Core Insights - The article discusses the evolution and challenges of bidding algorithms in real-time bidding (RTB) advertising systems, emphasizing the transition from traditional methods to advanced techniques like generative reinforcement learning [2][3][7]. Group 1: Evolution of Bidding Algorithms - The bidding algorithm has evolved through three generations: PID, MPC, and reinforcement learning (RL), each improving upon the previous in terms of adaptability and effectiveness in complex bidding environments [5][6][7]. - The introduction of generative reinforcement learning aims to enhance decision-making by utilizing historical bidding sequences for more accurate predictions [8][10]. Group 2: Challenges in Bidding - Key challenges faced by bidding algorithms include the need to manage daily budgets while minimizing conversion costs, the unpredictability of traffic and competitor behavior, and the complexity of sequential decision-making [5][6]. - The reliance on high-quality datasets poses a challenge, as simple exploration can lead to out-of-distribution (OOD) issues, necessitating efficient offline exploration mechanisms [12][14]. Group 3: GAVE Algorithm - The GAVE algorithm integrates score-based return-to-go (RTG) and value function-based action exploration to enhance model learning and address the challenges of data quality and exploration [18][19]. - Experimental results show that GAVE outperforms baseline algorithms in various budget settings, demonstrating its effectiveness in maximizing conversion value [22][25]. Group 4: CBD Algorithm - The CBD algorithm introduces Completer and Aligner modules to improve the alignment of generated sequences with optimization goals, addressing issues of sequence legality and preference alignment [29][31]. - Offline experiments indicate that CBD significantly outperforms other methods in total conversion value, validating its effectiveness in real-world applications [34][36]. Group 5: Future Directions - Future advancements in bidding technology are expected to focus on developing foundational models that leverage multi-scenario data and enhancing interpretability and decision-making capabilities through the integration of large language models [41].

整理 | 傅宇琪、褚杏娟 985 管培生吐槽每天拧螺丝 12 小时，车间实习期长达半年；深圳疆拓国庆前补班遭员工投诉，公司 反手调整放假制度，取消 14 天年假福利和所有额外假期；博世计划大规模裁员：规模或达数万人， 为省钱"别无选择"；特朗普签署行政令批准 TikTok 在美继续运营：字节跳动仍 100% 持股；800 美元 Meta 智能眼镜首秀发布会"翻车"，扎克伯格甩锅"网太差"；小米 17 系列开售 5 分钟刷新 2025 国产手 机首销全天销量纪录；谷歌高管出席高通骁龙 2025 峰会，暗示正在开发"安卓电脑"；OpenAI 甲骨文 软银扩大"星际之门"：投 4000 亿美元再建 5 座数据中心；马斯克旗下 xAI 向美国联邦政府提供 Grok 聊天机器人，定价仅 42 美分…… 行业热点 周鸿祎称有理由裁掉不用 AI 的员工 9 月 24 日，罗永浩与周鸿祎深度对谈。谈及 AI 在工作中的结合，周鸿祎表示，现在在给员工建立 一种认知，用了效果不好也要咬着牙坚持用。他提到，360 内部正在举办 AI 大赛，虽然目前没有因 为运用 AI 而大规模裁员，如果员工在被要求使用 AI 后仍然拒绝使用，那么 ...

Core Viewpoint - The article highlights the recent announcement of the partner team at Zhiyuan Robotics, showcasing the backgrounds and expertise of key members, which positions the company strongly in the AI and robotics industry [2][3]. Group 1: Company Leadership - Deng Taihua is the founder, chairman, and CEO of Zhiyuan Robotics, with over 20 years of experience at Huawei, where he led the development of the Kunpeng and Ascend AI computing ecosystems [2]. - Peng Zhihui, also known as "Zhi Hui Jun," is the co-founder, president, and CTO, who joined Huawei in 2020 and focused on Ascend AI chips and algorithms before establishing Zhiyuan Robotics in 2023 [2]. - Jiang Qingsong serves as a partner and senior vice president, responsible for marketing and service systems, with over 20 years of experience in the ICT sector, including roles at Huawei and Alibaba Cloud [2][3]. Group 2: Key Team Members - Yao Maoqing, partner and senior vice president, has a background in autonomous driving at Waymo and NIO, focusing on AI technology development and software R&D at Zhiyuan Robotics [3]. - Wang Chuang, partner and senior vice president, was part of the founding team at DJI's LiDAR product line and later served as CTO at Maher Innovation [3]. - Luo Jianlan, partner and chief scientist, has experience at GoogleX and Google DeepMind, contributing to significant advancements in robotic reinforcement learning [3].

Core Insights - The logistics industry, often seen as less attractive, is experiencing a transformation with the rise of intelligent logistics, emphasizing the need for seamless integration between cognitive decision-making and execution capabilities [3][4][6] - JD Logistics has introduced "Super Brain 2.0" and the "Wolf Pack" series, marking a shift from modular to integrated intelligence, creating a self-evolving intelligent ecosystem that can reshape the future of supply chain logistics [4][9] Group 1: Intelligent Core - Traditional logistics systems relied on passive decision support, but "Super Brain 2.0" has evolved into an active decision-making expert capable of real-time responses to complex physical world challenges [6][7] - The architecture of "Super Brain 2.0" is based on an Agentic framework, shifting from problem-driven to demand-driven decision-making, allowing for proactive optimization solutions [7][8] Group 2: Technological Innovations - JD Logistics has developed a digital twin logistics network for real-time monitoring and feedback, enabling continuous system iteration and self-evolution [8][9] - The system's multimodal fusion capability allows it to process complex, non-standard information and generate actionable instructions for execution terminals, enhancing operational efficiency [8][10] Group 3: Embodied Intelligence - The "Wolf" series, particularly "Embodied Wolves," addresses the long-standing challenge of automating the handling of non-standard packages, transforming logistics execution from mere tools to intelligent agents [10][12] - "Embodied Wolves" utilize advanced multimodal perception and decision-making systems to adaptively manage diverse package types, significantly improving operational accuracy and efficiency [11][12] Group 4: Competitive Landscape - JD Logistics stands out in the logistics sector by integrating AI with the entire supply chain, contrasting with competitors like SF Express, which focus on optimizing specific decision-making processes [14][19] - The "Super Brain + Wolf Pack" system exemplifies a comprehensive approach to logistics automation, enhancing efficiency across warehousing, sorting, transportation, and delivery [14][15] Group 5: Future Vision - JD Logistics aims to create a technology-driven, open, and sustainable logistics ecosystem, breaking down barriers between cognition and action to facilitate intelligent transformation across the supply chain [19][20] - The successful implementation of the "Super Brain + Wolf Pack" system is expected to yield significant cost reductions and efficiency gains, reinforcing JD Logistics' competitive edge in the market [19][20]

Core Viewpoint - The article discusses the current challenges and opportunities in the AI application market, emphasizing the need for effective connections between technology and business solutions, akin to how platforms like Dazhong Dianping (大众点评) helped consumers find suitable restaurants [4][6][9]. Group 1: Current AI Market Landscape - The AI application market is compared to the restaurant market a decade ago, highlighting the issue of information asymmetry [6][7]. - Despite a significant increase in AI-related projects on platforms like GitHub, with numbers rising from under 700,000 in 2020 to 1.81 million in 2023, only 25% of companies believe they have successfully implemented AI projects [8][9]. - The gap between technological advancements and commercial application is identified as a critical missing link in the current AI ecosystem [9]. Group 2: AI Infrastructure and Development - Alibaba Cloud announced major advancements in AI infrastructure, aiming to create a "super AI cloud," with the adoption rate of generative AI in China projected to rise from 8% in 2024 to 43% in 2025 [10][11]. - The need for application-level growth is emphasized, as foundational technologies are now mature [11]. Group 3: Challenges in AI Application Implementation - The AI Super Exchange hosted by Moduli Factory aims to address three main barriers to AI application deployment: unclear demand, lack of visibility for solutions, and inefficient matching between demand and supply [15][18]. - The exchange features a demand diagnosis platform, a real-time display of application features, and a matchmaking process for proposals and collaborations [17][18]. Group 4: Industry-Specific Solutions - Seven applications presented at the AI Super Exchange target specific industry pain points, such as: - Cloud operation automation, addressing the need for proactive maintenance in the industrial AI market, projected to reach $43.6 billion by 2024 [20][21]. - Intelligent bidding assistants that significantly reduce the time and error rates in the bidding process [26][28]. - AI-driven human resources solutions that shorten recruitment cycles and improve talent matching [30][31]. - Content creation tools that enhance efficiency for new media creators [34][36]. - Automation tools for repetitive office tasks, freeing up time for knowledge workers [37][38]. Group 5: Commercialization of AI Applications - Moduli Factory serves as an accelerator for AI application commercialization, providing exposure, user feedback, and industry connections to developers [44][49]. - The platform aims to bridge the gap between technology demos and commercial products, addressing the fact that 46.3% of companies are still hesitant to adopt AI due to a lack of suitable solutions [53][54]. Group 6: Developer Ecosystem and Future Opportunities - The "Autumn Competition" initiated by Moduli Factory is designed to create a self-reinforcing ecosystem for developers, offering support from model vendors, cost optimization, and guidance on sustainable business models [57][58]. - The article concludes by highlighting the historical opportunity for AI application developers to participate in this evolving landscape, as the focus shifts from technological breakthroughs to practical application [61][62].

Core Viewpoint - Microsoft is deepening its collaboration with Anthropic, integrating its AI models into the Copilot assistant, marking a significant shift away from its previous exclusive partnership with OpenAI [2]. Group 1: Partnership Developments - Starting from this week, Microsoft will incorporate Anthropic's AI models into its Copilot, which previously relied primarily on OpenAI's technology [2]. - On September 25, Microsoft CEO Satya Nadella announced this new partnership on the X platform [2]. - This agreement signifies a gradual "decoupling" from OpenAI, as Microsoft recently signed another agreement to apply Anthropic's AI technology in Office 365 applications like Word, Excel, and Outlook [2]. Group 2: AI Model Capabilities - The integration allows Copilot's commercial users to choose between OpenAI's deep reasoning models and Anthropic's Claude Opus 4.1 and Claude Sonnet 4 models for specific tasks [2]. - Claude Opus 4.1 is designed for complex reasoning, code writing, and deep architecture planning, while Claude Sonnet 4 is more suited for routine development tasks, large-scale data processing, and content generation [3]. Group 3: Industry Implications - Users have noted the significance of integrating both Claude and ChatGPT into a single enterprise platform, highlighting the importance of having multiple options for different tasks [3]. - The integration is seen as a challenge to the notion of a single optimal model in the AI field, indicating that the real competition in AI is just beginning [3].

Core Insights - JD.com is integrating AI technology into its ecosystem through a new app called "Ta Ta Ta," aiming to create a super application that combines various services and AI content communities [2][3][4]. Group 1: Product Features - The "Ta Ta Ta" app officially launched after a public beta in May, featuring a digital assistant, AI social circles, and smart hardware connectivity [3][4]. - The app demonstrates a strong integration with JD.com's internal services, functioning as a prototype for a super app [5]. - Users can interact with the digital assistant, which remembers previous conversations and provides tailored recommendations, such as medical advice and food delivery options [5][7][9]. Group 2: AI Capabilities - The app includes an "AI Circle" feature, allowing users to create personalized AI entities for interaction, and offers a "soul matching" function for one-on-one AI interactions [10]. - The digital assistant connects various services, including JD.com’s food delivery, health, and financial services, enhancing its utility as a comprehensive agent [12]. - The underlying technology, Joy AI, is noted for its superior reasoning capabilities, with models ranging from 3 billion to 750 billion parameters [13]. Group 3: Future Prospects - JD.com plans to expand the app's capabilities by potentially allowing external agents to join, enriching the functionality [13]. - The app aims to bridge online and offline experiences, enabling users to control physical devices through the app, fostering a two-way data sharing environment [13]. - JD.com envisions the "Ta Ta Ta" app as a key entry point in its strategy to build a trillion-dollar AI ecosystem over the next three years [17].

作者 | 棱镜 AI 浪潮席卷之下，企业技术领袖们无不摩拳擦掌，渴望将这些颠覆性技术融入自身的业务核心，抢占智能时代的制高点，不料却被现实狠狠地甩下一记 耳光。 PoC 时的惊艳还历历在目——自动报告生成、智能应答客服、代码辅助开发，一切都看起来那么完美。然而，当试图将这些能力嵌入核心业务系统时， 医疗团队发现，AI 助手会面不改色地编造根本不存在的药物方案；金融机构意识到，风控模型可能依据过时的条款做出百万级的误判；连最简单的客服 场景中，AI 都能把用户引导向一个早已下架的产品。这并非由于某个技术事故，而是生成式 AI 存在已久的幻觉问题。 清华大学计算机系教授指出，大模型在垂直领域知识与实时更新上是有局限的，特别是幻觉问题，已经成为大模型深入企业级应用的掣肘。因此，行业 迫切需要一种既保留大模型生成能力，又能对其输出进行确定性约束的方案。 9 月 25 日，由李国良教授作为技术顾问的数智引航团队，正式发布向量数据库 VexDB，能够支持百亿千维向量数据毫秒级查询，召回准确度高达 99% 以上，从数据基础设施层面为 AI 应用构建一个可信的知识基石。近日，在国际权威的 DABSTEP 非结构化数据分析测试 ...

Core Viewpoint - The article discusses the release of the Code World Model (CWM) by Meta, which aims to enhance code generation capabilities by integrating a deeper understanding of code execution, addressing the limitations of previous models that could generate syntactically correct code but failed in execution [4][10]. Group 1: Model Overview - CWM is the first open-source code world model with 32 billion parameters, designed to advance code generation research based on world models [4][5]. - Unlike traditional models that rely on static code training, CWM incorporates dynamic interaction data from Python interpreters and Docker environments to improve its understanding and reasoning about code [7][14]. - The model can simulate the step-by-step execution of code, understanding how variables change and what feedback the program receives [7][10]. Group 2: Performance Metrics - CWM achieved a score of 65.8% on the SWE-bench Verified task, outperforming all other open-source models of similar size and nearing GPT-4 levels [8]. - It scored 68.6% on LiveCodeBench, 96.6% on Math-500, and 76.0% on AIME 2024, showcasing its strong performance across various benchmarks [8]. Group 3: Training Methodology - The training of CWM involved three key phases: pre-training, mid-training, and post-training, utilizing supervised fine-tuning (SFT) and reinforcement learning (RL) [15][16]. - The model was pre-trained on 8 trillion tokens, followed by mid-training on code world modeling data with an additional 5 trillion tokens, enhancing its contextual understanding [15][16]. Group 4: Industry Context and Implications - The release of CWM marks a significant step in Meta's AI strategy, especially following the restructuring of its AI business [5][23]. - The model's development reflects a shift towards balancing open-source initiatives with commercial interests, as Meta navigates its AI strategy amidst organizational changes [26].

Core Insights - The 2025 Yunqi Conference will feature over 110 sessions and nearly 900 topics, attracting more than 2,000 global speakers, focusing on AI-driven technology foundations, expanding AI application scenarios, and reshaping productivity and collaboration in the AI era [2] Group 1: AI Super Exchange - The AI Super Exchange will debut as a significant highlight of the conference, representing an "industry-level supply and demand revolution" akin to a stock exchange for AI products [2] - The event will showcase cutting-edge AI applications, creative ideas, and technology sharing, facilitating face-to-face interactions among AI developers [2] Group 2: Event Details - The AI Super Exchange forum will take place on September 25, from 10:00 to 12:00, at the Yunqi Town, Hall 3 [3] - Participants can initiate demand for AI applications, whether clear industry needs or imaginative concepts, which will be displayed in real-time as dynamic bullet screens [3] Group 3: Engagement Opportunities - Attendees are encouraged to visit the Moli Workshop's booth at the conference to engage in interactive activities and receive limited-edition gifts [13]