据南方都市报报道，今天（3月4日）凌晨，阿里Qwen技术负责人林俊旸在社交平台上与千问 道别，称"我退出了"。 有业内人士猜测， 林俊旸的离开或由于其考核目标变成了日活跃用户数。 对于林俊旸离开后 谁 会 接 任 其 位 置 ， 该 业 内 人 士 也 提 到 或 是 一 位 来 自 谷 歌 Gemini 团 队 、 以 日 活 跃 用 户 数 （DAU）指标为导向的非核心人才。南都记者就此向阿里云方面求证，截至发稿尚无回复。 林俊旸与月之暗面创始人兼CEO杨植麟、腾讯首席AI科学家姚顺雨和清华大学教授、智谱创 始人兼首席科学家唐杰被在业内并称"基模四杰"， 其他几位均为月之暗面、腾讯、智谱等企 业的模型掌门人。 今年1月在中关村，林俊旸还提到，自己2021年做模型的时候，阿里芯片团队的人来问自己三 年后模型主流架构是不是Transformer和多模态，因为三年后模型才能流片。他表示，"三年之 后 ， 我 在 不 在 阿 里 巴 巴 ， 我 都 不 知 道 。 但 最 后 我 今 天 还 在 阿 里 巴 巴 ， 然 后 他 果 然 还 是 Transformer，还是多模态，我就非常懊悔为什么当时没有去催他去 ...

Core Viewpoint - The sudden resignation of Junyang Lin, the technical leader of Alibaba's Qwen model, raises concerns about the future of the Qwen team and its projects, as multiple team members express their sadness and gratitude for his contributions [3][5][6]. Group 1: Resignation and Team Impact - Junyang Lin announced his resignation on social media, stating "me stepping down. bye my beloved qwen," which has led to a wave of farewell messages from other key team members [3][4]. - Team members, including Qwen Code lead Hu Bin and core contributors, expressed their gratitude and pride in working with Lin, indicating his significant role in the team's achievements [3][5]. - Lin's departure is seen as a major loss for the Qwen team, with one member stating they cannot imagine Qwen without him [3][5]. Group 2: Recent Developments and Achievements - Just days before his resignation, Lin was actively involved in discussions about new ideas and recruitment for the Qwen team, showcasing his commitment to ongoing projects [4][5]. - The Qwen 3.5 model was recently open-sourced, which received positive feedback from notable figures like Elon Musk, who praised its "impressive intelligence density" [5][6]. - The team successfully completed their tasks for the Qwen 3.5 version ahead of schedule, highlighting Lin's leadership and support in achieving project goals [5]. Group 3: Speculations and Industry Insights - Speculations suggest that Lin's resignation may be linked to external talent recruitment, the advancement of Qwen's commercial goals, and internal team adjustments, although Alibaba has not commented on these matters [7]. - Former Alibaba VP Jia Yangqing emphasized the challenges of balancing open-source efforts with commercial priorities, noting that the tech industry has seen both successes and failures in this regard [7]. - The importance of retaining innovative talent and fostering community engagement is highlighted as essential for companies in the current competitive landscape [7].

Group 1 - Samsonite (01910) fell over 5% as it announced plans for a dual listing in the US via American Depositary Shares (ADS), with new shares priced at a discount of no more than 15% from the last closing price, leading to an estimated net dilution impact of about 4.0% after accounting for treasury stock [1] - Alibaba-W (09988) dropped over 4%, reaching a new low of 128.5 HKD, with a cumulative decline of over 20% in the past month, following the resignation of the technical head of the Qwen team, which may be linked to organizational adjustments [1] - Dongfang Electric (01072) saw a significant increase of over 13%, attributed to a milestone order of 20 units of 50MW gas turbine generator sets from a Canadian client, with a unit price of 200 million RMB and a gross margin of 40-50% [1] Group 2 - China Shipbuilding Defense (00317) rose by 3.15% as analysts noted that escalating conflicts in the Middle East could boost global military spending, positively impacting the valuation of missile, drone, air defense systems, and shipbuilding industries [2] - Mistral (02440) surged over 32% at one point, as it announced the launch of the world's first Pokémon trading card tokenized fund, expected to be available by March 2, 2026 [2] - COSCO Shipping Energy (01138) experienced a sharp decline of 22%, with a drop of 11.48% reported, following a significant decrease in oil tanker traffic through a critical global energy chokepoint, down over 95% from normal levels [2] Group 3 - Airline stocks fell again, with Cathay Pacific (00293) down 4.58%, Eastern Airlines (00670) down 3.66%, and China Southern Airlines (01055) down 2.48%, due to airspace closures in the Middle East following military actions, disrupting global flight operations [3] - Domestic insurance stocks continued their recent downward trend, with China Life (02628) down 6.46% and China Pacific Insurance (02601) down 4.25%, as analysts noted a lack of positive earnings forecasts from major insurers [4] - Oil stocks collectively declined, with CNOOC (02883) down 7.92% and PetroChina (00857) down 4.51%, following unusual trading activity and warnings about the uncertainty in international oil prices [4] Group 4 - US airline stocks saw a general decline, with United Airlines (UAL.US) down 0.65% and American Airlines (AAL.US) down 0.48%, as conflicts in the Middle East disrupted global flights, with major airports in Dubai and Doha closed for several days [5] - Chip stocks also fell, with Intel (INTC.US) down 5.27% and TSMC (TSM.US) down 4.33%, reflecting broader market concerns [5] - Gold and silver stocks experienced significant drops, with Gold Fields (GFI.US) down 11.58% and AngloGold Ashanti (AU.US) down 10.4%, as spot gold prices fell below $5020 per ounce [6]

Core Viewpoint - The sudden resignation of Junyang Lin, a key figure behind the Qwen project, marks a significant turning point for the company and the AI industry, as it signals the end of an era for Qwen's leadership and innovation [2][12][42]. Group 1: Junyang Lin's Departure - Junyang Lin announced his resignation from the Qwen project, which took many by surprise, especially after he had recently promoted the Qwen 3.5 small model series [2][4][9]. - His departure is seen as a loss for Qwen, as he was regarded as the "core brain" behind the open-source model series, which has been pivotal in the competitive AI landscape [21][42]. - Lin's career trajectory at Alibaba has been remarkable, rising from a senior algorithm engineer to a P10-level executive in just six years, reflecting the intense competition for top AI talent [32][34]. Group 2: Qwen Project and Team Changes - The Qwen project recently introduced the Qwen 3.5 small model series, which includes models with parameters of 0.8B, 2B, 4B, and 9B, emphasizing improved intelligence with reduced computational requirements [6][7]. - Alongside Lin, two other researchers from the Qwen team also announced their departures, indicating potential instability within the project [9][44]. - Binyuan Hui, another key member of the Qwen team, also left, having contributed significantly to the development of foundational models and programming intelligence [46]. Group 3: Lin's Background and Achievements - Junyang Lin, born in 1993, is noted for being Alibaba's youngest P10-level technical executive, showcasing his exceptional capabilities in the AI field [17][18]. - His academic background is unique, having transitioned from computer science to linguistics, which allowed him to leverage insights from language translation to enhance AI model development [24][25]. - Lin's contributions to the Qwen project include leading the team to open-source the Qwen series, positioning it competitively against major players like GPT and Gemini [38][42]. Group 4: Future Implications - Lin's resignation is viewed as a critical juncture for the Qwen project, potentially affecting its future direction and innovation capabilities [42]. - The departures of key team members may lead to challenges in maintaining the momentum and competitive edge that Qwen has established in the AI market [9][44].

3月3日，马云与阿里、蚂蚁的核心管理层来到杭州云谷学校，与校长、老师们畅谈AI带来的挑战和机会。 阿里巴巴集团主席蔡崇信、CEO吴泳铭、风险委员会主席邵晓锋、电商事业群CEO蒋凡，蚂蚁集团董事长井贤栋和CEO韩歆毅一同参加交 流。 马云表示，AI时代已经快速到来，对社会的冲击超出想象，我们大家谁都没有做好足够的准备，但是对十几岁的孩子来说，他们最有改变的 希望和机会，所以这次来到云谷学校，目的就是把阿里巴巴最近对AI越来越清晰的洞察和老师们分享。 目前，AI的迭代以周计算，能力还在不断增长。马云说："我们大家之所以一起到云谷来，就是想告诉大家，这个变化会来得非常快，要迅 速做出改变，帮助孩子们从现在开始学会和AI共存，适应这个巨大的变化。" 阿里巴巴董事长蔡崇信说，AI时代，思辨能力很重要，思辨能力不是会问问题，而是问对问题。未来很多事情机器都可以做，但是人和机器 的沟通，人和人的沟通能力可能是以后最重要的能力之一。 蚂蚁集团董事长井贤栋则表示，AI应该用来帮人做琐碎和频繁的事，让人有更多的时间去发展我们的特质，发展审美、创意和想象力。在用 好AI的同时，要避免AI成为扔不掉的拐杖，还是要保留独立思考的能力。 ...

Core Insights - The article discusses the latest updates from xbench regarding various AI models, particularly focusing on the BabyVision benchmark and the competitive landscape among leading models [1][14]. Group 1: Model Performance and Rankings - The latest leaderboard updates show that Doubao-Seed-2.0-pro ranks first among domestic models with an average score of 69.2, significantly outperforming its competitors in terms of output token cost, which is only one-fourth of Gemini 3 Pro's cost [5]. - Qwen3.5-plus achieved a score of 65.6, marking a notable improvement of 10.6 points from its predecessor, indicating a shift in focus towards stability and cost-effectiveness in model performance [7]. - GLM-5 scored 65.0, reflecting a 4.2 point increase from GLM-4.7, while maintaining high inference efficiency [8][9]. Group 2: Benchmarking and Evaluation - The BabyVision benchmark, developed by xbench in collaboration with various AI companies and researchers, has been adopted by several new models, showcasing its relevance in the industry [14]. - Doubao-Seed-2.0-pro leads the BabyVision leaderboard with a score of 62.60%, demonstrating its strong capabilities in multimodal visual understanding tasks [12]. - The competitive landscape is evolving, with models increasingly focusing on real-world agent tasks rather than just single-point benchmarks [28]. Group 3: Technological Advancements - Seed2.0, launched by ByteDance, enhances visual perception and reasoning capabilities, significantly improving the processing of complex documents and multimedia content [29][30]. - Qwen3.5 incorporates a hybrid attention mechanism and a sparse architecture, allowing for efficient deployment and improved inference throughput [33]. - GLM-5 introduces advanced capabilities in automated code generation and complex system reconstruction, marking a significant evolution in AI model functionality [34].

Core Insights - The article discusses the emergence of GUI Agents as a new paradigm in human-computer interaction, driven by the development of multimodal large models [1] - It highlights the challenges in building high-availability, cross-platform GUI Agents in real-world environments, including issues with data collection and the need for long-term memory and multi-Agent collaboration [2] Group 1: Technical Developments - Alibaba's Tongyi Lab has open-sourced the Mobile-Agent-v3.5 framework and the underlying GUI-Owl-1.5 model family to address technical barriers in deploying native GUI models [2][6] - The GUI-Owl-1.5 model family has achieved leading test results on over 20 mainstream GUI benchmarks, enabling unified control across desktop, mobile, and browser platforms [6] Group 2: Architectural Design - The architecture of GUI-Owl-1.5 decouples execution and reasoning, supporting edge-cloud collaboration with two model variants: Instruct for rapid response and Thinking for complex tasks [9] - The system allows for external tool invocation and supports the Model Context Protocol (MCP) for complex calculations or database queries [9] Group 3: Core Technical Principles - The performance of GUI-Owl-1.5 is attributed to enhancements in data pipeline construction, internal logic restructuring, and reinforcement learning algorithms [10] - A hybrid data pipeline has been developed to address the challenges of long trajectory synthesis, utilizing multimodal models for high-resolution UI screenshot generation [12][15] Group 4: Reinforcement Learning Innovations - The MRPO (Multi-platform Reinforcement Policy Optimization) algorithm addresses engineering bottlenecks in multi-platform reinforcement learning, overcoming issues like gradient conflicts and training instability [20][22] - The algorithm employs an online oversampling mechanism to maintain on-policy assumptions while enhancing sample diversity [22][28] Group 5: Evaluation and Performance - The GUI-Owl-1.5 model family has been rigorously evaluated across various dimensions, establishing new state-of-the-art (SOTA) benchmarks in the open-source domain [34] - The model demonstrated significant performance improvements on mobile and PC platforms, with the MRPO approach yielding better results compared to mixed training [35] Group 6: Grounding and Tool Collaboration - The model excels in grounding capabilities, achieving high accuracy in visual positioning tasks without cropping tools, and further improving with a two-stage Zoom-In strategy [41] - In complex scenarios, the model showcases its ability to perform cross-application collaboration and long-term memory management, effectively executing multi-step workflows [46][49] Group 7: Conclusion - The release of Mobile-Agent-v3.5 provides a comprehensive technical reference for developing engineering-grade Agents capable of executing long processes across multiple platforms [51] - The project has made its model weights, Agent framework source code, and online demo available on GitHub for community engagement and technical exchange [52]

Group 1 - The head of Alibaba's Qwen AI division, Lin Junyang, has announced his resignation two days after the company released updated products [1] - Qwen's mobile application experienced significant growth, with monthly active users increasing from 31.05 million in January to 203 million in February, ranking third globally behind OpenAI's ChatGPT and ByteDance's Doubao app [1] - The surge in user numbers is attributed to aggressive marketing campaigns by Chinese tech giants during the Lunar New Year holidays [1]

Core Viewpoint - The sudden resignation of Lin Junyang, a key figure in Alibaba's AI development, marks a significant transition for the Qwen project and raises concerns about its future direction and management [2][5][7]. Group 1: Lin Junyang's Background and Contributions - Lin Junyang, born in 1993, was the youngest P10-level technical expert at Alibaba, with a strong academic background in computer science and linguistics from Peking University [2]. - He joined Alibaba's DAMO Academy in 2019 and became a significant contributor to the M6 model team, which was established as the foundation for Alibaba's general large model technology [3][4]. - Under his leadership, the Qwen series of open-source models was launched, overcoming critical challenges in reinforcement learning and multilingual capabilities [4]. Group 2: Organizational Changes and Future Implications - Following Lin's departure, there are speculations about a shift in Qwen's management structure, with reports of increased oversight from Alibaba Cloud's CEO and the introduction of a new leader from the Gemini team [5]. - The potential transition of Qwen models towards a closed-source approach has been suggested, indicating a possible shift in Alibaba's strategy to align more closely with commercial models seen in competitors like Google Cloud [5]. Group 3: Industry Reactions and Community Impact - Lin's resignation has been met with widespread acknowledgment in the industry, with peers recognizing his contributions to the open-source community and the advancement of AI technologies [7][12]. - Comments from industry figures highlight the importance of leadership in fostering community engagement and innovation, emphasizing that the mission and vision of leaders are crucial for the future of projects like Qwen [12][13].

Core Viewpoint - The 2026 Mobile World Congress (MWC) highlights the significant shift towards AI technology, with Chinese companies increasingly focusing on the European market for AI applications and hardware integration [3][4]. Group 1: Chinese Companies at MWC - Approximately 350 Chinese companies participated in MWC 2026, marking a 21% increase from 288 last year, making them the largest overseas contingent after the US and Spain [3]. - The presence of Chinese firms reflects a transition from traditional manufacturing advantages to a focus on AI capabilities and hardware innovation [3][4]. Group 2: AI as the Core Technology - AI emerged as the central theme of MWC 2026, with Qualcomm's CEO emphasizing that AI is transforming human-computer interaction, shifting the focus from applications to intelligent agents [4]. - Honor's Robot Phone features an active camera structure that responds to voice commands, showcasing the integration of AI into hardware design [5][6]. Group 3: Innovations in Mobile Imaging and AI Integration - Vivo introduced a 400mm telephoto lens and modular accessories aimed at transforming smartphones into professional-grade filming tools, highlighting the use of AI to enhance imaging capabilities [8]. - TCL and Lenovo showcased AI-driven products, emphasizing the integration of AI into hardware rather than just software, which is a distinctive approach of Chinese manufacturers [8]. Group 4: Expansion of Robotics and Wearable AI Products - Chinese robotics companies, such as AGIBOT, showcased a range of robots designed for various applications, indicating a strategic focus on B2B markets and European partnerships [11][13]. - Alibaba's AI glasses debuted at MWC, with plans for additional AI wearable products, reflecting a broader strategy to penetrate global markets beyond traditional communication devices [14].