AI产业在过去一年来，也呈现五大新趋势。 张亚勤分析道，第一大趋势是从鉴别式AI到生成式AI，如今则走向智能体AI。 其中重要标志是，过去7个月间，智能体AI的任务长度翻倍、准确度超过50%，由此可以加速让智能体 应用到每个领域。 第二大趋势是过去一年来，在预训练阶段的规模定律（Scaling Law）已经放缓，更多工作转移到训练 后的如推理、智能体应用等阶段。 视频丨实习生唐娜斯 AI产业快速发展，正让诸多行业的迭代呈现加速度趋势。 2025骁龙峰会·中国期间，中国工程院外籍院士、清华大学智能产业研究院（AIR）院长张亚勤在演讲中 指出，新一代人工智能是原子、分子和比特的融合，是信息智能、物理智能和生物智能的融合。这将带 来巨大产业机遇。 从产业规模看，移动互联比PC互联时代至少大10倍；在工智能时代，整个产业规模将比前一代至少大 100倍。 同时具身智能也将快速爆发，预计在十年后的2035年，机器人有望比人类数量还多。 由此也延伸出第四大趋势，即AI风险正快速上升。"智能体出现后，让AI风险至少增加了一倍。"张亚勤 补充道，这尤其需要全球企业和政府对此投入更多精力，他本人也对此花了很多时间。 第五大趋势， ...

21世纪经济报道记者骆轶琪 北京报道 其中重要标志是，过去7个月间，智能体AI的任务长度翻倍、准确度超过50%，由此可以加速让智能体 应用到每个领域。 AI产业快速发展，正让诸多行业的迭代呈现加速度趋势。 2025骁龙峰会·中国期间，中国工程院外籍院士、清华大学智能产业研究院（AIR）院长张亚勤在演讲中 指出，新一代人工智能是原子、分子和比特的融合，是信息智能、物理智能和生物智能的融合。这将带 来巨大产业机遇。 从产业规模看，移动互联比PC互联时代至少大10倍；在工智能时代，整个产业规模将比前一代至少大 100倍。 AI产业在过去一年来，也呈现五大新趋势。 张亚勤分析道，第一大趋势是从鉴别式AI到生成式AI，如今则走向智能体AI。 第二大趋势是过去一年来，在预训练阶段的规模定律（Scaling Law）已经放缓，更多工作转移到训练 后的如推理、智能体应用等阶段。 "并不是说前沿模型就不需要了，整个智力上限还在不断往前走，但（迭代）速度比去年和前年是放缓 的。"张亚勤补充道，规模定律接下来有望在如智能体、视觉等其他领域重现。 在此过程中，过去一年来，推理成本降低了10倍，但智能体的复杂性，让算力同步上涨了10倍 ...

记者从业内了解到，一些端侧AI应用已经崭露头角。例如，当你想要利用AI为你制作一份旅游计划表 时，传统的AI会为你提供一份大众化攻略，但通过端侧AI，它可以自动识别你的日历表中哪些日期、 哪些时段已经安排了其他行程，在制作攻略时会主动避开这些时段，从而提供更合理的出行建议。 无独有偶，联发科在本周发布旗舰芯片时也对端侧AI功能着墨不少。该公司声称，其最新发布的旗舰 芯片率先支持Bitnet 1.58 Bit推理框架，端侧AI能力获得了史诗级的提升，大幅减少AI计算、图像识别 及自然语言处理能力运算对于云的需求，让以往需要依赖云端才能做到的4K超高分辨率文生图、128K 长文本处理、AI写真等功能得以在端侧实现。 近日，高通发布了年度旗舰芯片，在高通负责人的演讲中，频频提到一个关键词——端侧AI。 端侧AI是将AI模型部署在终端设备上，让终端设备具备本地的智能处理能力，无需依赖云端服务器就 能完成一些AI任务。与之相对应的是云侧AI，云侧AI是指将人工智能的模型训练和推理任务放在云端 服务器进行处理。 从名称上就可以感受到端侧与云侧两种AI处理模式的差异，在端侧处理，省去了信息在终端与云端的 交互时间，处理速度 ...

重磅推出的全新骁龙平台在终端侧AI计算、能效优化及多模态交互等多个方面实现突破，为本地大模型运行、实时语音翻译、图像生成等应用提供了有力 的硬件支撑，推动安卓旗舰从聚焦参数转向场景化服务提升。马蜂窝旅游研究院院长孙云蕾用直观的实例为与会嘉宾分享了移动端智能体AI旅行助手在旅 游场景中的全新体验。 孙云蕾为大家现场演示了智能体AI行程规划的功能，针对一位前往成都出差、已有会议安排和往返机票的用户，系统可通过小米日历自动识别空闲时段， 结合本地POI数据为其智能规划景点游览、餐饮推荐等行程。 9月24日至25日，2025骁龙峰会·中国在北京举行，马蜂窝作为旅游领域移动端智能体AI应用的合作伙伴亮相活动，展示其借助第五代骁龙®8至尊版移动平 台的端侧算力，以及小米的系统数据，规模化拓展旅游AI服务能力的前沿成果。 合作深度融合马蜂窝和小米技术优势，构建了从数据感知、本地计算到主动个性化服务的完整技术链路。结合马蜂窝的旅游行业垂直模型、旅游知识图谱、 RAG增强检索等多项技术，提供从行前攻略，到行中智能主动推荐、多语言翻译、餐厅预订等一站式服务。 ...

Core Insights - Qualcomm unveiled multiple chipsets at the Snapdragon Summit 2025, with AI capabilities being the main highlight of the new iterations [1][3] - The Snapdragon 8 series mobile platform features a 20% increase in CPU performance and a 23% enhancement in GPU graphics performance, making it the fastest in its category [1] - The Snapdragon X2 Elite Extreme processor integrates the third-generation CPU, offering up to 75% better performance than competitors at the same power consumption [3] AI Capabilities - The Snapdragon 8 series supports personalized AI assistants that provide customized operations across applications through continuous learning and real-time perception [1] - The NPU performance has improved by 37%, enabling advanced AI functionalities [1] Product Launches - Snapdragon X2 Elite is expected to launch in the first half of 2026, with a 31% performance increase at the same power consumption compared to previous models [3] - The GPU architecture of the Snapdragon X2 Elite Extreme shows a 2.3 times improvement in performance per watt compared to its predecessor [3] Industry Trends - Qualcomm's CEO highlighted six core trends in the AI industry, emphasizing a shift towards user-centric interfaces and the importance of AI agents in redefining user experiences across various smart devices [5][7] - A new computing architecture is necessary to support the transition to AI-driven experiences, requiring redesigns of operating systems, software, and chips [7] - The future of AI will involve seamless collaboration between edge and cloud processing, enhancing the capabilities of AI models through edge data training [7] Future Developments - The development of 6G technology is underway, expected to bridge the gap between cloud and edge computing, with pre-commercial terminals anticipated as early as 2028 [7]

面向手机，骁龙8系移动平台，将能支持 "真正的个性化智能体AI助手" ，能通过持续的终端侧学习、实时感知、多模态AI模型等深度理解用 户，提供跨应用的定制化操作。 明敏 发自 凹非寺 量子位 | 公众号 QbitAI 手机PC等终端芯片，在Agent变革面前也要被重塑了。 刚刚，高通发布全球最快Windows PC处理器、全球最快移动SoC处理器： 面向PC，高通首次推出专为超高端PC打造的骁龙X2 Elite Extreme，目标是 "轻松驾驭智能体AI体验" ； 高通CEO安蒙 也同步提出对AI趋势的六大理解，不仅对应此次最新发布，也暗暗透露高通将如何以智能体为核心颠覆个人计算体系： 如此思考与理解，在最新发布中已窥见一二。 一次发布，两个全球最快 骁龙X2 Elite系列 第五代骁龙8至尊版移动平台 AI是新的UI 以智能手机为中心转向以智能体为中心 需要构建全新计算架构体系 模型混合化发展 边缘数据相关性增强 6G将成为云边端之间的连接桥梁 PC最快：骁龙X2 Elite系列 骁龙X2 Elite系列采用3nm制程、第三代Oryon架构（12个Prime核+6个Performance核）。 各种峰值性 ...

HOME 窦德景 ◎记者 李兴彩 近日，在上证首席讲坛第二十三期节目上，复旦大学计算机学院特聘教授、北电数智首席科学家窦德景 就AI大模型的突破点和未来应用场景进行了深入浅出的分享，并同期接受了上海证券报记者的专访。 作为人工智能领域的资深学者与产业实践者，窦德景深耕AI领域二十余载，既见证了行业有起有落的 发展历程，也亲身参与了从技术研发到产业落地的全链条实践。在生成式AI掀起全球变革浪潮的当 下，他以横跨产学研的独特视角，解读中国AI发展的核心逻辑与未来机遇。 AI要突破必须扎根具体场景 从学术殿堂到产业一线，窦德景的履历勾勒出一条跨领域的产业和个人成长轨迹。 1996年，窦德景从清华大学电子工程系本科毕业后，赴耶鲁大学攻读电气工程硕士学位，随后又师从世 界著名人工智能学者德鲁·麦克德莫特（Drew Mcdermott）攻读人工智能方向的博士学位。此后，窦德 景历任斯坦福大学生物医学信息研究中心客座副教授、美国俄勒冈大学计算机和信息科学系正教授，发 表超过250篇论文，谷歌学术引用量超1.3万次，成为国际AI领域的知名学者。 2010年后，随着深度学习技术的突破，AI迎来第三次高潮，窦德景选择投身产业实践 ...

Core Insights - The demand for AI training and inference integrated machines is increasing as AI applications become more prevalent in various industries [1][4][5] - Companies like ZTE and Digital China are experiencing significant sales growth in their AI integrated machine products, indicating a strong market trend [2][7] Market Demand - Nearly 100 manufacturers have launched AI integrated machine products in the domestic market this year, including several listed companies [1][7] - The demand for training and inference integrated machines is driven by the need for private deployment in sectors with sensitive data, such as government and finance [3][8] Industry Applications - The integrated machines are being utilized across 15 industries, including government, education, healthcare, and telecommunications, with notable sales reported [2][7] - Specific applications include AI education tools, medical diagnostic systems, and automotive design solutions, showcasing the versatility of these machines [7] Future Outlook - The market for training and inference integrated machines is expected to grow significantly, with IDC predicting a 260% increase in the intelligent agent market by 2025 [4][5] - The integration of AI capabilities into various business processes is seen as essential for future development, with a focus on personalized solutions for different industries [5][6] Challenges - Companies face challenges in deploying integrated machines due to the complexity of AI ecosystems and the need for deep integration of hardware and software [9][10] - There is a need for improved scalability and cloud management to support the development of AI models and applications [9][10]

Core Insights - The demand for AI training and inference integrated machines is increasing as AI applications become more prevalent in various industries [1][4][5] - Companies like ZTE and Digital China are experiencing significant sales growth in their integrated training and inference machines [2][7] Market Trends - Nearly 100 manufacturers have launched integrated training and inference machine products in the domestic market this year, including several listed companies [1][7] - The integrated training and inference machine market is expected to grow significantly, driven by the need for AI applications across various sectors such as finance, government, and energy [8][9] Technology Development - The integrated training and inference machines support the entire process of large model training, inference, and application development, catering to the needs of enterprises for ready-to-use solutions [2][3] - The transition from training-focused machines to those that emphasize inference capabilities reflects the evolving landscape of AI technology [2][4] Industry Applications - Key sectors such as finance, government, and energy are showing strong demand for integrated training and inference machines, which are essential for building AI model training and real-time inference capabilities [8][9] - Companies are collaborating with educational institutions and healthcare providers to enhance AI applications in their respective fields [7] Challenges and Considerations - The deployment of integrated training and inference machines faces challenges related to the complexity of the AI ecosystem and the need for deep integration of hardware and software [9][10] - Companies are advised to enhance the scalability of integrated training and inference machines and incorporate cloud management systems to support the full lifecycle of AI model development [9][10]

Core Viewpoint - The insurance industry is undergoing a significant transformation driven by artificial intelligence (AI), particularly generative AI, which is reshaping workflows and enhancing customer interactions, leading to increased efficiency and personalized services [2][3][4]. Group 1: AI's Impact on the Insurance Industry - AI is fundamentally changing the insurance sector by improving risk identification and providing personalized support during customer crises [3]. - Generative AI's ability to process unstructured data allows for more personalized and human-like interactions, enhancing customer service [3][4]. - The integration of AI into core business functions, such as underwriting, claims processing, and customer service, is accelerating within insurance companies [3][4]. Group 2: Strategic AI Transformation - Successful AI transformation requires a comprehensive strategy that redefines key operational paradigms rather than piecemeal implementations [4]. - Companies must establish a future-oriented AI strategy that integrates technology capabilities into their operational mechanisms [4][5]. - The focus should be on end-to-end process reengineering rather than merely adding AI tools to existing workflows [4][5]. Group 3: AI Deployment and Management - The deployment of AI in insurance is not without challenges, including security risks, high costs, and cultural resistance [6]. - Effective change management is crucial for realizing both financial and non-financial returns from AI investments [6][7]. - Leading insurance companies are already leveraging AI to enhance their market position, with significant shareholder returns compared to their peers [7]. Group 4: Key Initiatives for AI Success - Companies should focus on six key initiatives to maximize AI potential: high-level collaboration, building a digital talent pool, creating scalable operational models, enhancing technology architecture, embedding data capabilities, and increasing resource investment [8][9][10][11][12][13]. - A clear AI transformation roadmap should prioritize business areas with significant optimization potential [14][15]. - The establishment of a robust data platform is essential for supporting AI systems and ensuring data quality and governance [45]. Group 5: Case Studies and Practical Applications - Leading insurance firms have successfully implemented AI in various areas, such as claims processing and sales automation, resulting in significant efficiency gains and cost savings [31][32]. - For instance, Aviva reduced claims assessment time by 23 days and improved accuracy in case assignment by 30% through AI deployment [31]. - Another company saw an increase in online transaction rates to 80% after introducing intelligent tools for customer quotes and policy issuance [31]. Group 6: Future Directions and Challenges - The insurance industry is poised for further transformation as generative AI continues to evolve, enhancing operational efficiency and customer engagement [16][19][22]. - Companies must address existing barriers, such as outdated systems and the need for modern infrastructure, to fully leverage AI capabilities [43][44]. - A culture of innovation and adaptability is necessary for employees to embrace new AI-driven workflows and maximize productivity [46][47].