正值大模型从"算力堆叠"走向"推理优先"的关键节点， 清华大学智能产业研究院（AIR）创始院长、中国工程院外籍院士张亚勤 提出： 编辑部 整理自 MEET2026 量子位 | 公众号 QbitAI 从ChatGPT到DeepSeek，AI正沿着 "智能+" 的路径进入新一轮浪潮。 新一轮人工智能，是信息智能、物理智能和生物智能的融合，本质上也是原子、分子和比特的融合。 也就是说，在规模定律持续发挥作用的前提下，当参数规模、数据体量与算力资源跨过某个阈值，智能就不再只停留在模式识别，而是开 始"涌现"—— 先是从鉴别式AI走向生成式AI，再从生成式AI走向以 智能体 为代表的新范式。 在本次 量子位MEET2026智能未来大会 上，他也将ChatGPT和DeepSeek，视作这一轮演进中的两个重要里程碑： 前者通过统一表征与token化，把文本、语音、图像乃至蛋白质、点云等数据纳入同一空间； 后者则以高效率、高性能、低价格和开源路径，把大模型从"预训练时代"推向以推理为核心的"DeepSeek时刻"。 至于未来5~10年的主战场，在他看来，将走向 "智能体互联网"时代 —— 基础大模型像操作系统一样在全球范围内 ...

今年的虎嗅F&M创新节也被AI"包围"了。 仅在11月22日这天，数十位AI领域的专家从不同维度向现场观众分享了AI给现实世界带来的改变，内容涵盖AI产业如今面临的瓶颈与挑战、AI如何重 构千行百业以及AI如何打开人类生命新边界等等。 开篇演讲中，北京智源人工智能研究院院长王仲远冷静地指出， 具身智能尚未迎来它的"ChatGPT时刻"，其还面临数据短缺、硬件不成熟、模型能力 弱、落地应用难四大瓶颈。 紧接着，阶跃星辰联合创始人朱亦博在现场分享了智能演进的路线图，他认为模型推理效率是决定 AI 大规模落地应用的 关键要素，而推理效率的提升需要产业上下游联合优化，至于下一代智能硬件，比拼的不是"硬件"而是"智能"。 我是谁，我从哪里来，我要到哪里去？ 这经典的人生三问成为人类在AI变革年代的核心困惑。人必须重新定义身份、重建内在叙事，并重新选择未来 方向。 在今年的虎嗅F&M创新节上，围绕AI的对话比任何一年都更直指本质——我们讨论的不仅是技术、产品或是行业趋势，而是一个更深刻的问题： AI将 如何重构"人"？ 在现场，感受AI热浪扑面而来 随后，一个崭新的对比视角由行云集成电路创始人季宇提出，他对比了计算机的发 ...

Core Viewpoint - The AI industry is transitioning from a "pyramid" structure, where chips capture most value, to an "inverted pyramid" structure, where models and applications generate significantly more value, indicating a healthier and sustainable future for AI [1][3]. Group 1: AI Industry Insights - The internalization of AI capabilities transforms it from a cost to a productivity driver, emphasizing the need for AI to integrate organically with tasks [3]. - Baidu's AI products, including the Wenxin 5.0 model, Kunlun chip, and the autonomous driving service "LuoBo KuaPao," showcase the company's long-term investment in AI [5][9]. - The Wenxin 5.0 model utilizes a unified autoregressive architecture for multimodal modeling, achieving global leading performance in various benchmarks [5][7]. Group 2: Technological Developments - Baidu's Kunlun chip has been validated internally and is being used for most inference tasks, with plans for new chip releases in 2026 and 2027 [7][8]. - The autonomous driving service "LuoBo KuaPao" has achieved over 250,000 fully autonomous orders weekly and has completed over 240 million kilometers of autonomous driving [9]. - Baidu's search engine has been significantly transformed by AI, with 70% of search results now generated by AI, moving from text-based to rich media content [10]. Group 3: Global Expansion and Market Position - Baidu's AI capabilities are being applied globally, with digital human technology launched in Brazil and autonomous driving services expanding in the Middle East [18]. - The company's AI strategy has led to a 40% increase in stock price since 2025, reflecting a market shift from model competition to application effectiveness [17][18]. - Baidu's transformation is not merely an addition of AI but a deep integration of AI capabilities into its business processes, enhancing productivity and efficiency across various sectors [18].

Core Insights - The 2025 Baidu World Conference emphasized the theme of "emergent effects" in AI, with Baidu's founder discussing the transition from "intelligent emergence" to "effect emergence" in AI technology [1] Industry Structure - The AI industry structure is shifting from an unhealthy "pyramid" model, where chips capture most value, to an "inverted pyramid" model, where models generate ten times the value of chips and AI applications create one hundred times the value, promoting a healthier industry ecosystem [1] AI Applications - AI's core value is realized when it is internalized within enterprises, enhancing decision-making, reducing costs, increasing profit margins, and shortening innovation cycles [2] - Baidu is actively restructuring its business using AI, with significant advancements showcased, including the release of the Wenxin 5.0 model, the new Kunlun chip, and various AI products [2] Key AI Products - The Wenxin 5.0 model is a new generation of native multimodal models that integrates language, image, video, and audio data from the training stage, achieving unified understanding and generation [3] - The new Kunlun chip includes two products, with the Tianchi 512 super node capable of training trillion-parameter models [3] - The "Luo Bo Kua Pao" service has surpassed 250,000 fully autonomous orders weekly and has completed over 240 million kilometers of autonomous driving globally [3] Digital Human Technology - Baidu's digital human technology, Huibo Xing, is being expanded globally, with significant adoption in Brazil and plans for further expansion into Southeast Asia and the US [4] - During the recent Double 11 shopping festival, 83% of live stream hosts utilized Huibo Xing, leading to a 119% increase in live streaming sessions and a 91% increase in GMV [4] Intelligent Agents - Baidu launched the self-evolving intelligent agent "Fa Mo," designed to find global optimal solutions by simulating evolutionary processes, which can outperform top algorithm experts [4] - The integration of AI into various industries is crucial for realizing its value and driving a productivity revolution, transforming "intelligent dividends" into "social dividends" [4]

Core Insights - The core message emphasizes the integration of AI as an inherent capability that transforms it from a cost into a productivity driver for both enterprises and individuals [1] Group 1: AI Integration and Business Transformation - Baidu is actively restructuring its business through AI, with its search engine undergoing significant AI-driven transformations, focusing on rich media content [1][9] - The AI industry structure is shifting from an unhealthy "pyramid" to a "reverse pyramid," where models generate ten times the value of chips, and AI applications can create a hundred times that value [1][3] Group 2: AI Product Innovations - Baidu introduced the next-generation "real-time interactive digital human" and upgraded its "秒哒" to version 2.0, showcasing advancements in digital human technology and self-evolving intelligent agents [3][11] - The latest Wenxin model 5.0 is a unified native multimodal model, excelling in multimodal understanding, creative writing, intelligent planning, and instruction adherence, placing it at a global leading level [7][8] Group 3: Performance Metrics and Market Expansion - The "萝卜快跑" service has achieved over 17 million global travel service instances, becoming the world's leading service in its category, with a weekly order count exceeding 250,000 [9][5] - Baidu's search results now feature a 70% coverage of rich media, indicating a significant shift from text-based results to a more visual and interactive format [9][10] Group 4: Future Outlook and Strategic Goals - Baidu plans to continue investing in cutting-edge models to push the limits of intelligence, with new products like the next-generation Kunlun chip and Tianchi supernodes set to launch next year [8][10] - The company aims to internalize AI as a native capability across various industries, fostering a productivity revolution that translates "intelligent dividends" into "social dividends" [15]

Core Insights - The article discusses the transformative impact of AI, highlighting its transition from "passive learning" to "active generation," marking a significant shift in the technology's capabilities and applications [1][2] - AI is seen as a driving force behind a new productivity revolution, with its integration into various industries accelerating over the past decade, particularly in the last five years [2][3] - The urgency of AI safety and governance is emphasized, with concerns about misinformation and the potential risks associated with AI's rapid development [7][8] Group 1: AI Development and Trends - AI is currently at a critical juncture, characterized by a "scale law" where increased data and computational power lead to significant advancements, including cognitive and generative capabilities [2] - The demand for AI applications is particularly high in sectors such as robotics, education, and healthcare, where AI can significantly enhance productivity and innovation [3][4] - The infrastructure for AI, including chips, data centers, and cloud computing, is rapidly evolving, although there are concerns about potential bubbles in investment similar to the internet boom of the late 1990s [3][6] Group 2: AI Infrastructure and Investment - Major tech companies are heavily investing in AI infrastructure to build comprehensive ecosystems, with varying strategies among firms focusing on application layers or specific industry solutions [5][6] - The global competition in AI models is intensifying, with a prediction that only a limited number of general-purpose models will emerge, while vertical or industry-specific models may present greater opportunities [5] Group 3: AI Safety and Governance - The rise of AI-generated content raises significant concerns about misinformation, with reports indicating that over 52% of English written content online is now AI-generated, surpassing human-created content [8] - The potential for systemic risks associated with AI's rapid advancement necessitates robust policies and regulations to establish clear boundaries for AI applications [7][8]

Core Viewpoint - Alibaba is building a large-scale AI infrastructure and investing in a super AI cloud to provide leading AI services to global developers through its full-stack technology accumulation [1][3] Group 1: AI Development Stages - The development of AI will go through three stages: intelligent emergence, general artificial intelligence (AGI), and superintelligence [1][3] - The first stage, intelligent emergence, has been the main theme in recent years, where AI learns from vast human knowledge to understand human intentions and solve real-world problems [3] - The second stage, AGI, is currently at its beginning, where AI agents can perform many tasks in the digital world and connect to the real world to operate some physical devices [3] - The third stage, superintelligence, will see AI connecting to vast amounts of raw data from the real world, possessing autonomous learning and self-iteration capabilities, ultimately surpassing human intelligence [3] Group 2: Alibaba Cloud's Positioning - Alibaba Cloud aims to be a full-stack AI service provider, offering world-leading intelligent capabilities and a globally distributed AI cloud computing network [3] - The transition from intelligent emergence to AGI and then to superintelligence represents a significant computational revolution, necessitating large-scale infrastructure and full-stack technology to meet massive computational demands [3]

Core Insights - The article discusses the evolution of AI from "cognition" to "action," highlighting the transition of Tang Jian from academia to industry, particularly in the fields of autonomous driving and embodied intelligence [1][2] - Tang Jian emphasizes the importance of experience-driven control methods over traditional mathematical modeling in complex environments, suggesting that AI systems can learn from historical data to make effective decisions [4][5] - The concept of a "world model" is introduced as essential for embodied intelligence, enabling robots to understand and predict their environment, thus enhancing their operational capabilities [13][14] Summary by Sections Transition from Academia to Industry - Tang Jian, a former tenured professor, shifted focus to practical applications of AI in industry, particularly in autonomous driving and robotics [1][3] - His experience in various companies, including Didi and Midea, has informed his approach to AI-driven system control [3][6] Experience-Driven Control - The article outlines the difference between traditional control methods and experience-driven approaches, with the latter relying on data and historical experiences rather than precise mathematical models [4][5] - This experience-driven philosophy is evident in autonomous driving applications, where end-to-end control merges perception, planning, and control into a single learning process [6][7] Embodied Intelligence and World Models - Tang Jian argues that embodied intelligence presents a higher complexity than autonomous driving, requiring robots to manage multiple joints and navigate dynamic environments [7][8] - The world model is described as a critical component for robots to understand and interact with the physical world, enabling them to perform tasks that require nuanced understanding and adaptability [14][15] - The article highlights the need for a world model to facilitate the development of robots that can generalize across various tasks and environments, which is crucial for their deployment in real-world scenarios [21][22] Future Directions and Challenges - The discussion includes the potential for world models to achieve a "DeepSeek moment" in embodied intelligence, drawing parallels to breakthroughs in AI performance under limited resources [9][10] - Tang Jian acknowledges the current limitations in data and model architecture, indicating that further iterations and improvements are necessary for the field to progress [2][13] - The article concludes with the assertion that the world model is not just a technical choice but a fundamental requirement for the advancement of embodied intelligence [13][22]

Core Insights - The article discusses the intersection of design and technology, highlighting innovative products developed by Tsinghua University's Future Lab, such as smart makeup mirrors and machines that create everyday items from coffee grounds [1] - The "WDCC2025 Intelligent Design Forum" held in Shanghai brought together experts from various fields to explore the future trends of intelligent design, emphasizing the shift towards "intelligent," "contextual," and "collaborative" design [1] - The rapid development of creative artificial intelligence is reshaping the design landscape, with a focus on the transition from "intelligent emergence" to "creative emergence" [2] Group 1 - The emergence of AI-generated art, such as the award-winning piece "Space Opera House," has sparked discussions about the capabilities of AI in creative fields [2] - Experts believe that while AI can assist designers by providing new perspectives, it may also pose threats to genuine creativity and understanding of the physical world [3] - The role of designers is evolving, with AI serving as a tool to enhance human creativity rather than replace it [3] Group 2 - Concerns have been raised about the limitations of AI in understanding cultural contexts, particularly in the training of models with insufficient Chinese language data [4] - The importance of high-quality datasets, especially those reflecting Chinese cultural values, is emphasized as crucial for the development of AI in design [4] - The potential for AI to distort creative processes and the need for a shift in human thinking to maintain uniqueness and autonomy in design are highlighted as current challenges [3][4]

Group 1 - The 2025 Yunqi Conference will be held from September 24 to 26, showcasing AI software products, new models, agent applications, and infrastructure hardware [1] - Alibaba Group's CEO, Wu Yongming, emphasized that the intelligent revolution will exceed human imagination, with general artificial intelligence (AGI) set to enhance human intelligence and liberate human potential [1] - Wu stated that achieving AGI is a certainty, marking the beginning of the journey towards super artificial intelligence (ASI), which aims to address major scientific challenges such as climate change and energy [1] Group 2 - Wu believes that large models will serve as the next generation operating system, while AI cloud will be the next generation computer, predicting that there may only be 5 or 6 super cloud computing platforms globally in the future [2] - Alibaba is actively advancing a 380 billion yuan investment in AI infrastructure and plans to increase this investment further [2] - By 2032, the energy consumption of Alibaba Cloud's global data centers is expected to increase tenfold compared to 2022, indicating an exponential increase in computing power investment [2]