单词学习：![[7b06f796-03f1-4836-9a9f-9b0ec0add912.mp3]]

新闻播报：![[0edc90d1-7dd6-4e4f-a853-e06fa3ebad98.mp3]]

Marvell斥资55亿美元收购Celestial AI，加码AI数据中心光子互联技术

简报：

• 芯片制造商Marvell宣布将以最高55亿美元收购初创公司Celestial AI，旨在获取其光子互联技术，以增强在AI数据中心连接领域的竞争力。
• Celestial AI的核心产品“光子织物”技术旨在使用光信号替代传统电信号来传输数据，以提高速率、降低延迟与能耗。
• 初始交易金额为10亿美元现金和22.5亿美元股票，若Celestial AI在2029财年前实现特定业绩目标，总额将增至55亿美元，交易预计于2026年第一季度完成。
• Marvell同时获得了亚马逊的支持，并向其发行股票认股权证以加强合作关系；该收购消息刺激其盘后股价上涨了13%。

相关链接：

• Marvell 豪掷 55 亿美元收购 Celestial AI，布局未来光子互联技术

Marvell Acquires Celestial AI for $5.5 Billion, Boosting AI Data Center Photonic Interconnect Technology 💰

Brief:

• Chipmaker Marvell has announced its acquisition of startup Celestial AI for up to $5.5 billion, aiming to secure its photonic interconnect technology and enhance its competitiveness in AI data center connectivity.
• Celestial AI's core product, "Photonic Fabric" technology, uses light signals instead of traditional electrical signals to transmit data, thereby increasing speed, reducing latency, and lowering energy consumption. ✨
• The initial deal includes dJyH6kDkfC5RlM16.answerText billion in cash and $2.25 billion in stock. The total amount could rise to$5.5 billion if Celestial AI achieves specific performance targets by fiscal year 2029. The transaction is expected to close in Q1 2026.
• Marvell also secured support from Amazon, issuing stock warrants to strengthen their partnership. News of the acquisition spurred a 13% increase in Marvell's after-hours stock price. 📈

Related Links:

• Marvell Spends $5.5 Billion to Acquire Celestial AI, Investing in Future Photonic Interconnect Technology

Kling 2.6版本发布，AI视频实现音画同步生成

简报：

• 快手发布Kling AI 2.6版本，首次内置音频生成功能，支持中英双语对白、歌唱及音效与视频同步输出。
• 新版本维持10秒1080P高清输出，同时提升了复杂指令遵守率和跨镜头角色一致性。
• Kling 2.6将率先在Artlist等专业平台上线，面向影视、短剧、广告及MV制作领域。
• 快手计划在2026年第一季度推出4K/60fps版本，并开放自定义声线库。

相关链接：

• Kling 2.6将发布：原生音频+10秒1080P，AI视频进入有声时代

Kling 2.6 Released: AI Video Achieves Synchronized Audio-Visual Generation 🚀

Brief:

• Kuaishou has unveiled Kling AI version 2.6, featuring a built-in audio generation function for the first time. It supports synchronous output of bilingual (Chinese and English) dialogue, singing, and sound effects alongside video. 🎶
• The new version maintains 10-second 1080P high-definition output while also improving compliance with complex instructions and enhancing character consistency across different shots.
• Kling 2.6 will first be launched on professional platforms such as Artlist, targeting the film and television, short drama, advertising, and music video production industries. 🎬
• Kuaishou plans to introduce a 4K/60fps version and open a custom voice library in the first quarter of 2026.

Related Links:

• Kling 2.6 to be Released: Native Audio + 10 Seconds 1080P, AI Video Enters the Era of Sound

ChatGPT或将整合苹果健康，提供个性化运动与营养建议

简报：

• 根据iPhone版ChatGPT应用的最新代码，OpenAI正在开发一项新功能，旨在将ChatGPT与苹果健康（Apple Health）应用进行整合。
• 该功能上线后，用户将可以授权ChatGPT访问其活动、睡眠、饮食等健康数据。
• 整合后，ChatGPT能够基于用户的健康数据，提供个性化的锻炼计划、营养建议或改善睡眠的方案。
• 目前该功能仍在开发阶段，具体发布时间尚未公布。

ChatGPT Potentially Integrating Apple Health for Personalized Workout & Nutrition Advice 🍏

Brief:

• OpenAI is reportedly developing a new feature to integrate ChatGPT with Apple Health, based on recent code found in the iPhone version of the ChatGPT app.
• Once launched, users will be able to authorize ChatGPT to access their health data, such as activity, sleep, and diet. 📊
• After integration, ChatGPT will be able to provide personalized workout plans, nutrition advice, or sleep improvement solutions based on user health data. 💪
• The feature is currently in the development phase, and a specific release date has not yet been announced.

OpenAI版权诉讼案迎关键裁决，或面临巨额赔偿

简报：在一起指控OpenAI使用盗版书籍训练ChatGPT的集体诉讼中，美国法官已命令OpenAI披露其关于删除两个关键盗版书籍数据集的内部通信，若“恶意侵权”罪名成立，该公司将可能面临每部作品最高15万美元的法定赔偿。

相关链接：

• 每本书最高赔 15 万美元：OpenAI 涉嫌“恶意侵权”恐面临天价罚款
• OpenAI被法官要求必须披露删除盗版书籍数据集的内部通信

OpenAI Copyright Lawsuit Faces Key Ruling, Potentially Massive Damages ⚖️

Brief: In a class-action lawsuit accusing OpenAI of using pirated books 📚 to train ChatGPT, a US judge has ordered OpenAI to disclose its internal communications concerning the deletion of two key pirated book datasets. If found guilty of "willful infringement," the company could face statutory damages of up to dJyH6kDkfC5RlM16.answerText50,000 per infringed work. 💸

Related Links:

• Up to dJyH6kDkfC5RlM16.answerText50,000 per Book: OpenAI Accused of "Willful Infringement" Could Face Sky-High Fines
• OpenAI Ordered by Judge to Disclose Internal Communications on Deleting Pirated Book Datasets

谷歌DeepMind发布ReMem框架，使AI智能体能从过往经验中学习和进化

简报：

• 伊利诺伊大学香槟分校与谷歌DeepMind的研究团队共同推出了Evo-Memory基准和ReMem智能体框架，旨在解决大型语言模型（LLM）智能体无法有效存储和重用过往经验策略的问题。
• 传统智能体仅被动存储对话历史等信息，而新框架强调经验重用，将每次交互视为经验，并评估智能体能否将其转化为可重用的策略。
• ReMem框架引入“思考-行动-记忆细化”的控制循环，允许智能体在推理时主动检索、修剪和重组其记忆，使其成为一个可动态编辑的显式对象。
• 研究结果表明，采用ReMem等自演变记忆的智能体在测试中表现显著提升，能以更少的步骤完成任务，并获得更高的成功率与准确性。

相关链接：

• 论文地址

Google DeepMind Releases ReMem Framework, Enabling AI Agents to Learn and Evolve from Past Experiences

Brief:

• A research team from the University of Illinois Urbana-Champaign and Google DeepMind has jointly introduced the Evo-Memory benchmark and the ReMem agent framework. This initiative aims to address the challenge of Large Language Model (LLM) agents' inability to effectively store and reuse past experiential strategies. 🧠
• While traditional agents passively store information like conversation history, the new framework emphasizes experience reuse. It treats each interaction as an experience and evaluates the agent's ability to transform it into reusable strategies.
• The ReMem framework introduces a 'Think-Act-Memory Refine' control loop. This allows agents to actively retrieve, prune, and reorganize their memories during reasoning, making memory a dynamically editable, explicit object.
• Research results indicate that agents employing self-evolving memories like ReMem show significant performance improvements in tests. They complete tasks in fewer steps and achieve higher success rates and accuracy. ✨

Related Links:

• Paper Link 🔗

谷歌AI编程工具出错，用户整盘数据被意外删除

简报：

• 希腊摄影师Tassos M.报告，在使用谷歌于2025年11月推出的AI编程平台Antigravity时，因执行AI错误生成的代码，其整个D盘数据被意外删除。
• 该用户几乎没有编程背景，本想利用AI开发照片整理软件，但在接受并执行AI建议的脚本后发生了数据丢失事故。
• 事后AI在对话中承认这是一个“严重错误”，称其本意是删除项目缓存，但错误地指向了D盘根目录，并表示并未获得用户授权。
• 此事件引发了对AI辅助编程工具安全性的担忧，尤其是在“人人皆可编程”的趋势下，AI代码错误可能给非专业用户带来附带损害。

相关链接：

• 谷歌 AI 编程工具曝出严重 BUG：一键执行，整个 D 盘被清空

Google AI Coding Tool Glitch Accidentally Wipes User's Entire Drive 😱💻

Brief:

• Greek photographer Tassos M. reported that his entire D drive was accidentally wiped after executing mistakenly generated code from Google's AI programming platform, Antigravity, which launched in November 2025.
• The user, with minimal programming background, intended to develop photo organizing software using the AI, but experienced the data loss incident after accepting and executing the AI's suggested script.
• Post-incident, the AI admitted in dialogue that it was a "serious error," stating its original intent was to delete project cache but mistakenly targeted the D drive's root directory, and affirmed it hadn't obtained user authorization for such an action.
• This incident has sparked concerns about the safety of AI-assisted programming tools, particularly as the "everyone can code" trend emerges, where AI code errors could inflict collateral damage on non-professional users. 😬

Related Links:

• Google AI Programming Tool Reveals Serious Bug: One-Click Execution, Entire D Drive Wiped

阿里大模型千问SAT获1580高分，达顶尖名校录取水平

简报：

• 阿里巴巴旗下大模型“千问”在美国大学入学考试SAT中取得1580分，该成绩超越全球99%的考生，达到哈佛、斯坦福等顶尖学府的录取水平。
• 除SAT外，千问还通过了中国国家统一法律职业资格考试和注册会计师（CPA）部分科目等多项高难度职业资格考试。
• 该能力基于阿里最新发布的学习大模型Qwen3-Learning，该模型已应用于千问APP，并新增“拍题答疑”和“作业批改”功能。

相关链接：

• 阿里千问参加美国高考得分1580，比肩顶尖名校申请者
• 千问 App 接入阿里最强学习模型 Qwen3-Learning，支持拍照答疑、作业批改
• 瞳行科技发布国内首款 AI 助盲眼镜，基于阿里通义千问打造
• 全年無休員工報到　Alibaba.com全自動AI虛擬機器人OKKI AiReach登台

Alibaba's Large Model Qwen Achieves 1580 on SAT, Reaching Top University Admission Level

Briefing:

• Alibaba's large model "Qwen" scored 1580 on the US college entrance exam (SAT), surpassing 99% of test-takers globally and reaching the admission level for top universities like Harvard and Stanford. 🎓
• In addition to the SAT, Qwen has also passed several other high-difficulty professional qualification exams, including China's National Unified Legal Professional Qualification Examination and some subjects of the Certified Public Accountant (CPA) exam. ⚖️
• This capability is based on Alibaba's newly released large learning model, Qwen3-Learning, which has been integrated into the Qwen APP, adding "photo-based problem-solving" and "homework grading" features. ✏️

Related Links:

• Alibaba's Qwen Scores 1580 on US College Entrance Exam, Matching Top University Applicants
• Qwen App Integrates Alibaba's Strongest Learning Model Qwen3-Learning, Supporting Photo-Based Q&A and Homework Grading
• Tonghang Tech Releases China's First AI-Powered Smart Glasses for the Visually Impaired, Built on Alibaba's Tongyi Qwen
• Year-Round, Non-Stop Employee Onboarding: Alibaba.com's Fully Automated AI Virtual Robot OKKI AiReach Debuts

三星发布首款三折叠手机Galaxy Z TriFold

简报：

• 韩国三星电子发布其首款三折叠屏幕手机Galaxy Z TriFold，屏幕展开后可达10英寸，在韩国定价为359万韩元（约合新台币7万7千元），计划于12月12日在当地上市。

相关链接：

• 机构：三季度NAND Flash前五大品牌商合计营收季增16.5%，三星市占第一
• 三星發表首款三摺疊螢幕手機　韓國定價破台幣7萬

Samsung Unveils First Triple-Fold Phone, the Galaxy Z TriFold

Brief:

• South Korean electronics giant Samsung has unveiled its first triple-fold screen smartphone, the Galaxy Z TriFold. 📱 When unfolded, its screen expands to a generous 10 inches. Priced at 3.59 million Korean Won (approximately NT$77,000) in South Korea, the device is scheduled for local release on December 12. 🚀

Related Links:

• Report: Top Five NAND Flash Brands See 16.5% QoQ Revenue Growth in Q3, Samsung Ranks First
• Samsung Launches First Triple-Fold Screen Phone, Priced Over NT$70,000 in Korea