Core Viewpoint - The controversy surrounding the restaurant chain "Xibei" involves allegations of using pre-prepared dishes, with significant public discourse initiated by influencer Luo Yonghao, who criticized the high cost and prevalence of such dishes in Xibei's offerings [1][4]. Group 1: Company Response - Xibei's founder, Jia Guolong, responded to the allegations, asserting that the restaurant does not use pre-prepared dishes as defined by national standards, emphasizing that their food is prepared on-site and not pre-packaged [1][4]. - Jia expressed that the terms "pre-prepared" and "pre-prepared dishes" are distinct, clarifying that their cooking process involves pre-preparation but does not equate to serving pre-packaged meals [1]. Group 2: Industry Context - The pre-prepared dish industry is rapidly evolving, driven by the need for cost reduction and efficiency in the restaurant sector, with a reported 8% decrease in overall costs for restaurants utilizing pre-prepared dishes [4]. - Major chain restaurants, including Xibei, are reported to have a pre-prepared dish usage rate exceeding 80%, indicating a significant trend towards standardization and industrialization in the food service industry [4][5]. - The growth of food delivery platforms has further accelerated the shift towards pre-prepared meals, with estimates suggesting that 70% of delivery merchants utilized pre-packaged meals in 2022 [5].

Core Viewpoint - The recent controversy surrounding the restaurant chain Xibei highlights public dissatisfaction with the use of pre-prepared dishes in the food industry, emphasizing the need for transparency and potential legislative action regarding food labeling [1]. Group 1: Public Reaction - On September 10, a public figure criticized Xibei for serving mostly pre-prepared dishes at high prices, calling for legislation to require restaurants to disclose the use of such dishes [1]. - Following the criticism, Xibei's founder announced plans to sue the critic, indicating a defensive stance from the company [1]. - The critic noted that over 90% of online comments were negative towards Xibei, suggesting a significant public backlash against the restaurant's practices [1]. Group 2: Industry Implications - The incident has sparked discussions about the transparency of pre-prepared food in the restaurant industry, with calls for consumers to have the right to know what they are being served [1]. - The controversy may accelerate legislative efforts to regulate the use of pre-prepared dishes in restaurants, reflecting a growing consumer demand for clarity in food sourcing [1].

9月12日消息，自巨星传奇获悉，巨星传奇与宇树科技共同打造的"巨星狗"将于9月13日在上海旅游节开 幕式花车巡游上首次公开亮相。 据了解这只"巨星狗"是明星IP+潮玩+科技的融合体，具备陪伴功能与社交属性。 新浪科技独家获取"巨星狗"穿衣视频。 责任编辑：李思阳 ...

Core Insights - Ant Group launched the world's first trusted connection technology framework for smart glasses, named gPass, at the 2025 Inclusion Bund Conference, focusing on "security, interaction, and connectivity" [1][2] - The current AI glasses industry is still in the basic interaction stage, with significant gaps compared to smartphones, due to limited hardware capabilities and a fragmented software ecosystem [1] - gPass aims to address the urgent need for a unified, secure, and interconnected infrastructure to advance AI glasses from basic tools to personal intelligent assistants [1][2] Security Capabilities - gPass incorporates financial-grade security features, ensuring user information safety through trusted identity circulation, end-to-end encryption, and device authentication [2] - The technology has successfully adapted to both Android and RTOS chip-level security solutions [2] Interaction Features - gPass integrates biometric technologies such as voiceprint, iris, and fingerprint recognition for seamless user verification, enhancing reliability and security [2] Connectivity Functions - gPass intelligently recognizes user context, enabling instant and secure message exchanges with various smart devices, featuring cross-device synchronization and multi-modal reminders [2] - The technology has been applied to brands like Rokid, Xiaomi, Quark, and Thunderbird, with future applications in healthcare, travel, and cultural sectors [2]

责任编辑：江钰涵 专题：2025 Inclusion·外滩大会 新浪科技讯 9月12日上午消息，蚂蚁集团CEO韩歆毅在2025Inclusion·外滩大会上表示，旗下AI健康应用 AQ是蚂蚁推动医疗健康普惠创新的一场重要探索，"我们立志，五年内，让人人都拥有值得托付的AI健 康管家，让每个医生都拥有专属的AI助手"，韩歆毅说。据悉，蚂蚁医疗健康将在本次大会上发布AQ智 能体开放平台，同时也将成立医疗健康实验室，进行AI赋能MDT多学科会诊等前沿探索。（罗宁） ...

9月11日晚间，华与华营销咨询有限公司董事长华杉发微博表示，西贝是拼了老命要把菜做好的企业， 贾国龙又是极其爱惜自己羽毛的人，罗永浩这样的污蔑，他绝对受不了。 责任编辑：李思阳 近日，"西贝"陷入的预制菜舆论风波。 9月10日，罗永浩发博公开吐槽连锁餐饮品牌西贝："好久没吃西贝了，今天下飞机跟同事吃了一顿，发 现几乎全都是预制菜，还那么贵。希望国家尽早推动立法，强制饭馆注明是否用了预制菜。" 9月11日，西贝创始人贾国龙在罗永浩就餐的餐厅对外界多方面问题进行回应，并称"将要起诉罗永 浩。" ...

Group 1 - The core viewpoint of the article is that Hello has announced its Robotaxi strategic plan and the launch of its first mass-produced Robotaxi model "HR1," aiming for large-scale deployment by 2026 and a fleet of over 50,000 vehicles by 2027 [1][2] - Hello plans to cover more than 10 cities with its Robotaxi service and achieve a scale of 10,000 vehicles in the first international city [1] - The company has signed strategic cooperation agreements with partners such as Qichen Automobile, Alibaba Cloud, Horizon Robotics, and Hesai to advance Robotaxi technology development and commercialization [1] Group 2 - Hello's Robotaxi employs an end-to-end integrated technology architecture, which enhances system response speed and decision-making consistency by mapping multi-sensor inputs directly to vehicle control commands [1] - To support this advanced architecture, Hello is building a technology foundation based on "big data, big computing power, and big models," aiming to create a high-quality dataset of no less than 10 million clips by 2026 [2] - The collaboration with Alibaba Cloud will establish a computing cluster capable of handling tens of thousands of vehicles for training and iterating autonomous driving models, utilizing the "Daoyu" model with over 10 billion parameters [2]

新浪科技讯 9月12日上午消息，支付宝在2025 Inclusion·外滩大会宣布推出国内首个"AI付"服务，面向AI 时代为智能体提供支付服务，并率先在瑞幸咖啡的AI点单助手"Lucky AI"上线，用户可在瑞幸支付宝小 程序或瑞幸咖啡App，用说话的方式完成下单并支付。这也是行业首次打通智能体内的下单与支付全链 路，实现无缝的AI服务体验。 以点咖啡为例，此前用户通过AI智能体下单，仍需跳转至支付页面再予手动确认。现在，用户在瑞幸 支付宝小程序或瑞幸App唤起"Lucky AI"，不仅可以说句话点咖啡，还可以直接说句"下单"，完成身份 核验后即支付。整个过程用户无需离开AI对话界面，像日常和店员聊天一样简单自然。（罗宁） 责任编辑：江钰涵 专题：2025 Inclusion·外滩大会 ...

Group 1: AI Technology and Investment - The investment and innovation in artificial intelligence (AI) have seen strong growth, with AI companies receiving approximately $90 billion in venture capital in Q2 2025 [2] - A McKinsey survey indicates that 92% of executives plan to increase AI investments over the next three years [2] - Major companies like Meta, Amazon, Alphabet, and Microsoft plan to invest $325 billion in AI infrastructure in 2025, a 46% increase from 2024 [2] Group 2: Organizational Transformation for AI - Successful AI transformation must be led by CEOs and driven by business needs rather than IT departments, focusing on profit rather than just application scenarios [2] - Organizations need to undergo restructuring rather than simple optimization, breaking down barriers and inertia to achieve real breakthroughs in "change management" [2] Group 3: Global Trade and Geopolitical Risks - Geopolitical instability and conflicts are identified as the top potential risks to global economic growth in the next 12 months, with trade policy changes also posing significant risks [3] - The proportion of global exports to GDP has fluctuated below 25% since 2010, halting a 50-year trend of continuous growth [3] Group 4: China's Globalization Stages - China's globalization process is categorized into three stages: 1.0 relying on low-cost manufacturing, 2.0 driven by overseas acquisitions, and 3.0 focusing on sustainable development as global corporate citizens [4] - Chinese companies need to shift from pure export transactions to establishing a global perspective, moving towards outputting intellectual property, expertise, and capabilities [4]

新浪科技讯 9月12日上午消息，在2025Inclusion·外滩大会上，蚂蚁集团与中国人民大学联合发布业界首 个原生MoE架构的扩散语言模型（dLLM）"LLaDA-MoE"。 责任编辑：江钰涵 实现数据显示，LLaDA-MoE模型性能效果在代码、数学、Agent等任务上领先于LLaDA1.0/1.5和Dream- 7B等扩散语言模型，接近或超越了自回归模型 Qwen2.5-3B-Instruct，仅激活 1.4B 参数即可实现等效3B 稠密模型的性能。 据了解，蚂蚁与人大团队攻关 3个月，在LLaDA-1.0基础上重写训练代码，并且基于蚂蚁自研分布式框 架 ATorch提供EP 并行等一系列并行加速技术，基于蚂蚁Ling2.0基础模型的训练数据，在负载均衡、噪 声采样漂移等核心难题上取得突破，最终采用 7B-A1B（总7B、激活1.4B）的MOE架构完成约 20T 数 据的高效训练。（罗宁） 据介绍，这款新模型通过非自回归的掩码扩散机制，首次通过原生训练的MoE在大规模语言模型中实现 了与Qwen2.5相当的语言智能（如上下文学习、指令遵循、代码和数学推理等），挑战了"语言模型必 须自回归"的主流认知。 ...