四川省大模型及人形机器人训练场揭牌 四川省人形机器人训练场由天府绛溪实验室牵头建设，依托清华大学、电子科技大学等高校顶尖科研资 源，联合四川省内重点人形机器人企业共建，聚焦人形机器人基础研究与关键技术突破，致力填补四川 省内人形机器人检测能力空白，加速技术攻关与产业转化。 四川省人形机器人训练场创新性采用"343"架构，即构建"研发试验－仿真训练－检测认证"三大核心支 撑平台，包括关键技术攻关平台、智能训练与仿真平台、检测认证标准化平台，覆盖"技术攻关→虚拟 验证→标准检测→场景反馈→迭代优化"全生命周期；覆盖养老、特种、零售、工业等4种以上典型应用 场景，建立严格的数据采集和质量控制规范；布局"本部－展示－外场"3体物理格局，其中，"本部"聚 焦具身智能人形机器人核心研发与训练工场，"展示"以场景展示与机器人示范应用基地为主，"外场"则 构建机器人室外训练验证场。 "四川省人形机器人训练场是实验室创新打造的，具有全国领先的'研发验证－仿真训练－检测认证'一 体化平台，并首创'数据+场景+训练'三位一体服务模式，将构建成渝地区双城经济圈人形机器人协同创 新枢纽，形成'核心研发在成都－场景验证辐射西部－技术输出 ...

作者 | 周智宇 编辑 | 张晓玲 张予彤，这位一度引起争议的金沙江创投前主管合伙人，以一个全新身份走向台前。 近日真格基金在清华大学举办的一场交流会上，张予彤首次以"Kimi总裁"的身份公开亮相。她负责的是 Kimi整体战略与商业化。 张予彤也借着这场演讲，回应了外界对于"独角兽资金不足、算力匮乏"的质疑，强调Kimi的效率优势。 从某种程度上来说，这也是场另类路演。 放眼望去，曾经并肩作战的"大模型六小虎"已在分岔路口渐行渐远：抢滩上市的急迫、无奈折叠万亿参 数雄心的妥协，以及被价格屠夫无情击穿底线的恐慌，共同交织成一幅残酷的众生相。在巨头围剿与资 本退出的双重夹击下，所有的技术信仰最终都必须兑换成财务报表上的数字。 张予彤走向台前，正是月之暗面试图穿越这片商业"无人区"的最后一搏，也预示着这场关乎生死的中场 战事，来到重要赛点。 走向台前 杨植麟需要张予彤。或者更准确地说，处于"中场战事"的月之暗面，急需一位懂资本、懂战略、更懂如 何把技术兑换成商业价值的操盘手。 这是一场跨越十年的重逢，也是一次角色的彻底重塑。作为清华系的"师姐"，张予彤曾是杨植麟上一家 创业公司循环智能的伯乐。 如今，她正式成为这家 ...

Core Insights - The "EVOLVE2025" summit highlighted the launch of a comprehensive enterprise-level intelligent agent roadmap by Zhongguancun KJ, featuring a "3+2+2" product matrix that includes three foundational platforms and two application platforms, aimed at accelerating the large-scale implementation of intelligent agents in various industries [1][2] Group 1: Intelligent Agent Development - The development of large models is rooted in the accumulation of smaller models and data modeling, emphasizing the need for data to be transformed into knowledge through the discovery of hidden patterns [1][2] - Intelligent agents integrate core capabilities such as perception, understanding, decision-making, and control, serving as key vehicles for technology implementation [1][2] - The evolution of intelligent agents is supported by foundational algorithms like deep learning and reinforcement learning, with a focus on enhancing efficiency through collaborative deployment across cloud, edge, and endpoint [1][2] Group 2: Industry Trends and Challenges - The need for precision and lightweight models in large model deployment is critical, with techniques like model distillation helping to reduce computational requirements [2] - There are technical risks such as "hallucinations" in natural language understanding, particularly in accurately grasping Chinese semantics, which remain a long-term challenge [2] - The future direction involves transitioning large models and intelligent agents from general-purpose to specialized applications tailored to specific industries and product scenarios [2] Group 3: AI Agent as a Central Hub - AI intelligent agents are seen as the central brain for enterprises, addressing issues like data silos and process fragmentation by connecting key elements such as people, resources, and systems [3] - Each connection made by intelligent agents generates new interaction data, which in turn iterates the model itself, leading to increased intelligence and value creation for enterprises [3] - The evolution from the internet to mobile internet and now to artificial intelligence represents an evolution of connectivity, with intelligent agents acting as super connectors within and outside organizations [2][3]

Core Insights - The article discusses the transformative impact of AI on various industries, emphasizing the shift towards a new paradigm characterized by "large models + large computing power + large data" as the foundation for AI 2.0 development [1][9] - It highlights the rapid growth of China's AI server market, which is expected to more than double by mid-2025 compared to the same period in 2024, driven by increasing demand [3][9] - The article outlines the evolution of large models, focusing on their deepening, broadening, and lengthening capabilities, which are essential for advancing AI technology [11][12] Industry Trends - The AI infrastructure development is entering a three-phase cycle: short-term acceleration, medium-term expansion, and long-term structural upgrades, necessitating a shift from merely stacking hardware to providing systematic services [10][12] - The penetration rate of AI in core products has increased from 5% in 2018 to an estimated 60% by next year, indicating a significant market evolution [12] - The global AI competition landscape is changing, with companies needing to balance between international and domestic GPU resources amid geopolitical tensions [5][12] Company Developments - Lenovo has introduced the "AI Factory" solution, aiming to streamline AI application development and deployment processes, transforming complex tasks into a standardized, efficient production line [15][16] - The Lenovo Wanquan Heterogeneous Intelligent Computing Platform 4.0 has been upgraded to support various computing scenarios, achieving a throughput of over 12,000 Tokens/s for local model deployment [16] - Lenovo's approach emphasizes the importance of a robust foundational support system to facilitate the transition from traditional computing centers to AI factories, helping businesses reduce initial barriers and operational costs [16]

Core Viewpoint - The emergence of a new Chinese company, Moer Thread, with a groundbreaking GPU has disrupted the long-standing foreign monopoly in the chip industry, potentially reshaping the computing power landscape [1] Group 1: Company Overview - Moer Thread's stock surged 468% on its debut, with an initial price of 114.28 yuan skyrocketing to 688 yuan within 10 minutes, generating significant excitement among investors [3] - It is the first fully functional GPU company to go public in China and represents the largest IPO on the Sci-Tech Innovation Board this year, marking a significant milestone in domestic computing power development [5] - The company has already produced five chip models and iterated four architecture generations, demonstrating its capability beyond mere concepts [10] Group 2: Technological Innovation - Unlike many domestic chip companies that focus on niche markets, Moer Thread has developed a full-featured GPU, akin to a "Swiss Army knife," capable of gaming, AI model training, scientific computing, and graphic design [8] - The founder, Zhang Jianzhong, has extensive experience in the GPU industry, having previously increased the market share of international giants in China from under 50% to over 80% [10] Group 3: Ecosystem Development - The success of a GPU is not solely dependent on hardware; it requires a robust software ecosystem and compatibility with various operating systems and graphic interfaces [12] - Moer Thread has invested in building its own MUSA architecture and a developer community, similar to Apple's App Store, to encourage software development for its GPUs [12][13] - The company has made significant efforts to overcome the challenges of ecosystem establishment, which requires time, reputation, and continuous optimization [13] Group 4: Investment Sentiment - Despite being in a loss-making phase, with over 5 billion yuan in losses over the past three years, institutional investors are heavily investing in Moer Thread, viewing it as a long-term bet on the future of domestic computing power [16][18] - The company has allocated 43 billion yuan to R&D, with 77% of its team focused on research, indicating a strong commitment to technological advancement [18] Group 5: Industry Context - Moer Thread's IPO is seen as a starting point for the domestic GPU market, with other companies like Muxi and Birun preparing for their own IPOs, indicating a burgeoning ecosystem [19] - The success of Moer Thread could lead to increased attention from developers and manufacturers towards domestic alternatives, marking a shift in the industry [19] Group 6: Future Outlook - The journey of Moer Thread signifies a new chapter in China's quest for technological sovereignty and industry security, emphasizing the importance of continuous innovation and ecosystem development [26]

Core Insights - Google officially launched Project Aura, an XR hardware reference design, in collaboration with Chinese AR hardware company XREAL, marking a significant advancement in Android XR operating system architecture [1][3] - Project Aura integrates Gemini AI with spatial awareness capabilities, transitioning AI from a "screen-based intelligence" to a "physical world intelligence" [1][3] Group 1: Project Aura Overview - Project Aura is positioned as the closest realization of the ideal Android XR hardware, enabling Gemini AI to construct environmental semantic maps and understand user intentions in real-time [3][5] - The strategic goal of Android XR is to create an open and unified extended reality platform, embedding AI into the real world [3][5] Group 2: Technical Innovations - Project Aura features a 70° optical field of view (FOV), the largest practical limit for consumer AR, enhancing the natural integration of digital content with the physical environment [5][6] - The X1S spatial computing chip, developed by XREAL, is optimized for spatial AI, providing a high-efficiency inference link from sensor input to semantic output [6][8] - Gemini AI is deeply integrated into the Android XR operating system, functioning as an OS-level service rather than a standalone application, enabling continuous and context-aware interactions [6][8] Group 3: Manufacturing and Future Outlook - Project Aura's core technology relies heavily on Chinese manufacturing, with key components like the X-Prism optical module and X1S chip developed and produced in China [8] - The global supply chain for Project Aura is based in the Yangtze River Delta, facilitating rapid hardware iteration cycles [8] - Project Aura is set to be commercially available by 2026, potentially transforming the XR industry from "display devices" to "spatial intelligent terminals" [8]

Core Viewpoint - In 2025, artificial intelligence (AI) is expected to become a fundamental infrastructure of society, evolving beyond expectations and reshaping various industries and productivity through human-machine collaboration [2]. Group 1: Embracing AI - The year 2025 will witness the global popularity of DeepSeek, the integration of agents across various industries, and accelerated iterations of large models, maintaining the heat of AI throughout the year [3]. - AI's impact is becoming evident as it transitions from a tool for efficiency enhancement to a partner in effectiveness, fundamentally restructuring human creative processes [4]. - AI's value lies in its ability to scale, significantly reducing production costs and rapidly increasing efficiency across multiple industries [4]. Group 2: AI in Marketing and Globalization - AI enables businesses to generate marketing content in multiple languages tailored to local characteristics, enhancing global outreach and marketing strategies [5]. - The dual-edged nature of AI technology raises concerns about job displacement and inequality, highlighting the complex and far-reaching societal impacts of AI [4]. Group 3: Education and Responsible AI Use - Current AI education should focus on three levels: user capability in utilizing AI systems, maintenance and optimization skills for AI systems, and fostering socially responsible research capabilities among creators [6]. - Embracing AI requires a commitment to using it correctly, emphasizing the importance of ethical considerations and risk management in AI deployment [6]. Group 4: Market Dynamics and AI Bubble - There is a growing debate about the existence of a bubble in AI valuations, with some companies experiencing inflated valuations amid unclear commercialization paths [7]. - Historical parallels with the internet boom suggest that while bubbles may exist, they are a normal phenomenon during the rise of new industries, leading to the emergence of major tech giants [7]. Group 5: Future of Robotics and AI Integration - The current enthusiasm for humanoid robots is met with skepticism regarding their practical applications, suggesting a focus on robots performing dangerous or repetitive tasks instead [8]. - Successful investment in AI and robotics requires a deep understanding of technological pathways, application scenarios, and business logic rather than merely chasing concepts [8]. - The development of large models should prioritize meeting urgent human needs, integrating knowledge into foundational models, and updating algorithms based on specific fields [8][9].

白皮书总结了上海数字公共服务的四个创新亮点。一是首创服务体系架构，率先提出"1+N+X"的架构， 通过全市统一数字公共服务平台，整合算力、数据、模型、软件等在内的资源，联合各主体构建中心辐 射、驿站补充的载体布局，打造面向企业的15分钟数智服务圈。二是联创服务载体格局，支持园区、协 会、科研机构等各类主体，立足自身的特色优势、创新打造服务中心和驿站。目前全市已经建设首批18 家中心、13家驿站，初步实现了"需求在哪、服务到哪"的数字服务新格局。三是构建数智服务资源库， 通过市级服务平台、中心驿站，为企业显著降低算力投入、高质量数据获取等方面的成本。四是打响数 字服务官品牌，企业既可以选择一名属地的数字服务官为其提供"一对一"的陪伴式服务，同时属地服务 官面临无法解决的需要跨区域、跨部门协调的复杂问题时，全域数字服务官可以到场支援、形成合力。 中国信通院副总工程师许志远表示，人工智能时代，大模型目前已初步具备一定能力，但在工业等行业 落地应用高度依赖数据赋能。数据将从提升全流程生产效率、促进人工智能的研发与落地、优化资源配 置方面等三方面发挥作用。 记者日前从"数字赋能·惠企共赢"上海数字公共服务主题论坛上获悉 ...

Core Insights - Liang Wenfeng has been recognized as one of the top ten scientists of 2025 by the prestigious journal Nature for his significant contributions to the AI field through the DeepSeek model [1][2] - DeepSeek has disrupted the AI industry by offering a cost-effective model that enhances the global presence of domestic large models, proving that high performance does not necessarily require extensive data or resources [4] Group 1: Recognition and Impact - Liang Wenfeng is described as a "Tech disruptor" by Nature, highlighting his dual identity as a financial expert and a pioneer in AI [3] - The DeepSeek model has achieved remarkable performance in the Agent evaluation, reaching the highest level among current open-source models [4] Group 2: Background of Liang Wenfeng - Liang Wenfeng was born in 1985 in Guangdong and excelled academically, eventually studying electronic information engineering at Zhejiang University [5] - He transitioned into quantitative investment in 2008, capitalizing on the emerging trend of quantitative trading in China [6] - In 2023, he established DeepSeek, focusing on general artificial intelligence (AGI) after recognizing the potential in large models ignited by ChatGPT [6] Group 3: Other Recognized Scientists - Mengran Du, another Chinese researcher, was also recognized for her discovery of the deepest known animal ecosystem on Earth, showcasing significant advancements in deep-sea research [8][10]

以下文章来源于特大号 ，作者特大明白 特大号 . IT B2B 特大号！每日八卦最香艳2B绯闻！ 英伟达H200要解禁了。 12月8日，美国总统特朗普表示将允许英伟达向中国等市场的「经批准客户」出口 H200 AI芯片 ， 并称该安排将附带国家安全条件，由美国商务部落实具体审批与细则。 特朗普还提到美国将从相关对华H200交易中收取约 25%的收入分成 ，这算盘打的噼哩啪啦响。 先不管这事儿后续会怎样，我们今天就来扒一扒，为什么看起来很牛的H200，却是最生不逢时的 一代。 后有Blackwell追兵： 当H200终于在2024年中期开始铺货时，英伟达已经发布了下一代核武器 Blackwell (B200/GB200) 。 对于很多大厂来说，既然手里已经有几万张 H100 了，与其费劲升级到提升只有1.4倍的H200， 不 如跳过这一代，直接等提升4倍的B200 。 | 特性 | H100 | H200 | B200 | B300 | H20 | | --- | --- | --- | --- | --- | --- | | | (标准版) | (大显存版) | (新一代) | (Ultra版) | (中国特 ...