单词学习：![[c9da17ed-415e-41e7-86af-099cf14a9e76.mp3]]

新闻播报：![[aa33ba06-1e0d-493c-809f-caee6eec0608.mp3]]

美团开源多模态大模型LongCat，多项测试超越闭源竞品

简报：

• 美团正式发布全新开源多模态大模型LongCat-Flash-Omni，该模型支持对文本、语音、图像、视频的实时融合处理。
• LongCat-Flash-Omni在发布时即在多项基准测试中超越了多个闭源竞品，实现了“开源即SOTA”（State-of-the-Art）。
• 模型采用创新的端到端统一架构，并借助Flash推理引擎，可在消费级GPU上实现近乎零延迟的流畅交互。
• 目前，该模型已在美团旗下平台免费开放，开发者可通过Hugging Face获取其权重。

Meituan Open-Sources Multimodal Large Model LongCat, Outperforming Closed-Source Competitors in Multiple Tests 🚀

Briefing:

• Meituan has officially released its new open-source multimodal large model, LongCat-Flash-Omni, which supports real-time integrated processing of text, speech, images, and video.
• Upon its release, LongCat-Flash-Omni immediately surpassed several closed-source competitors in multiple benchmark tests, achieving "open-source SOTA" (State-of-the-Art) 🤯.
• The model adopts an innovative end-to-end unified architecture, and leveraging the Flash inference engine, it can achieve nearly zero-latency smooth interaction on consumer-grade GPUs 💡.
• Currently, the model has been made freely available on Meituan's platforms, and developers can access its weights via Hugging Face.

苹果iOS 26.1更新：Apple Intelligence正式支持中文，Siri获重大升级

简报：

• 随着iOS 26.1正式版的更新，Apple Intelligence现已正式支持繁體中文等八种新语言，为iPhone、iPad及Mac带来包括即时翻译、书写工具、视觉智慧以及更自然灵活的Siri在内的多项深度整合AI功能，台湾大个人用户商務長林东闵评价称Siri已脱离“Silly阶段”。

相关链接：

• Apple Intelligence中文版正式降臨！專家：Siri脫離Silly階段了

Apple iOS 26.1 Update: Apple Intelligence Officially Supports Chinese, Siri Receives Major Upgrade

Briefing:

• With the official release of iOS 26.1, Apple Intelligence now officially supports eight new languages, including Traditional Chinese 🇨🇳. This update brings a host of deeply integrated AI features to iPhone, iPad, and Mac, including instant translation, writing tools, Visual Intelligence, and a more natural and flexible Siri. Lin Dong-Min, Business Unit Chief of Personal Users at Taiwan Mobile, commented that Siri has moved past its "silly phase" 🚀.

Related Link:

• Apple Intelligence Chinese Version Officially Arrives! Experts: Siri Has Moved Past Its Silly Phase

字节跳动发力具身智能，月薪最高12万招募人形机器人专家

简报：

• 字节跳动旗下火山引擎团队已启动高规格招聘，寻求“操作算法资深专家（具身智能）”，标志其正式进入人形具身机器人赛道。
• 该职位开出了每月9.5万元至12万元人民币的高薪，要求应聘者具备相关领域的硕士或博士学历。
• 岗位核心职责包括主导研发人形具身机器人的操作算法、参与具身大模型的预训练与部署，以及推动算法在硬件上的系统集成。
• 此举被行业视为字节跳动加入全球人形机器人研发竞赛的标志，意在与特斯拉、OpenAI等巨头竞争。

相关链接：

• 字节火山引擎启动高薪招聘，实质性进军“人形具身机器人”赛道

ByteDance Pushes into Embodied AI, Recruiting Humanoid Robot Experts with Salaries Up to ¥120,000/Month

Briefing:

• ByteDance's Volcano Engine team has launched a high-profile recruitment drive for a 'Senior Operations Algorithm Expert (Embodied AI),' signaling its formal entry into the humanoid embodied robot sector. 🤖
• The position offers a lucrative monthly salary ranging from RMB 95,000 to RMB 120,000, requiring applicants to hold a Master's or Ph.D. degree in a relevant field.
• Core responsibilities include leading the R&D of operational algorithms for humanoid embodied robots, participating in the pre-training and deployment of embodied large models, and driving the systemic integration of algorithms onto hardware platforms. 🧠
• This strategic move is widely regarded by the industry as ByteDance's entry into the global humanoid robot R&D race, positioning it to compete with tech giants such as Tesla and OpenAI. 🚀

Related Links:

• ByteDance's Volcano Engine Launches High-Paying Recruitment, Substantially Entering the 'Humanoid Embodied Robot' Sector

为训练擎天柱机器人，特斯拉让员工重复枯燥动作收集数据

简报：

• 根据报道，特斯拉在加州帕洛阿尔托设有一个秘密实验室，旨在通过收集大量人类行为数据来训练其Optimus（擎天柱）机器人。
• 实验室雇佣了数十名“数据收集者”，他们需佩戴含五个摄像头的头盔和重达40磅的背包，重复执行如举杯、擦桌子等日常任务，甚至包含跳舞等动作。
• 数据收集工作被描述为重复且枯燥，员工被要求在每8小时的班次中生成至少4小时的可用影像，如果动作被认为“不够人性化”，还可能面临处罚。
• 尽管特斯拉CEO马斯克计划在年底前生产5000台Optimus机器人，但其近期演示效果不佳，反应迟缓，常需依靠人工远程操控来完成展示。

Tesla Makes Employees Repeat Tedious Actions to Collect Data for Optimus Robot Training

Brief:

• According to reports, Tesla operates a secret lab in Palo Alto, California, dedicated to training its Optimus robots by collecting vast amounts of human behavioral data. 🤖
• The lab employs dozens of "data collectors" who must wear helmets equipped with five cameras and 40-pound backpacks. They repeatedly perform daily tasks like lifting cups, wiping tables, and even dancing. 🩰
• The data collection work is described as repetitive and tedious. Employees are required to generate at least four hours of usable footage during each 8-hour shift and could face penalties if their movements are deemed "not human-like enough." 😔
• Despite Tesla CEO Elon Musk's plan to produce 5,000 Optimus robots by the end of the year, recent demonstrations have been subpar, with slow reactions and frequent reliance on human teleoperation to complete tasks.

软银携手OpenAI成立合资公司，将推“晶睿智能”服务日本企业

简报：

• 软银集团与OpenAI于2025年11月5日宣布成立合资公司“SB OAI Japan GK”。
• 该合资公司将在日本市场独家推广名为“Crystal Intelligence”（晶睿智能）的企业级AI解决方案，预计2026年正式上市。
• “Crystal Intelligence”旨在通过集成的OpenAI技术与本地化服务，重塑日本企业的管理与运营。
• 软银将率先在内部部署并应用该技术，作为首家用户验证其效能。

相关链接：

• 软银与 OpenAI 宣布成立合资公司，明年推出企业级 AI 解决方案

SoftBank Partners with OpenAI to Form Joint Venture, Launching "Crystal Intelligence" for Japanese Businesses

Briefing:

• SoftBank Group and OpenAI announced on November 5, 2025, the establishment of their joint venture, "SB OAI Japan GK."
• The joint venture will exclusively promote an enterprise-grade AI solution named "Crystal Intelligence" in the Japanese market, with a formal launch expected in 2026. 🚀
• "Crystal Intelligence" aims to redefine management and operations for Japanese enterprises through integrated OpenAI technology and localized services. 💡
• SoftBank will be the first to deploy and apply this technology internally, serving as the initial user to validate its effectiveness. ✅

Related Links:

• SoftBank and OpenAI announce joint venture, to launch enterprise AI solution next year

微软 Azure 搭载英伟达芯片，AI 推理速度破百万 token

简报：

• 微软宣布其 Azure ND GB300 v6 虚拟机在运行 Meta Llama2 70B 模型时，推理速度达到每秒 110 万个 token，刷新了 AI 推理领域的行业纪录。
• 该虚拟机由英伟达 Blackwell Ultra GPU 提供动力，基于集成了 72 颗 GPU 和 36 颗 CPU 的 NVIDIA GB300 NVL72 系统。
• 此次性能结果已由独立机构 Signal65 认证，与上一代 NVIDIA GB200 相比，性能提升了 27%，而功耗仅增加 17%。
• 与更早的 NVIDIA H100 世代相比，其机架级推理性能实现了近 10 倍的提升。

相关链接：

• 每秒 110 万个 token！微软联手英伟达刷新 AI 推理纪录

Microsoft Azure, Powered by NVIDIA, Breaks Million-Token AI Inference Speed Barrier

Brief:

• Microsoft announced its Azure ND GB300 v6 virtual machine achieved an inference speed of 1.1 million tokens per second when running the Meta Llama2 70B model, setting a new industry record in AI inference. 🚀
• The virtual machine is powered by NVIDIA Blackwell Ultra GPUs, based on the NVIDIA GB300 NVL72 system, which integrates 72 GPUs and 36 CPUs.
• This performance result has been certified by independent agency Signal65, showing a 27% performance increase compared to the previous generation NVIDIA GB200, with only a 17% increase in power consumption. ✨
• Compared to the even earlier NVIDIA H100 generation, its rack-level inference performance has seen nearly a 10x improvement. 💡

Related Links:

• 1.1 Million Tokens Per Second! Microsoft Teams Up with NVIDIA to Break AI Inference Record

英伟达与德电信斥资10亿欧元，共建德国AI算力新高地

简报：

• 英伟达与德国电信宣布将共同投资10亿欧元，在德国建立一座大型数据中心，旨在增强欧洲的人工智能基础设施，以应对与美国和中国的技术竞争。
• 该项目将配备高达10,000颗图形处理器（GPU），预计将使德国的AI算力提升约50%，并由SAP提供技术支持。
• 数据中心预计在2026年第一季度投入运营，被认为是欧洲范围内规模最大的数据中心之一。
• 尽管投资巨大，但与美国动辄数十万颗GPU的数据中心计划相比，欧洲在AI基建方面仍有差距。

相关链接：

• 英伟达与德国电信携手投资 10 亿欧元建数据中心，提升德国 AI 算力 50%
• GTC25竟出現「假黃仁勳」直播　流量比輝達官方還高五倍！連YT都被騙
• 鸿海通过重磅 AI 算力建设预算，一年内最多投资 420 亿新台币

NVIDIA and Deutsche Telekom Invest €1 Billion to Build Germany's New AI Computing Hub

Brief:

• NVIDIA and Deutsche Telekom have announced a joint investment of €1 billion to build a large data center in Germany, aiming to strengthen Europe's artificial intelligence infrastructure and address technological competition with the US and China. 🌍
• The project will be equipped with up to 10,000 Graphics Processing Units (GPUs), expected to boost Germany's AI computing power by approximately 50%, with technical support provided by SAP.
• The data center is anticipated to become operational in the first quarter of 2026 and is considered one of the largest data centers within Europe. 🚀
• Despite the significant investment, Europe still lags behind in AI infrastructure compared to US data center projects, which often involve hundreds of thousands of GPUs.

Related Links:

• NVIDIA and Deutsche Telekom Join Forces in €1 Billion Data Center Investment, Boosting Germany's AI Computing Power by 50%
• A "Fake Jensen Huang" Appeared in a GTC25 Live Stream, Garnering Five Times More Traffic Than NVIDIA's Official Broadcast! Even YouTube Was Fooled
• Foxconn Approves Major AI Computing Power Construction Budget, Investing Up to NT$42 Billion Within One Year 💡