作者 | 允毅、木子 今年的 CES 真可谓是八仙过海，黄仁勋、苏姿丰、陈力武等"经典面孔"齐亮相； 不过台上谈的已不只限于显卡、算力和制 程，还在于 AI 接下来要被带去哪里。 在 AMD 的专场演讲中，苏妈甩出一个大胆判断： "未来五年内，将有 50 亿人每天使用 AI，超过世界人口的一半。" ——什么概念？就是这个增长速度将远超互联网早期阶段，自 ChatGPT 在 2022 年底发布以来，AI 活跃用户已从 100 万暴涨 至 10 亿 +。 值得一提的是，这场演讲还请来了"AI 教母"李飞飞。 李飞飞并不是来站台新品的，她和苏妈主要探讨空间智能和世界模型，这也是她已耕深 20 余年的领域。 谈到云端算力的未来，苏姿丰毫不掩饰 AMD 的野心： "全球人工智能运行在云端，而云端运行在 AMD 平台上。" 另外，她还指出，下一代 Instinct 数据中心 AI 加速器平台 MI500 系列，将在 2027 年推出并全面转向 2nm 工艺，并放出狠 话：希望借此在四年内 AI 芯片性能提升 1000 倍（远超摩尔定律啊...）。 与此同时，AMD 还在推动把 AI 从云端下放到本地，而他们的一个很核心的 ...

Core Insights - The article presents six key predictions regarding the evolution of enterprise intelligent agents by 2026, emphasizing the transition from single-point tools to comprehensive intelligent management systems [2][3][4]. Group 1: Evolution of Intelligent Agents - Prediction 1: Intelligent agents will evolve from "one person, one tool" to "one person, one team," enabling systemic intelligence across organizations and transforming top talent's capabilities into reusable business assets [5]. - Prediction 2: Multi-agent systems must possess three essential elements: AI Team (collaboration between humans and agents), allowing for flexible scaling of capabilities and reducing dependency on individual experts [6][7]. Group 2: China's Role and Strategic Implementation - Prediction 3: China is positioned to become a global leader in deploying multi-agent systems due to its complete industrial chain, leading open-source models, and vast market [8]. - Prediction 4: Successful AI transformation requires a "top-down" approach, where leadership drives systemic changes rather than isolated technical trials, emphasizing the need for leaders to understand AI's potential [9][10]. Group 3: Autonomous Evolution and Future Workforce - Prediction 5: Intelligent agents will contribute to the autonomous evolution of enterprise digital infrastructure, enhancing knowledge systems and decision-making processes [10]. - Prediction 6: By 2026, the focus of enterprise competition will shift from hiring to managing intelligent agents, with new roles such as "intelligent agent operators" emerging [11]. Group 4: Implementation Framework - The article outlines a three-step approach for enterprises to evolve their multi-agent systems: establishing a comprehensive strategy led by top management, utilizing Forward Deployed Engineers (FDE) to bridge organizational gaps, and fostering collaborative evolution through a mixed-model architecture [14][15][16]. Group 5: Technological Foundations and Future Directions - The foundation of enterprise multi-agent systems includes open-source models and industry-specific frameworks, aiming to create "super digital employees" that directly contribute to business objectives [17][18]. - The article concludes with a vision for the future of agents, highlighting the importance of safety, tool integration, task planning, and multi-model collaboration in enterprise environments [19][20].

Core Viewpoint - The article emphasizes that while AI has advanced in perception and decision-making, it has yet to master physical interaction, indicating a gap in the practical application of AI in robotics [2][3]. Group 1: Evolution of AI and Robotics - The transition from AI 1.0 to embodied intelligence marks a significant shift, with hardware and algorithms finally aligning, making 2025 a pivotal year for advancements in robotics [6][10]. - The rise of embodied intelligence is driven by improvements in hardware components, particularly the domestic production of key robot parts, which has reduced reliance on imports and improved cost control [8][9]. Group 2: Algorithm and Hardware Development - The article highlights that recent advancements in algorithms, such as Diffusion and Transformer models, have enabled robots to learn complex behaviors rather than relying solely on predefined rules [9][10]. - The collaboration between hardware and algorithm development is crucial, with the article suggesting that algorithm breakthroughs often lead to hardware advancements rather than the other way around [13][14]. Group 3: Methodology and Data Strategy - The company emphasizes a multi-modal approach in its development, integrating various sensory inputs beyond just visual data to enhance the robot's ability to perform tasks in real-world scenarios [18]. - A focus on high-quality, real-world data collection is essential, as even minor errors in robotic tasks can lead to significant failures, necessitating a rigorous data acquisition process [19][20]. Group 4: Benchmarking and Evaluation - The lack of a unified evaluation system in the industry is identified as a significant gap, prompting the company to invest in creating a benchmarking platform to facilitate comparisons and establish technical consensus [21][23]. - The goal is to create a clear methodology for evaluating robotic capabilities, which will help in validating and accumulating algorithmic advancements over time [23].

Core Insights - The article highlights a significant shift in AI technology, moving from understanding language to transforming the physical world, as announced by NVIDIA CEO Jensen Huang at CES 2026 [2] - NVIDIA has unveiled its latest technology roadmap for "Physical AI," aiming to create a comprehensive stack of computing and software systems to enable AI to understand, reason, and act in the real world [2] Group 1: AI Development and Breakthroughs - Huang emphasized the "dual platform migration," where computing shifts from traditional CPUs to GPU-centric accelerated computing, and application development transitions from predefined code to AI-based training [4] - In 2025, open-source models achieved key breakthroughs but still lagged behind advanced models by about six months, with explosive growth in model downloads as various sectors engage in the AI revolution [3][9] - The emergence of autonomous thinking agent systems in 2024 marks a pivotal development, with models capable of reasoning, information retrieval, and future planning [8] Group 2: Physical AI and New Models - NVIDIA's Physical AI models are categorized into three series: Cosmos World models for world generation and understanding, GROOT for general robotics, and the newly released AlphaMayo for autonomous driving [12] - AlphaMayo, an open-source AI model, enables autonomous vehicles to think like humans, addressing complex driving scenarios by breaking down problems and reasoning through possibilities [16][18] - GROOT 1.6, the latest open-source reasoning model for humanoid robots, enhances reasoning capabilities and coordination for executing complex tasks [22][24] Group 3: AI Supercomputing and Vera Rubin - NVIDIA introduced the Vera Rubin supercomputer, designed to meet the escalating computational demands of AI, with the first products expected to launch in late 2026 [32] - The Vera Rubin architecture features a collaborative design of six chips, providing 100 Petaflops of AI computing power, significantly enhancing performance and efficiency [40][42] - The system incorporates advanced cooling and security features, ensuring data protection and energy efficiency, which is crucial for modern AI workloads [47][49] Group 4: Ecosystem and Collaboration - NVIDIA's collaboration with Hugging Face connects a vast community of AI developers, facilitating the integration of NVIDIA's tools into existing workflows [30] - The launch of the Isaac Lab Arena provides a framework for safely testing robot skills in simulation, addressing the challenges of verifying robotic capabilities in real-world scenarios [27] - The open-source approach to AI and robotics is driving rapid advancements across various industries, with numerous companies leveraging NVIDIA's platforms for their next-generation AI systems [29]

Core Insights - Microsoft CEO Satya Nadella reflects on the company's AI progress over the past year and outlines a vision for 2026, emphasizing that it will be a pivotal year for AI development [2][4] - Nadella highlights the current limitations in AI's cultural and technical applications, indicating that the focus should shift from merely critiquing AI's flaws to designing systems that contribute positively to society [6][8] Group 1: AI Development Stages - Nadella states that the industry has moved past the initial exploration phase of AI and is entering a stage of widespread adoption, but acknowledges that many uncertainties remain [4] - He introduces the concept of "model lag," suggesting that the capabilities of AI models are advancing faster than their practical applications, which poses a challenge for realizing AI's full potential by 2026 [4][5] Group 2: Key Actions for AI Advancement - The first key action proposed by Nadella is to redefine AI as a "support framework" that empowers human potential rather than replacing it, emphasizing the importance of how people utilize AI tools [5] - The second action involves transitioning from single-model AI to multi-model systems that can collaborate effectively, incorporating memory functions and permission management to enhance real-world applications [5] - The third action addresses ethical considerations, stressing the need for AI to be evaluated based on its real-world impact on society and the environment, and the importance of consensus on these challenges [6] Group 3: Microsoft’s AI Strategy and Market Position - Microsoft has invested hundreds of billions in AI projects and infrastructure, aiming to solidify its core position in the industry’s software and hardware technology ecosystem [6] - The company is integrating AI into its Windows operating system and Office suite, although many promised features remain unfulfilled, leading to user dissatisfaction [12][15] - Despite the challenges, Microsoft sees AI integration as a crucial strategy to revitalize Windows and compete effectively in the cloud-native AI services market against companies like Google and Amazon [16]

作者 | 陈勇 在海量视频检索场景中，传统方法往往"重视觉、轻听觉"，且网络结构设计更多依赖经验与人工尝试，难以同时兼顾高效存储与快速检索。那么，是否 存在一种能够自动找到最优结构、并充分发挥多模态价值的方案？ 近日，来自北邮与北大的研究团队提出 AV-NAS，在多模态视频哈希领域首次引入神经架构搜索（NAS），构建了一个同时覆盖 Transformer 与 Mamba 的统一搜索空间。该方法不仅使模型能够自动发现最优的跨模态融合机制（Cross-Mamba），还揭示了一个颇具启发性的结论——在音频时序 建模任务中，看似简单的 "CNN + FFN" 结构竟然优于复杂的 Transformer 方案。 论文题目： AV-NAS: Audio-Visual Multi-Level Semantic Neural Architecture Search for Video Hashing 论文链接： https://dl.acm.org/doi/10.1145/3726302.3729899 代码链接： https://github.com/iFamilyi/AV-NAS 目前，AV-NAS 已被 SIGIR 2 ...

Core Insights - A senior Google engineer revealed that Anthropic's Claude Code was able to replicate a system that her team had spent a year developing in just one hour, highlighting the rapid advancements in AI programming capabilities [3][12]. Group 1: AI Programming Capabilities - The engineer, Jaana Dogan, described how she provided a brief problem statement to Claude Code, which generated a system closely resembling their year-long effort in just one hour [3][5]. - Dogan emphasized that while Claude Code is impressive, it is still not perfect and requires continuous iteration and refinement [7]. - The rapid evolution of AI programming tools has led to significant improvements in quality and efficiency, surpassing expectations for 2024 [9]. Group 2: Industry Reactions and Perspectives - The engineering community has shown polarized reactions to AI coding agents, with some expressing skepticism about the true capabilities of AI in programming [7][14]. - Concerns were raised that the efficiency gains from AI might lead companies to reduce workforce rather than reallocate engineers to higher-level tasks [17]. - Dogan's public praise for a competitor's product has sparked discussions about potential shifts in the industry and the nature of competition [12][13]. Group 3: Google and Anthropic Relationship - Google is a significant investor in Anthropic, holding approximately 14% of its shares and has invested around $3 billion in total [20][21]. - A partnership agreement between Google and Anthropic includes a commitment to provide up to 1 million TPU units, valued at hundreds of billions, to enhance AI capabilities [21]. - Dogan noted that the industry is not a zero-sum game, and acknowledging competitors' achievements can drive motivation and innovation [22].

Core Insights - The article discusses the recent appointment of Li Yuanqing, a former Huawei executive, to LeXiang Technology, where he will focus on innovation strategy and core technology development in the field of embodied intelligence [2][3] - Li emphasizes the importance of practical application and data in the development of embodied intelligence, predicting that the first widely adopted product in this field may emerge from China [3][24] Group 1: Industry Trends - There is a significant increase in investment in embodied intelligence from both tech giants and startups, driven by the maturity of technology and market expectations [6][7] - The current trend in the robotics sector is characterized by a strong linkage between primary and secondary markets, with listed companies signaling their entry into humanoid robotics to enhance their market value [6][7] - The stability and reliability of robots have improved significantly from 2024 to 2025, transitioning from mere demonstrations to market-ready products [8][9] Group 2: Technological Advancements - Key breakthroughs in embodied intelligence include the LocoFormer technology for local motion control and AnyTracker applications that allow robots to replicate human movements accurately [9][10] - Robots are now capable of completing simple tasks with a 100% success rate, a significant improvement from previous years [10][11] - The evolution of the technology stack for embodied intelligence is marked by advancements in local motion control and the integration of visual language navigation strategies [11][12] Group 3: Challenges and Opportunities - Major challenges for the large-scale application of embodied intelligence include high costs of core components and unclear product definitions in various scenarios [22][23] - The industry faces difficulties in integrating hardware and software technologies, leading to a lack of clarity in technical routes and supply chain adaptations [23][24] - The article suggests that the future of embodied intelligence may lie in a multi-robot collaboration model rather than a single universal intelligent agent [27][28] Group 4: Strategic Directions - Li's team aims to develop a functional product for home users, treating each household as a factory to enhance information and automation [25][26] - The company plans to leverage advanced spatial perception technology to build an information system for homes, integrating automation and intelligent interaction [26][24] - The article highlights the potential for new business models such as Robot as a Service (RaaS) and rental models to optimize the utilization of robotic systems [29][30]

行业热点 雷军：未来五年研发至少投入 2000 亿元，加大大模型投入 整理 | 傅宇琪、褚杏娟 雷军：未来五年研发至少投入 2000 亿元，加大大模型投入；微信回应安装包 10 多年膨胀数百倍；Anthropic 豪掷 210 亿美元购 谷歌 100 万块 TPU；比亚迪首次超越特斯拉，成全球最大电动汽车销售商；技术元老离场！腾讯 AI Lab 副主任俞栋离职；传快 手副总裁、基础大模型及推荐大模型负责人周国睿即将离职；新论文暗示 DeepSeek V4 已完成训练；Manus 武汉团队基本搬 离，核心业务人员迁往新加坡；"全球大模型第一股"来了！智谱发行市值达 511 亿港元；50 亿美元联姻！NVIDIA 正式收购 Intel 股份…… 1 月 3 日晚，小米集团董事长雷军在新年首场直播中透露，小米汽车 2026 年全年交付目标为 55 万辆，2025 年交付量超 41 万辆，超过原先计划的 30 万辆。而在今日早间发布的微博中，雷军称，"希望今年也能超额完成（目标）。" 雷军介绍小米规划聚焦三点：一是未来五年至少投入 2000 亿元坚持技术研发；二是加大对大模型的投入；三是坚持为人车家 全生态体验打造极 ...

Core Viewpoint - The article highlights the launch of the "Qiyuan Q1," a compact humanoid robot by Shangwei New Materials, marking the company's strategic entry into the personal robotics market. The robot aims to redefine the possibilities of small humanoid robots, making advanced robotic technology accessible for personal use and creativity [2][4]. Group 1: Product Features and Innovations - The Qiyuan Q1 is recognized as the world's smallest fully controllable humanoid robot, achieving significant breakthroughs in joint system miniaturization and application scenarios [2][4]. - The robot's size has been reduced to one-eighth of traditional models, and its weight has also decreased, enhancing its durability and stability during falls [4]. - The design allows for lower research and development costs, facilitating efficient iterations from virtual simulations to real-world applications [4]. Group 2: Target Audience - The Qiyuan Q1 is designed for three core user groups: researchers, creative enthusiasts, and general family users [6][7]. - For researchers and learners, it serves as a portable laboratory for embodied intelligence, enabling safe and cost-effective experimental validation [7]. - For creators and hobbyists, the robot offers an open-source structure for customization and programming, allowing users to easily create and modify robotic behaviors [8]. - For family users, the Qiyuan Q1 acts as an intelligent companion, capable of natural language interaction and educational support, enhancing daily life experiences [8]. Group 3: Strategic Vision - The launch of the Qiyuan Q1 is positioned as a strategic declaration for Shangwei New Materials, emphasizing the company's commitment to making high-end robotics technology accessible to the general public [2][4]. - The company invites global tech enthusiasts to participate in the co-creation and exploration of personal robotics, aiming to bring imaginative robotic experiences into everyday life [8].