一封警告信，将医疗器械企业康泰医学（300869.SZ）推到了合规风险的风口浪尖。 10月10日，康泰医学公告披露，于10月2日收到FDA（美国食品药品监督管理局）出具的警告信。核心 问题是，今年6月9日至12日，FDA对其位于河北秦皇岛的生产基地进行现场检查后认定，公司出口美国 的医疗器械产品在生产、包装、储存或安装环节，不符合美国联邦法规21 CFR Part 820医疗器械质量体 系规范（QSR）。 FDA此次行动基于6月的现场检查结果。这类检查通常针对已获得FDA注册资质的医疗器械企业，重点 核查其是否持续符合质量体系规范。 21 CFR Part 820作为核查的根本依据，是多数医疗器械在美国上市之前必须遵守、上市之后随时可能抽 查的基本要求，涵盖了从产品设计开发、生产过程控制、包装标签管理到储存运输的全链条。任何一个 环节的疏漏，诸如生产记录不完整、质量检测流程不规范、储存环境不达标等，都可能触发警告信。 基 于21 CFR Part 820的核查逻辑及警告解除流程示意图，来源：AI制图 目前，康泰医学公告未披露具体违规细节，也暂未公布具体整改计划。 从市场准入受限的实际影响来看，"拒绝产品进入美 ...

文 | 大模型之家 作为英国前首相，苏纳克还能选择"跳槽"美国AI公司，但对于欧洲本土AI产业而言，或许就没那么幸运 了…… 当地时间2025年10月9日，前英国首相里希·苏纳克（Rishi Sunak）正式宣布接受微软与人工智能初创公 司Anthropic的兼职高级顾问（Senior Advisor）职位。两家公司的声明与媒体报道同步披露：苏纳克的 职位为"内部、兼职"性质，并承诺不会在两年内为这些公司向政府游说，他表示将把报酬捐给其创办的 慈善项目。 昔日唐宁街10号的前主人，一位在任期内将AI安全提升至国家战略高度，并倾尽全力在布莱切利园举 办全球首届AI安全峰会、试图为英国抢占全球AI治理"盟主"地位的政治家，在卸下公职不过短短数月 后，便悄然转身，投入了美国科技资本的怀抱。 从唐宁街到硅谷：一场令人不安的"背书" 虽然英国商业任命咨询委员会（ACOBA）为苏纳克的这一任命套上了"枷锁"，规定他必须遵守限制， 包括在离任两年内不得代表两家公司游说英国政府，也不能利用其在政府任职期间获得的任何特权信 息。 但在欧洲本土的从业者与学者眼中，这更像是一场令人不安的"背书"。对外界而言，这并不仅是个人去 向的 ...

Core Insights - The trend of Chinese AI technology expanding into Southeast Asia has become prominent, with significant advancements in large models and intelligent agents driving this process [2] - The educational sector has emerged as a primary area for the application of AI technology, addressing challenges such as teacher shortages and resource imbalances in the region [3][4] Group 1: Technology Characteristics and Output Efficiency - The differences in AI development stages between China and Southeast Asia create a complementary relationship, facilitating accelerated technology output [2] - The efficiency of technology output has improved by 3-5 times, with system deployment time reduced from six months to two months due to the capabilities of large models [2] - The strategic implementation of "R&D in major cities, integration in Guangxi, and application in ASEAN" is shortening the technology landing path [2] Group 2: Educational Sector as a Priority Area - Southeast Asia's K12 education faces significant challenges, such as a teacher-student ratio of 1:40 in Indonesia, which is 2.5 times that of China, necessitating the addition of 4.5 million teachers by 2030 [3] - The low coverage of computer classrooms and the multilingual nature of the region exacerbate educational resource fragmentation, making AI real-time translation systems valuable for enhancing classroom interaction [3] - The structural contradictions in the education system are evident, with Indonesia lacking 120,000 science teachers and Vietnam facing a shortage of 5,000 AI-related faculty [3] Group 3: Collaborative Foundations and Trust - The long-standing educational exchange between China and Southeast Asia fosters a unique trust for technology output, with Chinese students making up 57% of international students in Malaysia [4] - The need for programming courses in Malaysia is high, with 300,000 university students requiring such training but only 2,000 qualified computer teachers available [5] Group 4: Systematic Cooperation and Policy Support - The collaboration between China and Southeast Asia has evolved from isolated applications to systematic outputs, with Tsinghua University's "Y-type education system" being implemented in Thailand [6] - Malaysia's National AI Roadmap 2021-2025 prioritizes "intelligent education," planning to invest 230 million MYR in a national education data platform [6] - Local government demand for educational solutions is driving technology transfer, with 21% of the AI scene demand list in Nanning focused on education [6] Group 5: Innovative Approaches to Technology Transfer - Chinese companies are adopting a different approach to technology transfer in Southeast Asia, focusing on empowering local teams rather than merely providing services [7] - The collaboration model includes a three-tier technology transfer system, enabling local engineers to independently iterate on AI tools [7][9] Group 6: Talent Development and Sustainable Growth - AI education cooperation is not a one-way output but aims to cultivate a shared talent pool, with plans to establish 10 AI joint laboratories in ASEAN by 2025-2030 [12] - The value of graduates involved in AI education projects is significantly higher, with starting salaries 40% above traditional computer science graduates [13] - The establishment of a national qualification framework for AI education in Malaysia reflects a trend towards co-building regional talent standards [13] Group 7: Business Sustainability and Value Creation - The core of business sustainability lies in value sharing, with Chinese companies transitioning from "technology providers" to "ecosystem builders" [14] - The profitability of AI education initiatives is linked to the value created for local partners, emphasizing a long-term collaborative approach [14]

10月11日消息，滴滴自动驾驶今天宣布完成D轮融资，本轮融资总额20亿元。 刚刚，中国最大出行平台滴滴的自动驾驶部门获新一轮融资，北京三大AI产业基金全面加持。 据悉，滴滴出行从2016年组建自动驾驶技术研发部门，是国内较早布局自动驾驶的企业之一。2019年8 月，滴滴宣布将自动驾驶部门升级为独立公司"滴滴沃芽"，新公司全面发力Robotaxi，专注于自动驾驶 研发、产品应用及相关业务拓展。 过去六年，滴滴自动驾驶加速获得融资和技术产品落地，投资方包括国际零部件巨头法雷奥集团和日本 软银愿景基金等。2024年10月，滴滴自动驾驶完成2.98亿美元C轮融资，投资方包括广汽集团领投、滴 滴出行参投。 实际上，随着 AI 大模型等新技术加速落地，AI 与自动驾驶的深度融合，已成为中国汽车产业发展的核 心方向，也是全球各大汽车制造商重点攻坚的核心技术"高地"。 一般情况下，自动驾驶功能需依靠车辆搭载的摄像头、毫米波雷达、激光雷达等传感器获取道路信息， 通过车载计算平台集成融合成以车身为中心的路况"鸟瞰图"，车辆的自动驾驶算法会以此"推理"出相应 行驶路径。相关数据在脱敏后也会通过互联网上传到云计算平台，"喂"给AI大 ...

Core Viewpoint - Dongpeng Beverage (Group) Co., Ltd. has submitted its H-share listing application for the second time, indicating its strong ambition to go public in Hong Kong despite previous setbacks [2][3]. Financial Performance - In 2024, the company achieved a revenue of 15.83 billion yuan, a year-on-year increase of 40.6%, and a net profit of 3.33 billion yuan, up 63.1% [2]. - For the first half of 2025, Dongpeng reported a revenue of 10.737 billion yuan, a growth of 36.37%, and a net profit of 2.375 billion yuan, an increase of 37.22% [2]. - Analysts predict that Dongpeng's total revenue for 2025 will exceed 20 billion yuan for the first time [2]. Market Position and Challenges - Dongpeng Beverage is a leading player in the Chinese functional beverage market, with a market capitalization of 160 billion yuan [2]. - The company has faced challenges in its Hong Kong listing process, including regulatory requirements related to foreign investment and data protection [3][6]. - The company's revenue is heavily concentrated in the energy drink segment, which accounted for over 80% of its income in recent years [5]. Product Structure and Growth - Dongpeng's core product, Dongpeng Special Drink, has seen its revenue contribution decline from 96.6% in 2022 to 84% in 2024, indicating a need for diversification [5]. - The gross profit margin has improved from 41.6% in 2022 to 44.1% in 2024, and further to 44.4% in the first half of 2025 [5]. International Expansion - Dongpeng has initiated efforts to explore overseas markets, particularly in Southeast Asia, by establishing a Hong Kong subsidiary and planning to build production bases [8][10]. - The company aims to cater to local tastes in Southeast Asia and has partnered with local distributors to enhance its market presence [10][11]. - Dongpeng is also investing in new production facilities, including a 1.2 billion yuan base in Hainan and a planned $200 million factory in Indonesia [10][11]. Strategic Goals - The Hong Kong IPO is seen as a strategic move to enhance capital strength, improve international brand image, and address the company's reliance on a single product category [10][12]. - The company recognizes the long-term impact of international market expansion on its overall strategy, despite the challenges it faces in competing with established brands like Red Bull and Monster [11][12].

Core Insights - The article explains the concept of Webhook in simple terms, comparing it to a "doorbell" for systems to notify each other in real-time, eliminating the need for constant polling [2][10][12]. Group 1: Understanding Webhook - Webhook is described as a "reverse" API that allows systems to send notifications to each other without the need for constant inquiries [10][12]. - The traditional API method requires users to actively check for updates, which is inefficient and resource-consuming [6][7]. - Webhook simplifies this process by allowing systems to push notifications when specific events occur, such as payment confirmations [12][14]. Group 2: Installation and Functionality - Setting up a Webhook involves three main steps: providing a Callback URL, specifying the events to subscribe to, and handling incoming notifications [17][20][23]. - The Callback URL acts as the "address" where notifications will be sent, and it must be configured in the system that will send the notifications [18][19]. - The system sends an HTTP POST request containing a Payload with relevant information when an event occurs [24][26]. Group 3: Common Pitfalls - Security is a major concern, as the Webhook URL is publicly accessible, making it vulnerable to unauthorized requests [29][30]. - Implementing signature verification is crucial to ensure that notifications are legitimate and from trusted sources [33][35]. - Handling duplicate notifications is necessary to prevent processing the same event multiple times, which can lead to errors [39][40]. Group 4: Practical Implementation - The article provides a step-by-step guide for setting up a Webhook receiver using Python and Flask, including code examples [26][50][56]. - It emphasizes the importance of using tools like Ngrok to expose local servers to the internet for testing purposes [62][63]. - Postman is recommended for sending test requests to verify the Webhook functionality [70][73]. Group 5: Automation with n8n - The article concludes by demonstrating how to integrate Webhook functionality into n8n for automated workflows, allowing for seamless communication between systems [75][88]. - It highlights the shift from a "pull" model to a "push" model in system interactions, enhancing efficiency and responsiveness [85].

文 | 逻辑学家 今年的Token2049恰逢十一假期，逛展之余也有了更多思索沉淀的时间。会展火爆一如往年，作为一名 有着深厚安全基因的从业者，为市场繁荣感到欣喜，也还是会被层出不穷的安全事件影响，思考如何构 筑更安全、更稳健的行业未来。这份思考既来自展会见闻，也源于团队在人工智能与数字资产一线的实 践与探索。遂成此文，谨供诸位参考探讨。 "国家级黑客"：数字资产安全新战场 这一态势催生了数字资产领域的"高级持续性威胁"（Advanced Persistent Threat）概念。与传统网络安全 中的APT相比，数字资产领域的APT具有三个更严峻的特征：其一，利害关系更直接，攻击目标直接锁 定可即时转移的巨额金融资产，攻击"投产比"极高；其二，攻击链条更短平，一旦私钥失守或合约被攻 破，资产瞬间流失，响应时间窗口极短；其三，攻击手法高度定制化，专门针对高净值个人、企业高管 进行长期、精准的社会工程学攻击，深度融合人性弱点与技术漏洞。 在实践中，AI与智能体技术有能力从个人到国家、从技术到运营形成立体化防护体系，构建数字资产 领域的"智能体军团"。 在个人层面，AI智能体扮演着"数字保镖"的角色。它能7x24小 ...

【中国白酒网】今年中秋、国庆再次揭露白酒线上线下"冰火两重天"现状：线下动销持续"冷淡"， 线上却格外热闹，美团闪购平台1499元的飞天茅台秒空，节前销量8倍增长。 这截然不同的市场表现背后，是即时零售正从电商平台的"独角戏"，演变为全行业参与的 "重头 戏"。从茅台与淘宝闪购达成合作、i茅台上线"即时配送"服务，到十余家白酒企业联合美团闪购推出即 时零售行业首个全链路保真体系，酒企正加速"上车"。 数据显示，酒类即时零售渗透率虽从2023年的1%起步，但业内预计2027年将飙升至6%，规模有望 冲击千亿级别。然而，这场渠道革命在拓宽竞争赛道的同时，也埋下了价格失控、经销商利润不断蒸发 的隐患，一场关于效率、利益与生态的博弈已然拉开序幕。毕竟在这场冒险实验背后也有不少人担心经 销商沦为平台的"送货工具人"。 01、现状：平台主导，多元模式共舞 即时零售以"小时达""分钟达"的极致履约效率，重构了零售逻辑。在白酒领域，这股浪潮已从早期 的垂类探索，发展为平台型、垂类型、仓店一体型等多模式并存的竞争格局，其中综合电商平台凭借资 源优势占据主导地位。 2014年酒仙网率先试水，推出"酒快到"平台，当年便喊出了"19 ...

不同于石油等自然资源，稀土产业的技术壁垒在于加工。从选矿到分离稀土氧化物，再到冶炼分离单一 金属及成品加工，每一步都有极高技术含量。稀土最终产品的纯度对永磁体等下游材料性能起决定性作 用。以当前热门的具身智能和新能源车为例，电动车的主驱动电机、车门雨刮器电机，以及机器人关节 电机，其内部永磁体的核心要素都是稀土。 进一步收紧稀土管制并非仅因双边经贸摩擦。20世纪80年代末，《外交季刊》发表《科学技术与国家主 权》一文，提出"科学、技术、经济和政治等各方面变化过快，使国家主权与实力构成的屏障开始改 变"。此后，"技术主权"逐渐成为"国家主权"概念集合中的新成员。在全球技术主权竞争日益白热化的 当下，实现战略领域核心技术与产业链自主掌控、捍卫"技术主权"，具有事关未来生存发展的重大战略 意义，而稀土正是中国最具代表性的"技术主权"之一。 稀土出口管制升级：技术主权再强化！10月9日，商务部发布第61号和第62号公告，对稀土全产业链相 关技术出口实施审批监管，形成更广泛的管制闭环。公告显示，前者将含中国成分的境外稀土物项纳入 许可管理，填补了"转口用于军事敏感领域"的漏洞；后者全面管控稀土全产业链技术出口，从开采工 ...

【钛媒体综合】中央网信办、国家发展改革委近日联合印发《政务领域人工智能大模型部署应用指 引》，为各级政务部门提供人工智能大模型部署应用的工作导向和基本参照。 《指引》强调场景牵引。政务部门可围绕政务服务、社会治理、机关办公和辅助决策等工作中的共性、 高频需求，因地制宜、结合实际，选择典型场景进行人工智能大模型探索应用。 《指引》强调规范部署。政务部门应根据不同政务场景需求与现有技术基础，审慎选择人工智能大模型 实施路径。应以统筹集约的方式开展政务领域人工智能大模型部署，地市应在省（自治区、直辖市）统 一要求下开展部署应用，县级及以下原则上应复用上级的智能算力和模型资源开展应用和服务。应探索 构建"一地建设、多地多部门复用"的集约化部署模式，统筹推进政务大模型部署应用，防止形成"模型 孤岛"。应加强政务数据治理，持续提升数据质量，支撑政务大模型的优化训练。 钛媒摘声： AI芯片的核心技术瓶颈在于数据传输成本过高，必须尽量避免数据传输，用计算来替代 它。现在TPU和GPU都采用这个设计思路，它们会做大量运算，同时尽量减少数据传输，但 这种做法也有极限，当想追求更高的并行规模时，效率会不升反降，甚至最终成为性能瓶 ...