在吴泳铭"AI驱动"的核心战略下，阿里正进一步将顶尖人才向AI基础模型研发的核心战场集结，而多模态交互则被视为下一阶段AI突破的关键隘口。 作者 | 小 猫 编辑 | 硬 AI 正值全球科技巨头在人工智能领域展开激烈军备竞赛之际，阿里巴巴内部的排兵布阵再次出现关键变动。 硬AI获悉，近期备受瞩目的AI顶尖科学家、阿里集团副总裁许主洪（Steven Hoi）已从智能信息事业群首席科学家的职位上，转岗至阿里集团的核心AI研发机构 ——通义实验室。 阿里方面向硬AI证实了这一消息，并表示许主洪将负责多模态交互模型方向的研究，后续向通义实验室负责人、阿里云CTO周靖人汇报。 这一内部调动释放出重要信号：在吴泳铭"AI驱动"的核心战略下，阿里正进一步将顶尖人才向AI基础模型研发的核心战场集结，而多模态交互则被视为下一阶段AI 突破的关键隘口。 对于许主洪而言，这次转岗意味着他将从更贴近C端应用的"前线阵地"转向更为核心和 底层的"研发心脏"。 时间拉回至今年2月，这位在学术界和工业界均享有盛誉的AI大牛（IEEE Fellow、斯坦福大学评选的"全球前1%的AI科学家"）正式加盟阿里，在当时引发了业内不 小的轰动。他最 ...

Core Insights - The future of image models is expected to evolve similarly to LLMs, transitioning from creative tools to information retrieval tools [4] - Multi-modal interaction will be crucial, focusing on understanding user intent and adapting to various interaction modes [4][20] - The integration of "world knowledge" from LLMs into image models is a significant application direction for enhancing user assistance [14] Group 1: Trends and Developments - Image models are anticipated to become more proactive and intelligent, capable of using text and images flexibly based on user queries [4][14] - Users' expectations for instant, high-quality outputs from models are often unrealistic, highlighting the need for iterative processes [18] - The design of user interfaces (UI) for model products is currently undervalued, with a need for better integration of various modalities to enhance usability [4][18] Group 2: User Interaction and Experience - The "blank canvas dilemma" is a significant challenge, necessitating clear communication of what actions are possible within the interface [5][20] - Simplifying operations for ordinary users is essential, with a focus on visual guidance and examples to facilitate understanding [17] - Social sharing plays a key role in overcoming the "blank canvas dilemma," as users are inspired by others' creations [17] Group 3: Model Evaluation and Aesthetics - User feedback is critical for evaluating model performance, with a focus on aesthetic quality and meeting user needs [21][22] - Meeting aesthetic demands is challenging and requires deep personalization to provide useful suggestions [26] - The future may see a shift towards more personalized models, but current expectations are likely to remain at the prompt level [27] Group 4: Future Directions and Integration - The development of "Omni Models" that can handle multiple tasks is a likely trend, with shared technologies between image and video models [40] - Traditional tools and AI models are expected to coexist, with each serving different user needs based on the complexity of tasks [35][37] - The integration of AI into existing workflows, such as enhancing presentation tools, is a promising area for future development [38]

Core Insights - The 2025 International Automotive Intelligent Cockpit Conference was held in Suzhou, focusing on AI-enabled cockpit innovation and the future ecosystem of human-vehicle interaction [1] - The conference featured discussions on key technologies, international standards, and talent cultivation in the intelligent cockpit sector, with participation from 800 experts and representatives [1][3] Industry Development - The global automotive industry is undergoing a historic transformation driven by technology and innovation, with China's intelligent cockpit sector leading due to its technological and market advantages [3] - The future development of intelligent cockpits requires a strong foundation in core technologies like cockpit models and high-performance chips, emphasizing user-centric design and international collaboration [3] Regional Insights - Jiangsu Province, as a major automotive industry cluster, has established a complete innovation system in areas like vehicle-mounted chips and intelligent cockpit solutions, positioning Suzhou as a key player in the intelligent connected vehicle sector [3] - The establishment of the Yangtze River Delta Technology Exchange Center is a result of strategic cooperation between the China Automotive Engineering Society and the Suzhou government, aimed at enhancing regional automotive industry development [4] Research Findings - The average score for intelligent cockpit evaluations reached 6.78, with most models scoring above 6, indicating significant advancements in technology and consumer experience [5] - The report on the intelligent cockpit standard system aims to establish a framework by 2026 and to lead global standards by 2035, filling gaps in key technology standards [5] Technological Innovations - The integration of technologies like Tesla's FSD and China's "vehicle-road-cloud" model is highlighted, with a focus on cost-effective solutions and the need for clearer business models in the intelligent connected vehicle sector [6] - New network security solutions, such as the Multi-Identification Network (MIN), are proposed to enhance data security and privacy in intelligent cockpits [7] User Experience and Interaction - The concept of "happy space" is introduced by Li Auto, emphasizing the role of intelligent cockpits in differentiating automotive brands through advanced interaction systems [8] - Companies like Zebra Zhixing are focusing on user value-driven transformations in intelligent cockpits, leveraging AI to create personalized user experiences [8] Future Directions - Unity's advancements in 3D real-time rendering technology are set to enhance intelligent cockpit functionalities, including music visualization and interactive navigation [9] - The industry is moving towards a more human-centered approach in cockpit design, exploring new applications like in-car gaming and meditation spaces [9]

9月16日，华为举办智能世界2035系列报告发布会。 华为常务董事汪涛发表了"探索未知，跃见未来"的主题演讲，正式发布智能世界2035系列报告，包括《智能 世界2035》和《全球数智化指数2025》报告两大研究成果，展望了未来十年的关键技术趋势以及这些技术对 教育、医疗、金融、制造、电力等行业带来的改变和影响，并帮助全球各国量化数智化发展进程。 趋势一： AGI将是未来十年最具变革性的驱动力量 ，但仍需克服诸多核心挑战，方能实现AGI奇点突破。因 此，走向物理世界是AGI形成的必由之路。 趋势二： 随着大模型的发展，AI智能体将从执行工具演进为决策伙伴，驱动产业革命。 趋势三： 开发模式迎来变革， 人机协同编程成为主流 。人类将更专注于顶层设计和创新思考，而把繁琐的编 码执行工作，交给高效的AI来完成。 趋势四： 交互方式正从图形界面转向自然语言，并向着融合人类五感的多模态交互演进。用户通过语音、手 势等方式与数字世界互动，获得深度沉浸的体验。 趋势五： 手机App正从独立的功能实体，转变为由AI智能体驱动的服务节点。用户只需给出指令，AI智能体 将调用相关服务节点，为用户提供极致体验。 趋势六： 随着世界模 ...

趋势二：随着大模型的发展，AI智能体将从执行工具演进为决策伙伴，驱动产业革命。 趋势三：开发模式迎来变革，人机协同编程成为主流。人类将更专注于顶层设计和创新思考，而把繁琐 的编码执行工作，交给高效的AI来完成。 据华为官微消息，9月16日，华为举办智能世界2035系列报告发布会。正式发布智能世界2035系列报 告，包括《智能世界2035》和《全球数智化指数2025》报告两大研究成果。 其中，《智能世界2035》系列报告，详细解读了通往智能世界2035的十大技术趋势： 趋势一：AGI将是未来十年最具变革性的驱动力量，但仍需克服诸多核心挑战，方能实现AGI奇点突 破。因此，走向物理世界是AGI形成的必由之路。 趋势五：手机App正从独立的功能实体，转变为由AI智能体驱动的服务节点。用户只需给出指令，AI智 能体将调用相关服务节点，为用户提供极致体验。 趋势六：随着世界模型等关键技术突破，全新的L4+自动驾驶汽车将会走入人们的生活，成为"移动第 三空间"。 趋势七：2035年全社会的算力总量将增长10万倍，计算领域将突破传统冯• 诺依曼架构的束缚，在计 算架构、材料器件、工程工艺、计算范式四大核心层面实现颠覆性创新 ...

Core Insights - Huawei released the "Intelligent World 2035" series report, which includes key technology trends and their impacts on various industries over the next decade [1][3] Group 1: Key Technology Trends - Trend 1: Artificial General Intelligence (AGI) is expected to be the most transformative force in the next decade, but significant challenges must be overcome to achieve AGI breakthroughs [3] - Trend 2: AI agents will evolve from execution tools to decision-making partners, driving an industrial revolution [3] - Trend 3: Development models will transform, with human-AI collaborative programming becoming mainstream, allowing humans to focus on top-level design and innovation [3] - Trend 4: Interaction methods will shift from graphical interfaces to natural language and multi-modal interactions, enhancing user experience [3] - Trend 5: Mobile apps will transition from standalone functions to AI-driven service nodes, providing users with optimized experiences [4] Group 2: Future Projections - Trend 6: L4+ autonomous vehicles will become part of daily life, creating a "mobile third space" [4] - Trend 7: By 2035, total computing power is projected to increase by 100,000 times, leading to disruptive innovations in computing architecture and paradigms [4] - Trend 8: Data will become the "new fuel" for AI development, with storage capacity needs expected to grow 500 times by 2025, accounting for over 70% of AI storage [4][5] - Trend 9: The number of connected objects will expand from 9 billion people to 900 billion intelligent agents, marking a shift from mobile internet to intelligent agent internet [5] - Trend 10: Energy will be a critical factor for AI development, with renewable energy expected to surpass 50% of total energy generation by 2035 [5] Group 3: Societal Impact - By 2035, AI is predicted to help prevent over 80% of chronic diseases, shifting health management from passive treatment to proactive prevention [6] - Over 90% of Chinese households are expected to have smart robots, leading to immersive technological transformations in home environments [6] - AI-driven autonomous decision-making organizations will reshape production paradigms, with AI application rates exceeding 85% and productivity improvements of 60% [6] - The global digital economy is being invigorated by the technological revolution, with over 70 countries releasing AI strategies [6][7] Group 4: Global Digital Intelligence Index (GDII) - The GDII framework maps traditional economic factors to the digital world, focusing on data, ICT talent, and digital production tools as core elements [7] - The model includes key indicators such as data scale, network connectivity, computing power, and ICT skills, aimed at providing quantitative references for national digital economic development [7] - Huawei aims to collaborate with global partners to leverage opportunities in the digital economy and contribute to a better intelligent world [7]

Core Viewpoint - The auxiliary driving function in the smart car industry is losing its appeal due to regulatory restrictions and technological limitations, prompting companies to seek new strategies for growth and innovation [1][2][8] Group 1: Challenges in Auxiliary Driving - The ban on L2/L2+ systems on certain highways highlights the regulatory push against auxiliary driving, which is seen as a response to safety concerns following accidents linked to these technologies [1][2] - The limitations of current auxiliary driving systems have been exposed, as they struggle to recognize stationary vehicles and other complex scenarios, leading to a loss of consumer trust [2][5] - New regulations require clearer communication from companies regarding the limitations of auxiliary driving features, moving away from misleading marketing tactics [2][8] Group 2: Technological Evolution - The shift from auxiliary driving to a focus on multi-modal interaction represents a significant evolution in how vehicles understand their environment, enhancing safety and decision-making capabilities [4][7] - AI models are being developed to improve the vehicle's ability to predict and respond to various driving scenarios, significantly enhancing safety measures [5][7] - The integration of high-quality data into AI training is crucial for overcoming the challenges faced by auxiliary driving systems, particularly in recognizing unconventional stationary objects [7][8] Group 3: Market Dynamics and Future Directions - The industry is transitioning from a focus on flashy features to a more holistic approach that emphasizes safety and ecosystem integration, driven by new regulations [8][9] - Companies are encouraged to build trust with consumers through transparency and real-time updates on system capabilities, which can lead to increased usage of auxiliary driving features [8][9] - The future of smart vehicles lies in their ability to function as part of a broader urban efficiency infrastructure, transforming the role of car manufacturers into operators of transportation efficiency [9]

Core Insights - The emergence of AI-generated content is revolutionizing the video production landscape, as demonstrated by the short film "Kira," which was created with minimal cost and time using various AI tools [2][4][6] - The rapid growth of user engagement and revenue in AI video generation platforms, particularly Kuaishou's Keling, indicates a significant shift in the industry towards AI-assisted content creation [8][17][27] Group 1: AI Video Generation - The short film "Kira" was produced for only $500 and gained significant viewership on platforms like YouTube and Bilibili, showcasing the potential of AI in content creation [2][4] - Hashem AI-Ghaili, the creator of "Kira," utilized multiple AI tools for scriptwriting, image processing, video editing, and sound design, highlighting the collaborative capabilities of AI technologies [4][6] - Keling, a video generation model by Kuaishou, reported an annual recurring revenue (ARR) exceeding $100 million, surpassing competitors like MiniMax, which projected $70 million for 2024 [7][17] Group 2: User Growth and Market Dynamics - Keling's user base grew from 6 million to over 45 million within a year, indicating a strong market demand for AI video generation tools [15][40] - The introduction of features like "multi-image reference" and "motion brush" in Keling has significantly improved user experience and content quality, leading to increased user retention and satisfaction [11][15][28] - The competitive landscape is intensifying, with companies like ByteDance and Google entering the market, indicating a broader acceptance and investment in AI video generation technologies [23][43] Group 3: Technological Advancements - Keling's development of a multi-modal visual language (MVL) allows users to interact with the model using various inputs, enhancing the creative process [15][38] - The introduction of features aimed at improving controllability and consistency in video generation, such as "first and last frame" functionality, has been well-received by creators [11][35] - The industry is witnessing a shift from skepticism to embracing AI tools, as evidenced by the integration of AI in traditional media workflows and the emergence of new job roles related to AI content creation [42][43]

Core Viewpoint - The article discusses the rapid evolution and commercialization of AI-generated video content, highlighting the success of creators like Hashem AI-Ghaili and the advancements in video generation technology, particularly through the company KuaLing, which has achieved significant user growth and revenue in a competitive landscape [6][11][12]. Group 1: AI Video Generation Success - Hashem AI-Ghaili created the short film "Kira" using multiple AI tools, costing only $500 and taking 12 days to produce, contrasting with traditional high-budget productions [6][7]. - KuaLing's annual revenue surpassed $100 million as of March 2023, with user numbers growing from 6 million to 4.5 million in a short span, indicating strong market demand [11][20]. - The video generation sector is experiencing rapid growth, with KuaLing outperforming competitors like MiniMax and Tencent in user acquisition and revenue generation [12][22]. Group 2: Technological Advancements - KuaLing has introduced several innovative features in its video generation models, such as "first and last frame" functionality, which enhances the coherence of generated videos [14][46]. - The introduction of multi-modal interaction capabilities allows users to upload images and videos as references, significantly improving the controllability and quality of the generated content [19][50]. - The company has successfully integrated user feedback into its product development, leading to significant improvements in user experience and satisfaction [47][58]. Group 3: Market Dynamics and Competition - The competitive landscape for AI video generation is intensifying, with new entrants like ByteDance's Jimo and Luma AI rapidly gaining traction [25][26]. - KuaLing's market share in video generation tools is substantial, but maintaining this position will require continuous innovation and adaptation to user needs [23][25]. - The industry is witnessing a shift in perception, with AI tools being embraced as valuable assets rather than threats, leading to the emergence of new job roles focused on AI content creation [61][62]. Group 4: Future Directions - KuaLing plans to explore the development of AI agents to automate the video creation process, further lowering barriers for users and enhancing creative workflows [66]. - The company envisions a future where AI-generated content not only serves existing media formats but also creates new, interactive content forms [68].

Core Concept - The article discusses the development of AI digital assistants that enhance human-computer interaction by simulating human communication, aiming to provide natural and efficient service support in daily scenarios [1] Natural Language Interaction System - The dialogue interface utilizes multi-turn conversation technology, enabling context semantic understanding and intent recognition. Users can input requests via text or voice, with the system automatically correcting and completing key information [2] - The response module is designed to express human-like responses, matching emojis and tone words to the conversation content to avoid mechanical replies [2] Task Management and Scheduling - The digital assistant can parse complex user requests and break them down into executable steps. For example, if a user inputs "prepare for a weekend family gathering," the system generates a shopping list, venue setup suggestions, and a schedule [4] - The scheduling module synchronizes with the user's mobile calendar, setting reminders and detecting conflicts, automatically suggesting adjustments when overlapping events are detected [4] Preference Model and Service Recommendations - Based on historical dialogue data, the digital assistant can proactively push relevant services. For instance, if a user frequently inquires about fitness plans, the system will regularly send workout tutorials and dietary suggestions [5] - Recommended content spans various categories, including lifestyle services, learning resources, and entertainment activities, with each recommendation accompanied by a brief description and action entry [5] Multimodal Interaction Expansion - In addition to basic text interaction, the digital assistant supports simple gesture recognition and emotional feedback. Users can express satisfaction through a thumbs-up gesture, which the system records to enhance similar recommendations [6] - The visual presentation adopts a 2.5D cartoon style to avoid discomfort from excessive realism, maintaining a consistent hairstyle and outfit for brand recognition while reducing cognitive load [6] Privacy Protection and Permission Management - Dialogue data is secured with end-to-end encryption, allowing users to choose data retention periods. The permission settings page offers detailed control options, such as allowing calendar access while prohibiting contact list access [7] - Sensitive operations require secondary verification, such as entering a preset password or biometric information to modify schedule arrangements [7] Visual Standards and Adaptation Optimization - The interface design adheres to brand color standards, primarily using a light blue color scheme to create a technological feel. Key operation buttons are sized no less than 44px to ensure accurate touch response across different devices [8] - Animation frame rates are maintained above 30fps to prevent lag during interactions. Testing shows that optimized versions have reduced the error rate by 40% among elderly users [8] - Through the collaborative operation of these functional modules, the AI digital assistant can establish a complete link of "demand understanding - task breakdown - service push," balancing technical advancement with emotional value to provide users with an efficient and warm digital assistance experience [8]