Core Viewpoint - The future of AI technology is projected to evolve significantly, with robots potentially outnumbering humans in various sectors, including factories and households, as outlined by Zhang Yaqin, the director of Tsinghua University's Institute of Intelligent Industry Research (AIR) [1]. AI Technology Development Directions - AI large models are seen as a cornerstone of digitalization 3.0, with key development directions including multi-modal intelligence, autonomous intelligence, edge intelligence, physical intelligence, and biological intelligence [1][8]. - The transition from "digitalization 1.0" and "2.0" to "digitalization 3.0" involves a shift from small models to large models and from single-modal to multi-modal systems, indicating a broad application of AI across various industries [2]. Five Evolution Trends of AI Large Models - Large models and generative AI are expected to be the main technologies and industrial routes over the next decade, driving innovation and transformation [5]. - The ecosystem of AI will be significantly larger than that of personal computing and mobile internet, with foundational large models coexisting with vertical and edge models [6]. - Key elements of large models include tokenization and scaling laws, which enhance the model's ability to process diverse data types and improve performance with increased parameters and data [7]. Autonomous Intelligence - Autonomous intelligence will lead to personalized intelligent agents capable of self-planning, coding, and optimizing tasks, achieving high levels of autonomy and self-iteration [8]. - New algorithmic frameworks are necessary to overcome current inefficiencies and high energy consumption in existing algorithms, with potential breakthroughs expected in the next five years [9]. Path to General Artificial Intelligence - General artificial intelligence is anticipated to be realized within 15 to 20 years, with significant advancements expected in information intelligence, physical intelligence, and biological intelligence [10]. Future of Autonomous Driving - Autonomous driving is projected to be a key application of physical intelligence in the next five years, with safety levels expected to exceed human drivers by at least ten times [11]. - Large models and generative AI will enhance the generalization capabilities of Level 4 autonomous driving systems by generating high-quality edge case data and improving scenario simulation [12]. - The integration of multi-modal sensor data and end-to-end training will enable real-time collaboration between cloud-based large models and vehicle-specific models [13]. - Future autonomous driving applications will focus on single-vehicle intelligence, with a "vehicle-road-cloud" integration to ensure safety and optimize traffic flow [14]. - By 2025, autonomous driving may reach a pivotal moment, with 10% of new vehicles expected to have Level 4 capabilities by 2030 [15].

出品｜搜狐科技 作者｜杨锦 在大模型的热闹之外，朱松纯不惧做一个少数派。 他认为，科学本质是用简约的模型解释复杂的现象，比如杨振宁、爱因斯坦这些科学家，他们构建世界的模型里可能只有一两个参数。而今天的大模型，动 辄百亿甚至千亿级的参数，还不可解释，"从这个角度看，确实是非常丑陋的。" 但他也承认，从工程的角度来讲，大模型又能够在某些方面产生比较好的结果。 他和北京通用人工智能研究院（通研院）及北大几位老师最新合编的图书《通用人工智能标准、评级、测试与架构》，针对通用人工智能提出了系统的标 准、评级、测试与架构体系，对大模型也适用。 什么是智能？怎么样才算是通用了？在这套评测体系下，智能体不仅要完成任务，还需要自主定义任务，这意味着，智能体没办法像在其他评测体系上一 样"刷榜"。 过去20年，刷榜像附着在人工智能发展史上的藤壶，他本人也曾长时间地站在数据驱动和刷榜的一线。2004年，朱松纯和另一位全球计算机领域顶级科学家 沈向洋，在他的湖北家乡创建了莲花山研究院，是最早规模性地做大数据标注的机构。 刷榜是针对性地做端到端训练，而通用泛化的任务，是一个无穷的任务，比如训练机器人学会抓杯子，这很快，但问题是，稍微换 ...

再度实现万亿元规模的年营收后，中国移动2025年在云计算领域的战略正在进一步明晰。 在日前于苏州举行的"2025中国移动云智算大会"上，中国移动宣布，已建成全国首个覆盖"通算算力、智能算力、量子算力、超算算力"四算融合的算力网络 【通算算力8.5EFLOPS（FP32），智能算力43EFLOPS（FP16），量子算力1138Qubit，超算算力800PFLOPS】，算力总规模占全国的1/6，以此为国家一体 化算力平台的建设思路提供借鉴。 对此，中国移动董事长杨杰在会上指出，通用人工智能技术的创新突破和广泛应用，正在带来"两个新变化"：一是算力总量的指数增长，未来3年，我国智 能算力规模增长超2.5倍，年均复合增速近40%；二是算力结构的显著变化，推理算力需求将超过训练算力需求，未来3年，推理算力年复合增速将达到训练 算力的近4倍，到2028年，推理算力规模将超过训练算力规模。这一进程将给算网发展注入新的动能，推动产业进入新一轮增长周期。 因此，杨杰认为，AI发展呈现"两个规模效应"。一是技术能力的规模效应，随着高水平算法、高性能算力、高质量数据的持续投入，AI整体能力将实现指数 级增长，每年提升约10倍。二是 ...

Group 1 - The company reported a revenue of 1.22 billion yuan for the year 2024, representing a year-on-year growth of 2.80% [1] - Despite the revenue growth, the company did not achieve profitability due to impairment losses in its spatiotemporal data information business, although operational conditions improved compared to the previous year [1] - The net cash flow from operating activities for 2024 was 102 million yuan, showing a significant increase of 377.43% year-on-year [1] Group 2 - The company has focused on developing high-precision positioning equipment and solutions, with revenue from core businesses such as precise spatiotemporal perception equipment, Beidou high-precision industry applications, and intelligent driving and navigation control applications all showing growth [1] - The revenue from high-precision positioning equipment and solutions now accounts for 86.21% of the company's total revenue [1] - The intelligent driving vehicle-mounted high-precision business grew by approximately 40% year-on-year, with a substantial increase in orders for vehicle-mounted software and hardware products, exceeding 100% year-on-year [1] Group 3 - The company is actively enhancing its overseas brand presence through its HITARGET and SATLAB brands, establishing a subsidiary in Hungary to serve as a technical and warehousing center for European operations [2] - The company successfully won a bid for geological disaster monitoring and hydrological monitoring projects from the Ugandan government, marking a significant entry into the African market [2]

一上台，商汤科技董事长兼CEO 徐立就感叹，"如果三个月不更新自己的认知，可能就会被淘汰。" 4月10日，商汤举办2025技术交流日，徐立正式发布全新升级的"日日新SenseNova V6"（以下简称"日日 新V6"）大模型体系。 在徐立看来，多模态模型和通用人工智能的发展，画上约等号，以计算机视觉起家的商汤，从视觉能力 到原生多模态模型的布局，则是自然延伸。 商汤科技联合创始人兼大模型首席科学家林达华向《21CBR》记者表示，公司去年5、6月份就在做多模 态的探索，到了9、10月，技术路线基本跑通。 林达华称，之所以专注多模态推理，而非纯文本赛道的竞争，在于坚信未来的交互，必然是多模态的。 日日新V6，作为拥有超6000亿参数的MoE原生多模态通用大模型，凭借单一模型就可以完成文本、多 模态等各类任务。 其技术能力上的突破，重在四个方面： 长思维链：超过200B高质量多模态长思维链数据，最长64K思维链；数理能力：数据分析能力大幅领先 GPT-4o；推理能力：多模态深度推理国内第一，对标OpenAI o1；全局记忆：率先在国内突破长视频理 解，支持10分钟的视频理解及深度推理。 值得一提的是，长记忆。林达华 ...

Core Viewpoint - The article discusses the evolution of artificial intelligence (AI) from its inception to the current state, emphasizing the transition from computational intelligence to embodied intelligence, and speculating on the future of general artificial intelligence [2][5][15]. Group 1: Evolution of AI - AI has undergone at least three significant phases since the term was coined in 1956, with notable advancements occurring around 2010 and the release of models like ChatGPT in 2022 [5]. - The concept of embodied intelligence is emerging as a new frontier, where machines not only process information but also interact with the physical world [5][12]. Group 2: Embodied Intelligence - Embodied intelligence is not limited to humanoid robots; it refers to a methodology or developmental stage where machines possess physical forms and can engage with their environment [6][11]. - The article highlights the importance of sensory perception in embodied intelligence, suggesting that true intelligence requires a physical body to interact with the world [13][14]. Group 3: Challenges and Future Directions - Traditional Turing tests are deemed insufficient for evaluating AI's capabilities, as they do not account for the complexity of real-world interactions and emotional understanding [8][10]. - The evolution of embodied intelligence is expected to be accelerated by modern technology, allowing for more rapid development of complex functionalities compared to human evolution [15].

Core Viewpoint - The article discusses the ongoing debate in the embodied intelligence industry regarding the commercialization prospects, highlighting a divide between cautious investors and optimistic supporters of the technology [2][8]. Group 1: Investment and Financing - Starry Sky Technology, an embodied intelligence robotics company, recently completed A2 and A3 rounds of financing totaling over 300 million RMB, following a previous A round of nearly 300 million RMB, bringing the total financing to 1 billion USD [2][3]. - The investment was led by K2VC, with participation from Lenovo Capital, Haier Capital, IDG Capital, and Hillhouse Capital, indicating strong interest from both industry and venture capital [2][8]. Group 2: Industry Concerns and Responses - Concerns were raised by Zhu Xiaohu, General Manager of Jingshan Capital, regarding the commercial viability of embodied intelligence, suggesting that current customer profiles are largely speculative and that high prices limit market demand [3][4]. - In response, Starry Sky Technology clarified that Jingshan Capital had exited its angel investment in September 2024 and emphasized that there had been no direct communication between the two parties [6]. Group 3: Perspectives on Future Development - Supporters of embodied intelligence, including industry leaders and scholars, argue that the technology is on the verge of significant breakthroughs, with predictions that robots will become as ubiquitous as smartphones in the coming years [9][10]. - Tsinghua University Professor Zhang Yaqin noted that while humanoid robots may take 5-10 years to mature, vertical applications in logistics and food processing could see advancements within 1-2 years [9]. - Elon Musk predicts that by 2040, the number of humanoid robots could reach 10 billion globally, with prices dropping below $20,000, positioning them as foundational to human civilization [10][11]. Group 4: Technological Optimism and Market Dynamics - The article highlights a dichotomy in the industry, with some companies engaging in price wars to capture market share, while others, like Starry Sky Technology, focus on a technology-business model integration to validate their value in industrial applications [8]. - The CEO of Starry Sky Technology, Gao Jiyang, proposed a "one brain, multiple forms" strategy aimed at overcoming commercialization challenges, claiming a 10-fold reduction in single-task learning costs and a 100% increase in simulation training efficiency [7][8].

算力的"回弹效应"加速显现，释放算力服务新需求。通用人工智能技术的创新突破、广泛应用，将大幅提升算力使用效率，引发算力领域的"杰文斯悖论"， 带来"两个新变化"。一是算力总量的指数增长，未来3年，我国智能算力规模增长超2.5倍，年均复合增速近40%。二是算力结构的显著变化，推理算力需求 将超过训练算力需求，未来3年，推理算力年复合增速将达到训练算力的近4倍，到2028年，推理算力规模将超过训练算力规模。这一进程将给算网发展注入 新的动能，推动产业进入新一轮增长周期。 连接的"加速效应"不断拓展，催生信息消费新形态。高速、移动、安全、泛在的网络，推动"人机物"多元主体的全面连接，加速数算智的系统融入，引 发"比特×瓦特"的融合聚变，实现能力和价值的显著倍增，催生信息消费"新三样"等新业态新模式。一是AI智能终端，云边端算力、数据和模型在网络的带 动下加速整合贯通，显著提升AI智能体的性能，深度嵌入各类终端设备，打造能够提供专家级、个性化服务的"生活助理"，未来AI智能体数量将超过人的数 量。二是智能网联汽车，依托高性能算力和低时延网络，车路云协同不断深化，加速端到端智能驾驶的落地，拓展人类数智生活的"第四空间" ...

Core Insights - Starry Sea has successfully completed A2 and A3 rounds of financing, raising over 300 million RMB, bringing total financing since 2025 to nearly 100 million USD [1][6] - The valuation of Starry Sea's A3 round has doubled compared to the A1 round earlier this year, with ongoing discussions for a new round at a valuation of 5 billion RMB [3][6] - The lead investor, KKR Fund, aims to leverage its global ecosystem resources to accelerate Starry Sea's development in the field of embodied intelligence [3][8] Financing Details - The A2 and A3 rounds were led by KKR Fund, with participation from Lenovo Capital, Haier Capital, and existing investors such as IDG Capital and Hillhouse Capital [5] - The A1 round, completed in February, raised nearly 300 million RMB, led exclusively by Ant Group, with additional investments from existing shareholders [5][6] - Cumulatively, the financing since the angel round in February 2024 has reached approximately 100 million USD [6] Valuation and Growth - Starry Sea's rapid valuation growth is attributed to its comprehensive business lines, including models, entities, and commercialization [6] - The company is recognized for its strong capabilities across the entire supply chain, making it one of the few companies in China with end-to-end AI algorithm capabilities and commercial validation [6] - If the valuation reaches 5 billion RMB, Starry Sea will position itself as a leader in the second tier of the industry, with top-tier companies valued over 10 billion RMB [6] Industry Outlook - KKR Fund expresses confidence in the potential of embodied intelligence in the global market, viewing it as a key pathway to achieving general artificial intelligence [8] - The sector is attracting significant investment, with multiple high-profile funding rounds announced recently, indicating a growing interest in embodied intelligence [8][9] - According to KPMG, the number of financing events in the humanoid robot sub-sector has increased significantly, with projections indicating a market size of 6 trillion RMB by 2050 in China [9]

每经记者 张蕊 每经编辑 陈星 每经4月2日电（记者张蕊） 4月1日，中国电子信息产业发展研究院（又称赛迪研究院）在京举办2025赛迪论坛，本次论坛 以"新型工业化：新动能 新篇章"为主题。 在当日下午举办的"十五五"规划分论坛上，赛迪研究院电子信息研究所所长陈渌萍作了《"十五五"时期我国通用人工智能 产业发展趋势研究》成果发布。 不过，陈渌萍也提到我国通用人工智能产业面临的重大挑战。首先，大模型技术在数据-算法-算力领域尚存瓶颈。其次， 大模型伦理安全仍是全球治理的重要挑战。此外，旺盛的大模型应用场景需求加剧人才供需矛盾。 论坛现场展示的机器人 图片来源：每经记者 张蕊 摄 在瓶颈方面，从数据看，高质量专业数据集缺乏，行业内数据共享难度较高，数据标准和治理保障体制不完善，数据流通 利用基础较为薄弱；从算法看，模型自身存在不可解释性和可靠性风险，在医疗、法律等特定领域应用受限；从算力看， 算力供给仍不充分、不平衡，能源消耗瓶颈明显。 封面图片来源：每经记者 张蕊 摄 在核心技术迭代演进方面，陈渌萍提到，AI专用芯片和量子计算的突破将推动计算硬件实现高效能、低功耗和绿色化的跨 越式发展。数据质量大幅提升，跨领域 ...