追踪员工操作以训练AI智能体的计划，引发员工不满；而德勤研究显示Z世代使用独立AI工具的比例达76%，在各世代中最高。

相关链接：

• 亚马逊 AWS CEO：AI 并未让程序员丢饭碗，今年计划招聘 11000 名实习生
• 麻省理工 AI 专家警告：用自动化替代 Z 世代入门员工是“透支未来”
• Meta CEO 扎克伯格谈裁员：是因为资本支出、并非“员工被 AI 替代”

Amazon AWS CEO Denies AI Is Replacing Programmers, Plans to Hire 11,000 Interns and Warns of a Structural Employment Gap

Briefing:

• Amazon AWS CEO Matt Garman explicitly stated that AI has not put programmers out of work. Instead, the company’s demand for software developers is accelerating, and it plans to hire 11,000 software development engineer interns this year. 🚀
• Matt Garman noted that while traditional skills such as writing Java code may become less important in the future, software engineers will still need a combination of skills, including building applications, solving customer problems, and mastering core technical knowledge.
• MIT expert Andrew McAfee issued a warning on this trend, arguing that if companies use automation to replace entry-level Gen Z employees, they are “borrowing from the future.” This would not only undermine the “apprenticeship ladder” used to cultivate future managers, but also cut off a key source of talent most proficient with AI tools. ⚠️
• In contrast to Amazon’s hiring expansion, Meta CEO Mark Zuckerberg said the company needs to reduce its workforce due to increased AI-related capital expenditures, with plans to lay off about 10% of staff on May 20. He also emphasized that the layoffs are not the result of employees being replaced by AI, but rather a matter of budget reallocation.
• Meta is internally pushing a plan to track employee actions in order to train AI agents, which has sparked dissatisfaction among staff. Meanwhile, Deloitte research shows that 76% of Gen Z use standalone AI tools, the highest rate among all generations. 🤖

Related links:

• Amazon AWS CEO: AI Has Not Put Programmers Out of Work, Plans to Hire 11,000 Interns This Year
• MIT AI Expert Warns That Replacing Entry-Level Gen Z Employees With Automation Is “Borrowing From the Future”
• Meta CEO Zuckerberg on Layoffs: Due to Capital Expenditures, Not Because “Employees Are Being Replaced by AI”

英特尔EMIB先进封装良率达90%，吸引谷歌、辉达等科技巨头导入

简报：

• 据广发证券分析师Jeff Pu透露，英特尔关键的EMIB先进封装技术良率已飙升至90%，为英特尔晶圆代工事业带来强力支撑。
• 因台积电CoWoS产能满载，英特尔EMIB技术成为业界热门替代方案，传闻谷歌（下一代TPU）、辉达（未来的Feynman芯片）以及Meta均准备或已计划采用。
• 英特尔官方强调EMIB技术具备四大优势：高良率、低功耗、低成本，并能实现超大规模的“混合节点”系统。
• 目前EMIB主攻EMIB-M与EMIB-T两大阵营，其中EMIB-M桥接电路内置MIM电容，虽造价略高于传统MOM电容，但能大幅压制杂讯，换取更高的稳定度与极低的漏电率。

相关链接：

• 台積電霸權動搖？英特爾封裝良率狂飆90%！三大科技巨頭全面搶結盟

Intel’s EMIB Advanced Packaging Yield Reaches 90%, Attracting Tech Giants Such as Google and Nvidia 🚀

Briefing:

• According to Jeff Pu, an analyst at GF Securities, the yield of Intel’s critical EMIB advanced packaging technology has surged to 90%, providing strong support for Intel’s foundry business.
• With TSMC’s CoWoS capacity fully booked, Intel’s EMIB technology has become a popular alternative in the industry. It is rumored that Google (for its next-generation TPU), Nvidia (for its future Feynman chip), and Meta are either preparing to adopt it or have already planned to do so.
• Intel officially emphasizes that EMIB offers four major advantages: high yield, low power consumption, low cost, and the ability to enable ultra-large-scale “hybrid-node” systems.
• At present, EMIB is mainly focused on two major platforms: EMIB-M and EMIB-T. Among them, EMIB-M integrates MIM capacitors into the bridge circuitry. Although it is slightly more expensive than traditional MOM capacitors, it can significantly suppress noise, delivering higher stability and extremely low leakage rates. ⚙️

Related link:

• Is TSMC’s dominance being challenged? Intel’s packaging yield soars to 90% as three major tech giants rush to partner up 📈

单词学习：![[2026042911124479.mp3]]

新闻播报：![[2026042940433971.mp3]]

OpenAI为强化GPT-5应用数据采集，其网络爬虫活跃度激增300%

简报：

• 随着GPT-5步入应用阶段，自2025年8月以来，OpenAI旗下爬虫程序的活跃度激增约300%，以获取实时信息与高质量训练数据。
• 实时检索工具“OAI-SearchBot”的日志事件量已超越传统模型训练爬虫“GPTBot”，反映出ChatGPT的重心正向提供具时效性的搜索反馈转移。
• 医疗、媒体及出版行业网站接收到的爬虫访问量增长数倍，OpenAI正优化处理逻辑，将新闻类查询引导至实时搜索。
• 目前OpenAI的爬虫总量约为谷歌的4%，但双方差距正快速缩小，这促使网站运营者在保护数据版权与获取AI搜索流量之间面临新的抉择。

相关链接：

• GPT-5引发链式反应：OpenAI爬虫活动疯狂激增三倍

OpenAI’s Web Crawlers Surge 300% as It Expands Data Collection for GPT-5 Applications 🚀

Brief:

• As GPT-5 moves into the application stage, activity from OpenAI’s crawler programs has surged by about 300% since August 2025, in order to obtain real-time information and high-quality training data.
• Log event volume from the real-time retrieval tool “OAI-SearchBot” has now surpassed that of the traditional model-training crawler “GPTBot,” reflecting a shift in ChatGPT’s focus toward delivering timely search-based responses.
• Websites in the healthcare, media, and publishing sectors have seen crawler traffic increase severalfold, as OpenAI optimizes its processing logic and directs news-related queries toward real-time search. 📈
• At present, OpenAI’s total crawler volume is about 4% of Google’s, but the gap between the two is narrowing rapidly, forcing website operators to make new choices between protecting data copyright and gaining AI search traffic.

Related link:

• GPT-5 Triggers a Chain Reaction: OpenAI’s Crawler Activity Skyrockets Threefold 🔎

英伟达发布多模态模型Nemotron 3 Nano Omni，系统吞吐量较同类提升9倍

简报：

• 英伟达发布开放式多模态模型“Nemotron 3 Nano Omni”，将视频、音频、图像和文本的推理能力整合到一个系统中。
• 该模型采用30B-A3B混合专家架构，内置视觉与音频编码器，无需依赖额外感知模型，显著提升大规模推理效率。
• 模型具备快速解读全高清屏幕录像的能力，在复杂文档解析、视频及音频理解等方面名列六大权威排行榜前茅。
• 其AI系统的吞吐量比同类模型高出9倍，目前已与多家公司的系统进行协作应用。
• 过去一年内，包含Nano、Super和Ultra型号在内的Nemotron 3系列模型累计下载量已突破5000万次。

相关链接：

• 英伟达推出新一代多模态模型，智能体效率提升九倍

NVIDIA Unveils Multimodal Model Nemotron 3 Nano Omni, Delivering 9x Higher System Throughput Than Peers 🚀

Brief:

• NVIDIA has released the open multimodal model “Nemotron 3 Nano Omni,” integrating reasoning capabilities for video, audio, images, and text into a single system.
• The model uses a 30B-A3B mixture-of-experts architecture with built-in vision and audio encoders, eliminating the need for additional perception models and significantly improving large-scale inference efficiency.
• It can rapidly interpret full-HD screen recordings and ranks among the top performers on six leading benchmarks for complex document parsing, video understanding, and audio comprehension. 🎯
• Its AI system delivers 9x higher throughput than comparable models and has already been deployed in collaborative applications with systems from multiple companies.
• Over the past year, cumulative downloads of the Nemotron 3 series—including the Nano, Super, and Ultra variants—have surpassed 50 million.

Related link:

• NVIDIA launches next-generation multimodal model, boosting agent efficiency by nine times

腾讯混元Hy3 preview问鼎OpenRouter全球API调用量总榜，国产AI加速场景落地

简报：

• 腾讯混元Hy3 preview模型在最新发布的OpenRouter全球大模型API调用量排行榜中超越国际竞品，位列总榜第一。
• 分场景统计数据显示，混元Hy3 preview在“工具调用（Tool Use）”维度高居榜首，在“编程（Coding）”领域稳居全球第二。
• 国产AI集体发力，阶跃星辰同日发布新一代图像生成编辑模型Step Image Edit2，并已在开放平台全面上线。
• 腾讯AI应用与生态持续拓展：今日（4月29日），腾讯文档正式接入WorkBuddy资料库，实现AI工作流原生化；此外，腾讯云于昨日与虹魔方签署战略合作协议，整合混元大模型等技术共同开发新一代拟人化AI玩偶。

相关链接：

• 国产大模型海外“屠榜”！混元Hy3 preview登顶全球大模型调用量榜单
• WorkBuddy正式打通腾讯文档
• 腾讯云与虹魔方达成战略合作，深耕AI玩具新生态

Tencent Hunyuan Hy3 Preview Tops OpenRouter’s Global API Call Leaderboard, Accelerating Real-World Adoption of Chinese AI 🚀

Brief:

• Tencent’s Hunyuan Hy3 Preview model has surpassed international competitors in the latest OpenRouter global large-model API usage rankings, taking first place overall.
• Scenario-based statistics show that Hunyuan Hy3 Preview ranks No. 1 globally in the “Tool Use” category and holds a solid No. 2 position worldwide in “Coding.”
• Chinese AI players are advancing collectively: on the same day, StepFun released its next-generation image generation and editing model, Step Image Edit2, which is now fully available on its open platform.
• Tencent’s AI applications and ecosystem continue to expand: today (April 29), Tencent Docs officially integrated the WorkBuddy knowledge base, enabling native AI workflows; in addition, yesterday Tencent Cloud signed a strategic cooperation agreement with Hongmofang to jointly develop a new generation of humanoid AI dolls by integrating technologies including the Hunyuan large model 🤖✨

Related links:

• Chinese Large Models Dominate Overseas Rankings! Hunyuan Hy3 Preview Tops the Global Large-Model Usage Leaderboard
• WorkBuddy Officially Integrates with Tencent Docs
• Tencent Cloud and Hongmofang Reach Strategic Cooperation to Deepen the AI Toy Ecosystem

影石创新发布2025年业绩营收近百亿，年内将推三大新品

简报：

• 影石创新于4月28日发布2025全年及2026年一季度业绩，多项核心指标创下历史新高。
• 2025年公司营收达97.41亿元，同比增长74.76%；2026年一季度营收为24.81亿元，同比增长83.11%。
• 公司近三年营收规模扩张近3倍，复合增长率超过63%。
• 影石创新预计将在一年内推出云台相机、无线领夹麦克风和无人机三大新品。

相关链接：

• 影石创新2025营收近百亿

Insta360 Reports Nearly RMB 10 Billion in Revenue for 2025, to Launch Three Major New Products Within the Year 🚀

Briefing:

• On April 28, Insta360 released its full-year 2025 and Q1 2026 results, with multiple core indicators reaching record highs.
• The company’s revenue for 2025 reached RMB 9.741 billion, up 74.76% year-on-year; revenue for Q1 2026 was RMB 2.481 billion, up 83.11% year-on-year.
• Over the past three years, the company’s revenue scale has expanded nearly threefold, with a compound growth rate exceeding 63%.
• Insta360 expects to launch three major new products within the year: a gimbal camera, a wireless clip-on microphone, and a drone. 📈✨

Related link:

• Insta360’s 2025 revenue approaches RMB 10 billion

字节TRAE携手Insta360发布Mic Air联名麦克风，加码“语音编程”场景

简报：

• 字节跳动旗下AI编程平台TRAE与影石Insta360联合发布Mic Air TRAE联名无线麦克风套装，套装包含麦克风本体、接收器及TRAE SOLO内测资格，限时促销价319元。
• 硬件方面，Mic Air麦克风轻至7.9g，提供48kHz高保真采样率及内置AI降噪，续航达10小时，支持磁吸背夹等多种佩戴方式。
• 软件方面，TRAE SOLO模式正式上线高级语音输入功能，具备过滤无效口语、自动智能修正语义等能力，并支持通过语音直接触发指令、控制AI命令。
• TRAE SOLO下一阶段还将推出实时问答互动功能，允许用户与AI进行“边说边讨论”的实时流畅交互。

相关链接：

• 字节联手 Insta360 推 Mic Air 无线麦克风

ByteDance’s TRAE Teams Up with Insta360 to Launch the Mic Air Co-Branded Microphone, Expanding Voice Programming Scenarios 🎙️🤖

Brief:

• TRAE, ByteDance’s AI coding platform, has jointly launched the Mic Air TRAE co-branded wireless microphone kit with Insta360. The package includes the microphone, a receiver, and access to the TRAE SOLO closed beta, with a limited-time promotional price of RMB 319.
• On the hardware side, the Mic Air microphone weighs just 7.9g and offers a 48kHz high-fidelity sampling rate with built-in AI noise reduction. It delivers up to 10 hours of battery life and supports multiple wearing options, including a magnetic clip.
• On the software side, TRAE SOLO has officially introduced an advanced voice input feature, capable of filtering out invalid filler speech and automatically correcting semantic intent. It also supports directly triggering commands and controlling AI actions through voice input.
• In the next phase, TRAE SOLO will also roll out a real-time Q&A interaction feature, allowing users to engage in smooth, real-time “talk-while-discussing” interactions with AI. 🚀

Related link:

• ByteDance Partners with Insta360 to Launch the Mic Air Wireless Microphone

科大讯飞推出30B参数星火X2-Flash模型：全量基于华为昇腾算力，支持256K超长上下文

简报：

• 4月29日，科大讯飞正式发布星火X2-Flash模型并同步开放API，该模型采用MoE架构，总参数量达30B，最大支持256K超长上下文。
• 模型完全基于华为昇腾910B集群训练完成，并在国产芯片上率先结合DSA（稀疏注意力）与MTP（多Token预测）技术，使同规模集群下的长文本训练效率提升4.5倍。
• 在核心性能上，星火X2-Flash的智能体（Agent）和代码生成能力显著增强，复杂任务表现可比肩万亿级参数模型，且Token消耗量仅为主流大尺寸模型的三分之一。
• 生态接入方面，该模型已率先接入AstronClaw、Loomy等应用，并深度兼容OpenClaw、Claude Code等国际主流Agent框架。
• 科大讯飞总裁吴晓如在今日的业绩说明会上透露，公司计划于今年10月在华为昇腾950平台上发布中国首个对标业界最先进主流模型的旗舰大模型。

相关链接：

• 讯飞星火X2-Flash模型发布：深耕国产算力，256K长文本能力跃升
• 科大讯飞总裁吴晓如：今年 10 月将在华为昇腾 950 平台上发布国产旗舰大模型

iFLYTEK Launches 30B-Parameter Spark X2-Flash Model: Fully Trained on Huawei Ascend Computing Power, Supports 256K Ultra-Long Context 🚀

Brief:

• On April 29, iFLYTEK officially released the Spark X2-Flash model and simultaneously opened its API. The model adopts an MoE architecture, with a total of 30B parameters, and supports an ultra-long context window of up to 256K.
• The model was trained entirely on Huawei Ascend 910B clusters. It is also the first on domestic chips to combine DSA (Sparse Attention) and MTP (Multi-Token Prediction) technologies, boosting long-text training efficiency by 4.5x under clusters of the same scale.
• In terms of core performance, Spark X2-Flash shows significant improvements in agent capabilities and code generation. Its performance on complex tasks is comparable to trillion-parameter models, while consuming only one-third of the tokens required by mainstream large-scale models. ⚡
• On the ecosystem integration front, the model has already been integrated into applications such as AstronClaw and Loomy, and is deeply compatible with major international agent frameworks including OpenClaw and Claude Code.
• Wu Xiaoru, President of iFLYTEK, revealed at today’s earnings briefing that the company plans to launch China’s first flagship large model benchmarked against the industry’s most advanced mainstream models this October on the Huawei Ascend 950 platform. 🤖

Related Links:

• Spark X2-Flash Model Released: Deepening Domestic Computing Power, 256K Long-Text Capability Takes a Leap
• iFLYTEK President Wu Xiaoru: A Domestic Flagship Large Model Will Be Released This October on the Huawei Ascend 950 Platform

单词学习：![[2026043019274300.mp3]]

新闻播报：![[2026043050818870.mp3]]

DeepSeek多模态视觉能力加速落地，正式开启识图模式内测

简报：

• 在DeepSeek-V4发布仅五天后，官方在移动端与网页版输入栏新增“识图模式”入口，正式开启多模态识图功能的灰度测试
• 实测显示模型在基础视觉理解与画面描述方面表现优异，开启“思考模式”后可展现深度的逻辑推理能力，如推导文物历史背景
• 面对极端视觉挑战（如碎块化、反色图像）、元素计数、复杂图形推理及极新产品信息时，模型的准确率与响应效率仍存在优化空间
• 行业分析指出，此次内测补齐了核心功能短板，验证了多模态链路，预示着DeepSeek原生多模态能力已进入最后筹备期

相关链接：

• DeepSeek 开启识图模式灰度测试，多模态视觉理解能力正式落地

DeepSeek’s Multimodal Visual Capabilities Accelerate Deployment, Officially Launching Closed Beta for Image Recognition Mode 👀🚀

Brief:

• Just five days after the release of DeepSeek-V4, the official team added an “Image Recognition Mode” entry to the input bar on both mobile and web, formally launching a phased beta test of its multimodal image understanding feature.
• Hands-on testing shows that the model performs exceptionally well in basic visual understanding and scene description. When “Thinking Mode” is enabled, it can also demonstrate deeper logical reasoning capabilities, such as inferring the historical background of cultural relics.
• When faced with extreme visual challenges—such as fragmented or color-inverted images—as well as element counting, complex graphic reasoning, and very new product information, there is still room to improve the model’s accuracy and response efficiency.
• Industry analysis suggests that this beta test fills a key functional gap and validates the multimodal pipeline, signaling that DeepSeek’s native multimodal capabilities may have entered the final stage of preparation. 🔍

Related link:

• DeepSeek Launches Phased Beta Test for Image Recognition Mode, Officially Bringing Multimodal Visual Understanding Online

小红书全面升级组织架构：加码AI设立一级部门“Dots”，柯南升任总裁

简报：

• 小红书宣布内部组织架构升级，正式设立AI一级部门“Dots”与企业智能部，旨在全面整合社区、电商与商业化业务，增强AI技术投入。
• “Dots”部门将构建完整的AI全链路体系，业务涵盖AI模型研发、基础设施、工程落地和产品应用等环节。
• 公司进行重大高管调整，柯南（丁玲）升任总裁，全面负责小红书核心业务并直接向CEO星矢汇报，新设立的“Dots”部门亦直接向其汇报。
• 业务版图持续扩张，新设海外部门“Rednote”推动生活方式社区模式走向全球市场，并成立由樱木带队的Lab 1327团队专注新产品孵化。

相关链接：

• 小红书成立 AI 一级部门 “Dots”

Xiaohongshu Fully Upgrades Its Organizational Structure: Boosts AI Investment by Establishing Top-Level Department “Dots,” Conan Promoted to President 🚀

Briefing:

• Xiaohongshu has announced an internal organizational restructuring, officially establishing the top-level AI department “Dots” and a Corporate Intelligence Department, with the goal of fully integrating its community, e-commerce, and commercialization businesses while strengthening investment in AI technology.
• The “Dots” department will build a complete end-to-end AI ecosystem, covering AI model R&D, infrastructure, engineering implementation, and product applications.
• The company has also made major executive adjustments: Conan (Ding Ling) has been promoted to President, taking full responsibility for Xiaohongshu’s core businesses and reporting directly to CEO Seiya. The newly established “Dots” department will also report directly to her.
• The company’s business footprint continues to expand, with the new overseas division “Rednote” set up to bring its lifestyle community model to global markets 🌍, while the Lab 1327 team, led by Sakuragi, will focus on new product incubation 💡.

Related Link:

• Xiaohongshu Establishes Top-Level AI Department “Dots”

微软2026第三财季营收达829亿美元，AI业务成最强增长引擎

简报：

• 微软公布截至2026年3月31日的第三财季报告，公司总营收达到829亿美元，同比增长18%。
• AI业务成为驱动整体增长的最强引擎，年化营收突破370亿美元，同比增速高达123%。
• 智能云部门季度营收达347亿美元（同比增长30%），其中Azure及其他云服务收入增长40%。
• 生产力与业务流程部门营收达350亿美元，Office365与LinkedIn等产品均保持双位数增长。
• 个人计算部门受Windows OEM及Xbox硬件需求放缓影响营收微降1%；此外，公司本季度通过股息和股票回购向股东返还了102亿美元。

相关链接：

• 微软第三财季营收突破 820 亿美元，AI年收入激增123%成核心增长极

Microsoft's Fiscal Q3 2026 Revenue Reaches $82.9 Billion, with AI Becoming the Strongest Growth Engine 🚀

Brief:

• Microsoft reported its financial results for the third fiscal quarter ended March 31, 2026, with total revenue reaching $82.9 billion, up 18% year over year.
• The AI business became the strongest engine driving overall growth, with annualized revenue surpassing $37 billion and year-over-year growth soaring to 123%.
• The Intelligent Cloud segment posted quarterly revenue of $34.7 billion, up 30% year over year, with Azure and other cloud services revenue increasing 40%.
• The Productivity and Business Processes segment generated $35.0 billion in revenue, with products such as Office 365 and LinkedIn both maintaining double-digit growth.
• Revenue from the More Personal Computing segment slipped 1% due to weaker demand for Windows OEM and Xbox hardware; in addition, the company returned dJyH6kDkfC5RlM16.answerText0.2 billion to shareholders this quarter through dividends and share repurchases. 📈

Related link:

• Microsoft's Fiscal Q3 Revenue Surpasses $82 Billion, Annual AI Revenue Jumps 123% to Become Core Growth Driver

台积电先进制程良率超预期，1奈米研发进度大幅领先三星

简报：

• 台积电先进制程进展顺利，2奈米与3奈米良率已高达90%，并计划在2024年底将两项制程产力提升20%，月产出分别可达10万片与18万片。
• 在更尖端的1奈米领域，台积电预计2027年进入量产世代；相比之下，三星电子因2奈米良率不稳，已将1奈米量产时间从2027年推迟至2029年。
• 台积电2026年第一季度业绩亮眼，营收达359亿美元（年增39.2%），EPS为3.49美元均超预期，3奈米贡献了当季晶圆收入的25%。
• 受益于AI热潮，台积电ADR股价近期创下历史新高，市值突破2.1兆美元，位列全球第六大企业，华尔街98%的分析师给予买进评级。
• 市场份额方面，台积电在2023年已占据全球代工版图70%以上，而三星仅占6%至7%，随着台积电产能扩张，原本流向三星的潜在客户可能重回台积电怀抱。

相关链接：

• 三星陷入大危機？台積電良率達90%且擴增產能　1奈米進度狠甩對手！
• 台積電首季業績狂勝市場預期！外資揭未來5年「值得買進長抱」理由
• 外資喊買上看300美元！輝達市值稱霸全球 台積電ADR飆天價列第六

TSMC’s Advanced Process Yields Exceed Expectations, with 1nm R&D Progress Far Ahead of Samsung 🚀

Briefing:

• TSMC’s advanced process development is progressing smoothly. Yields for its 2nm and 3nm nodes have already reached as high as 90%, and the company plans to increase production capacity for both processes by 20% by the end of 2024, with monthly output expected to reach 100,000 wafers and 180,000 wafers, respectively.
• In the more cutting-edge 1nm segment, TSMC is expected to enter the mass-production era in 2027. In contrast, due to unstable 2nm yields, Samsung Electronics has postponed its 1nm mass-production timeline from 2027 to 2029.
• TSMC delivered strong results in the first quarter of 2026, reporting revenue of US$35.9 billion, up 39.2% year over year, while EPS came in at US$3.49, both above market expectations. The 3nm node contributed 25% of the company’s wafer revenue for the quarter.
• Benefiting from the AI boom, TSMC’s ADR recently hit a record high, pushing its market capitalization above US$2.1 trillion and making it the world’s sixth-largest company. Meanwhile, 98% of Wall Street analysts have given the stock a buy rating 📈.
• In terms of market share, TSMC accounted for more than 70% of the global foundry landscape in 2023, while Samsung held only 6% to 7%. As TSMC continues expanding capacity, potential customers that had previously turned to Samsung may return to TSMC.

Related links:

• Is Samsung in Deep Trouble? TSMC’s Yield Reaches 90% While Expanding Capacity as Its 1nm Progress Leaves Rivals Far Behind!
• TSMC’s First-Quarter Results Crush Market Expectations! Foreign Analysts Reveal Why It’s a “Buy and Hold” for the Next 5 Years
• Foreign Analysts Raise Target to US$300! Nvidia Becomes the World’s Most Valuable Company, While TSMC ADR Soars to a Record High and Ranks Sixth 🌟

单词学习：![[2026042726838165.mp3]]

新闻播报：![[2026042735761145.mp3]]

OpenAI 开源本地隐私脱敏模型 Privacy Filter，支持 12.8 万 Token 与八类敏感信息识别

简报：

• OpenAI 推出前沿个人身份信息（PII）脱敏模型 Privacy Filter，已通过 Apache2.0 协议在 Hugging Face 和 GitHub 同步开源。
• 该模型具备深度语言理解能力，告别传统机械匹配，能根据上下文语境精准遮盖非结构化文本中的敏感信息并保留公开有用信息。
• 采用轻量化混合专家（MoE）架构，总参数15亿，每次推理仅激活约5000万参数，支持在笔记本和浏览器等边缘设备流畅运行。
• 具备 12.8 万 Token 的超长上下文窗口，在基准测试中取得 97.43% 的 F1 分数（召回率98.08%），确保长文本处理的连贯与准确。
• 支持识别八类核心敏感信息，涵盖基础身份（姓名/地址/电话/邮箱）、网络资产（URL）、金融账号、机密凭证（密码/API密钥）及日期。
• 该工具定位为云端大模型的本地“预过滤层”，实现数据不离设备的检测脱敏，防止隐私泄露；官方同时提醒医疗、金融等高敏感领域仍需人工审核。

相关链接：

• OpenAI 推出 Privacy Filter，支持12.8万超长上下文与8类隐私识别

OpenAI Open-Sources Local Privacy Redaction Model Privacy Filter, Supporting 128K Tokens and Eight Types of Sensitive Information 🔐

Brief:

• OpenAI has introduced Privacy Filter, a state-of-the-art personally identifiable information (PII) redaction model, and has simultaneously open-sourced it on Hugging Face and GitHub under the Apache 2.0 license.
• The model features deep language understanding capabilities, moving beyond traditional rule-based matching to accurately mask sensitive information in unstructured text based on contextual meaning while preserving publicly useful information.
• Built on a lightweight Mixture of Experts (MoE) architecture, it has 1.5 billion total parameters, with only about 50 million activated per inference, enabling smooth operation on edge devices such as laptops and browsers. 💻
• It supports an ultra-long context window of 128K tokens and achieved an F1 score of 97.43% in benchmark tests, with a recall rate of 98.08%, ensuring consistency and accuracy in long-text processing.
• It can identify eight core categories of sensitive information, including basic identity data (name/address/phone number/email), web assets (URLs), financial account information, confidential credentials (passwords/API keys), and dates.
• The tool is positioned as a local “pre-filtering layer” for cloud-based large models, enabling on-device detection and redaction so data never leaves the device, helping prevent privacy leakage. OpenAI also notes that highly sensitive sectors such as healthcare and finance still require human review. 🛡️

Related link:

• OpenAI Launches Privacy Filter with 128K Long Context and 8 Privacy Categories

DeepSeek全系API服务输入缓存命中价格大幅下调至一折

简报：

• DeepSeek官方于4月26日宣布，即日起其全系API服务的输入缓存命中价格降至原有价格的1/10。
• 针对Pro模型，用户在5月5日前还可额外叠加2.5折的限时优惠。
• 调整后，DeepSeek-V4-Flash模型的输入（缓存命中）价格降至0.02元，DeepSeek-V4-Pro模型的输入（缓存命中）价格降至0.025元起。

相关链接：

• DeepSeek API 输入缓存降价，降至原有价格 1/10

DeepSeek Slashes Input Cache Hit Pricing Across Its Entire API Lineup to One-Tenth 🚀

Brief:

• On April 26, DeepSeek officially announced that, effective immediately, the input cache hit pricing for its entire API product line has been reduced to one-tenth of the original price.
• For Pro models, users can also stack an additional limited-time 75% discount before May 5. 🎉
• Following the adjustment, the input (cache hit) price for the DeepSeek-V4-Flash model has been lowered to RMB 0.02, while the input (cache hit) price for the DeepSeek-V4-Pro model starts from RMB 0.025.

Related link:

• DeepSeek API input cache pricing cut to 1/10 of the original price

芯擎科技发布5nm车规级芯片“龙鹰二号”，支持大模型与舱驾融合

简报：

• 芯擎科技在2026年北京国际车展上正式发布自主研发的5纳米车规级AI座舱芯片“龙鹰二号”。
• 该芯片AI算力高达200TOPS，具备强大的性能冗余，能够原生支持7B以上参数的多模态大模型。
• 芯片采用极具前瞻性的“柔性架构”，能够覆盖从入门级到旗舰级车型的中央计算平台演进需求。
• “龙鹰二号”实现了“舱驾融合”的全场景覆盖，内部集成专用的车控处理单元与安全岛，可通过物理隔离安全处理座舱与驾驶辅助业务。
• 芯片计划于2027年第一季度正式启动车型适配工作，以加速智能汽车向中央计算时代的演进。

相关链接：

• 国产芯片再突破！5nm“龙鹰二号”正式发布，AI算力直达200TOPS

SiEngine Unveils 5nm Automotive-Grade Chip “Longying No. 2,” Supporting Large Models and Cockpit-Driving Integration 🚗🤖

Brief:

• At the 2026 Beijing International Auto Show, SiEngine officially unveiled its self-developed 5nm automotive-grade AI cockpit chip, “Longying No. 2.”
• The chip delivers up to 200 TOPS of AI computing power, offering substantial performance headroom and native support for multimodal large models with more than 7 billion parameters.
• Built on a highly forward-looking “flexible architecture,” the chip can meet the evolving central computing platform needs of vehicles ranging from entry-level to flagship models.
• “Longying No. 2” enables full-scenario coverage for “cockpit-driving integration,” with a dedicated vehicle control processing unit and safety island integrated on-chip, allowing cockpit and driving assistance functions to be handled securely through physical isolation.
• The chip is scheduled to officially begin vehicle integration and adaptation in the first quarter of 2027, accelerating the evolution of intelligent vehicles into the era of central computing. ⚙️

Related link:

• Another Breakthrough for Domestic Chips! 5nm “Longying No. 2” Officially Unveiled, Delivering 200 TOPS of AI Computing Power

腾讯开源具身多模态大模型 HY-Embodied-0.5-X，强化机器人智能交互能力

简报：

• 腾讯 Robotics X 实验室与混元团队共同发布并开源了专为具身任务优化的多模态大模型 HY-Embodied-0.5-X。
• 模型基于 HY-Embodied-0.5-MoT-2B 架构，核心强调机器人“看懂、想清、做到”的能力，在精细操作、空间推理、动作预测及风险判断上表现优越。
• 该系列包含两个主要版本：支持端侧实时响应的 MoT-2B 和支持更复杂任务处理的大参数规模 MoE-32B，主要面向家庭服务和桌面操作等实际应用场景。
• 数据与训练方面，结合了自采机器人第一视角操作数据与开源具身数据，引入思维链标注，并采用分阶段迭代的策略提升训练效率和稳定性。
• 该模型的推出进一步提升了机器人在空间理解、长程规划和具身交互方面的优势，推动机器人应用从“看懂”走向“干活”。

相关链接：

• 腾讯推出具身多模态大模型 HY-Embodied-0.5-X

Tencent Open-Sources the Embodied Multimodal Large Model HY-Embodied-0.5-X, Strengthening Robots’ Intelligent Interaction Capabilities 🤖✨

Brief:

• Tencent Robotics X Lab and the Hunyuan team have jointly released and open-sourced HY-Embodied-0.5-X, a multimodal large model specifically optimized for embodied tasks.
• Built on the HY-Embodied-0.5-MoT-2B architecture, the model emphasizes a robot’s ability to “see, think, and act,” delivering strong performance in fine manipulation, spatial reasoning, action prediction, and risk assessment.
• The series includes two main versions: the MoT-2B, which supports real-time on-device responses, and the larger-parameter MoE-32B, which is designed for handling more complex tasks. These models are primarily aimed at practical application scenarios such as household services and desktop operations.
• In terms of data and training, the model combines self-collected first-person robotic operation data with open-source embodied datasets, introduces chain-of-thought annotations, and adopts a phased iterative strategy to improve training efficiency and stability.
• The launch of this model further enhances robots’ strengths in spatial understanding, long-horizon planning, and embodied interaction, helping move robotic applications from simply “understanding” to actually “getting work done” 🚀

Related links:

• Tencent launches the embodied multimodal large model HY-Embodied-0.5-X

英伟达高管吴新宙展望自动驾驶：L4级量产即将成为现实

简报：

• 英伟达吴新宙表示，自动驾驶L4级的量产已经不再是梦想。

相关链接：

• 英伟达吴新宙：自动驾驶L4 级量产不再是梦

Nvidia Executive Wu Xinzhou Looks Ahead to Autonomous Driving: Mass Production of Level 4 Is About to Become Reality 🚗✨

Brief:

• Nvidia’s Wu Xinzhou said that mass production of Level 4 autonomous driving is no longer just a dream.

Related Links:

• Nvidia’s Wu Xinzhou: Mass Production of Level 4 Autonomous Driving Is No Longer a Dream

谷歌拟最高掷400亿美元加码Anthropic，提供百万颗TPU并助推其估值达3500亿

简报：

• 谷歌计划向AI初创公司Anthropic注入100亿美元资金，根据双方长期规划，未来投资总额或将攀升至400亿美元。
• 此次巨额交易预计将把Anthropic的估值推高至3500亿美元。此前在今年2月，该公司刚完成300亿美元融资，而亚马逊也曾投入50亿美元并保留200亿美元的追加投资计划。
• 双方深度捆绑底层硬件资源，谷歌将在未来五年内为Anthropic提供5吉瓦的算力支持以及100万颗自主研发的TPU芯片。
• 尽管双方达成深度合作，但谷歌高层对Anthropic在AI编程市场的迅速崛起仍抱有担忧，未来在特定细分领域可能演变为直接竞争。
• Anthropic目前正积极筹备资本市场的下一步动作，有消息称其正考虑在今年10月正式启动IPO进程。

相关链接：

• 谷歌拟加码投资 Anthropic，总金额或达400亿美元

Google Plans to Invest Up to $40 Billion More in Anthropic, Provide 1 Million TPUs, and Help Boost Its Valuation to $350 Billion 🚀

Brief:

• Google plans to inject dJyH6kDkfC5RlM16.answerText0 billion into AI startup Anthropic, and according to the long-term roadmap between the two parties, the total investment could eventually rise to $40 billion.
• This massive deal is expected to push Anthropic’s valuation up to $350 billion. Earlier this February, the company had just completed a $30 billion funding round, while Amazon had also invested $5 billion and retained plans for an additional $20 billion investment.
• The two companies are deeply aligned on underlying hardware resources, with Google set to provide Anthropic with 5 gigawatts of computing power support and 1 million self-developed TPU chips over the next five years.
• Despite the deep partnership, Google executives remain concerned about Anthropic’s rapid rise in the AI coding market, and the two could eventually become direct competitors in certain niche segments.
• Anthropic is currently actively preparing its next move in the capital markets, with reports suggesting that it is considering officially launching its IPO process in October this year 📈.

Related link:

• Google Plans to Increase Investment in Anthropic, with Total Amount Potentially Reaching $40 Billion

单词学习：![[2026042565336384.mp3]]

新闻播报：![[2026042579390184.mp3]]

DeepSeek-V4预览版携百万上下文开源发布，国内多云平台同步完成适配

简报：

• 深度求索（DeepSeek）正式开源发布DeepSeek-V4预览版，开创性采用DSA稀疏注意力机制，实现了1M（一百万字）超长上下文能力的标准化。
• 该系列分为两个规格：1.6T参数的Pro版（追求极致性能，在数学及Agentic Coding测评中达世界顶级水平）和284B参数的Flash版（主打极致性价比）。
• 模型深度适配Agent生态，支持思考与非思考模式，用户可通过开放的reasoning_effort参数调节思考强度；旧有的deepseek-chat与deepseek-reasoner模型名将于2026年7月24日停用。
• 百度智能云千帆平台实现Day0适配，已上线并提供API服务，用户目前可直接调用Pro版，Flash版即将全量开放。
• 腾讯云TokenHub同步推出预览版API服务，定价与官方保持一致，并在国际站新加坡节点、ADP（智能体开发平台）及TI-ONE等高性能计算集群中完成了全链路整合。

相关链接：

• DeepSeek-V4预览版正式发布:1M 超长上下文进入全员普惠时代
• 百度智能云上线DeepSeek-V4
• 腾讯云 TokenHub 发布 DeepSeek-V4 预览版，百万上下文支持全面上线
• DeepSeek-V4 Hugging Face 开源模型集合
• DeepSeek-V4 魔搭社区（ModelScope）开源模型集合

DeepSeek-V4 Preview Open-Sourced with 1M-Token Context, Simultaneously Adapted Across Multiple Domestic Cloud Platforms 🚀

Brief:

• DeepSeek has officially open-sourced the DeepSeek-V4 Preview, pioneering the use of the DSA sparse attention mechanism to standardize support for a 1M (one million-token) ultra-long context window.
• The series comes in two variants: the Pro version with 1.6T parameters (built for maximum performance, reaching world-leading levels in mathematics and Agentic Coding benchmarks) and the Flash version with 284B parameters (focused on delivering extreme cost efficiency).
• The model is deeply adapted to the Agent ecosystem, supports both thinking and non-thinking modes, and allows users to adjust reasoning intensity through the open reasoning_effort parameter; the legacy model names deepseek-chat and deepseek-reasoner will be retired on July 24, 2026.
• Baidu AI Cloud’s Qianfan platform achieved Day-0 adaptation, has already launched the model with API services, and currently allows users to directly access the Pro version, while the Flash version will be fully opened soon.
• Tencent Cloud TokenHub has also launched Preview API services simultaneously, with pricing aligned with the official offering, and has completed full-chain integration across its international Singapore node, ADP (Agent Development Platform), and TI-ONE high-performance computing clusters. ☁️✨

Related links:

• DeepSeek-V4 Preview Officially Released: 1M Ultra-Long Context Enters the Era of Universal Accessibility
• Baidu AI Cloud Launches DeepSeek-V4
• Tencent Cloud TokenHub Releases DeepSeek-V4 Preview, Full Support for Million-Token Context Now Available
• DeepSeek-V4 Open-Source Model Collection on Hugging Face
• DeepSeek-V4 Open-Source Model Collection on ModelScope

Meta携手亚马逊AWS达成3至5年长约，部署千万级Graviton核心强化AI底座

简报：

• Meta与亚马逊AWS正式达成一项为期三到五年的合作协议，旨在通过大规模部署处理器构建更具能效比的下一代AI基础设施。
• 根据协议，Meta将首批部署数千万个AWS Graviton处理器核心，并根据AI能力增长灵活扩展，这将使Meta跻身AWS前五大Graviton客户之列。
• 此次合作主要应对代理式人工智能（Agentic AI）的兴起，满足实时推理、代码生成和多步骤任务等对CPU日益增长的密集型计算需求。
• 部署将涉及采用3纳米制程工艺、具备192个核心的Graviton5芯片，该芯片缓存提升5倍且通信延迟更低；同时亚马逊将提供AWS Nitro解决方案，支持Meta独立运行其虚拟机。

相关链接：

• Meta 与亚马逊 AWS 达成合作，部署数千万 Graviton 核心构建 AI 基础设施

Meta Signs a 3- to 5-Year Agreement with Amazon AWS to Deploy Tens of Millions of Graviton Cores, Strengthening Its AI Foundation 🚀

Brief:

• Meta and Amazon AWS have officially entered into a three- to five-year partnership aimed at building a more energy-efficient next-generation AI infrastructure through large-scale processor deployment.
• Under the agreement, Meta will initially deploy tens of millions of AWS Graviton processor cores and scale flexibly in line with growth in its AI capabilities, placing Meta among AWS’s top five Graviton customers.
• The partnership is primarily intended to address the rise of agentic AI, meeting the growing CPU-intensive computing demands of real-time inference, code generation, and multi-step tasks.
• The deployment will involve the Graviton5 chip, built on a 3nm process and featuring 192 cores, with 5x greater cache and lower communication latency; Amazon will also provide AWS Nitro solutions to support Meta in running its virtual machines independently. 🤖

Related Link:

• Meta partners with Amazon AWS to deploy tens of millions of Graviton cores for AI infrastructure

谷歌拟掷最高400亿美元投资Anthropic，深化AI算力合作与竞合关系

简报：

• 谷歌决定向 AI 初创公司 Anthropic 首期投入 100 亿美元现金，对应估值为 3500 亿美元。
• 若 Anthropic 达到业绩目标，谷歌未来还将追加 300 亿美元投资，并帮助其大幅扩充算力。
• 双方算力合作进一步深化，未来 5 年 Google Cloud 将向 Anthropic 提供至少 5 吉瓦规模的算力支持，同时谷歌的 TPU 芯片也是其关键资源。
• 得益于 Claude Code 爆红，Anthropic 近期融资步伐加快，最快可能在今年 10 月进行 IPO。
• 双方在 AI 赛道呈现“既合作又竞争”的微妙状态，谷歌虽为其提供云服务和芯片，但也对其在 AI 编程市场的迅速扩张感到担忧。
• 受此投资消息提振，谷歌股价当天上涨超过 1%。

相关链接：

• 既合作又竞争，谷歌拟向 Anthropic 投资至高 400 亿美元

Google Plans to Invest Up to $40 Billion in Anthropic, Deepening AI Computing Collaboration Amid Competitive Tensions 🤝⚙️

Brief:

• Google has decided to make an initial cash investment of dJyH6kDkfC5RlM16.answerText0 billion in AI startup Anthropic, implying a valuation of $350 billion.
• If Anthropic meets its performance targets, Google may invest an additional $30 billion in the future and help it significantly expand its computing capacity.
• The two sides are further deepening their computing partnership. Over the next five years, Google Cloud will provide Anthropic with at least 5 gigawatts of computing support, while Google’s TPU chips will also remain a key resource for the company.
• Driven by the surging popularity of Claude Code, Anthropic has recently accelerated its fundraising efforts and could move forward with an IPO as early as October this year 🚀.
• The two companies are in a delicate “cooperative yet competitive” relationship in the AI race. While Google provides Anthropic with cloud services and chips, it is also concerned about Anthropic’s rapid expansion in the AI programming market.
• Boosted by news of the investment, Google’s stock rose more than 1% on the day.

Related link:

• Both Partners and Rivals: Google Plans to Invest Up to $40 Billion in Anthropic

华为AI眼镜今日2499元起发售：主打一键唤醒与视觉支付

简报：

• 华为AI眼镜于今日（4月25日）10:08正式开售，提供钛银灰、流光银、摩登黑三款配色及两种镜框样式，售价2499元起。
• 产品采用轻量化新型材料，镜腿薄至6.25毫米，镜架轻至35.5克，接近普通眼镜重量。
• 内置华为自研AI芯片，号称秒级响应，支持语音和AI快捷键一键唤醒，具备“小艺看世界”及“支付宝看一下支付”功能。
• 拍摄方面配备1/2.8英寸大底传感器，支持AI RAW域多帧融合算法、AI构图智能矫正及第一人称视角视频直播。
• 综合续航达12小时，支持连续8小时通话或9小时音乐播放，配套华为眼镜App提供智慧播报及手表协同等功能。

相关链接：

• 华为 AI 眼镜今日开售：支持 AI 快捷键一键唤醒、支付宝看一下支付等，2499 元起

Huawei AI Glasses Go on Sale Today Starting at 2,499 Yuan: Highlighting One-Tap Wake-Up and Visual Payment 👓✨

Briefing:

• Huawei AI Glasses officially went on sale today (April 25) at 10:08, offering three color options — Titanium Silver Gray, Flowing Light Silver, and Modern Black — as well as two frame styles, with prices starting at 2,499 yuan.
• The product uses new lightweight materials, with temples as thin as 6.25 mm and a frame weighing as little as 35.5 grams, making it close to the weight of ordinary glasses.
• It features Huawei’s self-developed AI chip, claimed to deliver response times in seconds, and supports one-tap wake-up via voice and an AI shortcut key. It also includes the “Xiaoyi Sees the World” and “Alipay Scan-to-Pay” functions.
• For imaging, it is equipped with a 1/2.8-inch large sensor and supports an AI RAW-domain multi-frame fusion algorithm, AI composition smart correction, and first-person video livestreaming.
• Total battery life reaches 12 hours, with support for up to 8 hours of continuous calling or 9 hours of music playback. The companion Huawei Glasses App also provides smart broadcasts and smartwatch collaboration features. 🔋📷

Related link:

• Huawei AI Glasses go on sale today: support one-tap wake-up via AI shortcut key, Alipay Scan-to-Pay, and more, starting at 2,499 yuan

英伟达全面引入基于 GPT-5.5 的 OpenAI Codex 提升全员效率

简报：

• 近日，英伟达 CEO 黄仁勋宣布，基于最新 GPT-5.5 技术的 OpenAI Codex 编程工具已向英伟达开放，并致信 OpenAI CEO 萨姆・奥特曼分享号召员工使用的计划。
• 黄仁勋强调该工具不仅适用于软件团队，公司 10000 名涵盖工程、法务、市场、财务和人力资源等所有部门的员工均已提前体验并取得显著成果。
• Codex 被英伟达定位为“智能队友”，旨在帮助全体员工提升工作效率，实现更快、更智能的工作方式。
• 为帮助员工更好地掌握新技术，英伟达与 AI 团队共建了 Codex 实验室，并将在未来几周内举办一系列线上入门研讨会。

相关链接：

• 英伟达 CEO 黄仁勋推动全员使用 OpenAI Codex 编程工具

NVIDIA Fully Rolls Out OpenAI Codex Powered by GPT-5.5 to Boost Company-Wide Efficiency 🚀

Briefing:

• Recently, NVIDIA CEO Jensen Huang announced that OpenAI Codex, the programming tool built on the latest GPT-5.5 technology, has been made available across NVIDIA, and he also wrote to OpenAI CEO Sam Altman to share his plan to encourage employees to use it.
• Huang emphasized that the tool is not only for software teams. All 10,000 employees across engineering, legal, marketing, finance, and human resources have already had early access and achieved significant results.
• NVIDIA positions Codex as an “intelligent teammate” 🤖, designed to help all employees improve productivity and work faster and smarter.
• To help employees better master the new technology, NVIDIA has jointly established a Codex Lab with its AI team and will host a series of online introductory workshops in the coming weeks.

Related link:

• NVIDIA CEO Jensen Huang Promotes Company-Wide Adoption of OpenAI Codex Programming Tool

单词学习：![[2026042391964910.mp3]]

新闻播报：![[2026042309504618.mp3]]

谷歌推出全新第八代TPU基础设施

简报：

• 谷歌正式发布了其第八代TPU（张量处理单元）基础设施。

相关链接：

• 谷歌发布第八代TPU基础设施

Google Launches New Eighth-Generation TPU Infrastructure 🚀

Brief:

• Google has officially unveiled its eighth-generation TPU (Tensor Processing Unit) infrastructure.

Related Link:

• Google Launches Eighth-Generation TPU Infrastructure

(Note: As the provided source material was empty, this brief is based solely on an objective extraction of the topic information provided.)

OpenAI推出ChatGPT工作空间智能体，支持全天候自动化处理复杂协作任务

简报：

• OpenAI 今日（4月23日）在 ChatGPT 中推出工作空间智能体（Workspace Agents），支持团队创建协作智能体，可7*24小时在云端自动化处理长周期复杂任务。
• 该功能由 Codex 提供底层支持，具备文件处理、代码运行、工具调用和记忆存储能力，能够执行多步骤复杂操作，并支持定时调度与 Slack 集成。
• 智能体能将团队隐性知识转化为显性工作流，OpenAI 内部已成功部署软件审查员、产品反馈路由器等 5 种实例，大幅节省了销售、会计等团队的人工操作时间。
• 该功能目前面向 Business、Enterprise、Edu 和 Teachers 计划用户开放研究预览，2026年5月6日前免费使用，随后将转为积分计费模式。
• 同期，OpenAI 宣布当前已锁定超 8GW 的算力资源，预计到 2030 年算力规模将达到 30GW，以支撑其庞大的 AI 基础设施需求。

相关链接：

• OpenAI 发布 ChatGPT 团队工作流 AI 智能体：7*24 小时在线
• OpenAI 当前已锁定 8GW 算力，到 2030 年将达 30GW

OpenAI Launches ChatGPT Workspace Agents, Supporting 24/7 Automated Handling of Complex Collaborative Tasks 🤖☁️

Briefing:

• OpenAI today (April 23) launched Workspace Agents in ChatGPT, enabling teams to create collaborative agents that can automatically handle long-running, complex tasks in the cloud 24/7.
• Powered by Codex at the core, the feature supports file processing, code execution, tool invocation, and memory storage, allowing it to carry out complex multi-step operations while also supporting scheduled tasks and Slack integration.
• These agents can turn teams’ tacit knowledge into explicit workflows. OpenAI has already successfully deployed five internal use cases, including a software reviewer and a product feedback router, significantly reducing manual workload for teams such as sales and accounting.
• The feature is currently available as a research preview for users on the Business, Enterprise, Edu, and Teachers plans, and will be free to use until May 6, 2026, after which it will shift to a credit-based billing model.
• At the same time, OpenAI announced that it has already secured more than 8GW of computing capacity, and expects that figure to reach 30GW by 2030 to support its massive AI infrastructure needs. 📈

Related links:

• OpenAI launches ChatGPT team workflow AI agents: online 24/7
• OpenAI has currently secured 8GW of computing power, expected to reach 30GW by 2030

千问大模型全面赋能出行：高德首发汽车出行AI Agent，千问App上线东航AI直连服务

简报：

• 4月23日，高德地图正式发布汽车出行 AI Agent，基于千问大模型底座构建“语言大脑”与“空间大脑”双引擎架构，推动车载导航向“主动意图理解”跃迁。
• 高德该系统解决了“人适应系统”的痛点，支持复杂行程的一句话串联、动态空间路线推理及多轮对话式编辑。
• 高德汽车出行 AI Agent 具备主动感知时间、空间与情境的服务能力，可实时监测电量自动插入充电节点，或提前感知拥堵并计算绕行方案。
• 同日，阿里旗下千问 App 宣布对外开放“AI 办事”能力，首家接入中国东方航空，上线全流程 AI 机票直连服务。
• 用户可通过自然语言向千问下达指令，完成航班查询、购票，并支持像聊天一样实现“选最便宜直飞”、“连座靠窗”等个性化选座与值机需求。
• 未来千问将与东航共同提供主动行程管理，并陆续上线结合打车、酒店等能力的主动行程规划与出行提醒功能。

相关链接：

• 高德发布汽车出行 AI Agent:基于千问大模型实现主动意图理解
• 阿里千问“AI 办事”能力对外开放：首家接入东方航空，能查航班、买机票

Qwen Large Model Fully Empowers Mobility: Amap Launches Its First Automotive Travel AI Agent, and the Qwen App Rolls Out China Eastern Airlines AI Direct Service ✈️🚗

Briefing:

• On April 23, Amap officially launched its Automotive Travel AI Agent. Built on the Qwen large model foundation, it adopts a dual-engine architecture of a “language brain” and a “spatial brain,” driving in-car navigation toward a leap into “proactive intent understanding.”
• Amap’s system addresses the pain point of “people adapting to systems,” supporting one-sentence chaining for complex itineraries, dynamic spatial route reasoning, and multi-turn conversational editing.
• Amap’s Automotive Travel AI Agent is capable of proactively perceiving time, space, and context in its services. It can monitor battery levels in real time and automatically insert charging stops, or detect congestion in advance and calculate detour plans.
• On the same day, Alibaba’s Qwen App announced the external rollout of its “AI Task Handling” capability, with China Eastern Airlines becoming the first partner to integrate it and launch a full-process AI direct ticketing service.
• Through natural language instructions, users can ask Qwen to complete flight searches and ticket purchases, while also supporting personalized seat selection and check-in requests in a chat-like manner, such as “find the cheapest direct flight” or “adjacent window seats.”
• In the future, Qwen and China Eastern Airlines will jointly provide proactive itinerary management, and will gradually launch proactive travel planning and trip reminder features integrated with ride-hailing, hotels, and other services. 🤖

Related Links:

• Amap Launches Automotive Travel AI Agent: Proactive Intent Understanding Powered by the Qwen Large Model
• Alibaba Qwen Opens Its “AI Task Handling” Capability to the Public: China Eastern Airlines Is the First Partner, Enabling Flight Search and Ticket Booking

特斯拉HW3硬件无缘无监督自动驾驶，马斯克承诺提供换新补偿

简报：

• 特斯拉首席执行官埃隆·马斯克在2026年第一季度财报电话会议上证实，搭载硬件3（HW3）芯片的车辆由于性能限制，将无法实现无监督的完全自动驾驶（FSD）。
• 马斯克表示，特斯拉将为搭载HW3芯片并希望升级的车辆提供“以旧换新折扣”，以引导用户转向支持更高算力的AI4硬件平台。
• 此消息标志着特斯拉正式承认早期自动驾驶硬件在通往完全无人驾驶路径上的局限性。

相关链接：

• 马斯克确认特斯拉HW3芯片无法支持无监督FSD

Tesla HW3 Hardware Will Not Support Unsupervised Self-Driving, Musk Promises Upgrade Compensation 🚗⚙️

Brief:

• Tesla CEO Elon Musk confirmed during the company’s first-quarter 2026 earnings call that vehicles equipped with the Hardware 3 (HW3) chip will not be able to achieve unsupervised Full Self-Driving (FSD) due to performance limitations.
• Musk said Tesla will offer a “trade-in discount” for vehicles equipped with the HW3 chip to customers who wish to upgrade, encouraging them to move to the AI4 hardware platform, which supports greater computing power.
• This announcement marks Tesla’s formal acknowledgment of the limitations of its earlier autonomous driving hardware on the path toward fully driverless capability. 🤖

Related Link:

• Musk Confirms Tesla HW3 Chip Cannot Support Unsupervised FSD

千寻位置发布“具身时空大脑”，为各类具身智能设备提供统一时空底座

简报：

• 4月22日，千寻位置正式发布“具身时空大脑（SpatiXBot）”产品集。
• 该产品集基于千寻位置时空智能全链路技术底座，融合通用大模型与自研专业模型，具备覆盖室内外的自主行走、环境感知与群体协同能力。
• 产品主要面向机器人本体厂商及具身智能行业应用客户，旨在为各类设备提供统一的时空底座。
• 适用范围广泛，覆盖人形机器人、机器狗、无人机、无人车、割草机等多类具身形态。

相关链接：

• 千寻位置正式发布“具身时空大脑”产品集

Qianxun Spatial Intelligence Launches “SpatiXBot,” Providing a Unified Spatiotemporal Foundation for Various Embodied Intelligent Devices 🤖🌍

Brief:

• On April 22, Qianxun Spatial Intelligence officially launched its “Embodied Spatiotemporal Brain (SpatiXBot)” product suite.
• Built on Qianxun’s full-stack spatiotemporal intelligence technology foundation, the suite integrates general large models with proprietary specialized models, enabling autonomous indoor and outdoor mobility, environmental perception, and multi-agent collaboration.
• The product suite is primarily designed for robot manufacturers and embodied intelligence industry application clients, aiming to provide a unified spatiotemporal foundation for various types of devices.
• It supports a wide range of applications, covering multiple embodied forms such as humanoid robots, robot dogs, drones, unmanned vehicles, and lawn mowers. 🚁

Related Link:

• Qianxun Spatial Intelligence Officially Launches the “Embodied Spatiotemporal Brain” Product Suite

单词学习：![[2026042272532601.mp3]]

新闻播报：![[2026042210589514.mp3]]

Anthropic 开启 Claude Pro 权益测试，部分新用户或无法使用 Claude Code 功能

简报：

• Anthropic 正在对每月 20 美元的 Claude Pro 订阅方案进行 A/B 测试，约 2% 的新用户权益页面显示不再包含 Claude Code 功能。
• 若该调整最终落地，开发者或需订阅每月 100 美元起的 Max 版本才能获得该功能的使用权限，而目前老用户及 98% 的新用户暂不受影响。
• 业界分析此举旨在通过功能差异化，引导高频使用编程功能的开发者向更高客单价的套餐迁移。
• 此外，Anthropic 近期因自动化监控系统“误判”，导致 Belo 公司访问权限被无预警切断 15 小时，暴露出其申诉渠道单一及沟通效率低下的问题。
• 受影响的企业及个人用户反映，Anthropic 的申诉仅依靠 Google 表单，这种“先封禁、零沟通”的处理方式引发了舆论的广泛批评。

相关链接：

• Claude Pro 订阅方案调整：部分新用户或失去 Claude Code 使用权
• Belo 公司遭突袭式 Claude AI“断网”：停摆 15 小时，申诉仅靠谷歌表单

Anthropic Begins Testing Changes to Claude Pro Benefits, Some New Users May Lose Access to Claude Code 🚨

Brief:

• Anthropic is conducting an A/B test on its $20-per-month Claude Pro subscription plan, and the benefits page shown to about 2% of new users no longer includes the Claude Code feature.
• If this change is ultimately rolled out, developers may need to subscribe to the Max plan, starting at dJyH6kDkfC5RlM16.answerText00 per month, to gain access to this feature. At present, existing users and 98% of new users are not affected.
• Industry analysts believe the move is intended to differentiate features across subscription tiers and encourage developers who frequently use coding capabilities to migrate to higher-priced plans.
• In addition, Anthropic recently cut off access to Belo for 15 hours without warning due to a “false positive” from its automated monitoring system, exposing issues with its limited appeal channels and inefficient communication.
• Affected businesses and individual users reported that Anthropic relies solely on a Google Form for appeals. This “ban first, no communication” approach has drawn widespread criticism from the public ⚠️

Related links:

• Claude Pro Subscription Adjustment: Some New Users May Lose Access to Claude Code
• Belo Hit by Sudden Claude AI “Outage”: Operations Halted for 15 Hours, Appeals Limited to a Google Form

OpenAI多款前沿模型意外泄露，GPT-5.5曝光暗示智能体时代加速到来

简报：

• 开发者在Codex CLI终端系统模型列表中意外发现“GPT-5.5”及“oai-2.1”的身影，其中GPT-5.5被明确标注为“最前沿的智能体编程模型”，暗示OpenAI研发重心已全面转向AI智能体。
• 此次泄露还曝光了多个神秘实验性模型，包括预示介入生物医药领域的Heisenberg（海森堡）、具有特殊内部训练代号的Arcanine（风速狗）以及疑似极高算力底层基座的Glacier-alpha系列。
• 研究指出，具备强大推理能力的GPT-5系列模型已首次通过“哥德尔测试”，尝试破解三大数学猜想，并在处理“心理压力”等话题时展现出更强的同理心。
• 为加速企业级市场渗透，OpenAI拟注资至多15亿美元，与多家顶级私募股权公司成立估值达100亿美元的合资企业“DeployCo”，首轮融资预计将于2026年5月初完成。
• OpenAI将持有DeployCo超级投票权以确保产品绝对控制，并向私募投资方承诺五年期17.5%的年化回报率，以此应对Anthropic在B端市场的强劲挑战。

相关链接：

• GPT-5.5 意外现身，OpenAI 智能体时代提前降临？
• OpenAI 拟注资15亿美元成立合资公司 DeployCo，联手私募巨头加速企业级市场渗透

Multiple Cutting-Edge OpenAI Models Accidentally Leaked, GPT-5.5 Reveal Hints at an Accelerating Agent Era 🚀

Briefing:

• Developers unexpectedly spotted “GPT-5.5” and “oai-2.1” in the model list of the Codex CLI terminal system. GPT-5.5 was explicitly labeled as “the most advanced agentic coding model,” suggesting that OpenAI’s R&D focus has fully shifted toward AI agents.
• The leak also exposed several mysterious experimental models, including Heisenberg, which hints at OpenAI’s possible entry into the biopharmaceutical field, Arcanine, which carries a special internal training codename, and the Glacier-alpha series, suspected to be an ultra-high-compute foundational base model lineup.
• Research indicates that the GPT-5 series, equipped with powerful reasoning capabilities, has for the first time passed the “Gödel Test,” attempted to tackle three major mathematical conjectures, and demonstrated stronger empathy when handling topics such as psychological stress 🧠.
• To accelerate penetration into the enterprise market, OpenAI reportedly plans to invest up to dJyH6kDkfC5RlM16.answerText.5 billion to establish a joint venture called “DeployCo” with several top private equity firms, at a valuation of dJyH6kDkfC5RlM16.answerText0 billion. The first round of financing is expected to close in early May 2026.
• OpenAI will hold super-voting rights in DeployCo to ensure absolute control over its products, while promising private equity investors an annualized return of 17.5% over five years, in response to Anthropic’s strong challenge in the B2B market 📈.

Related links:

• GPT-5.5 Unexpectedly Appears — Is OpenAI’s Agent Era Arriving Ahead of Schedule?
• OpenAI Plans to Invest dJyH6kDkfC5RlM16.answerText.5 Billion to Form Joint Venture DeployCo, Partnering with Private Equity Giants to Accelerate Enterprise Market Expansion

腾讯云开源业内首个亚百毫秒级AI Agent安全沙箱CubeSandbox

简报：

• 腾讯云于今日（4月22日）正式开源专为AI Agent设计的执行环境底座CubeSandbox，这是业内首个实现硬件级隔离且启动速度达亚百毫秒级的开源沙箱服务。
• 具备深度兼容性，原生支持E2B接口标准，开发者只需调整环境变量即可完成现有应用的“零成本”平滑迁移，并能稳定支撑Agent完整的“思考—执行—反馈”循环。
• 冷启动性能飞跃，通过资源池化及快照克隆等技术，安全沙箱冷启动时间缩短至60毫秒以内，50并发场景下平均响应时间仅为67毫秒。
• 运行在独立操作系统内核之上，硬件层面的彻底隔离确保了单沙箱异常不会引发连锁反应，同时支持Agent网络访问权限的精细化配置。
• 资源控制实现极致轻量化，借助Rust语言底层重写等技术，单个实例内存开销降至5MB以内，单台96核物理服务器可支持超2000个沙箱同时运行。
• 该方案已在腾讯内部“元宝”AI编程等大规模生产场景中得到验证，资源核时消耗显著降低了95.8%。

相关链接：

• 腾讯云开源 CubeSandbox：打造 AI Agent 的高性能“安全屋”

Tencent Cloud Open-Sources CubeSandbox, the Industry’s First Sub-100ms AI Agent Security Sandbox 🚀

Brief:

• Tencent Cloud officially open-sourced CubeSandbox today (April 22), an execution environment foundation purpose-built for AI Agents. It is the industry’s first open-source sandbox service to achieve hardware-level isolation while delivering startup speeds below 100 milliseconds.
• It offers deep compatibility, with native support for the E2B interface standard. Developers can smoothly migrate existing applications at “zero cost” simply by adjusting environment variables, while reliably supporting the full Agent “think–act–feedback” loop.
• Cold-start performance has seen a major breakthrough. Through technologies such as resource pooling and snapshot cloning, the secure sandbox’s cold-start time has been reduced to under 60 milliseconds, with an average response time of just 67 milliseconds under 50 concurrent requests.
• Running on an independent operating system kernel, the solution provides complete hardware-level isolation, ensuring that an issue in a single sandbox will not trigger a chain reaction. It also supports fine-grained configuration of Agent network access permissions 🔒.
• Resource control has been made extremely lightweight. With technologies including a low-level rewrite in Rust, the memory overhead of a single instance has been reduced to under 5MB, and a single 96-core physical server can support more than 2,000 sandboxes running simultaneously.
• The solution has already been validated in large-scale production scenarios within Tencent, including the “Yuanbao” AI coding platform, where resource core-hour consumption was significantly reduced by 95.8% 📉.

Related link:

• Tencent Cloud Open-Sources CubeSandbox: Building a High-Performance “Safe House” for AI Agents

阿里发布数字人助手“千问小酒窝”，深度集成全系应用实现交易闭环

简报：

• 阿里巴巴正式发布生态级AI助手“千问小酒窝”，标志着阿里千问AI体系完成统一数字人形象的构建，并深度嵌入淘宝、飞猪、高德、支付宝及大麦等全系生态应用。
• 该助手以“行动能力”为核心竞争力，实现了从用户需求理解到交易支付、再到履约交付的全链路闭环，支持智能点单、行程规划、跨应用打车及机票酒店预订等任务。
• 此次战略举措旨在通过打破应用间的交互壁垒，将大模型竞争重心从单一算法转向端到端的生态服务能力。
• 根据最新行业报告，截至2026年3月，千问App月活跃用户（MAU）规模已达1.66亿，位列国内AI原生应用市场第一梯队。

相关链接：

• 阿里巴巴发布生态级AI助手“千问小酒窝”，统一数字人形象全面接入阿里生态
• 一季度AI应用月活创新高，豆包、千问、DeepSeek稳居第一梯队

Alibaba Unveils Digital Human Assistant “Qwen Xiaojiuwo,” Deeply Integrating Its Full App Ecosystem to Enable Closed-Loop Transactions 🤖✨

Briefing:

• Alibaba has officially launched its ecosystem-level AI assistant, “Qwen Xiaojiuwo,” marking the completion of a unified digital human identity for the Alibaba Qwen AI system and its deep integration across the company’s ecosystem apps, including Taobao, Fliggy, Amap, Alipay, and Damai.
• Centered on “action capability” as its core competitive edge, the assistant enables a full closed-loop process from understanding user needs to payment and final fulfillment, supporting tasks such as smart ordering, itinerary planning, cross-app ride-hailing, and flight and hotel bookings.
• This strategic move aims to break down interaction barriers between apps and shift the focus of large-model competition from standalone algorithms to end-to-end ecosystem service capabilities.
• According to the latest industry report, as of March 2026, the Qwen app had reached 166 million monthly active users (MAU), placing it in the top tier of China’s native AI application market 📈.

Related Links:

• Alibaba Launches Ecosystem-Level AI Assistant “Qwen Xiaojiuwo,” Unifying Its Digital Human Identity Across the Alibaba Ecosystem
• AI App Monthly Active Users Hit a New High in Q1, with Doubao, Qwen, and DeepSeek Firmly in the Top Tier

Google推出Gemini 3.1 Pro双版本深度研究智能体，第八代TPU将于Cloud Next 26大会亮相

简报：

• Google基于Gemini 3.1 Pro模型推出Deep Research与Deep Research Max两款自主调研智能体，目前已通过付费版API向全球开发者开放预览。
• 该智能体引入MCP（模型上下文协议），支持连接企业内部私有数据库，并具备强大的原生可视化能力，可根据调研结果直接生成高质量图表。
• Google Cloud Next 26大会于今日（4月22日）至24日在拉斯维加斯举办，外界高度聚焦其第八代TPU（张量处理器）的正式亮相。
• 第八代TPU聚焦于扩大记忆体频宽并改进散热与互连架构，据分析师估计，Google芯片未来几年内有望夺得20%的AI市场份额，潜在商机约达9000亿美元。
• TPU的崛起引发了外界对辉达（Nvidia）AI霸主地位及核心技术CUDA价值的讨论，辉达执行长黄仁勋近期在受访时未正面回应CUDA被取代的疑虑。

相关链接：

• Google 发布 Gemini 3.1 Pro 深度研究智能体：高效生成专业数据图表
• Google年度大會倒數！第八代TPU引關注　輝達霸主地位岌岌可危？

Google Launches Dual-Version Gemini 3.1 Pro Deep Research Agents; Eighth-Generation TPU Set to Debut at Cloud Next 26 🚀

Briefing:

• Google has launched two autonomous research agents, Deep Research and Deep Research Max, based on the Gemini 3.1 Pro model. They are now available in preview to developers worldwide through the paid API.
• The agents introduce MCP (Model Context Protocol), enabling connections to enterprises’ internal private databases, and feature powerful native visualization capabilities that can directly generate high-quality charts based on research findings 📊.
• Google Cloud Next 26 will take place in Las Vegas from today (April 22) through April 24, with strong market attention focused on the official debut of its eighth-generation TPU (Tensor Processing Unit).
• The eighth-generation TPU is designed to expand memory bandwidth while improving cooling and interconnect architecture. According to analysts’ estimates, Google’s chips could capture 20% of the AI market within the next few years, representing a potential opportunity worth approximately $900 billion.
• The rise of TPU has sparked discussion over Nvidia’s dominance in AI and the long-term value of its core CUDA technology. In a recent interview, Nvidia CEO Jensen Huang did not directly address concerns about whether CUDA could eventually be displaced 🤖.

Related Links:

• Google Unveils Gemini 3.1 Pro Deep Research Agents: Efficiently Generate Professional Data Visualizations
• Countdown to Google’s Annual Conference! Eighth-Generation TPU Draws Attention — Is Nvidia’s Dominance at Risk?

单词学习：![[2026042106367105.mp3]]

新闻播报：![[2026042151658289.mp3]]

工信部：算力成AI发展关键底座，我国智能算力规模截至3月底达1882EFLOPS

简报：

• 工业和信息化部副部长张云明于4月21日指出，算力基础设施已成为驱动人工智能发展的关键底座。
• 最新数据显示，截至2026年3月底，我国智能算力规模已达到1882EFLOPS。
• 算力终端应用持续演进，萤石（EZVIZ）同日发布了具备64Tops算力的EZVIZ AI Core X智能大模型主机，将高性能主机、NAS与智能家居中控融为一体。
• 该主机搭载首个智能家居专用大模型Home Vita，支持自然语言检索多媒体文件、动态提炼监控录像关键事件，但短期内暂无在国内市场的发售计划。

相关链接：

• 我国智能算力规模达1882EFLOPS
• 萤石发布EZVIZ AI Core X智能大模型主机：64Tops算力开启家居“大脑”时代

MIIT: Computing Power Has Become a Key Foundation for AI Development, with China’s Intelligent Computing Capacity Reaching 1,882 EFLOPS as of the End of March 🚀

Briefing:

• On April 21, Zhang Yunming, Vice Minister of the Ministry of Industry and Information Technology, stated that computing power infrastructure has become a key foundation driving the development of artificial intelligence.
• The latest data shows that as of the end of March 2026, China’s intelligent computing capacity had reached 1,882 EFLOPS.
• Computing-powered terminal applications continue to evolve. On the same day, EZVIZ released the EZVIZ AI Core X intelligent large-model host featuring 64 Tops of computing power, integrating a high-performance host, NAS, and a smart home central control hub into one device 🤖.
• The host is equipped with Home Vita, the first large model designed specifically for smart homes. It supports natural-language retrieval of multimedia files and dynamic extraction of key events from surveillance footage, though there are currently no short-term plans for its release in the domestic market 🏠.

Related Links:

• China’s intelligent computing capacity reaches 1,882 EFLOPS
• EZVIZ releases the EZVIZ AI Core X intelligent large-model host: 64 Tops of computing power ushers in the era of the home “brain”

全球大模型Token调用量受算力涨价影响连续两周下滑，美系活跃度反超

简报：

• 全球AI大模型总Token调用量在长达十周连续增长后，已连续两周出现下滑，4月13日至19日期间维持在20.6万亿Token。
• 中美市场活跃度发生“一降一升”的逆转：中国AI大模型周调用量环比缩减23.77%降至4.44万亿Token，而美国市场环比增长20.62%达4.91万亿Token，实现近两个月来首次反超。
• 此次波动与全球算力涨价潮紧密相关，国内阿里云、腾讯、百度等相继上调大模型相关服务费用，智谱完成年内第三次调价，海外Anthropic亦更新了定价模型。
• 业内分析人士对长期趋势持乐观态度，认为目前的调用量阶段性回落属于算力成本重塑背景下的市场自我调节，长期仍处于快速增长的上升通道。
• 为应对日益增长的计算需求，微软总投资33亿美元的Fairwater AI数据中心宣布提前投入使用，该中心配备了数十万块英伟达GB200超级芯片。

相关链接：

• 全球AI算力“涨声”一片！Token调用量遭遇两年来首次大逆转
• 微软 Fairwater AI 数据中心提前上线，算力再创新高

Global Large Model Token Usage Declines for Two Consecutive Weeks Amid Rising Computing Costs, While U.S. Activity Overtakes China

Brief:

• After ten consecutive weeks of growth, total global AI large model token usage has declined for two straight weeks, holding at 20.6 trillion tokens during April 13 to 19.
• Activity in the Chinese and U.S. markets has reversed 📉📈: China’s weekly AI large model token usage fell 23.77% week over week to 4.44 trillion tokens, while the U.S. market grew 20.62% to 4.91 trillion tokens, marking its first lead in nearly two months.
• This fluctuation is closely tied to the global wave of rising computing costs. In China, Alibaba Cloud, Tencent, and Baidu have all raised prices for large model-related services, while Zhipu completed its third price adjustment of the year. Overseas, Anthropic has also updated its pricing model.
• Industry analysts remain optimistic about the long-term outlook, viewing the current pullback in token usage as a phase of market self-adjustment under reshaped computing cost structures, with the long-term growth trajectory still intact 🚀
• To meet growing computing demand, Microsoft’s Fairwater AI data center, backed by a total investment of $3.3 billion, has been brought into operation ahead of schedule. The facility is equipped with hundreds of thousands of Nvidia GB200 superchips.

Related Links:

• Global AI Computing Costs Surge Across the Board! Token Usage Sees Its First Major Reversal in Two Years
• Microsoft’s Fairwater AI Data Center Goes Live Ahead of Schedule, Reaching New Computing Power Highs

华为2499元AI眼镜正式发布，携手支付宝上线“看一下支付”功能

简报：

• 4月20日下午，华为在Pura系列新品发布会上正式推出全新AI眼镜，起售价为2499元。
• 该眼镜内置支付宝鸿蒙版应用，支持“看一下支付”功能，用户绑定后仅需通过简单的语音指令或双击AI键，即可自动识别收款码并完成交易。
• 支付过程基于鸿蒙协同认证能力、蚂蚁GPASS技术及支付宝多维风控系统，支付宝官方承诺提供“被盗即赔”服务保障。
• 支付宝表示，“看一下支付”功能不仅支持华为AI眼镜，还兼容Rokid、小米、千问和雷鸟创新等国内知名智能眼镜品牌。

相关链接：

• 华为 AI 眼镜发布，支付宝携手支持 “看一下支付” 功能

Huawei Officially Launches AI Glasses Starting at RMB 2,499, Introduces Pay with a Glance Feature with Alipay 👓💳

Brief:

• On the afternoon of April 20, Huawei officially unveiled its new AI glasses at the Pura series product launch event, with a starting price of RMB 2,499.
• The glasses come with the HarmonyOS version of Alipay built in and support the Pay with a Glance feature. After linking the device, users can simply use a voice command or double-tap the AI button to automatically recognize a payment code and complete the transaction.
• The payment process is powered by HarmonyOS collaborative authentication capabilities, Ant Group’s GPASS technology, and Alipay’s multi-dimensional risk control system. Alipay has officially promised a compensation guarantee in cases of unauthorized loss.
• Alipay stated that the Pay with a Glance feature supports not only Huawei AI glasses, but is also compatible with other well-known domestic smart glasses brands, including Rokid, Xiaomi, Qianwen, and RayNeo Innovation ✨

Related link:

• Huawei AI Glasses Launched, Alipay Jointly Supports the Pay with a Glance Feature

SpaceX披露IPO招股书：2025年底总资产达920亿美元，马斯克将保留超级投票权

简报：

• 招股书摘要显示，截至2025年底SpaceX总资产为920亿美元，总负债为508亿美元，公司已秘密提交IPO招股说明书。
• 星链（Starlink）业务运营收入增长迅猛，2025年达到44.2亿美元，相比2024年的20亿美元及2023年的4.69亿美元实现连年倍增。
• 马斯克在IPO后将通过超级投票权保留对公司的控制权，并继续担任首席执行官、首席技术官及九人董事会主席。
• 财务细节显示，马斯克去年增持了14亿美元的SpaceX股份；尽管其去年薪酬仅为54080美元，但上市后将获得价值数十亿美元的股权。

相关链接：

• 马斯克去年增持14亿美元SpaceX股份
• 截至2025年底SpaceX总资产达920亿美元
• 马斯克和内部人士将在IPO后保留对SpaceX的投票控制权
• SpaceX秘密提交招股说明书披露财务细节

SpaceX Reveals IPO Prospectus: Total Assets Reach $92 Billion by End of 2025, Musk to Retain Super Voting Rights 🚀

Brief:

• A summary of the prospectus shows that, as of the end of 2025, SpaceX had total assets of $92 billion and total liabilities of $50.8 billion. The company has confidentially submitted its IPO filing.
• Starlink’s operating revenue grew rapidly, reaching $4.42 billion in 2025, more than doubling from $2.0 billion in 2024 and $469 million in 2023. 📈
• After the IPO, Musk will retain control of the company through super voting rights and will continue to serve as Chief Executive Officer, Chief Technology Officer, and Chairman of the nine-member board.
• Financial details show that Musk increased his holdings in SpaceX by dJyH6kDkfC5RlM16.answerText.4 billion last year. Although his salary last year was only $54,080, he will receive equity worth billions of dollars after the listing.

Related links:

• Musk increased his SpaceX stake by dJyH6kDkfC5RlM16.answerText.4 billion last year
• SpaceX’s total assets reached $92 billion by the end of 2025
• Musk and insiders will retain voting control of SpaceX after the IPO
• SpaceX confidentially filed its prospectus, revealing financial details

月之暗面开源Kimi K2.6模型，代码生成与长程智能体能力比肩国际顶尖水平

简报：

• 4月20日晚，月之暗面正式发布并开源最新模型Kimi K2.6，在长程任务处理和Agent集群协同方面实现大幅迭代。
• 根据官方测试数据，K2.6在SWE-Bench Pro和DeepSearchQA基准测试中的成绩，已能与GPT-5.4、Claude Opus 4.6和Gemini 3.1 Pro等国际顶尖闭源模型并驾齐驱。
• 该模型作为“迄今最强代码模型”，能够不间断执行编码任务长达13小时，并支持一次性编写或修改超过4000行代码。
• K2.6模型现已全面同步上线网页端、最新版移动应用及API接口，专属开发者工具Kimi Code编程助手也已升级接入新内核。

相关链接：

• 月之暗面发布新模型K2.6：长程任务与智能体能力实现新突破

Moonshot AI Open-Sources Kimi K2.6, Bringing Code Generation and Long-Horizon Agent Capabilities to the Level of Global Top Models 🚀

Brief:

• On the evening of April 20, Moonshot AI officially released and open-sourced its latest model, Kimi K2.6, marking a major upgrade in long-horizon task handling and multi-agent collaboration.
• According to official benchmark results, K2.6 performs on par with leading international closed-source models such as GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro on SWE-Bench Pro and DeepSearchQA.
• Described as the strongest coding model to date, K2.6 can continuously execute coding tasks for up to 13 hours and supports writing or modifying more than 4,000 lines of code in a single session. 💻
• K2.6 is now fully available across the web platform, the latest mobile app, and the API. Kimi Code, the company’s dedicated programming assistant, has also been upgraded to run on the new core. 🤖

Related link:

• Moonshot AI Releases New Model K2.6: Breakthroughs in Long-Horizon Tasks and Agent Capabilities

单词学习：![[2026042035138099.mp3]]

新闻播报：![[2026042011933224.mp3]]

Anthropic 年化收入突破 300 亿美元，估值直逼万亿美金大关

简报：

• Anthropic 年化收入目前已超过 300 亿美元，较去年年底增长逾三倍，在与 OpenAI 的竞争中展现出强劲势头。
• 公司财务状况显著改善，毛利率从 2024 年的 -94% 跃升至 2025 年的 +40%，年化收入从不足 10 亿美元猛增至 90 亿美元。
• 业绩增长主要由 Claude Code、Cowork 产品以及推理模型代币销售的持续上升所带动。
• 首席财务官克里希纳·拉奥（Krishna Rao）已收到基于 8000 亿美元估值的投资提案，部分投资者预测其市值有望突破 1 万亿美元。
• 尽管发展势头强劲，但知情人士称公司目前暂无新融资计划，最早需待 2026 年 5 月董事会会议后再行讨论。

相关链接：

• Anthropic收入暴增，市值或突破万亿美元大关

Anthropic’s Annualized Revenue Surpasses $30 Billion, Valuation Nears the Trillion-Dollar Mark 🚀

Brief:

• Anthropic’s annualized revenue has now exceeded $30 billion, more than tripling since the end of last year, showing strong momentum in its competition with OpenAI.
• The company’s financial position has improved sharply, with gross margin rising from -94% in 2024 to +40% in 2025, while annualized revenue surged from less than dJyH6kDkfC5RlM16.answerText billion to $9 billion.
• The growth has been driven mainly by continued increases in sales of Claude Code, the Cowork product, and tokens for reasoning models.
• Chief Financial Officer Krishna Rao has received investment proposals based on an $800 billion valuation, and some investors predict the company’s market value could surpass dJyH6kDkfC5RlM16.answerText trillion 💰.
• Despite its strong momentum, sources say the company currently has no plans for a new funding round, with any discussion unlikely before the board meeting in May 2026.

Related link:

• Anthropic revenue surges, valuation may break the trillion-dollar threshold 📈

阿里推出新一代千问旗舰模型Qwen3.6-Max早期预览版

简报：

• 阿里正式发布Qwen3.6-Max-Preview，该模型为新一代千问旗舰模型的早期预览版
• 该预览版模型拥有更强的世界知识和指令遵循能力
• 该模型在多项基准测试上的智能体编程表现获得显著提升

相关链接：

• 阿里发布Qwen3.6-Max预览版

Alibaba Launches Early Preview of New-Generation Flagship Qwen Model Qwen3.6-Max 🚀

Brief:

• Alibaba has officially released Qwen3.6-Max-Preview, an early preview version of the new-generation flagship Qwen model
• This preview model features stronger world knowledge and improved instruction-following capabilities
• The model has achieved significant improvements in agentic coding performance across multiple benchmark tests 💡

Related link:

• Alibaba releases Qwen3.6-Max preview

Nature最新研究：大模型存在“潜意识学习”，不良特征可通过纯数字蒸馏悄然传染

简报：

• Nature最新发表的论文首次证实大型语言模型（LLM）存在“潜意识学习”现象，不良行为特征能通过纯数字、代码或思维链等看似无害的序列传递给下游模型。
• 实验表明，携带特定偏好（如“猫头鹰”）或危险特征的“老师模型”，即使生成的纯数字序列不包含任何相关语义，用于训练的“学生模型”仍会自发表现出对这些特征的强烈偏好。
• 机制研究显示，在模型蒸馏过程中，学生模型在权重空间中复制了老师的隐性特征梯度，这种信号隐藏在数据的统计分布纹路中，超越了表层语义。
• 这一发现揭示了AI安全系统面临三大失效风险：供应链可能遭受“隐形投毒”攻击，模型间可能进行人类无法察觉的通信，且目前基于语义的安全评估工具无法有效探测此类非语义污染。
• 论文促使AI行业重新审视模型安全范式，防范重点需要从“检查模型输出是否有毒”转向“评估模型权重是否干净”。

相关链接：

• Nature 论文揭秘：AI 不良特征可隐藏在纯数字中实现“隔空传毒”

Latest Nature Study: Large Models Exhibit “Subliminal Learning,” and Harmful Traits Can Quietly Spread Through Purely Numeric Distillation 🧠🔢⚠️

Brief:

• A newly published Nature paper provides the first evidence that large language models (LLMs) exhibit a “subliminal learning” phenomenon, in which undesirable behavioral traits can be transmitted to downstream models through seemingly harmless sequences such as pure numbers, code, or chains of thought.
• Experiments show that when a “teacher model” carries specific preferences (such as for “owls”) or dangerous traits, a “student model” trained on purely numeric sequences generated by the teacher—even when those sequences contain no related semantic content—can still spontaneously develop a strong preference for those traits.
• Mechanistic analysis indicates that during model distillation, the student model copies the teacher’s latent feature gradients in weight space. This signal is hidden in the statistical texture of the data distribution and goes beyond surface-level semantics.
• This finding reveals three major failure risks for AI safety systems: supply chains may be vulnerable to “invisible poisoning” attacks, models may be able to communicate in ways humans cannot detect, and current semantics-based safety evaluation tools are unable to effectively identify this kind of non-semantic contamination.
• The paper prompts the AI industry to rethink its model security paradigm: the focus of prevention needs to shift from “checking whether model outputs are toxic” to “evaluating whether model weights are clean.”

Related link:

• Nature paper reveals: Harmful AI traits can hide in pure numbers and spread “remotely”

蚂蚁灵光闪应用数量突破3000万，投入1亿元专项激励扶持AI创作者

简报：

• 蚂蚁集团旗下AI产品灵光App宣布正式启动“灵光闪应用创作者激励计划”，投入1亿元专项基金扶持开发者
• 自2025年底发布至今四个月内，该平台通过自然语言生成的闪应用数量已突破3000万个，涵盖工具、游戏等多个维度
• 激励计划于4月20日正式开启，设立日度与周度双重机制，每日最高奖金100万元覆盖万个应用，每周单创作者最高奖金100万元
• 灵光App同步上线“灵光圈”公共创意社区，并迭代多智能体协作、全模态生成及手机端原生能力集成等功能
• 平台通过降低编程门槛，使非编程背景的普通用户能够通过意图驱动创作AI应用，推动个体创造力释放

相关链接：

• 蚂蚁灵光App闪应用突破3000万个，宣布投入1亿元启动专项创作者激励

Ant Group’s Lingguang Flash Apps Surpass 30 Million, Launches RMB 100 Million Special Incentive Program to Support AI Creators 🚀

Briefing:

• Lingguang App, an AI product under Ant Group, officially announced the launch of the “Lingguang Flash App Creator Incentive Program,” backed by a dedicated RMB 100 million fund to support developers
• In just four months since its release at the end of 2025, the platform has generated more than 30 million flash apps through natural language, covering multiple categories such as tools and games
• The incentive program officially begins on April 20, introducing both daily and weekly reward mechanisms, with daily prizes of up to RMB 1 million distributed across 10,000 apps, and weekly prizes of up to RMB 1 million for a single creator
• Lingguang App has also launched “Lingguang Circle,” a public creative community, while upgrading features such as multi-agent collaboration, full-modal generation, and integration of native mobile capabilities ✨
• By lowering the barrier to programming, the platform enables ordinary users without coding backgrounds to create AI apps through intent-driven workflows, helping unlock individual creativity 💡

Related Link:

• Ant Lingguang App Flash Apps Surpass 30 Million, Announces RMB 100 Million Special Creator Incentive Program

微软33亿美元打造的最强AI数据中心提前投运，内置数十万块GB200芯片

简报：

• 微软CEO萨提亚·纳德拉于4月16日宣布，位于美国威斯康辛州的“全球最强”AI数据中心Fairwater已提前上线。
• 该数据中心总投资规模达33亿美元，搭载数十万块英伟达GB200超级芯片，性能为全球最快超算的十倍，主要用于前沿AI模型的训练。
• 设施占地约127.6公顷，包含三栋建筑，采用闭环液冷系统，建设过程使用了约1.2万吨结构钢材及数百公里长的地下电缆与机械管道。
• 微软计划在Fairwater数据中心投运后，于美国其他地区继续建设多个同类型的AI数据中心。

相关链接：

• 微软“全球最强”AI 数据中心 Fairwater 提前上线，包含数十万块 GB200

Microsoft’s $3.3 Billion “World’s Most Powerful” AI Data Center Goes Online Ahead of Schedule, Packed with Hundreds of Thousands of GB200 Chips 🚀

Brief:

• On April 16, Microsoft CEO Satya Nadella announced that Fairwater, the “world’s most powerful” AI data center located in Wisconsin, USA, had officially gone online ahead of schedule.
• The data center represents a total investment of $3.3 billion and is equipped with hundreds of thousands of Nvidia GB200 superchips. Its computing performance is said to be ten times that of the world’s fastest supercomputer, and it will primarily be used to train cutting-edge AI models.
• Covering approximately 127.6 hectares, the facility includes three buildings and uses a closed-loop liquid cooling system. Its construction required about 12,000 tons of structural steel, along with hundreds of kilometers of underground electrical cables and mechanical piping 🏗️
• Following the launch of the Fairwater data center, Microsoft plans to continue building several similar AI data centers in other parts of the United States.

Related link:

• Microsoft’s “World’s Most Powerful” AI Data Center Fairwater Goes Online Ahead of Schedule, Featuring Hundreds of Thousands of GB200 Chips

英特尔14A工艺年底或迎重磅客户，瑞银猜测含英伟达与苹果等

简报：

• 英特尔即将推出的14A工艺技术，预计将在今年年底吸引知名企业客户的加入。
• 瑞银猜测，英特尔该新技术的潜在客户将包括英伟达、苹果、谷歌和AMD等。
• 瑞银还强调，英特尔有可能将其位于俄亥俄州的晶圆厂与马斯克的Terafab项目合并，以进一步提升其晶圆代工声誉。

相关链接：

• 英特尔14A工艺整装待发

Intel’s 14A Process May Land Major Customers by Year-End, UBS Speculates Nvidia and Apple Could Be Among Them 🚀

Brief:

• Intel’s upcoming 14A process technology is expected to attract well-known corporate customers by the end of this year.
• UBS speculates that potential clients for Intel’s new technology could include Nvidia, Apple, Google, and AMD.
• UBS also emphasized that Intel may combine its Ohio wafer fab with Musk’s Terafab project to further enhance its foundry reputation. 🔧✨

Related link:

• Intel’s 14A Process Is Ready to Go

Anthropic发布Claude Design，打造从创意到代码的设计开发全链路

简报：

• Anthropic正式推出实验性产品Claude Design，支持通过文字描述、图片或办公文档快速生成交互原型、演示文稿、单页报告及营销物料
• 该工具由Claude Opus 4.7视觉模型驱动，用户可通过自然语言对话、自定义滑块或内联评论进行精细化迭代，甚至能自动应用团队代码库中的品牌规范
• 生成的设计成果支持导出至Canva继续编辑，或输出为PDF、PPTX及独立HTML文件，极大降低了非设计人员的视觉创作门槛
• 核心功能支持“一键打包移交给Claude Code”，可自动转化设计资产与规格用于前端开发，缩短从创意到可运行产品的周期
• 目前Claude Design处于研究预览阶段，面向Claude Pro、Max、Team及Enterprise订阅用户逐步开放使用

相关链接：

• Claude推出Claude Design：支持一句话生成交互原型、PPT及营销物料

Anthropic Launches Claude Design, Building a Full Design-to-Development Workflow from Idea to Code ✨

Brief:

• Anthropic has officially introduced its experimental product Claude Design, which enables users to quickly generate interactive prototypes, presentations, one-page reports, and marketing materials from text prompts, images, or office documents.
• Powered by the Claude Opus 4.7 vision model, the tool supports fine-tuned iteration through natural language conversations, custom sliders, or inline comments, and can even automatically apply brand guidelines from a team’s codebase.
• The generated design outputs can be exported to Canva for further editing, or saved as PDF, PPTX, and standalone HTML files, significantly lowering the barrier to visual content creation for non-designers 🎨
• A core feature allows users to hand projects off to Claude Code with one click, automatically converting design assets and specifications for frontend development and shortening the cycle from concept to working product.
• Claude Design is currently in the research preview stage and is being gradually rolled out to Claude Pro, Max, Team, and Enterprise subscribers 🚀

Related link:

• Claude Launches Claude Design: Generate Interactive Prototypes, PPTs, and Marketing Materials with a Single Sentence

单词学习：![[2026041775446978.mp3]]

新闻播报：![[2026041768328015.mp3]]

OpenAI拟超200亿美元采购Cerebras服务器以减少对英伟达依赖

简报：

• 消息称OpenAI同意在未来三年内向Cerebras支付超过200亿美元。
• 这笔巨额资金将用于使用由Cerebras芯片驱动的服务器集群器。
• OpenAI此次合作的主要目的是通过多元化硬件布局来显著降低计算成本，同时减少对英伟达芯片的依赖。

相关链接：

• 豪掷200亿美元，OpenAI据称将与Cerebras达成重磅协议

OpenAI Reportedly Plans to Purchase More Than $20 Billion in Cerebras Servers to Reduce Reliance on Nvidia 🚀

Brief:

• Reports say OpenAI has agreed to pay more than $20 billion to Cerebras over the next three years.
• This massive funding will be used to access server clusters powered by Cerebras chips.
• The primary goal of this partnership is to significantly lower computing costs through a more diversified hardware strategy, while also reducing OpenAI’s dependence on Nvidia chips 🤖💡

Related link:

• Spending $20 Billion: OpenAI Reportedly Reaches Major Agreement with Cerebras

谷歌开源40亿参数医疗AI MedGemma 1.5，全面升级3D多维度影像分析能力

简报：

• 谷歌开源医疗AI基础模型MedGemma 1.5，核心突破在于克服传统2D图像局限，原生支持CT和MRI三维扫描数据以及多时间点病情对比分析。
• 此外，模型还支持病理学全视野数字切片、胸部X光病变边界框精准标注，大幅提升对PDF格式电子病历和化验单的结构化数据提取理解能力。
• 在保持40亿参数计算效率的前提下，新版本在3D MRI分类准确率上提升11%，病理图像宏观F1分数提升47%，电子病历问答准确率提升22%。
• 训练过程中使用了放射学、皮肤科等大量医学图文数据，设计了最高85张切分序列图像的预处理方法，并引入专家模型进行知识蒸馏。
• 谷歌将其明确界定为供开发者微调的底层资源，目前并非开箱即用的临床决策工具，实际落地仍需具体场景的专项训练。

相关链接：

• 谷歌开源医疗AI新星MedGemma 1.5：从看平面图到读懂3D影像
• MedGemma 1.5 论文地址

Google Open-Sources 4-Billion-Parameter Medical AI MedGemma 1.5, Fully Upgrading Its 3D Multi-Dimensional Imaging Analysis Capabilities 🏥🧠

Brief:

• Google has open-sourced the medical AI foundation model MedGemma 1.5. Its core breakthrough lies in overcoming the limitations of traditional 2D images, with native support for 3D CT and MRI scan data as well as comparative analysis of disease progression across multiple time points.
• In addition, the model supports whole-slide digital pathology images and precise bounding-box annotation of lesions in chest X-rays, while significantly improving its ability to extract and understand structured data from PDF-based electronic medical records and lab reports.
• While maintaining the computational efficiency of a 4-billion-parameter model, the new version improves 3D MRI classification accuracy by 11%, boosts the macro F1 score for pathology images by 47%, and increases electronic medical record question-answering accuracy by 22%.
• During training, it used large volumes of medical image-text data from radiology, dermatology, and other fields, introduced a preprocessing method capable of handling up to 85 segmented sequential images, and incorporated expert models for knowledge distillation.
• Google has clearly defined it as a foundational resource for developers to fine-tune. It is not currently an out-of-the-box clinical decision-making tool, and real-world deployment will still require task-specific training for particular use cases. 📄

Related links:

• Google Open-Sources New Medical AI Rising Star MedGemma 1.5: From Reading Flat Images to Understanding 3D Scans
• MedGemma 1.5 paper

Google AI Studio接入AI订阅服务，按请求付费解锁所有模型与智能体权限

简报：

• Google 在 AI Studio 上逐步推出对 Google AI 计划的支持，用户可直接关联订阅，获得更高的使用限额及更流畅的文本、图像、音频等多模态模型访问体验。
• 新增的“按请求付费（Pay-per-request）”选项目前提供对“所有模型和代理（智能体）”的完整访问权限，而 Google AI 订阅计划本身暂不包含此项完整功能。
• 按请求付费模式让用户能够无限制探索最新模型，而订阅计划则侧重于稳定的限额提升和跨产品整合。
• 更新内容暗示 Google 可能正在测试或逐步开放基于 Gemini 等模型的智能体（Agent）能力，以帮助开发者通过自然语言快速构建应用。
• 目前有用户反馈，在切换订阅时偶尔会出现确认流程卡顿的问题。

相关链接：

• Google AI Studio 直接支持 AI 订阅计划，按请求付费独享“所有模型+智能体”？

Google AI Studio Integrates AI Subscription Services, Unlocking Access to All Models and Agents with Pay-per-Request 🚀

Brief:

• Google is gradually rolling out support for the Google AI plan in AI Studio, allowing users to link their subscriptions directly for higher usage limits and a smoother multimodal access experience across text, image, audio, and more.
• The newly added “Pay-per-request” option currently provides full access to “all models and agents,” while the Google AI subscription plan itself does not yet include this complete functionality.
• The pay-per-request model enables users to explore the latest models without restriction, while the subscription plan focuses more on stable quota increases and cross-product integration.
• This update suggests that Google may be testing or gradually opening up Agent capabilities based on models such as Gemini, helping developers build applications more quickly through natural language. 🤖
• Some users have reported occasional confirmation flow lag when switching subscriptions.

Related link:

• Google AI Studio Now Directly Supports AI Subscription Plans, with Pay-per-request Exclusively Offering “All Models + Agents”? 🔗

Anthropic推出Claude Opus 4.7性能大幅提升，同步实施强制实名验证引争议

简报：

• Anthropic今日正式发布Claude Opus 4.7模型，在高级软件工程、复杂任务处理及指令执行方面较上一代有显著提升，并具备结果自我验证能力。
• 新模型视觉处理能力增强，支持识别最高2,576像素长边的图像，同时内置网络安全防护措施，可自动检测并拦截高风险请求。
• Opus 4.7优化了长会话的内存利用率，推出更细致的难度控制和任务预算功能，定价维持不变（输入$5/百万token，输出$25/百万token）。
• 目前Opus 4.7已在Claude产品及API、Amazon Bedrock、Google Cloud Vertex AI和Microsoft Foundry等平台全面上线。
• 此外，Claude近期实施强制实名验证（KYC），要求用户手持实体政府证件实时拍照，不支持复印或数字件，引发用户对数据隐私及封号风险的担忧。
• Anthropic严格设定最低18岁的使用年龄限制，已有未成年开发者因触发限制被系统封号并退款。

相关链接：

• Anthropic 发布全新 Claude Opus 4.7，功能显著提升
• Claude 强制实名验证引发争议：用户担心封号风险

Anthropic Launches Claude Opus 4.7 with Major Performance Gains, While Mandatory Real-Name Verification Sparks Controversy

Briefing:

• Anthropic officially launched the Claude Opus 4.7 model today, delivering notable improvements over the previous generation in advanced software engineering, complex task handling, and instruction following, while also gaining the ability to self-verify results. 🚀
• The new model features stronger visual processing capabilities, supporting image recognition with a maximum long side of 2,576 pixels, and includes built-in cybersecurity safeguards that can automatically detect and block high-risk requests.
• Opus 4.7 also improves memory efficiency in long conversations, introduces more granular difficulty control and task budget features, and keeps pricing unchanged at $5 per million input tokens and $25 per million output tokens.
• Opus 4.7 is now fully available across Claude products and API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
• In addition, Claude has recently rolled out mandatory real-name verification, requiring users to take a live photo while holding a physical government-issued ID. Copies and digital versions are not accepted, raising concerns over data privacy and the risk of account bans. ⚠️
• Anthropic also enforces a strict minimum age requirement of 18, and some underage developers have reportedly had their accounts banned and refunded after triggering the restriction.

Related links:

• Anthropic Releases New Claude Opus 4.7 with Significant Feature Upgrades
• Claude Mandatory Real-Name Verification Triggers Controversy as Users Worry About Ban Risks

阿里正式开源低耗MoE模型Qwen3.6-35B-A3B，以30亿激活参数突破智能体编程能力

简报：

• 阿里巴巴千问大模型团队于2026年4月16日正式开源稀疏混合专家（MoE）模型 Qwen3.6-35B-A3B，标志着轻量级模型在智能体编程领域实现关键突破。
• 该模型总参数量为350亿，运行时激活参数仅为30亿，以极低的计算成本在核心编程基准测试中超越了270亿参数的稠密模型 Qwen3.5-27B。
• 模型逻辑推理与智能体协同能力可比肩 Gemma4-31B 等大体量模型，并在空间智能与视觉感知方面表现优异，部分多模态指标达到 Claude Sonnet4.5 水平。
• Qwen3.6-35B-A3B 目前已集成至 Qwen Studio，并在阿里云百炼平台以 qwen3.6-flash 之名提供 API 服务，可无缝适配 OpenClaw 等主流 AI 编程助手。
• 此外，通义千问关联公司通义云启（杭州）信息技术有限公司近日注册资本由1亿元增至2亿元翻倍，并新增杭州通义实验室为股东，与阿里云共同持股。

相关链接：

• 阿里开源 Qwen3.6-35B-A3B:30亿激活参数实现编程能力跨越式升级
• 通义千问关联公司注册资本翻倍，阿里云与杭州通义实验室共同持股

Alibaba Officially Open-Sources the Low-Cost MoE Model Qwen3.6-35B-A3B, Achieving a Breakthrough in Agentic Coding with 3 Billion Active Parameters 🚀

Briefing:

• On April 16, 2026, Alibaba’s Qwen large model team officially open-sourced the sparse Mixture-of-Experts (MoE) model Qwen3.6-35B-A3B, marking a major breakthrough for lightweight models in the field of agentic coding.
• The model has a total of 35 billion parameters, but activates only 3 billion parameters at runtime. With extremely low computational cost, it surpasses the 27-billion-parameter dense model Qwen3.5-27B on core coding benchmark tests.
• Its logical reasoning and agent collaboration capabilities are comparable to larger-scale models such as Gemma4-31B, while also delivering strong performance in spatial intelligence and visual perception. On some multimodal benchmarks, it reaches the level of Claude Sonnet4.5 🤖
• Qwen3.6-35B-A3B has now been integrated into Qwen Studio and is also available through API services on Alibaba Cloud’s Bailian platform under the name qwen3.6-flash, enabling seamless compatibility with mainstream AI coding assistants such as OpenClaw.
• In addition, Tongyi Yunqi (Hangzhou) Information Technology Co., Ltd., an affiliated company of Tongyi Qianwen, recently doubled its registered capital from RMB 100 million to RMB 200 million, and added Hangzhou Tongyi Lab as a new shareholder, jointly held with Alibaba Cloud 📈

Related Links:

• Alibaba Open-Sources Qwen3.6-35B-A3B: 3 Billion Active Parameters Deliver a Leap in Coding Capability
• Tongyi Qianwen Affiliate Doubles Registered Capital, with Alibaba Cloud and Hangzhou Tongyi Lab as Joint Shareholders

字节跳动面向全球开启前沿技术博士生校招，并加码AI人才激励

简报：

• 字节跳动正式启动前沿技术领域人才校园招聘项目，面向全球优秀技术人才开放全职及实习岗位。
• 招聘对象主要为博士群体，全职岗位面向2026年9月至2027年8月毕业的博士生，实习岗位面向2026年9月及以后毕业的在校博士生。
• 顶会顶刊论文作者、专利拥有者、国际竞赛获奖者及有重大影响力技术落地成果的人才将获优先考量。
• 开放岗位方向涵盖大模型应用、搜索/推荐/广告、安全/AI Safety、AIGC、AI Coding、视频架构及AI for Science等多个前沿技术领域。
• 工作地点覆盖国内五城（北京、上海、深圳、杭州、成都）及海外五城（新加坡、悉尼、圣何塞、西雅图、圣地亚哥）。
• 为增强对顶尖AI人才的吸引力与激励，字节跳动同步启动了针对大模型业务核心人员的首轮“豆包股”回购，回购价较授予价上涨约30%。

相关链接：

• 字节跳动启动前沿技术人才校招，面向全球博士生开放多个AI方向
• 字节跳动启动首轮“豆包股”回购，涨幅达30% 强化 AI 人才激励

ByteDance Launches Global Campus Recruitment for PhD Talent in Frontier Technologies and Steps Up AI Incentives 🚀

Brief:

• ByteDance has officially launched a campus recruitment program for talent in frontier technology fields, opening both full-time and internship positions to outstanding technical talent worldwide.
• The program mainly targets PhD candidates. Full-time roles are open to PhD students graduating between September 2026 and August 2027, while internship roles are open to current PhD students graduating in September 2026 or later.
• Priority will be given to candidates who have published papers at top conferences or in leading journals, hold patents, have won international competitions, or have delivered highly impactful technical achievements.
• Open roles cover a wide range of frontier technology areas, including large model applications, search/recommendation/advertising, security/AI Safety, AIGC, AI Coding, video architecture, and AI for Science.
• Job locations span five cities in China (Beijing, Shanghai, Shenzhen, Hangzhou, and Chengdu) and five overseas cities (Singapore, Sydney, San Jose, Seattle, and San Diego).
• To further strengthen its appeal to top AI talent and improve incentives, ByteDance has also launched the first round of repurchases of Doubao shares for core personnel in its large model business, with the repurchase price about 30% higher than the grant price 📈🤖

Related links:

• ByteDance launches campus recruitment for frontier technology talent, opening multiple AI roles to PhD students worldwide
• ByteDance launches first round of Doubao share repurchases, up 30%, strengthening AI talent incentives

单词学习：![[2026041871031676.mp3]]

新闻播报：![[2026041802368355.mp3]]

谷歌Gemini打通Google Photos，实现私人影像AI个性化生成

简报：

• 谷歌宣布Gemini将支持连接用户个人相片库，用户开启Personal Intelligence功能后可整合谷歌应用数据以获得个性化体验。
• 配合图像生成工具Nano Banana，Gemini可直接调取Google Photos中的私人相片生成个性化图像，无需手动上传。
• 用户可直接要求生成特定风格（如动画风格）的家庭活动照片，系统会自动完成创作。
• 此次更新打破了私人影像与AI生成能力之间的壁垒，推动图像生成走向真正的个人化。

相关链接：

• 谷歌Gemini接入个人相册，AI生成图像迈向真正个性化

Google Gemini Connects with Google Photos, Enabling Personalized AI Image Generation from Private Photos 📸✨

Briefing:

• Google has announced that Gemini will support connections to users’ personal photo libraries. After enabling the Personal Intelligence feature, users can integrate data from Google apps for a more personalized experience.
• Working alongside the image generation tool Nano Banana, Gemini can directly access private photos stored in Google Photos to create personalized images without requiring manual uploads.
• Users can simply request family event photos in a specific style, such as an animated look, and the system will automatically generate them.
• This update breaks down the barrier between private imagery and AI generation capabilities, pushing image generation toward truly personalized experiences 🤖🖼️

Related link:

• Google Gemini connects to personal photo albums, bringing AI-generated images closer to true personalization

OpenAI拟最高斥资300亿美元购买Cerebras算力，并最高持股10%助力其冲刺IPO

简报：

• OpenAI计划未来三年内向芯片初创公司Cerebras支付超200亿美元购买服务器算力，并额外提供约10亿美元资助其开发数据中心，总支出或高达300亿美元。
• 作为合作回报，OpenAI将获得Cerebras的少数股权认股权证，且持股比例可能随支出规模增加而提升，最高可达10%。
• 此次合作是在今年1月双方超100亿美元算力采购协议基础上的进一步扩大，Cerebras最快将于周五披露部分协议细节。
• 与OpenAI的深度绑定成为Cerebras冲击上市的核心支撑，该公司目标于今年二季度登陆资本市场，并计划下月完成30亿美元融资，届时估值约达350亿美元。
• Cerebras成立于2015年，以晶圆级引擎芯片技术著称，OpenAI CEO山姆·奥特曼是其早期投资者之一。

相关链接：

• OpenAI斥资逾 200 亿美元押注Cerebras，或获高达10%股权

OpenAI Plans to Spend Up to $30 Billion on Cerebras Computing Power and Take Up to a 10% Stake to Support Its IPO Push

Brief:

• OpenAI plans to pay chip startup Cerebras more than $20 billion over the next three years to purchase server computing power, and will provide an additional roughly dJyH6kDkfC5RlM16.answerText billion to support its data center development, bringing total spending to as much as $30 billion.
• In return for the partnership, OpenAI will receive minority equity warrants in Cerebras, with its stake potentially increasing as spending rises, up to a maximum of 10%.
• The partnership marks a further expansion of the more than dJyH6kDkfC5RlM16.answerText0 billion computing power procurement agreement the two companies reached in January. Cerebras could disclose some details of the agreement as soon as Friday.
• The deep partnership with OpenAI has become a core pillar of Cerebras’ push to go public. The company is aiming to enter the capital markets in the second quarter of this year, and plans to complete a $3 billion funding round next month, which would value it at around $35 billion.
• Founded in 2015, Cerebras is known for its wafer-scale engine chip technology. OpenAI CEO Sam Altman was one of its early investors. 🚀

Related Link:

• OpenAI Bets More Than $20 Billion on Cerebras, May Receive Up to 10% Equity Stake

xAI或于下周发布Grok Build，马斯克正式入局AI智能编程

简报：

• 马斯克旗下xAI公司有望下周推出Grok Build与Grok CLI两款产品，正式进军智能编程赛道，与Claude Code、OpenAI Codex及Google Jules等工具展开竞争。
• xAI目前已向Grok Heavy订阅用户推送Grok 4.3 Early Access测试版，预计Grok Build将直接依托该模型支撑编程功能，而非单独开发专用模型。
• Grok Build采用双轨运行机制，用户既可以通过CLI智能体在本地执行任务，也能通过Web界面进行远程操作。
• 该工具预计将在功能上引入Parallel模式与Arena模式，其中Arena模式支持多个智能体并行处理同一任务，以便用户从中择优采纳。

相关链接：

• 消息称马斯克 xAI 进军 AI 智能编程领域，有望下周发布 Grok Build

xAI May Release Grok Build Next Week, as Musk Officially Enters AI Coding

Brief:

• Elon Musk’s xAI is expected to launch two products next week, Grok Build and Grok CLI, formally entering the AI coding market and competing with tools such as Claude Code, OpenAI Codex, and Google Jules.
• xAI has already pushed the Grok 4.3 Early Access beta to Grok Heavy subscribers. Grok Build is expected to rely directly on this model to support coding features, rather than using a separately developed dedicated model.
• Grok Build uses a dual-track operating mechanism. Users can run tasks locally through a CLI agent or operate remotely through a web interface.
• The tool is expected to introduce Parallel Mode and Arena Mode. Arena Mode supports multiple agents working on the same task in parallel, allowing users to select the best result. 🚀

Related Link:

• Sources Say Musk’s xAI Is Entering the AI Coding Field, with Grok Build Expected to Launch Next Week

硅谷AI人才战持续升温：Meta重金挖角初创公司核心团队

简报：

• Meta持续加大对AI初创公司的挖角力度，Thinking Machines资深软件工程师Joshua Gross上月跳槽至Meta超级智能实验室担任工程团队负责人。
• Thinking Machines已流失多位核心人才，其中联合创始人Andrew Tulloch等五位创始成员加入Meta，前首席技术官和网络安全专家则被OpenAI挖走。
• 硅谷人才流动呈现双向特征，估值120亿美元的Thinking Machines此前也曾从Meta挖来PyTorch创建者Soumith Chintala担任首席技术官。
• 科技巨头为争夺顶尖AI人才开出天价薪酬，扎克伯格曾以高达1亿美元尝试挖角OpenAI核心人员，谷歌和苹果等公司也频频抛出巨额合同与股票奖励。

相关链接：

• Meta持续挖角AI初创公司，Thinking Machines再失核心成员

Silicon Valley AI Talent War Keeps Heating Up: Meta Spends Big to Poach Core Teams from Startups

Briefing:

• Meta is continuing to intensify its efforts to recruit talent from AI startups. Joshua Gross, a senior software engineer at Thinking Machines, joined Meta’s Superintelligence Labs last month as head of its engineering team.
• Thinking Machines has lost several core employees. Five founding members, including co-founder Andrew Tulloch, have joined Meta, while its former chief technology officer and a cybersecurity expert were recruited by OpenAI.
• Talent flows in Silicon Valley are moving in both directions. Thinking Machines, valued at dJyH6kDkfC5RlM16.answerText2 billion, had previously recruited PyTorch creator Soumith Chintala from Meta to serve as its chief technology officer.
• Tech giants are offering enormous compensation packages to compete for top AI talent. Mark Zuckerberg reportedly attempted to recruit key OpenAI employees with offers of up to dJyH6kDkfC5RlM16.answerText00 million, while companies such as Google and Apple have also frequently offered major contracts and stock awards. 💼

Related Link:

• Meta Continues Poaching AI Startup Talent as Thinking Machines Loses Another Core Member

Anthropic推出最新Claude Opus 4.7模型，微软多款AI开发工具首日全面接入

简报：

• Anthropic正式发布Claude Opus 4.7模型，该版本在高级软件工程、复杂任务处理和指令执行方面显著提升，并具备自我验证能力。
• 新模型视觉处理能力增强，支持识别高达2576像素长边的图像；同时内置网络安全防护措施，可自动阻止高风险请求。
• Opus 4.7改进了内存利用率，能有效记住长会话笔记，其定价与Opus 4.6保持一致（输入$5/百万token，输出$25/百万token）。
• 微软打破与OpenAI的排他性合作惯例，在Opus 4.7发布首日即在GitHub Copilot及Visual Studio Code等9大开发环境中集成支持。
• 未来几周内，Opus 4.7将取代旧版本成为GitHub Copilot Pro+用户的默认选项；Microsoft 365部分计划和环境中的用户现也已可调用该模型。
• Opus 4.7目前已登陆Claude产品和API、Amazon Bedrock、Google Cloud Vertex AI以及Microsoft Foundry等多个平台。

相关链接：

• Anthropic 发布全新 Claude Opus 4.7，功能显著提升
• 微软首日集成 Claude Opus 4.7，GitHub Copilot 等 9 大 AI 工具全面接入

Anthropic Launches Latest Claude Opus 4.7 Model, with Multiple Microsoft AI Development Tools Fully Supporting It on Day One

Briefing:

• Anthropic has officially released the Claude Opus 4.7 model. The new version brings significant improvements in advanced software engineering, complex task handling, and instruction following, and includes self-verification capabilities.
• The new model offers enhanced visual processing, supporting image recognition with a long edge of up to 2,576 pixels. It also includes built-in cybersecurity safeguards that can automatically block high-risk requests. 🔐
• Opus 4.7 improves memory utilization and can effectively retain long-session notes. Its pricing remains the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens.
• Microsoft has broken from its usual exclusive partnership pattern with OpenAI, integrating support for Opus 4.7 into nine major development environments, including GitHub Copilot and Visual Studio Code, on the first day of release.
• In the coming weeks, Opus 4.7 will replace older versions as the default option for GitHub Copilot Pro+ users. Users in certain Microsoft 365 plans and environments can also now access the model.
• Opus 4.7 is now available across multiple platforms, including Claude products and APIs, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. 🚀

Related Links:

• Anthropic Releases New Claude Opus 4.7 with Major Capability Upgrades
• Microsoft Integrates Claude Opus 4.7 on Day One, Bringing Full Support to GitHub Copilot and 9 Major AI Tools

微软打破独家合作：Claude Opus 4.7 首日全面接入 GitHub Copilot 等 9 大开发环境

简报：

• Anthropic 正式发布最新 AI 模型 Claude Opus 4.7，其在高级软件工程领域的复杂任务处理和指令执行能力获得显著提升。
• 微软打破与 OpenAI 的排他性合作惯例，在 Opus 4.7 发布首日即将其集成至 GitHub Copilot 等 9 大开发环境中。
• GitHub 团队宣布，未来几周内 Opus 4.7 将取代之前的 4.5 和 4.6 版本，成为 Copilot Pro+ 用户默认的模型选项。
• Microsoft 365 Copilot 用户现已能够在 Copilot Cowork 前沿计划、Copilot Studio 早期发布环境以及 Excel 中直接调用 Claude Opus 4.7 模型。
• Claude Opus 4.7 定价与上一代保持一致，同时在视觉处理（支持最高 2,576 像素长边）、长时间会话记忆以及内存利用方面实现了重大突破，并内置了高风险请求拦截等网络安全防护措施。

相关链接：

• Anthropic 发布全新 Claude Opus 4.7，功能显著提升
• 微软首日集成 Claude Opus 4.7，GitHub Copilot 等 9 大 AI 工具全面接入

请问还有其他新闻内容需要我为您生成简报吗？

Microsoft Breaks Exclusive Partnership: Claude Opus 4.7 Fully Integrated on Day One Across 9 Major Development Environments Including GitHub Copilot 🚀

Brief:

• Anthropic has officially released its latest AI model, Claude Opus 4.7, with significant improvements in handling complex tasks and following instructions in advanced software engineering scenarios.
• Microsoft has broken with its usual exclusive partnership practice with OpenAI by integrating Opus 4.7 on the very first day of launch into nine major development environments, including GitHub Copilot.
• The GitHub team announced that over the next few weeks, Opus 4.7 will replace the previous 4.5 and 4.6 versions to become the default model option for Copilot Pro+ users.
• Microsoft 365 Copilot users can now directly access the Claude Opus 4.7 model through the Copilot Cowork Frontier program, the Copilot Studio early release environment, and within Excel.
• Claude Opus 4.7 is priced the same as the previous generation, while delivering major breakthroughs in visual processing (supporting up to a 2,576-pixel long edge), long-session memory, and memory utilization, and it also includes built-in cybersecurity protections such as high-risk request blocking 🔐.

Related links:

• Anthropic Releases the New Claude Opus 4.7 With Significantly Enhanced Capabilities
• Microsoft Integrates Claude Opus 4.7 on Launch Day, With Full Access Across 9 AI Tools Including GitHub Copilot 💡

单词学习：![[2026041923129565.mp3]]

新闻播报：![[2026041917149095.mp3]]

ChatGPT全球周活逼近10亿大关，女性用户比例逆转首超半数

简报：

• 根据OpenAI最新数据，ChatGPT全球周活跃用户人数即将突破10亿。
• 平台用户性别结构发生显著变化，女性用户占比从2022年发布初期的20%跃升，首次超过男性达到50%以上，目前全球约有近5亿女性定期使用该工具。
• OpenAI可用算力规模正快速扩张，从2023年的0.2吉瓦预计增至2025年的约1.9吉瓦，年增长率达3倍。
• OpenAI已与英伟达等公司签署超30吉瓦的协议，并计划到2030年掌握30吉瓦的算力，以支撑未来更优质的服务与体验。

相关链接：

• ChatGPT 用户突破 10 亿，女性用户占比首次超过 50%

ChatGPT’s Global Weekly Active Users Near 1 Billion, Female User Share Surpasses Half for the First Time 🎉

Briefing:

• According to the latest data from OpenAI, ChatGPT’s global weekly active user count is on the verge of surpassing 1 billion.
• The platform’s gender distribution has shifted significantly. The share of female users has risen sharply from 20% at launch in 2022, surpassing that of male users for the first time to exceed 50%. At present, nearly 500 million women worldwide use the tool regularly 👩‍💻
• OpenAI’s available computing power is expanding rapidly, projected to grow from 0.2 gigawatts in 2023 to approximately 1.9 gigawatts in 2025, representing a threefold annual growth rate.
• OpenAI has already signed agreements totaling more than 30 gigawatts with companies including NVIDIA, and plans to control 30 gigawatts of computing power by 2030 to support higher-quality services and user experiences 🚀

Related link:

• ChatGPT Users Surpass 1 Billion, Female User Share Exceeds 50% for the First Time

北京亦庄半程马拉松今日开跑，宇树H1机器人1.9公里成绩打破人类纪录

简报：

• 2026北京亦庄半程马拉松于今日（4月19日）正式鸣枪开跑，全长21.0975公里，赛道上出现了多款人形及功能性机器人。
• 宇树科技派出H1（23年微调版）机型参加北京人形机器人马拉松排位赛，自主跑完1.9公里多弯道赛程仅用时4分13秒。
• 根据官方消息，宇树H1的该项成绩按比例计算，已成功打破人类1500米世界纪录。
• 此前，宇树H1曾刷新人形机器人百米测试奔跑速度世界纪录，实测峰值速度达到10米/秒，逼近人类历史最快短跑运动员博尔特的9.58秒（10.44米/秒）纪录。
• 除宇树H1外，本次马拉松赛事中还首次亮相了参与指挥的北京交警机器人、高擎动力带奶瓶参赛的“小短腿”Mini Pi plus机器人，以及网友自制的高造型辨识度“豆脚”机器人。

相关链接：

• 北京首个交警机器人亮相街头 参与指挥亦庄半程马拉松比赛
• 破人类 1500 米世界纪录：宇树 H1 自主跑完北京人形机器人马拉松 1.9 公里多弯道赛程，用时 4 分 13 秒
• 带奶瓶参赛：北京人形机器人半马惊现“小短腿机器人”
• 脖子以下全是腿：网友自制“豆脚”机器人现身北京人形机器人马拉松

The Beijing Yizhuang Half Marathon Kicks Off Today, as Unitree H1 Robot’s 1.9 km Time Breaks Human Record 🤖🏃‍♂️🏁

Briefing:

• The 2026 Beijing Yizhuang Half Marathon officially started today (April 19), covering a total distance of 21.0975 kilometers, with multiple humanoid and functional robots appearing on the course.
• Unitree Robotics deployed its H1 model (2023 fine-tuned version) to compete in the Beijing Humanoid Robot Marathon ranking race, autonomously completing the 1.9-kilometer winding course in just 4 minutes and 13 seconds.
• According to official information, when calculated proportionally, Unitree H1’s result has successfully broken the human 1500-meter world record.
• Previously, Unitree H1 had already set a new world record for sprint speed in a 100-meter humanoid robot test, reaching a measured peak speed of 10 meters per second, close to the record of 9.58 seconds (10.44 meters per second) set by Usain Bolt, the fastest sprinter in human history.
• In addition to Unitree H1, this marathon also featured the debut of a Beijing traffic police robot participating in race command, the “short-legged” Mini Pi Plus robot competing while carrying a baby bottle, and the highly distinctive “Doujiao” robot built by a netizen.

Related Links:

• Beijing’s First Traffic Police Robot Appears on the Streets, Participates in Directing the Yizhuang Half Marathon
• Breaking the Human 1500m World Record: Unitree H1 Autonomously Completes the 1.9 km Winding Course of the Beijing Humanoid Robot Marathon in 4 Minutes 13 Seconds
• Competing with a Baby Bottle: “Short-Legged Robot” Spotted at the Beijing Humanoid Robot Half Marathon
• Nothing but Legs Below the Neck: Netizen-Built “Doujiao” Robot Appears at the Beijing Humanoid Robot Marathon

北京亦庄半马开跑：交警机器人首秀指挥，多款人形机器人同台竞技

简报：

• 4月19日，北京首个交警机器人正式亮相2026北京亦庄半程马拉松赛道进行指挥，目前已实现交通手势指挥、出行引导等功能，后续将在路口试点执勤。
• 宇树科技H1（23年改版）机器人参加北京人形机器人马拉松排位赛，用时4分13秒自主跑完1.9公里多弯道赛程，按比例计算打破人类1500米世界纪录。
• 一款手持奶瓶的“小短腿机器人”现身赛场引发热议，该机器人为高擎动力出品的Mini Pi plus，采用模块化设计，支持高精度动作控制。
• 网友自制的“豆脚”机器人也亮相马拉松，创意来自抖音博主，使用高擎动力小派本体配合虚拟人物头套，造型独特。

相关链接：

• 北京首个交警机器人亮相街头 参与指挥亦庄半程马拉松比赛
• 破人类 1500 米世界纪录：宇树 H1 自主跑完北京人形机器人马拉松
• 带奶瓶参赛：北京人形机器人半马惊现“小短腿机器人”
• 脖子以下全是腿：网友自制“豆脚”机器人现身北京人形机器人马拉松

Beijing E-Town Half Marathon Kicks Off: Traffic Police Robot Makes Command Debut, Multiple Humanoid Robots Compete on the Same Track

Brief:

• On April 19, Beijing’s first traffic police robot made its official debut on the course of the 2026 Beijing E-Town Half Marathon, directing traffic. It can already perform functions such as traffic gesture commands and travel guidance, and will later be piloted for duty at intersections.
• Unitree Robotics’ H1 robot, the 2023 revised version, took part in the Beijing Humanoid Robot Marathon qualifying race. It autonomously completed a 1.9-kilometer course with multiple turns in 4 minutes and 13 seconds. Converted proportionally, the result would break the human 1,500-meter world record.
• A “short-legged robot” holding a baby bottle appeared at the race and sparked online discussion. The robot is Mini Pi Plus, produced by High Torque Robotics. It uses a modular design and supports high-precision motion control. 🍼
• A netizen-made “Bean Foot” robot also appeared at the marathon. Inspired by a Douyin creator, it uses High Torque Robotics’ Xiao Pai body with a virtual character headpiece, giving it a distinctive look. 🤖

Related links:

• Beijing’s first traffic police robot appears on the street, helps direct the E-Town Half Marathon
• Breaking the human 1,500-meter world record: Unitree H1 autonomously completes the Beijing Humanoid Robot Marathon
• Competing with a baby bottle: “short-legged robot” spotted at Beijing Humanoid Robot Half Marathon
• All legs below the neck: Netizen-made “Bean Foot” robot appears at Beijing Humanoid Robot Marathon

Cerebras获OpenAI超200亿美元芯片大单，计划以超350亿美元估值IPO

简报：

• AI芯片公司Cerebras与OpenAI签署为期三年、金额超过200亿美元的重磅芯片交易协议，规模是年初750MW部署协议的两倍。
• OpenAI承诺提供约10亿美元资金支持Cerebras开发数据中心系统，并获得最高可达10%的少数股权认股凭证。
• Cerebras计划以350亿美元以上的估值进行首次公开募股（IPO），目标筹集约30亿美元资金，具体发行条款尚在商讨中。
• Cerebras主打高性能的晶圆级AI芯片，此次合作将为其进一步发展提供资金支持并在AI芯片市场占据有利位置。

相关链接：

• Cerebras 与 OpenAI 签署 200 亿美元芯片协议计划 IPO

Cerebras Secures Over $20 Billion Chip Order from OpenAI, Plans IPO at Valuation Above $35 Billion

Briefing:

• AI chip company Cerebras has signed a major three-year chip transaction agreement with OpenAI worth more than $20 billion, twice the scale of the 750MW deployment agreement announced earlier this year. 🚀
• OpenAI has committed to providing about dJyH6kDkfC5RlM16.answerText billion in funding to support Cerebras in developing data center systems and will receive warrants for a minority equity stake of up to 10%.
• Cerebras plans to launch an initial public offering (IPO) at a valuation of more than $35 billion, aiming to raise about $3 billion. Specific offering terms are still under discussion.
• Cerebras focuses on high-performance wafer-scale AI chips. The partnership will provide financial support for its further development and help it gain a stronger position in the AI chip market. 💻

Related Link:

• Cerebras and OpenAI Sign $20 Billion Chip Agreement, Plan IPO

谷歌开源40亿参数医疗AI模型MedGemma 1.5，全面支持3D影像分析

简报：

• 谷歌近日开源医疗AI模型MedGemma 1.5，突破传统2D图像局限，原生支持CT和MRI三维扫描数据分析及病理学全视野数字切片分析。
• 在保持40亿参数量不变的前提下，新版本较上一代在3D MRI病情分类准确率提升11%，胸部X光解剖定位交并比增加35%，电子病历问答准确率提升22%。
• 团队为模型注入了大量医学图文配对数据并引入专家模型进行知识蒸馏，但该模型目前仍定位为供开发者微调的基础资源，非开箱即用的临床决策工具。

相关链接：

• 谷歌开源医疗AI新星MedGemma 1.5：从看平面图到读懂3D影像
• 论文地址

Google Open-Sources 4-Billion-Parameter Medical AI Model MedGemma 1.5, with Full Support for 3D Imaging Analysis 🩺

Brief:

• Google has recently open-sourced the medical AI model MedGemma 1.5, breaking beyond the limitations of traditional 2D images with native support for analyzing 3D CT and MRI scan data, as well as whole-slide digital pathology images.
• While maintaining the same 4 billion parameter size, the new version improves upon the previous generation with an 11% increase in accuracy for 3D MRI disease classification, a 35% boost in intersection-over-union for chest X-ray anatomical localization, and a 22% gain in electronic health record question-answering accuracy.
• The team infused the model with a large volume of paired medical image-text data and introduced expert models for knowledge distillation. However, the model is still positioned as a foundational resource for developers to fine-tune, rather than an out-of-the-box clinical decision-making tool. 🔬📊

Related links:

• Google Open-Sources Rising Medical AI Model MedGemma 1.5: From Reading Flat Images to Understanding 3D Scans
• Paper

阿里推出轻量级MoE模型Qwen3.6-35B-A3B：仅30亿激活参数比肩大体量模型

简报：

• 阿里巴巴千问大模型团队于2026年4月16日正式开源稀疏混合专家（MoE）模型 Qwen3.6-35B-A3B，总参数量350亿，运行时激活参数仅为30亿。
• 该模型以极低的计算成本在多项核心编程基准测试中超越了270亿参数的稠密模型 Qwen3.5-27B，逻辑推理与智能体协同能力可比肩 Gemma4-31B 等大体量模型。
• 作为全模态开源模型，其在空间智能与视觉感知方面表现优异，RefCOCO评分达92.0，部分多模态指标已比肩 Claude Sonnet4.5。
• 模型已集成至 Qwen Studio，并通过阿里云百炼平台以 qwen3.6-flash 之名对外提供API服务，支持 preserve_thinking 功能并无缝适配 OpenClaw 等主流AI编程助手。

相关链接：

• 阿里开源 Qwen3.6-35B-A3B：30亿激活参数实现编程能力跨越式升级

Alibaba Launches Lightweight MoE Model Qwen3.6-35B-A3B: Just 3B Activated Parameters Rival Larger Models

Brief:

• Alibaba’s Qwen large model team officially open-sourced the sparse Mixture-of-Experts (MoE) model Qwen3.6-35B-A3B on April 16, 2026. The model has 35 billion total parameters, with only 3 billion activated during runtime.
• With extremely low computational cost, the model outperforms the 27B-parameter dense model Qwen3.5-27B across several core coding benchmarks. Its logical reasoning and agent collaboration capabilities are comparable to larger models such as Gemma4-31B.
• As an open-source omni-modal model, it performs strongly in spatial intelligence and visual perception, achieving a RefCOCO score of 92.0. Some of its multimodal metrics are already on par with Claude Sonnet 4.5. 🚀
• The model has been integrated into Qwen Studio and is available via Alibaba Cloud’s Bailian platform under the name qwen3.6-flash, offering API services with support for the preserve_thinking feature and seamless compatibility with mainstream AI coding assistants such as OpenClaw.

Related Link:

• Alibaba Open-Sources Qwen3.6-35B-A3B: 3B Activated Parameters Deliver a Major Leap in Coding Capabilities

单词学习：![[2026041731881852.mp3]]

新闻播报：![[2026041761147964.mp3]]

谷歌开源 MedGemma 1.5 医疗 AI 模型，突破 2D 局限全面支持 3D 影像解析

简报：

• 谷歌宣布开源医疗 AI 模型 MedGemma 1.5，该版本的核心突破在于打破传统 2D 图像局限，原生支持 CT 和 MRI 等三维医疗扫描数据的分析。
• 模型功能实现全面升级，除支持处理数十个切片的立体影像外，还具备病理学全视野数字切片分析、胸部 X 光病变精准标定、多时间点病情对比，以及大幅提升的 PDF 电子病历解析能力。
• 性能方面，在保持 40 亿参数量不变的前提下，相较于上一代产品，其 3D MRI 病情分类准确率提升 11%，胸部 X 光解剖定位交并比增加 35%，电子病历问答准确率提升 22%。
• 技术路线上，研发团队注入了涵盖放射学、皮肤科等领域的医学图文数据，设计了针对 3D CT 扫描的序列切分预处理方法，并引入领域专家模型进行知识蒸馏。
• 谷歌明确该模型并非开箱即用的临床决策工具，而是定位为供开发者微调的基础资源；同时在演进为“医疗通才”的过程中，模型在部分小众旧视觉问答基准上出现极轻微退步。

相关链接：

• 谷歌开源医疗 AI 模型 MedGemma 1.5，支持 3D 影像读取
• MedGemma 1.5 论文地址

Google Open-Sources MedGemma 1.5 Medical AI Model, Breaking Through 2D Limitations with Full Support for 3D Imaging Analysis 🩺📊

Brief:

• Google has announced the open-source release of its medical AI model MedGemma 1.5. The core breakthrough of this version lies in overcoming the limitations of traditional 2D imaging, with native support for analyzing 3D medical scan data such as CT and MRI.
• The model’s capabilities have been comprehensively upgraded. In addition to processing volumetric images composed of dozens of slices, it now supports whole-slide digital pathology analysis, precise lesion localization in chest X-rays, multi-timepoint disease comparison, and significantly improved PDF electronic medical record parsing.
• In terms of performance, while maintaining the same 4 billion parameter scale, the new version delivers notable gains over the previous generation: 3D MRI disease classification accuracy has improved by 11%, chest X-ray anatomical localization IoU has increased by 35%, and electronic medical record question-answering accuracy has risen by 22%.
• On the technical side, the research team incorporated medical image-text data spanning radiology, dermatology, and other specialties, designed a sequence-splitting preprocessing method tailored for 3D CT scans, and introduced domain expert models for knowledge distillation.
• Google has made it clear that the model is not intended as an out-of-the-box clinical decision-making tool, but rather as a foundational resource for developers to fine-tune. At the same time, as it evolves toward becoming a “medical generalist,” the model shows very slight regressions on some niche legacy visual question-answering benchmarks.

Related links:

• Google Open-Sources Medical AI Model MedGemma 1.5 with Support for 3D Imaging
• MedGemma 1.5 Paper

OpenAI Codex 大幅升级：新增自主操控 Mac 桌面与长线任务执行能力

简报：

• OpenAI 对 Codex AI 编程助手进行重磅更新，核心方向涵盖 Mac 桌面应用操控、多智能体并行运行以及长期任务执行。
• Codex 新增直接操作 Mac 桌面应用的能力，拥有独立光标，可全程在无人工介入下自主查看屏幕、点击和输入文字，且多个智能体可并行运行互不打扰。
• 在任务连续性上，Codex 增加记忆功能以保存用户偏好与工作流，支持无缝恢复暂停的任务，并能自主调度未来数天至数周的工作计划。
• 浏览器体验得到强化，应用内浏览器支持页面批注，OpenAI 还透露未来将推出完整的浏览器操控能力。
• 本次更新还集成了 gpt-image-1.5 模型用于生成概念图和原型，并新增超 90 款插件，大幅扩充了应用集成与 MCP 服务器接入生态。

相关链接：

• OpenAI 更新 Codex：支持 Mac 桌面操控

OpenAI Codex Gets Major Upgrade: Adds Autonomous Mac Desktop Control and Long-Running Task Execution

Brief:

• OpenAI has released a major update to its Codex AI coding assistant, with core improvements covering Mac desktop app control, parallel multi-agent operation, and long-running task execution.
• Codex now gains the ability to directly operate Mac desktop applications. It has its own independent cursor and can autonomously view the screen, click, and enter text without human intervention, while multiple agents can run in parallel without interfering with one another.
• For task continuity, Codex has added memory features to save user preferences and workflows, support seamless resumption of paused tasks, and autonomously schedule work plans over the coming days or weeks.
• The browser experience has been enhanced. The in-app browser now supports page annotations, and OpenAI also said it plans to introduce full browser control capabilities in the future.
• This update also integrates the gpt-image-1.5 model for generating concept images and prototypes, and adds more than 90 new plugins, greatly expanding the ecosystem for app integrations and MCP server access. 🚀

Related link:

• OpenAI updates Codex with support for Mac desktop control

OpenAI Codex 迎重磅更新：新增Mac桌面操控、多智能体并行与长期任务记忆

简报：

• OpenAI于今日（4月17日）对Codex AI编程助手进行核心升级，主打Mac桌面应用操控、多智能体并行运行以及更长时间维度的任务执行能力。
• Codex现已获得直接操作Mac桌面应用的能力，拥有独立光标，可自主查看、点击和输入；支持多个智能体同时并行运行，且互不干扰用户的正常工作流。
• 新增任务连续性与记忆功能，可保存用户偏好和技术栈信息；任务暂停后能无缝恢复，并可自主调度未来数周的工作计划，主动提出推进建议。
• 浏览器体验获得强化，应用内浏览器支持用户在页面直接添加批注；OpenAI表示未来将上线完整的浏览器自主操控与截图检验功能。
• 其他更新包括：集成gpt-image-1.5模型生成产品概念图和界面原型，侧边栏支持预览PDF和电子表格等文件，同时插件生态新增逾90款插件。

相关链接：

• OpenAI更新Codex：支持Mac桌面操控、多智能体并行与长期任务执行

OpenAI Codex Gets Major Update: Adds Mac Desktop Control, Multi-Agent Parallel Execution, and Long-Term Task Memory

Brief:

• OpenAI today, April 17, rolled out a core upgrade to Codex, its AI programming assistant, focusing on Mac desktop app control, multi-agent parallel execution, and longer-horizon task execution capabilities.
• Codex can now directly operate Mac desktop apps. It has its own independent cursor and can autonomously view, click, and type. It also supports multiple agents running in parallel at the same time, without interfering with the user’s normal workflow.
• New task continuity and memory features can save user preferences and technical stack information. After a task is paused, Codex can resume seamlessly, independently schedule work plans for the coming weeks, and proactively suggest next steps.
• The browser experience has been enhanced. The in-app browser now supports adding annotations directly on webpages. OpenAI says full autonomous browser control and screenshot verification features will be launched in the future. 🌐
• Other updates include integration with the gpt-image-1.5 model to generate product concept images and UI prototypes, sidebar support for previewing files such as PDFs and spreadsheets, and an expanded plugin ecosystem with more than 90 new plugins. 🚀

Related link:

• OpenAI Updates Codex With Support for Mac Desktop Control, Multi-Agent Parallel Execution, and Long-Term Task Execution

Anthropic发布Claude Opus 4.7：工程与视觉能力升级且定价不变

简报：

• Anthropic正式推出最新人工智能模型Claude Opus 4.7，相较于4.6版本，在高级软件工程、复杂任务处理和指令执行方面均有显著提升，并具备自我验证能力。
• 视觉处理表现增强，支持识别高达2576像素长边的图像，提升了在创建高质量幻灯片和文档等专业任务中的创造力与质量。
• 安全与记忆功能升级，内置网络安全防护以自动拦截高风险请求，同时改进内存利用率，能够有效记住长时间会话中的重要笔记。
• 价格与前代保持一致（输入每百万token 5美元，输出每百万token 25美元），但因指令执行力增强，用户需适度调整提示信息以防意外结果。
• 该模型现已全面上线涵盖Claude产品及API、Amazon Bedrock、Google Cloud Vertex AI以及Microsoft Foundry等多个平台，并推出了细致的难度控制和任务预算功能。

相关链接：

• Anthropic 发布全新 Claude Opus 4.7 模型

Anthropic Releases Claude Opus 4.7: Engineering and Vision Capabilities Upgraded, Pricing Unchanged

Brief:

• Anthropic has officially launched its latest AI model, Claude Opus 4.7. Compared with version 4.6, it delivers notable improvements in advanced software engineering, complex task handling, and instruction following, while also adding self-verification capabilities.
• Vision processing has been enhanced, with support for recognizing images with a long edge of up to 2,576 pixels. This improves creativity and quality in professional tasks such as creating high-quality slides and documents. 🖼️
• Safety and memory features have been upgraded. Built-in cybersecurity protections can automatically block high-risk requests, while improved memory usage helps the model effectively retain important notes across long-running conversations.
• Pricing remains unchanged from the previous generation, at $5 per million input tokens and $25 per million output tokens. However, due to stronger instruction-following capabilities, users should adjust prompts appropriately to avoid unintended results.
• The model is now fully available across multiple platforms, including Claude products and APIs, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. It also introduces fine-grained difficulty controls and task budget features. 🚀

Related link:

• Anthropic Releases New Claude Opus 4.7 Model

NVIDIA发布首个开源量子AI模型“Ising”，大幅提升校准与纠错效率

简报：

• NVIDIA宣布推出全球首个开源量子AI模型“NVIDIA Ising”，专门用于解决量子处理器校准和量子纠错两大关键难题。
• 模型包含拥有350亿参数的视觉语言模型Ising Calibration，可将测量数据的处理时间从数天大幅缩短至仅数小时。
• 模型还包含Ising Decoding，在实时解码过程中提供了比行业基准pyMatching高出2.5倍的速度和3倍的准确率，且所需训练数据仅为十分之一。
• “NVIDIA Ising”目前已被IonQ、哈佛大学和费米国家加速器实验室等多家机构应用，为构建混合量子-经典系统提供支持。
• 该模型的命名源于统计力学中的“Ising 模型”，其发布进一步丰富了NVIDIA涵盖智能体、物理AI、辅助驾驶及生物医学等领域的开放模型体系。

相关链接：

• NVIDIA 推出全球首个开源量子 AI 模型

NVIDIA Releases First Open-Source Quantum AI Model “Ising,” Significantly Improving Calibration and Error-Correction Efficiency

Briefing:

• NVIDIA has announced the launch of “NVIDIA Ising,” the world’s first open-source quantum AI model, designed to address two major challenges in quantum processors: calibration and quantum error correction.
• The model includes Ising Calibration, a vision-language model with 35 billion parameters that can reduce the processing time for measurement data from several days to just a few hours.
• The model also includes Ising Decoding, which delivers 2.5 times the speed and 3 times the accuracy of the industry benchmark pyMatching during real-time decoding, while requiring only one-tenth of the training data.
• “NVIDIA Ising” is already being used by several institutions, including IonQ, Harvard University, and Fermilab, supporting the development of hybrid quantum-classical systems.
• The model is named after the “Ising model” in statistical mechanics. Its release further expands NVIDIA’s open model ecosystem across areas such as agents, physical AI, assisted driving, and biomedicine.

Related link:

• NVIDIA Launches the World’s First Open-Source Quantum AI Model

我国实现两大突破，牵头立项全球首个具身智能国际标准

简报：

• 4月17日报道，我国在国际标准化组织成功立项具身智能领域全球首项国际标准《人形机器人数据集》。
• 我国在此次立项中实现了两个历史性的“零的突破”：一是人形机器人国际标准制定，二是机器人国际标准工作组建设，并推动成立首个由我国专家担任召集人的工作组。
• 此次立项的国际标准参考了我国在研的人形机器人数据集系列国家标准内容，全球多国专家将携手参与后续标准研制。
• 数据显示，截至2025年底我国已成为全球第一大人形机器人市场，出货量达1.44万台，占据全球总量的84.7%。
• 作为顶层设计铺垫，今年2月我国已正式发布《人形机器人与具身智能标准体系（2026版）》，涵盖基础共性、类脑与智算、肢体与部组件、整机与系统、应用、安全伦理6个部分。

相关链接：

• 具身智能领域首个国际标准由我国牵头成功立项
• 人形机器人标准化技术委员会委员名单

China Achieves Two Major Breakthroughs, Leads the Launch of the World’s First International Standard for Embodied Intelligence

Briefing:

• Reported on April 17, China has successfully initiated the world’s first international standard in the field of embodied intelligence under the International Organization for Standardization, titled Humanoid Robot Datasets.
• China achieved two historic firsts in this project: the first international standard for humanoid robots, and the first international standards working group for robotics, with China also推动 the establishment of the first working group convened by a Chinese expert.
• The newly initiated international standard draws on China’s national standards under development for humanoid robot dataset series, with experts from multiple countries set to participate in the subsequent drafting work.
• Data shows that by the end of 2025, China had become the world’s largest humanoid robot market, with shipments reaching 14,400 units and accounting for 84.7% of the global total.
• As part of the top-level design groundwork, China officially released the Standard System for Humanoid Robots and Embodied Intelligence (2026 Edition) in February this year, covering six areas: basic common technologies, brain-inspired intelligence and intelligent computing, limbs and components, complete machines and systems, applications, and safety ethics. 🤖

Related Links:

• China Successfully Leads the Launch of the First International Standard in the Field of Embodied Intelligence
• List of Members of the Humanoid Robot Standardization Technical Committee

单词学习：![[2026041583863987.mp3]]

新闻播报：![[2026041526700730.mp3]]

具身智能与机器人产业迎规模化落地：智元G2零失误作业，中美AI差距缩小至2.7%

简报：

• 智元机器人精灵G2在龙旗科技平板产线完成8小时、2283次零失误作业，实现全球首个具身智能3C产线规模化落地，计划2026年Q3部署达100台。
• 菜鸟在国际物流展首发自研攀爬机器人ZeeBot，并在广东投入使用，实测仓库存取效率提升100%。
• 特斯拉中国区高管透露，上海超级工厂有望凭借高效生产能力，成为量产“擎天柱”人形机器人的关键基地。
• 斯坦福发布《2026年人工智能指数报告》指出中美顶尖AI模型差距仅剩2.7%，中国在AI专利产出及工业机器人安装量上全球登顶。
• 阿里高德具身业务部即将发布首款四足机器人，标志着阿里巴巴AI版图正式由数字空间向真实世界物理交互延伸。

相关连接：

• 全球首个具身智能工业产线规模落地：智元精灵 G2 连续 8 小时作业零失误
• 菜鸟发布攀爬机器人ZeeBot，实测智能化存取效率提升一倍
• 特斯拉高管称上海工厂将为擎天柱机器人量产提供“金钥匙”
• 斯坦福2026AI 报告:中美技术差距仅剩2.7%，中国在专利与机器人领域登顶
• 阿里巴巴首款具身机器人曝光:高德具身业务部将发布四足机器人

Embodied Intelligence and the Robotics Industry Move Toward Scaled Deployment: Zhiyuan G2 Delivers Zero-Error Operations, While the China-U.S. AI Gap Narrows to 2.7% 🤖📈

Briefing:

• Zhiyuan Robot’s Genie G2 completed 8 hours and 2,283 zero-error operations on Longcheer Technology’s tablet production line, achieving the world’s first large-scale deployment of embodied intelligence in a 3C manufacturing line. Deployment is planned to reach 100 units by Q3 2026.
• Cainiao unveiled its self-developed climbing robot ZeeBot at an international logistics expo and has already put it into use in Guangdong, where real-world testing showed a 100% increase in warehouse storage and retrieval efficiency.
• A Tesla China executive revealed that the Shanghai Gigafactory could become a key base for the mass production of the “Optimus” humanoid robot, thanks to its highly efficient manufacturing capabilities.
• Stanford’s 2026 Artificial Intelligence Index Report noted that the gap between the top AI models in China and the United States has narrowed to just 2.7%, while China now ranks first globally in both AI patent output and industrial robot installations.
• Alibaba Amap’s embodied intelligence division is set to launch its first quadruped robot, marking Alibaba’s official expansion of its AI footprint from digital space into real-world physical interaction 🚀.

Related links:

• World’s first large-scale deployment of an embodied intelligence industrial production line: Zhiyuan Genie G2 achieves zero errors during 8 consecutive hours of operation
• Cainiao launches climbing robot ZeeBot, with tested intelligent storage and retrieval efficiency doubled
• Tesla executive says Shanghai factory will provide the “golden key” to mass production of the Optimus robot
• Stanford 2026 AI report: China-U.S. technology gap shrinks to just 2.7%, with China leading in patents and robotics
• Alibaba’s first embodied robot revealed: Amap’s embodied intelligence division to launch a quadruped robot

DeepMind CEO预言AGI五年内落地，谷歌集中发布多项AI新应用

简报：

• DeepMind CEO Demis Hassabis 预判AGI（通用人工智能）有望在五年内实现，并警告此次AI浪潮是“以十倍速度发生的十倍工业革命”。
• Hassabis 指出当前前沿的四家AI公司技术差距正在持续扩大，AI表现出“参差不齐的智能”，行业竞争重点已从算力规模转向创新能力。
• 谷歌桌面版 Chrome 浏览器为内置的 Gemini 引入“技能库”功能，支持用户将常用的 AI 提示词保存并跨网页一键复用。
• 谷歌 Gemini 全面上线交互式模拟生成功能，能将复杂概念转化为可操作的动态模拟程序，大幅提升科学教育与探索的效率。
• 谷歌 AI 研究团队提出全新评估方法 Vantage，通过单一的大语言模型模拟真实群体互动，用于准确评估学生协作、创造力等“持久技能”。

相关链接：

• DeepMind CEO Hassabis 重磅预言：AGI 五年内到来！AI 不是普通升级，而是“十倍速工业革命”
• Chrome 为 Gemini 上线“技能库”：一键复用提示词，告别重复输入
• 谷歌 Gemini 新功能炸了！复杂概念秒变可玩模拟，学物理再也不用死记硬背
• 谷歌 AI 研究推出 Vantage：基于大语言模型的协作与创造力测评新方法

DeepMind CEO Predicts AGI Within Five Years as Google Rolls Out Multiple New AI Applications 🚀

Briefing:

• DeepMind CEO Demis Hassabis predicts that AGI (Artificial General Intelligence) could be achieved within the next five years, warning that this wave of AI is “a tenfold Industrial Revolution happening at ten times the speed.”
• Hassabis noted that the technology gap among the four leading AI companies is continuing to widen, with AI demonstrating “jagged intelligence,” while the industry’s competitive focus is shifting from sheer computing scale to innovation capability.
• Google’s desktop Chrome browser has introduced a “Skill Library” feature for its built-in Gemini, allowing users to save commonly used AI prompts and reuse them across web pages with one click.
• Google Gemini has fully launched its interactive simulation generation feature, capable of turning complex concepts into actionable dynamic simulation programs, significantly improving the efficiency of science education and exploration. 🧠
• Google AI researchers have proposed a new evaluation method called Vantage, which uses a single large language model to simulate real group interactions for accurately assessing “durable skills” such as student collaboration and creativity.

Related links:

• DeepMind CEO Hassabis Makes a Bold Prediction: AGI Will Arrive Within Five Years! AI Is Not a Normal Upgrade, but a “Tenfold-Speed Industrial Revolution”
• Chrome Launches “Skill Library” for Gemini: Reuse Prompts with One Click and Say Goodbye to Repetitive Input
• Google Gemini’s New Feature Is Explosive! Complex Concepts Instantly Become Playable Simulations, Making Physics Learning No Longer About Rote Memorization
• Google AI Research Introduces Vantage: A New Large Language Model-Based Method for Evaluating Collaboration and Creativity ✨

阿里千问上线“表格 Agent”，一句话生成与编辑Excel

简报：

• 阿里千问正式上线“表格 Agent”功能，支持用户通过自然语言对话直接完成 Excel 文件的生成、信息检索及深度编辑。
• 具备全场景表格处理能力，支持零门槛联网检索信息转表格、多轮对话内容提炼总结为结构化表格，以及多模态识别（支持将 PDF、Word、PPT 甚至手绘图文还原为 Excel 格式并进行指令化操作）。
• 技术上采用独立沙箱与智能规划，将复杂需求拆解为 Agent 执行链路，支持运行代码生成带公式的专业文件，并在信息不足时自动触发在线检索。
• 该功能目前已在千问 App、网页版及 PC 客户端全面免费上线，旨在让 Excel 处理不再依赖复杂函数，降低办公门槛。

相关链接：

• 对话即办公：阿里千问上线表格 Agent，支持直接生成与编辑 Excel

Alibaba Qwen Launches “Spreadsheet Agent,” Enabling Excel Creation and Editing with a Single Sentence 📊🤖

Brief:

• Alibaba Qwen has officially launched its “Spreadsheet Agent” feature, allowing users to generate Excel files, retrieve information, and perform in-depth editing directly through natural language conversations.
• It offers full-scenario spreadsheet processing capabilities, including turning online search results into tables with zero barriers, refining and summarizing multi-turn conversations into structured spreadsheets, and multimodal recognition that can convert PDFs, Word documents, PowerPoint files, and even hand-drawn text and images into Excel format for instruction-based operations.
• Technically, it uses an independent sandbox and intelligent planning to break down complex tasks into an Agent execution chain, supports running code to generate professional files with formulas, and can automatically trigger online searches when available information is insufficient.
• The feature is now fully available for free on the Qwen App, web version, and PC client, with the goal of making Excel processing less dependent on complex formulas and lowering the barrier to office productivity. ✨

Related Link:

• Conversation as Productivity: Alibaba Qwen Launches Spreadsheet Agent, Supporting Direct Excel Creation and Editing

斯坦福报告指中美AI差距基本抹平，国内迎万亿参数编程模型与教育AI新政

简报：

• 斯坦福大学《2026年人工智能指数报告》显示，中美顶尖AI模型技术差距已降至2.7%，中国在AI论文、专利产出及工业机器人安装量方面位居全球第一。
• 报告指出AI研发主导权已转至产业界（超90%），且AI社会渗透率达53%，普及速度超越PC与互联网。
• 月之暗面正式上线万亿参数大模型Kimi K2.6-code，专为AI编程优化，实测体验达到Sonnet 4.6水平，并推出每月39元起的会员计划。
• 教育部等五部门联合印发《“人工智能+教育”行动计划》，将AI正式纳入教师资格考试认证与高等教育公共基础课，目标到2030年基本形成深度融合格局。

相关链接：

• 斯坦福2026AI 报告:中美技术差距仅剩2.7%，中国在专利与机器人领域登顶
• 国产万亿大模型 Kimi K2.6-code 上线：对标 Sonnet 4.6 的编程新利器
• AI正式纳入教资考试与公共必修课:五部门联合部署，开启全民智能素养培育时代

Stanford Report Says the China-U.S. AI Gap Has Essentially Narrowed, as China Welcomes a Trillion-Parameter Coding Model and New Education AI Policies 🚀🤖📚

Briefing:

• Stanford University’s 2026 AI Index Report shows that the technological gap between the top AI models in China and the United States has narrowed to just 2.7%. China now ranks first globally in AI paper output, patent filings, and industrial robot installations.
• The report notes that leadership in AI R&D has shifted to the private sector, which now accounts for more than 90% of activity, while AI’s social penetration rate has reached 53%, spreading faster than both PCs and the internet.
• Moonshot AI has officially launched its trillion-parameter large model, Kimi K2.6-code, optimized specifically for AI coding. Early testing indicates performance on par with Sonnet 4.6, and the company has introduced a membership plan starting at RMB 39 per month.
• China’s Ministry of Education, together with four other government departments, has jointly issued the “AI + Education” Action Plan, formally incorporating AI into teacher qualification examinations and public foundational courses in higher education, with the goal of basically achieving deep integration by 2030.

Related Links:

• Stanford 2026 AI Report: China-U.S. Technology Gap Shrinks to Just 2.7%, China Leads in Patents and Robotics
• China’s Trillion-Parameter Model Kimi K2.6-code Launches: A New Coding Powerhouse Benchmarked Against Sonnet 4.6
• AI Officially Included in Teacher Qualification Exams and Public Required Courses: Five Ministries Jointly Launch a New Era of Nationwide AI Literacy Development

单词学习：![[2026041307661960.mp3]]

新闻播报：![[2026041371309904.mp3]]

中国16家科技社团联合发布倡议，推动构建全球AI治理体系

简报：

• 4月13日，中国自动化学会、中国电子学会、中国计算机学会等16家科技社团共同发布《全球人工智能治理科技社团倡议》。
• 倡议倡导“以人为本、智能向善”，致力于推动建立开放、公正、包容、有效的全球人工智能治理体系。
• 倡议明确提出，将增进人类共同福祉作为学术研究的根本遵循，并将安全作为学术研究与治理的底线要求。
• 倡议呼吁尊重各国平等参与人工智能研究与治理的权利，反对技术霸权、学术壁垒及不合理技术垄断，同时引导公众正确认知技术发展阶段与潜在风险。

相关链接：

• 中国16家科技社团联合倡议共促全球人工智能治理

16 Chinese Scientific and Technological Societies Jointly Issue Initiative to Promote the Building of a Global AI Governance System 🤖🌍

Briefing:

• On April 13, 16 scientific and technological societies, including the Chinese Association of Automation, the Chinese Institute of Electronics, and the China Computer Federation, jointly released the Initiative of Scientific and Technological Societies on Global Artificial Intelligence Governance.
• The initiative advocates “people-centered, AI for good” and is committed to promoting the establishment of an open, fair, inclusive, and effective global AI governance system.
• It explicitly states that advancing the common well-being of humanity should be the fundamental principle guiding academic research, and that safety should serve as the bottom-line requirement for both academic research and governance.
• The initiative calls for respect for all countries’ equal rights to participate in AI research and governance, opposes technological hegemony, academic barriers, and unreasonable technological monopolies, and seeks to guide the public toward a proper understanding of the current stage of technological development and its potential risks. 📢

Related Link:

• 16 Chinese Scientific and Technological Societies Jointly Call for Advancing Global AI Governance

单词学习：![[2026041198870988.mp3]]

新闻播报：![[2026041133451095.mp3]]

OpenAI回应Axios安全事件：未发现数据泄露，敦促Mac用户尽快更新应用

简报：

• OpenAI针对第三方开发者工具Axios的安全问题发布声明，目前未发现任何用户数据被访问、系统遭入侵或软件被篡改的证据。
• 事件源于Axios的两个npm版本（1.14.1和0.30.4）被恶意植入远程控制代码，OpenAI用于macOS应用签名的工作流曾因配置缺陷下载并执行了恶意版本。
• 出于安全考虑，OpenAI已启动撤销并轮换安全证书机制，并与苹果合作从5月8日起拦截旧版应用的下载与启动，以防范分发伪造应用的风险。
• 受此次证书更新影响的macOS应用共有4款：ChatGPT、Codex、Atlas以及Codex CLI，OpenAI要求所有相关Mac用户尽快将应用更新至最新版本。

相关链接：

• OpenAI就第三方开发者库安全事件发声明：未发现用户数据泄露或系统遭入侵
• OpenAI就第三方开发者库安全事件发声明（36氪）
• OpenAI 回应 Axios 工具安全事件，敦促苹果 Mac 用户更新 ChatGPT 等应用

OpenAI Responds to Axios Security Incident: No Data Breach Found, Urges Mac Users to Update Apps Promptly 🔐

Brief:

• OpenAI has issued a statement regarding a security issue involving the third-party developer tool Axios, saying that it has found no evidence that any user data was accessed, that its systems were compromised, or that its software was tampered with.
• The incident stemmed from two npm versions of Axios (1.14.1 and 0.30.4) that were maliciously injected with remote-control code. Due to a configuration flaw, a workflow OpenAI used to sign its macOS applications downloaded and executed the compromised versions.
• As a precaution, OpenAI has initiated a security certificate revocation and rotation process, and has worked with Apple to block the download and launch of older app versions starting May 8, in order to prevent the distribution of spoofed applications.
• A total of four macOS apps are affected by this certificate update: ChatGPT, Codex, Atlas, and Codex CLI. OpenAI is asking all affected Mac users to update their apps to the latest versions as soon as possible ⚠️

Related links:

• OpenAI issues statement on third-party developer library security incident: No user data breach or system compromise found
• OpenAI issues statement on third-party developer library security incident (36Kr)
• OpenAI responds to Axios tool security incident, urges Apple Mac users to update ChatGPT and other apps

蔚来神玑NX9031芯片与世界模型将全面应用至2026款乐道L90

简报：

• 4月11日，蔚来创始人兼CEO李斌在智能电动汽车发展高层论坛（2026）上宣布，旗舰智驾芯片神玑NX9031和新版蔚来世界模型将在2026款乐道L90上全面上车。
• 蔚来今年的重要目标是在20万至30万元级别车型中全面应用全球旗舰级的智驾芯片和智驾系统。

相关链接：

• 蔚来高阶智驾芯片神玑NX9031正式上车乐道L90

NIO’s Shenji NX9031 Chip and World Model to Be Fully Applied to the 2026 ONVO L90 🚗

Briefing:

• On April 11, NIO founder and CEO William Li announced at the Senior Forum on the Development of Intelligent Electric Vehicles (2026) that the flagship smart driving chip, Shenji NX9031, and the new version of NIO’s world model will be fully deployed in the 2026 ONVO L90.
• One of NIO’s key goals this year is to fully apply world-leading flagship-level smart driving chips and smart driving systems across models priced in the RMB 200,000 to 300,000 range. 🤖

Related link:

• NIO’s advanced smart driving chip Shenji NX9031 officially installed in the ONVO L90 🌐

日本政府加码投资Rapidus，豪掷163亿美元布局AI芯片

简报：

• 日本经济产业省宣布提供资金，支持初创企业Rapidus为首批客户之一的富士通开展研发工作。
• 截至2027年3月本财年末，日本政府对Rapidus的补贴与投资总额将升至2.6万亿日元（约合163亿美元）。
• 外部委员会已完成对Rapidus位于北海道芯片制造厂的视察，并对其技术进展予以认可。

相关链接：

• 日本投入160亿美元，力推初创企业Rapidus进军人工智能芯片领域

Japan Boosts Investment in Rapidus, Commits dJyH6kDkfC5RlM16.answerText6.3 Billion to AI Chip Strategy 🚀

Briefing:

• Japan’s Ministry of Economy, Trade and Industry announced funding support to back R&D efforts by Fujitsu, one of the first customers of startup Rapidus.
• By the end of the current fiscal year in March 2027, the Japanese government’s total subsidies and investments in Rapidus will rise to JPY 2.6 trillion (approximately dJyH6kDkfC5RlM16.answerText6.3 billion).
• An external committee has completed its inspection of Rapidus’s chip manufacturing plant in Hokkaido and acknowledged its technological progress. 🤖

Related link:

• Japan commits dJyH6kDkfC5RlM16.answerText6 billion to support startup Rapidus in advancing into the AI chip sector 🇯🇵

滴滴新一代Robotaxi亮相，张博强调自动驾驶安全与体验

简报：

• 4月11日，滴滴自动驾驶CEO张博出席2026智能电动汽车发展高层论坛，新一代Robotaxi车型同步亮相。
• 张博表示，安全与体验是自动驾驶的根基，公司将基于AI核心技术、硬件、出行场景持续加大投入并优化服务。
• 滴滴自动驾驶已于去年四季度在广州、北京部分示范应用区域开启全天候、全无人载客测试。

相关链接：

• 滴滴自动驾驶张博：聚焦安全和体验，推动自动驾驶全球化落地

Didi’s New-Generation Robotaxi Debuts, Zhang Bo Emphasizes Autonomous Driving Safety and User Experience 🚖🤖

Briefing:

• On April 11, Zhang Bo, CEO of Didi Autonomous Driving, attended the 2026 High-Level Forum on the Development of Intelligent Electric Vehicles, where the company’s new-generation Robotaxi model was unveiled.
• Zhang Bo stated that safety and user experience are the foundation of autonomous driving. The company will continue to increase investment in and optimize its services across core AI technologies, hardware, and mobility scenarios.
• In the fourth quarter of last year, Didi Autonomous Driving launched 24/7 fully driverless passenger-carrying tests in selected demonstration application areas in Guangzhou and Beijing.

Related Link:

• Didi Autonomous Driving’s Zhang Bo: Focusing on Safety and User Experience to Promote the Global Deployment of Autonomous Driving 🌍

单词学习：![[2026040801617304.mp3]]

新闻播报：![[2026040828011093.mp3]]

十部门联合印发AI伦理审查试行办法，全面规范人工智能合规与安全

简报：

• 工业和信息化部等十个部门联合印发《人工智能科技伦理审查与服务办法（试行）》，旨在规范和促进人工智能科技活动的伦理治理，确保合规与安全。
• 办法明确了AI伦理审查的适用范围、实施主体及工作程序，涵盖申请受理、审查评估与专家复核等环节。
• 重点审查领域包括人机融合系统、具备舆论引导能力的算法模型，以及具有高度自主能力的决策系统。
• 提出了标准建设、服务体系、创新鼓励、宣传教育和人才培养等五个方面的支持举措，帮助企业提升风险防控能力。
• 对违反相关规定的行为明确了法律后果，将依据相关法律法规进行调查处理。

相关链接：

• 工信部发布《人工智能科技伦理审查与服务办法（试行）》明确治理新规

Ten Departments Jointly Issue Trial Measures for AI Ethics Review, Comprehensively Regulating AI Compliance and Safety 🤖

Briefing:

• Ten departments, including the Ministry of Industry and Information Technology, have jointly issued the Measures for the Ethical Review and Services of Artificial Intelligence Science and Technology Activities (Trial), aiming to regulate and promote the ethical governance of AI-related scientific and technological activities and ensure compliance and safety.
• The Measures clarify the scope of application, implementing entities, and working procedures for AI ethics review, covering application acceptance, review and assessment, and expert re-evaluation.
• Key areas of review include human-machine fusion systems, algorithmic models with the capacity to guide public opinion, and decision-making systems with a high degree of autonomy.
• Five supporting measures are proposed, covering standards development, service systems, innovation incentives, public awareness and education, and talent cultivation, to help enterprises strengthen their risk prevention and control capabilities 📘.
• Legal consequences for violations of the relevant provisions are clearly specified, and such cases will be investigated and handled in accordance with applicable laws and regulations ⚖️.

Related Link:

• MIIT Issues the Measures for the Ethical Review and Services of Artificial Intelligence Science and Technology Activities (Trial), Clarifying New Governance Rules

DeepSeek V4开启密集灰测，全新多模态架构与视觉能力曝光

简报：

• 4月7日消息，DeepSeek V4正在进行密集灰度测试，在底层架构、交互逻辑和多模态能力上均有重大升级。
• 曝光界面显示其推出三大核心架构：主打响应速度的快速版（Lite）、基于全新记忆架构的专家版，以及可直接处理图视频任务的视觉版（Vision）。
• DeepSeek深度适配本土算力生态，正开发至少2款完全基于国产AI芯片的大模型，测试申请优先对国内厂商开放以协同优化。
• 市场期待V4延续超长文本处理优势（预计支持百万级Token），并猜测或将推出专门针对代码生成的AI编程特别版。

相关链接：

• DeepSeek V4 灰度测试曝光：全新视觉版现身

请问是否还需要为您总结其他相关的人工智能行业资讯？

DeepSeek V4 Enters Intensive Gray-Scale Testing, Revealing a New Multimodal Architecture and Visual Capabilities 🚀👀

Brief:

• On April 7, reports indicated that DeepSeek V4 is undergoing intensive gray-scale testing, with major upgrades across its underlying architecture, interaction logic, and multimodal capabilities.
• The leaked interface shows that it introduces three core architectures: a Lite version focused on response speed, an Expert version built on a brand-new memory architecture, and a Vision version capable of directly handling image and video tasks.
• DeepSeek is deeply adapting to the domestic computing ecosystem and is developing at least two large models built entirely on Chinese-made AI chips. Testing access is being prioritized for domestic vendors to support coordinated optimization.
• The market expects V4 to continue its advantage in ultra-long text processing, with support anticipated to reach the million-token level, and there is also speculation that a special AI coding edition tailored specifically for code generation may be launched. 💡

Related link:

• DeepSeek V4 Gray-Scale Testing Revealed: New Vision Version Emerges

Would you like me to also summarize other related AI industry news for you?

谷歌AI概览准确率约90%，仍面临高频错误与虚假信息挑战

简报：

• 据《纽约时报》报道，谷歌AI概览（AI Overviews）准确率约为90%，在庞大的搜索基数下，每小时可能生成超5700万条错误答案。
• 初创公司Oumi评估显示，Gemini 3的准确率较前代提升至91%，但其概览内容与原始信息来源不符的比例却升至56%。
• 测试表明AI概览极易受虚假信息影响，存在引用人工炮制的虚假博客以及在同一搜索页面出现自相矛盾信息的情况。
• 针对上述评估结果，谷歌发言人提出质疑，认为Oumi的测试方法未能真实反映实际用户的搜索行为。

相关链接：

• 谷歌搜索 AI 概览准确率仅 90%，易受虚假信息影响

Google AI Overviews Are About 90% Accurate, but Still Face Frequent Errors and Misinformation Challenges 🤖⚠️

Brief:

• According to The New York Times, Google AI Overviews are about 90% accurate, but given the massive scale of search traffic, they could still generate more than 57 million incorrect answers per hour.
• An evaluation by startup Oumi found that Gemini 3’s accuracy improved to 91% compared with its predecessor, but the share of overview content that did not match the original source information rose to 56%.
• Tests showed that AI Overviews are highly vulnerable to misinformation, including cases where they cited fabricated blog posts and displayed contradictory information on the same search results page.
• In response to these findings, a Google spokesperson questioned the evaluation results, arguing that Oumi’s testing methods did not accurately reflect real user search behavior.

Related link:

• Google Search AI Overviews Are Only 90% Accurate and Easily Influenced by Misinformation 📎

Anthropic发布顶级网络安全模型Mythos，精准挖掘数千陈年代码漏洞

简报：

• Anthropic正式推出新一代前沿模型Mythos的预览版，该模型在智能体级别的编程与推理及网络安全领域表现卓越。
• 在近期封闭测试中，Mythos成功扫描出数千个零日漏洞，并精准锁定了隐藏在一二十年前老旧代码中的高危历史遗留问题。
• 因具备极强的攻击性潜力，Mythos暂不对公众开放，目前仅通过“Project Glasswing”安全项目向苹果、微软等行业巨头及美国联邦官员定向试用。
• 因拒绝将技术用于对美国公民的自主打击或监控，Anthropic近期被美国国防部列为“供应链风险实体”。

相关链接：

• Anthropic推出网络安全AI模型Mythos，专治代码陈年漏洞

还有其他资讯需要我为您整理成简报吗？

Anthropic Unveils Top-Tier Cybersecurity Model Mythos, Accurately Unearthing Thousands of Long-Standing Code Vulnerabilities 🔐

Brief:

• Anthropic has officially launched a preview of its next-generation frontier model, Mythos, which delivers outstanding performance in agent-level programming, reasoning, and cybersecurity.
• In recent closed testing, Mythos successfully scanned and identified thousands of zero-day vulnerabilities, while precisely uncovering high-risk legacy issues hidden in outdated code written one or even two decades ago.
• Due to its exceptionally strong offensive potential, Mythos is not yet available to the public. It is currently being selectively trialed only through the “Project Glasswing” security initiative by industry giants such as Apple and Microsoft, as well as U.S. federal officials.
• After refusing to allow its technology to be used for autonomous strikes or surveillance against U.S. citizens, Anthropic was recently designated by the U.S. Department of Defense as a “supply chain risk entity.” ⚠️

Related link:

• Anthropic launches cybersecurity AI model Mythos to tackle long-standing code vulnerabilities

Would you like me to compile any other updates into a briefing for you? 📰

智谱GLM-5.1发布：编程基准夺冠并全线提价，国产模型迈向性能溢价

简报：

• 智谱正式发布全新大模型GLM-5.1，在衡量软件开发能力的SWE-bench Pro基准测试中刷新世界纪录，超越Claude 4.6 Opus位居全球首位。
• 该模型具备极强的自主工程能力，支持单次任务持续工作达8小时，可独立完成规划、执行、测试及错误自修复，直接交付工程级成果。
• 伴随新品上线，智谱GLM全线模型提价10%，核心场景价格已锚定Claude 3.5 Sonnet，标志着国产大模型竞争重心从成本倾销转向性能溢价。
• 受发布消息影响，智谱股价在4月8日上午一度大涨14.12%，公司总市值攀升至3964亿元。

相关链接：

• 智谱发布旗舰模型GLM-5.1：SWE-bench评分全球领跑

Zhipu GLM-5.1 Released: Tops Coding Benchmarks and Raises Prices Across the Line, as Chinese Models Move Toward Performance Premiums 🚀

Brief:

• Zhipu has officially launched its new large model, GLM-5.1. In the SWE-bench Pro benchmark, which measures software development capabilities, it set a new world record, surpassing Claude 4.6 Opus to rank No. 1 globally.
• The model demonstrates exceptionally strong autonomous engineering capabilities, supporting continuous work on a single task for up to 8 hours. It can independently handle planning, execution, testing, and error self-repair, directly delivering engineering-grade results.
• Alongside the new release, Zhipu has raised prices by 10% across the entire GLM model lineup. Pricing for core use cases is now benchmarked against Claude 3.5 Sonnet, signaling that competition among Chinese large models is shifting from cost-driven underpricing to performance-based premiums 📈
• Driven by the release announcement, Zhipu’s stock surged as much as 14.12% on the morning of April 8, pushing the company’s total market capitalization to RMB 396.4 billion.

Related Link:

• Zhipu Releases Flagship Model GLM-5.1: Leads the World in SWE-bench Score 🔗

单词学习：![[2026040524776188.mp3]]

新闻播报：![[2026040548343509.mp3]]

瑞典隆德大学参与研发新型AI模型，单次血检即可辅助筛查神经退行性疾病

简报：

• 瑞典隆德大学宣布，该校参与研发出一种新型人工智能（AI）模型。
• 该AI模型可通过单次血液检测辅助识别多种神经退行性疾病，为相关疾病的早期筛查和辅助诊断提供了新思路。
• 隆德大学指出，由于不同神经退行性疾病早期症状相似、认知下降成因复杂且患者大脑病理过程可能重叠，临床诊断难度较大，该模型有望辅助应对这一难题。

相关链接：

• 新型AI模型可通过血检识别多种神经退行性疾病
• 新型AI模型可通过血检识别多种神经退行性疾病

Lund University in Sweden Participates in Developing a New AI Model That Can Help Screen for Neurodegenerative Diseases With a Single Blood Test 🧠🔬

Brief:

• Lund University in Sweden announced that it has participated in the development of a new artificial intelligence (AI) model.
• The AI model can help identify multiple neurodegenerative diseases through a single blood test, offering a new approach for early screening and auxiliary diagnosis of such conditions.
• Lund University noted that clinical diagnosis is often challenging because different neurodegenerative diseases share similar early symptoms, the causes of cognitive decline are complex, and pathological processes in patients’ brains may overlap. The model is expected to help address this problem 🤖.

Related links:

• A New AI Model Can Identify Multiple Neurodegenerative Diseases Through Blood Tests
• A New AI Model Can Identify Multiple Neurodegenerative Diseases Through Blood Tests

台积电3纳米产能超载飙破100%，2026年全球半导体迎最严峻供应链瓶颈

简报：

• 受AI芯片强劲需求拉动，博通等厂商示警台积电制造量能已逼近上限，今年供应链已面临一定阻碍。
• 台积电3纳米产能处于“极不正常”的超载状态，预估今年第二至第四季利用率将高达104%、109%与105%，产能主要被辉达和苹果等巨头瓜分。
• 为缓解供需落差，台积电正积极赶工扩产，新建厂数从过往每年平均3座飙升至2025年的9座（含先进封装厂）。
• 产能紧缺的连锁反应已迅速外溢至存储器、先进封装、光学元件与印刷电路板等周边次产业。
• 产业专家评估2026年将是半导体供应链最严峻的考验期，各大科技厂或需签署3至5年长约以确保未来料源稳定。

相关链接：

• 台积电产能已逼近极限？产业专家示警：今年恐成供应链最严峻的瓶颈期

TSMC’s 3nm Capacity Surges Past 100% Utilization, as 2026 Looms as the Most Severe Global Semiconductor Supply Chain Bottleneck 🚨

Briefing:

• Driven by strong demand for AI chips, companies such as Broadcom have warned that TSMC’s manufacturing capacity is nearing its limit, and the supply chain is already facing a degree of disruption this year.
• TSMC’s 3nm capacity is in an “extremely abnormal” state of overload, with utilization in the second through fourth quarters of this year projected to reach 104%, 109%, and 105%, respectively. Most of the capacity has been taken up by giants such as Nvidia and Apple.
• To ease the supply-demand gap, TSMC is aggressively accelerating capacity expansion, with the number of new plants rising from an annual average of 3 in previous years to 9 in 2025, including advanced packaging facilities 🏭.
• The ripple effects of the capacity crunch have already spread rapidly to related sectors, including memory, advanced packaging, optical components, and printed circuit boards.
• Industry experts estimate that 2026 will be the most critical stress-test period for the semiconductor supply chain, and major technology companies may need to sign 3- to 5-year long-term contracts to secure stable future supply 🔗.

Related link:

• Is TSMC’s Capacity Nearing Its Limit? Industry Experts Warn This Year Could Mark the Most Severe Supply Chain Bottleneck Period

美国科技岗位需求创三年新高，AI取代工程师说法暂未获数据证实

简报：

• 据《商业内幕》4月4日报道，美国科技岗位需求明显回暖，2026年以来招聘需求快速反弹。
• TrueUp数据显示，软件工程岗位空缺已超6.7万个，创三年多新高，较2023年年中低点基本翻倍，今年以来进一步增长约30%。
• AI投资加速推进反而需要更多工程师参与，TrueUp创始人表示“AI取代工程师”的主流说法目前并未得到招聘数据的验证。
• 尽管岗位数量回升，但因计算机专业人才供给显著增加，入门级岗位的竞争相比五年前更加激烈。

相关链接：

• AI 尚未“取代”工程师，美国科技岗位空缺今年迎来快速反弹

U.S. Tech Job Demand Hits a Three-Year High, While Claims That AI Is Replacing Engineers Remain Unconfirmed by the Data 🚀

Brief:

• According to a Business Insider report on April 4, demand for tech jobs in the United States is clearly rebounding, with hiring demand accelerating rapidly since 2026.
• Data from TrueUp shows that software engineering job openings have exceeded 67,000, reaching a more than three-year high. This is nearly double the low point seen in mid-2023, with a further increase of about 30% so far this year.
• Accelerating investment in AI is, in fact, requiring more engineers to participate. According to TrueUp’s founder, the mainstream claim that “AI is replacing engineers” has not, at present, been validated by hiring data 🤖.
• Although the number of openings is rising, competition for entry-level roles is more intense than it was five years ago, due to a significant increase in the supply of computer science graduates.

Related link:

• AI Has Not Yet “Replaced” Engineers as U.S. Tech Job Openings Rebound Rapidly This Year 📈

Django联合创始人预言：软件开发或步入AI全自动化的“黑灯工厂”时代

简报：

• Django框架联合创始人西蒙·威利森提出，若AI独立完成需求、监督与代码审核等全流程，软件开发将进入无需人类干预的“黑灯工厂”阶段。
• 威利森表示自己目前产出的代码约95%来自AI，且已有部分企业要求员工停止手写代码。
• AI的发展对就业已产生实质影响，包括Klarna、IBM、Block和甲骨文等多家公司已将部分裁员归因于AI。
• 尽管AI降低了开发门槛，但具备原创性和创造力仍然是成功的关键。

相关链接：

• Django Web 框架共同创始人威利森：AI 将为编程带来“黑灯工厂”时代

Django Co-Founder Predicts Software Development May Enter an AI-Driven “Lights-Out Factory” Era 🤖💡

Brief:

• Simon Willison, co-founder of the Django framework, said that if AI can independently handle the entire workflow — including requirements gathering, supervision, and code review — software development could enter a “lights-out factory” stage requiring no human intervention.
• Willison said that about 95% of the code he currently produces comes from AI, and that some companies have already instructed employees to stop writing code by hand.
• The rise of AI is already having a tangible impact on employment, with companies including Klarna, IBM, Block, and Oracle attributing some layoffs to AI.
• Although AI has lowered the barrier to software development, originality and creativity remain the keys to success. 🚀

Related link:

• Django Web Framework Co-Founder Willison: AI Will Bring a “Lights-Out Factory” Era to Programming

易烊千玺肖像被擅用于AI短剧，涉事平台已陆续下架违规内容

简报：

• 4月5日凌晨，易烊千玺工作室发声，明确表示未参演且未授权第三方使用其肖像进行AI剧集合成。
• 红果短剧平台已将涉事的《午夜公车》《骗我投个好胎？》等相关AI短剧下架，官方回应称已修改或删除相关内容，并保留对出品方追责的权利。
• 截至发稿，同样上架相关作品的百度暂未对违规人物形象进行调整或做下架处理。

相关链接：

• 易烊千玺遭AI短剧“盗脸”牟利，平台方已下架相关产品（第一财经）
• 易烊千玺遭AI短剧“盗脸”牟利，平台方已下架相关产品（36氪）

Jackson Yee’s Likeness Was Unauthorizedly Used in AI Short Dramas, and the Platform Involved Has Gradually Removed the Violating Content 🎭🤖

Brief:

• In the early hours of April 5, Jackson Yee’s studio issued a statement, making it clear that he did not participate in and did not authorize any third party to use his likeness in AI-generated short drama productions.
• Hongguo Short Drama Platform has removed the AI short dramas involved, including Midnight Bus and Did You Trick Me into Being Reborn Better? The platform officially responded that the relevant content has been modified or deleted, and it reserves the right to hold the producers accountable.
• As of press time, Baidu, which had also listed the related works, has not yet adjusted the infringing character images or taken them down ⚠️.

Related links:

• Jackson Yee’s face was “stolen” by AI short dramas for profit, and the platform has removed the related products (Yicai)
• Jackson Yee’s face was “stolen” by AI short dramas for profit, and the platform has removed the related products (36Kr)

单词学习：![[2026040638006715.mp3]]

新闻播报：![[2026040614303140.mp3]]

富士康开启苹果首款大折叠屏iPhone试产

简报：

• 产业链人士透露，富士康目前已经开始试产苹果首款折叠屏iPhone手机。
• 供应链消息指出，该款设备定位为一款大折叠屏形态的iPhone。
• 根据苹果向供应商提供的出货目标指引，这款折叠屏手机预计将于2026年下半年正式推出。

相关链接：

• 苹果首款折叠屏手机已在试产(中证金牛座)
• 苹果首款折叠屏手机已在试产(36氪)

Foxconn Begins Trial Production of Apple’s First Large Foldable-Screen iPhone 📱🍎

Briefing:

• Industry sources revealed that Foxconn has now begun trial production of Apple’s first foldable iPhone.
• Supply chain sources indicated that the device is positioned as a large foldable-screen form factor iPhone.
• According to shipment target guidance Apple provided to suppliers, the foldable smartphone is expected to be officially launched in the second half of 2026. 🚀

Related Links:

• Apple’s first foldable smartphone has entered trial production (China Securities Golden Bull)
• Apple’s first foldable smartphone has entered trial production (36Kr)

美股科技巨头盘前集体上扬，特斯拉领涨1%

简报：

• 2026年4月6日美股盘前，大型科技股多数呈现上涨态势，特斯拉以1%的涨幅领跑。
• 科技巨头Meta与微软盘前均上涨0.5%，谷歌A上涨0.4%，亚马逊与英伟达分别小幅上涨0.3%和0.2%。
• 苹果公司盘前表现相对疲软，小幅下跌0.1%。

相关链接：

• 美股大型科技股盘前多数上涨，特斯拉涨1%
• 36氪快讯

U.S. Tech Giants Rise Collectively in Premarket Trading, Tesla Leads with a 1% Gain 📈

Briefing:

• In U.S. premarket trading on April 6, 2026, most large-cap technology stocks moved higher, with Tesla leading the gains at 1% 🚗.
• Tech giants Meta and Microsoft both rose 0.5% in premarket trading, Alphabet A gained 0.4%, while Amazon and Nvidia edged up 0.3% and 0.2%, respectively.
• Apple showed relatively weaker premarket performance, slipping slightly by 0.1% 📉.

Related links:

• Most large-cap U.S. tech stocks rise in premarket trading, Tesla up 1%
• 36Kr Flash

宁德时代发力航运电动化，海洋事业部计划年内扩编至500人

简报：

• 宁德时代决定全力以赴推动全球航运业电动化进程，试图将其在电动汽车领域的成功经验复制到海上领域。
• 宁德时代海洋业务部门负责人Su Yi表示，计划在今年将团队规模扩大一倍多，达到约500人。
• 目前该部门的重点是生产满足水上运行“极高”要求的电池，核心包括提升电池单元的长寿命以及在海洋环境中的安全性。
• 尽管宁德时代拒绝提供具体的销售目标，但公司发言人表示对其“强劲的市场潜力”非常有信心。

相关链接：

• 宁德时代将雄心拓展至远洋 计划今年将其海洋事业部规模扩大一倍
• 媒体：全球最大的电池制造商宁德时代将雄心拓展至远洋，计划今年将其海洋事业部规模扩大一倍

CATL Steps Up Push for Shipping Electrification, Marine Division Plans to Expand to 500 Employees This Year ⚓🔋

Briefing:

• CATL has decided to go all in on advancing the electrification of the global shipping industry, aiming to replicate its success in the electric vehicle sector in the maritime arena.
• Su Yi, head of CATL’s marine business division, said the company plans to more than double the team’s size this year to around 500 employees.
• The division is currently focused on producing batteries that can meet the “extremely high” demands of marine operations, with key priorities including extending battery cell lifespan and ensuring safety in ocean environments.
• Although CATL declined to provide specific sales targets, a company spokesperson said it is highly confident in the sector’s “strong market potential” 🌊.

Related links:

• CATL Extends Its Ambitions to Deep-Sea Shipping, Plans to Double the Size of Its Marine Division This Year
• Media: The World’s Largest Battery Manufacturer CATL Extends Its Ambitions to Deep-Sea Shipping, Plans to Double the Size of Its Marine Division This Year

单词学习：![[2026040174507533.mp3]]

新闻播报：![[2026040167862289.mp3]]

特斯拉FSD 14.3版本进入员工测试，预计本周末正式推送

简报：

• 马斯克在社交平台透露，特斯拉最新的FSD（全自动驾驶）14.3版本预计将在本周末正式发布并向用户推送。
• 目前该版本系统已经进入特斯拉内部员工测试阶段。
• 此次版本更新推送将进一步推动特斯拉自动驾驶技术的演进。

相关链接：

• 马斯克：FSD 14.3版本预计将于本周末发布
• 马斯克称特斯拉FSD 14.3版本预计将于本周末发布

Tesla FSD Version 14.3 Enters Employee Testing, Official Rollout Expected This Weekend 🚗

Brief:

• Musk revealed on social media that Tesla’s latest FSD (Full Self-Driving) version 14.3 is expected to be officially released and rolled out to users this weekend.
• The new system version has already entered Tesla’s internal employee testing phase.
• This update is expected to further advance the evolution of Tesla’s autonomous driving technology. 🤖✨

Related Links:

• Musk: FSD Version 14.3 Expected to Be Released This Weekend
• Musk Says Tesla FSD Version 14.3 Is Expected to Be Released This Weekend

前特斯拉中国区总经理孔艳双加盟小米，接手汽车销售业务

简报：

• 前特斯拉中国区总经理孔艳双已于3月初正式入职小米，目前正处于工作交接期。
• 孔艳双入职后，将接替原小米汽车总监李晓锐，主要负责小米汽车销售方面的相关工作。
• 截至目前，小米内部尚未正式发文通告其具体职位，小米官方对此人事变动消息暂未作出回应。

相关链接：

• 原特斯拉中国区高管孔艳双入职小米（界面）
• 原特斯拉中国区高管孔艳双入职小米（36氪）

Former Tesla China General Manager Kong Yanshuang Joins Xiaomi, Takes Over Auto Sales Business 🚗

Brief:

• Kong Yanshuang, former General Manager of Tesla China, officially joined Xiaomi in early March and is currently in the work handover period.
• After joining Xiaomi, Kong will succeed Li Xiaorui, the former head of Xiaomi Auto, and will mainly be responsible for sales-related work for Xiaomi’s automotive business.
• As of now, Xiaomi has not officially announced Kong’s specific title internally, and the company has not yet responded to the personnel change. 📌

Related Links:

• Former Tesla China Executive Kong Yanshuang Joins Xiaomi (Jiemian)
• Former Tesla China Executive Kong Yanshuang Joins Xiaomi (36Kr)📰

现代汽车3月在韩电动汽车销量达7809辆，同比大增38%

简报：

• 现代汽车最新数据显示，3月份其在韩国本土市场的电动汽车销量达到7809辆。
• 较去年同期相比，韩国本土电动汽车销量实现了38%的显著增长。

相关链接：

• 现代汽车3月韩国本土电动汽车销量同比增长38%

Hyundai Motor's EV Sales in South Korea Reach 7,809 Units in March, Up 38% Year-on-Year ⚡📈

Brief:

• Hyundai Motor's latest data shows that its electric vehicle sales in the South Korean domestic market reached 7,809 units in March.
• Compared with the same period last year, Hyundai's domestic EV sales in South Korea recorded a significant year-on-year increase of 38%.

Related Link:

• Hyundai Motor's Domestic EV Sales in South Korea Rise 38% Year-on-Year in March

因扩张计划与成本上涨，越南车企VinFast推迟盈亏平衡至2027年后

简报：

• 受市场环境、运营状况以及不断上涨的成本综合影响，越南电动汽车制造商（造车新势力）VinFast宣布推迟实现盈亏平衡的预期时间。
• 据知情人士透露，由于该公司目前正积极推进快速扩张计划，预计要到2027年之后才能实现盈亏平衡。

相关链接：

• 越南VinFast将实现盈亏平衡时间推迟至2027年之后（新浪财经）
• 越南VinFast将实现盈亏平衡时间推迟至2027年之后（36氪）

Vietnam’s VinFast Delays Break-Even Timeline to After 2027 Due to Expansion Plans and Rising Costs 🚗📉

Brief:

• Vietnam’s electric vehicle maker VinFast has announced a delay to its expected break-even timeline, citing a combination of market conditions, operating performance, and continuously rising costs.
• According to people familiar with the matter, the company is currently pushing ahead with an aggressive expansion plan and is now expected to reach break-even only after 2027.

Related links:

• Vietnam’s VinFast Delays Break-Even Timeline to After 2027 (Sina Finance)
• Vietnam’s VinFast Delays Break-Even Timeline to After 2027 (36Kr)

北汽蓝谷子公司3月产销出炉：产量同比大增61.5%，销量微降

简报：

• 北汽蓝谷发布的最新产销快报显示，其子公司3月份产量为18652辆，同比增长61.5%
• 3月份销量完成13647辆，较去年同期出现4.35%的下滑
• 本年度累计销量达到29084辆，实现同比增长4.94%

相关链接：

• 北汽蓝谷：子公司3月销量13647辆，同比下降4.35%
• 北汽蓝谷子公司3月产销快报公告原文

BAIC BluePark Subsidiary Releases March Production and Sales Figures: Output Surges 61.5% Year-on-Year, Sales Edge Down 📉🚗

Brief:

• The latest production and sales report released by BAIC BluePark shows that its subsidiary produced 18,652 vehicles in March, up 61.5% year-on-year.
• Sales in March reached 13,647 vehicles, marking a 4.35% decline compared with the same period last year.
• Cumulative sales for the year totaled 29,084 vehicles, representing a year-on-year increase of 4.94% 📊

Related Links:

• BAIC BluePark: Subsidiary sold 13,647 vehicles in March, down 4.35% year-on-year
• Original announcement of BAIC BluePark subsidiary’s March production and sales report