安全与体验双升级 重构汽车价值链交互逻辑 技术落地多点突破 算力、模型、场景全面赋能 在技术应用与产业实践环节，论坛嘉宾们结合企业案例，展示了AI在汽车研发、生产、产品等全链条 的落地成果。百度智能云副总裁、汽车业务部总经理高果荣详细介绍了百度在VLA（多模态智能驾 驶）领域的技术布局：依托百度百舸5.0平台与昆仑芯超节点，实现VLA多模态训练效率大幅提升；通 过45万公里高精地图与数据合成技术，显著降低数据标注成本；借助文心大模型与全流程数据闭环工具 链，构建从生成到仿真的一体化支持体系。"我们将以更高效的算力、更丰富的数据、更先进的模型与 工具，与行业伙伴共同加速智能辅助驾驶技术的跃升。"高果荣表示。 吉利控股集团首席数字官姚滨晖，从三大维度解析AI价值。在产品智能层面，聚焦辅助驾驶、智能座 舱等核心场景，提升车辆智能化体验；在企业智能层面，将AI应用贯穿研产供销服全业务链及战略、 人力等中后台环节，优化经营效率；在行业智能层面，通过旗下工业互联网平台"广域铭岛"，将内部验 证成熟的AI实践对外输出，赋能全行业智能化升级。 针对当前AI应用中的痛点，长城汽车（601633）技术中心副总经理荣雪东坦言，车企 ...

Core Insights - The forum highlighted the theme of "Car-Cloud Collaboration Driving the Leap in Intelligent Assisted Driving Technology," emphasizing the role of AI and cloud computing in the automotive industry [1] - Experts agreed that AI is driving a deep restructuring of the industrial value chain, from reshaping smart cockpit experiences to enhancing efficiency across the entire R&D, production, and marketing chain [1] Group 1: Strategic Integration of AI in Automotive - The deep integration of AI with the automotive industry is becoming a key driver of industry transformation, enhancing China's global competitiveness in the automotive sector [2] - Three integration strategies were proposed: 1. Car-cloud collaboration as the core path for AI and automotive integration, expanding new service segments including data, computing power, models, and simulations [2] 2. The automotive industry should enhance its understanding and application of AI technologies, particularly in intelligent driving, necessitating a reassessment of technology strategies [2] 3. Automotive companies should accelerate their transformation into AI-driven tech companies, capable of developing and producing various intelligent terminal products [2] Group 2: Trends in Competition and Data Utilization - Competition is shifting from single-point technology comparisons to "system efficiency + ecological collaboration," requiring the integration of internal and external resources to enhance user experience [3] - Data has evolved from being an "important resource" to a "core competitive advantage," with computing power being essential for unlocking data value, indicating a sustained high growth in data reliance and computing needs over the next two years [3] Group 3: AI Empowerment in R&D and Industry Applications - AI is driving industry implementation from point solutions to comprehensive applications, with advancements in multi-modal training and significant improvements in training efficiency through platforms like Baidu's [4] - The use of high-precision maps and data synthesis technology has significantly reduced labeling costs and improved efficiency [4] - Baidu's integration of large models and complete data closed-loop toolchains supports a seamless transition from generation to simulation [4] Group 4: AI Value Dimensions - AI's value can be categorized into three dimensions: 1. Product intelligence, enhancing vehicle smart features like assisted driving and smart cockpits [5] 2. Enterprise intelligence, covering all business activities related to company operations, including management and support functions [5] 3. Industry intelligence, leveraging AI practices to empower the entire industry through commercialized outputs [5] Group 5: Challenges and Future Directions - Current AI applications in R&D face challenges such as "tool silos," "data breakpoints," and "disconnected processes," limiting their effectiveness [6] - Future efforts will focus on transitioning from "technology-driven" to "business value-driven" approaches, integrating AI with simulation to enhance design iterations [6] - AI must evolve from being an optional enhancement to an indispensable asset in the automotive industry [6] Group 6: Safety and User Experience Transformation - AI is not only enhancing R&D but also transforming automotive safety systems and user experiences, with companies addressing regulatory compliance and cybersecurity challenges [7] - The establishment of vehicle security operation centers and AI-enabled log analysis has significantly improved alert processing efficiency [7] - The evolution of in-car voice interaction is moving towards an end-to-end processing model, enhancing the naturalness and efficiency of user interactions [7] Group 7: Implementation Framework for AI in Automotive - The integration of AI and the automotive industry is essential for industry development, relying on the establishment of car-cloud collaboration mechanisms, deep application of AI technologies, and the technological transformation of automotive companies [8][9]

Core Insights - Tesla is dissolving its Dojo supercomputer team and integrating its technology into the FSD vehicle chips, marking a shift from independent chip development to a more cost-effective approach [2][3] - The decision to partner with Samsung for chip manufacturing indicates a strategic pivot towards collaboration rather than in-house development, which could reshape the autonomous driving industry [3][6] Company Strategy - The restructuring aims to reduce costs and improve efficiency by merging the training chip (Dojo) and inference chip (HW series) teams, allowing for the development of AI5 and AI6 chips that can handle both training and inference tasks [6][10] - The AI5 chip has already been designed and is being produced by TSMC, while the AI6 chip will integrate the Dojo training module and be manufactured by Samsung, enhancing performance and reducing development time [6][8] Market Impact - The new "training-inference unified" architecture is expected to redefine the hardware paradigm in autonomous driving, allowing Tesla vehicles to act as mobile data centers and reducing reliance on third-party computing platforms [7][10] - Analysts predict that if Tesla's FSD penetration increases from 35% to 60% by 2027, the combined effects of cost reductions and increased market share could add $500 billion to Tesla's market valuation [11] Competitive Landscape - The shift in strategy comes as competitors like Nvidia and Waymo are rapidly advancing their own technologies, making it crucial for Tesla to innovate quickly to maintain its competitive edge [9][11] - The integration of training and inference capabilities within a single chip is seen as a potential industry trend, prompting other companies to explore similar architectures [10]