这一年，D.E. Shaw 为计算机行业做了两个贡献。一个副总裁带队，做出了当时罕见的免费电子邮件产 品 Juno，成功上市；另一个副总裁离职，带着自己和老板讨论产生的好点子开车去了西雅图，做出了全 世界的电商鼻祖、市值超过 20000 亿美元的亚马逊。 30 年后，又有一家量化公司的 "副业" 影响整个计算机行业：管理数百亿元的中国头部量化公司幻方， 推出大语言模型 DeepSeek R1，没花一分钱营销就震撼全球，用户涌来的速度甚至快过早年的抖音。 贝索斯创办亚马逊，或者梁文锋造出 DeepSeek 的主要原因自然不是因为他们做过量化，而是因为他们 骨子里都是创业者。但量化投资这个极度追求人才密度且极度保密的行业文化，确实提供了适合大模型 研发的环境。 招来一群聪明人不必然导致创新，叠加一个简单的环境才够。量化公司证明了这一点，DeepSeek 则证明 这也适用于大模型研发。 剥离主观因素，在数据里挖掘规律 从十万次交易到千亿参数的 AI 进化。 文 丨 孙海宁 编辑 丨 黄俊杰 1994 年，量化公司是当时最神秘最热门的技术公司，他们雇用数学家和物理学家，成批买来高性能计算 机做交易。这个行业里的标杆公 ...

Core Viewpoint - The article discusses the rapid development and popularity of the intelligent agent sector, particularly highlighting the impact of the Manus product and the emergence of open-source projects like OWL and OpenManus, which have sparked a wave of innovation and competition in the field [1][2][3]. Group 1: Manus and Its Impact - Manus has significantly influenced the intelligent agent landscape, leading to a surge in both open-source and commercial closed-source products [1]. - The official social media account of Manus faced a temporary ban but has since resumed, promising more demonstrations and updates [12]. - Manus has gained traction internationally, with strategies such as distributing invitation codes to influencers and users [13][14]. Group 2: Open-Source Projects - The OWL project, developed by the CAMEL-AI team, has integrated the DeepSeek model into a multi-agent collaboration framework, showcasing its capabilities [3][4]. - OWL achieved an average score of 58.18 in the GAIA benchmark, ranking first among open-source projects [5][6]. - The CAMEL-AI team expressed confidence in improving their scores in the GAIA benchmark, despite some gaps in Level 2 and Level 3 scores compared to competitors [7]. Group 3: GAIA Benchmark - The GAIA benchmark, created by Meta AI, Hugging Face, and AutoGPT teams, consists of over 450 complex questions designed to evaluate the capabilities of intelligent agent systems [24][25]. - The benchmark is divided into three levels of difficulty, with Level 1 requiring simple problem-solving and Level 3 demanding advanced capabilities [26][27]. - Manus scored 57.7% in Level 3, significantly outperforming other systems, while its Level 2 score was close to that of commercial systems [28][29]. Group 4: User Experiences and Market Trends - Users have reported high satisfaction with Manus, noting its ability to accurately gather personal information and perform complex tasks [18][19][20]. - The willingness to pay for Manus is higher among international users compared to domestic ones, as it offers a more affordable alternative to other high-end AI solutions [17]. - The article highlights a growing interest in agent-related projects on platforms like GitHub, indicating a trend towards the development of specialized intelligent agents in various fields [8][9].

Investment Rating - The report does not explicitly state an investment rating for the database industry. Core Insights - The Chinese distributed transaction database software market is projected to reach $1.5 billion in the first half of 2024, reflecting an 18.5% year-on-year growth, with public cloud market share at 61.2% [4][38] - OceanBase and GoldenDB are gaining traction in the market, with OceanBase scoring over 700 points and GoldenDB showing significant improvements in its latest version [7][10] - The integration of large language models (LLMs) with database technologies is highlighted as a key trend, showcasing practical applications and collaborations [5][19] Summary by Sections 1. February Database Rankings Interpretation - OceanBase leads the rankings with a score of 753.90, followed by PolarDB at 632.21 and GaussDB at 630.44 [8][9][11] - GoldenDB and Kingbase also show strong performances, with scores of 621.23 and 611.62 respectively, indicating their growing market presence [10][11] 2. Database Industry News and Dynamics - The report notes the release of Oracle's Exadata X11M, which boasts over a 55% performance improvement compared to its predecessor [43][44] - The market is increasingly competitive, with major players like Alibaba Cloud, Tencent, and Huawei dominating the landscape [38][40] 3. LLM + Database - The report discusses the synergy between LLMs and databases, emphasizing the role of vector databases as optimal partners for LLM applications [5][19] 4. Typical Cases of Chinese Database Products - The report highlights significant procurement activities in January 2025, with domestic databases winning contracts exceeding 100 million yuan, particularly in the finance and government sectors [23][24][27] - Notable projects include the procurement of GoldenDB by Guangfa Bank for approximately 34.89 million yuan and various projects involving OceanBase [23][24][27]

Core Viewpoint - The government work report emphasizes the need for a balanced approach to economic growth, focusing on stability while promoting progress, amidst a complex external environment and internal challenges [2][4][3]. Group 1: Economic Situation Assessment - The report highlights that the economic recovery is solid, driven by macroeconomic policies, but acknowledges increased external pressures from geopolitical tensions and trade challenges [2][3]. - It identifies internal issues such as insufficient effective demand and difficulties faced by some enterprises, alongside new concerns regarding social welfare and local government finances [2]. Group 2: Development Goals - The economic growth target for this year is set at around 5%, consistent with previous years, aiming to balance short-term needs with long-term development goals [5]. - Employment targets remain at 12 million new urban jobs, with a focus on addressing structural employment issues, particularly among youth and migrant workers [6][7]. Group 3: Macroeconomic Policies - Fiscal policy is set to be more proactive, with a total fiscal space expanding to 13.86 trillion yuan, including a record deficit rate of 4.0% [9][10]. - Monetary policy will remain moderately accommodative, with an emphasis on maintaining liquidity and aligning social financing growth with economic growth and inflation targets [12][13]. Group 4: Key Initiatives - Consumer spending is prioritized as a key driver for economic growth, with initiatives to boost consumption through various measures, including a doubling of funds for old-for-new consumer goods programs [15]. - The report emphasizes the importance of technological innovation and industrial upgrades, particularly in emerging sectors like AI and quantum technology, to enhance productivity [16][17]. Group 5: Risk Management - The report outlines a focus on managing risks in real estate, local government debt, and small financial institutions, advocating for a gradual approach to risk resolution [23][26][27]. - Specific measures include controlling new real estate developments and enhancing transparency in local government debt management [24][26]. Group 6: Capital Market Outlook - The report indicates a shift in the A-share market from concept-driven to performance-driven dynamics, with a focus on technology sectors benefiting from policy support [32][34]. - It anticipates stable long-term performance for the A-share market, with an emphasis on enhancing the capital market's stability and value through reforms [33][36].

Core Viewpoint - The Mobile World Congress (MWC) 2025 is evolving to focus more on AI technologies rather than traditional communication, resembling the Consumer Electronics Show (CES) [6][4]. Group 1: Event Overview - MWC 2025 officially opened on March 3 in Barcelona, showcasing major tech companies like Lenovo, Huawei, Xiaomi, Google, Samsung, and LG [3]. - The event is increasingly showcasing what technologies can do for users rather than just the technologies themselves [6]. Group 2: Company Highlights - Xiaomi made a significant impact at MWC with its SU7 Ultra, which garnered over 19,000 orders shortly after its release, overshadowing other products like the Xiaomi 15 Ultra [8]. - Google had a prominent presence with its Android, Google Cloud, and Google Pixel exhibits, utilizing its Gemini model to attract attendees [10]. - Notably absent from the event was DeepSeek, a major language model that has gained attention in China, indicating a potential gap in AI solution showcases [11]. Group 3: Technological Innovations - Lenovo showcased a new foldable laptop with an external screen, continuing its trend of innovative display technologies [17]. - The "Magic Bay" technology from Lenovo allows for multiple screen configurations, enhancing user experience by providing additional display options [18]. - Lenovo's Tiko device, which serves as an AI assistant with interactive capabilities, represents a shift towards more personalized computing experiences [20][21]. Group 4: Industry Trends - The shift in focus at MWC reflects a broader trend in the tech industry where AI is becoming a central theme, influencing hardware and software development [4][22]. - The evolution of personal computing devices is being driven by AI advancements, suggesting a significant transformation in how these devices will function in the future [22].

Core Viewpoint - Tencent Yuanbao has rapidly ascended to the top of the free app download rankings in China, indicating strong user growth and engagement in the AIGC application sector [1][3]. Group 1: User Growth and Market Position - As of March 3, Tencent Yuanbao ranked first in the free app download chart, surpassing DeepSeek and positioning itself as the fastest-growing AIGC app [1][3]. - On February 22, Tencent Yuanbao experienced a significant jump of over 100 places in the download rankings, indicating a surge in user interest [3]. Group 2: Product Features and Innovations - Tencent Yuanbao launched a desktop version on March 1, supporting both Windows and macOS, which enhances user experience by allowing image reading and intelligent dialogue [5]. - The desktop version integrates advanced capabilities, enabling users to analyze images and documents, thereby improving reading efficiency [5][6]. - Future updates for the desktop version will include features like word search and translation, as well as screenshot inquiries [7]. Group 3: Integration with DeepSeek - Tencent Yuanbao has integrated multiple models, including DeepSeek-R1 and DeepSeek-V3, enhancing its ability to understand images and documents [15]. - The integration of DeepSeek's capabilities with Tencent's multi-modal understanding technology allows for a more comprehensive analysis of images beyond simple text recognition [14][13]. - This innovation reflects a shift from merely utilizing existing model capabilities to creating differentiated value through product innovation [16]. Group 4: Strategic Adjustments and Industry Trends - Tencent has proactively embraced the trend of integrating DeepSeek across its product lines, demonstrating agility in its strategic adjustments [18]. - The company has incorporated DeepSeek into various products, including WeChat, Tencent Documents, and QQ Music, expanding its application across its extensive user base [19][20]. - The integration of DeepSeek into Tencent's financial services and enterprise communication tools enhances the professionalism and timeliness of these services [21][22]. Group 5: Competitive Landscape - Tencent's extensive C-end user base and diverse product matrix position it well to accelerate the practical application of large models in various scenarios [24]. - The industry anticipates that Tencent's innovations will lead to new AI application experiences beyond traditional Q&A formats, leveraging its vast user engagement [24].

Core Viewpoint - Nvidia's CEO Jensen Huang expressed excitement about the potential demand for AI inference, which is expected to far exceed current large language models (LLMs), potentially requiring millions of times more computing power [1][5]. Group 1: AI Inference and Demand - The demand for inference will significantly increase, especially for long-thought inference AI models, which may require several orders of magnitude more computing power than pre-training [5]. - Nvidia's Blackwell architecture is designed for inference AI, improving inference performance by 25 times compared to Hopper while reducing costs by 20 times [6][34]. - The DeepSeek-R1 inference model has generated global enthusiasm and is an outstanding innovation, being open-sourced as a world-class inference AI model [1]. Group 2: Financial Performance and Projections - Nvidia reported record revenue of $39.3 billion for the fourth quarter, a 12% quarter-over-quarter increase and a 78% year-over-year increase, exceeding expectations [32]. - The data center revenue for fiscal year 2025 is projected to be $115.2 billion, doubling from the previous fiscal year [32]. - Nvidia's CFO Colette Kress expects profit margins to improve once Blackwell production increases, with margins projected to be in the mid-70% range by the end of 2025 [2][11]. Group 3: Product Development and Supply Chain - The supply chain issues related to the Blackwell series chips have been fully resolved, allowing for the next training and subsequent product development to proceed without hindrance [1]. - Blackwell Ultra is planned for release in the second half of 2025, featuring improvements in networking, memory, and processors [16][60]. - Nvidia's production involves 350 factories and 1.5 million components, achieving $11 billion in revenue last quarter [8][53]. Group 4: Market Dynamics and Growth Areas - The global demand for AI technology remains strong, with the Chinese market's revenue remaining stable [20][68]. - Emerging fields such as enterprise AI, agent AI, and physical AI are expected to drive long-term demand growth [14][24]. - Nvidia's full-stack AI solutions will support enterprises throughout the entire AI workflow, from pre-training to inference [25]. Group 5: Infrastructure and Future Outlook - The current AI infrastructure is still utilizing various Nvidia products, with a gradual update expected as AI technology evolves [26][27]. - Nvidia's CUDA platform ensures compatibility across different generations of GPUs, facilitating a flexible update process [28]. - The company anticipates significant growth in data center and gaming businesses in the first quarter, driven by strong demand for Blackwell [44].

Core Viewpoint - The article reflects on the recent wave of AI startups that have failed or been acquired, highlighting the harsh realities of the AI industry and the challenges faced by companies in this rapidly evolving landscape [2][29]. Group 1: AI Startup Failures - From November 2022 to July 2024, approximately 80,000 AI-related companies in China have disappeared, indicating a significant contraction in the sector [2]. - The article memorializes companies that were once promising but ultimately succumbed to market pressures before the AI revolution fully materialized [2]. Group 2: Case Studies of Failed Companies - **Wave Intelligence**: Founded by a young entrepreneur, the company quickly gained traction with significant funding and product launches but was ultimately acquired by OPPO, with its founder moving to the tech giant [3][4]. - **Afiniti**: An established AI unicorn that matched customers with service representatives, Afiniti declared bankruptcy after 18 years due to a lack of profitability and internal scandals [5][6]. - **Eagle Eye Wisdom**: This company aimed to digitize traditional Chinese medicine but collapsed shortly after being acquired by a public company, highlighting the fragility of even well-backed startups [8][9]. - **Huaxia Chip**: Founded in 2014, this company aimed for complete independence in chip design but faced bankruptcy in 2024 due to financial mismanagement despite technological achievements [15][16]. - **Stability AI**: Known for its open-source model, the company struggled to monetize its technology and faced leadership changes, leading to a precarious financial situation [20][21]. - **Character.AI**: Initially seen as a competitor to OpenAI, the company faced a leadership exodus and was acquired by Google, reflecting the trend of startups being absorbed by larger firms [26][27]. Group 3: Industry Insights - The article emphasizes that many AI startups are unable to survive the transition from innovation to sustainable business models, often leading to acquisitions by larger companies as a means of survival [20][29]. - The narrative suggests that the failures of these companies serve as cautionary tales for future entrepreneurs in the AI space, underscoring the importance of aligning technological aspirations with commercial viability [29][30].

"从开源论文、开源代码出发，现在已经进化到开源思维链了嘛！" 文丨Andrew Lu 注释丨贺乾明 程曼祺 2 月 18 日，Kimi 和 DeepSeek 同一天发布新进展，分别是 MoBA 和 NSA，二者都是对 "注意力机 制"（Attention Mechanism）的改进。 今天，MoBA 的一位主要研发同学 Andrew Lu 在知乎发帖，自述研发过程的三次踩坑，他称为 "三入思过 崖"。他在知乎的签名是"新晋 LLM 训练师"。 这条回答下的一个评论是："从开源论文、开源代码出发，现在已经进化到开源思维链了嘛。" 注意力机制之所以重要，是因为它是当前大语言模型（LLM）的核心机制。回到 2017 年 6 月那篇开启 LLM 革命的 Transformer 八子论文，标题就是：Attention Is All You Need（注意力就是你所需要的一 切），该论文被引用次数至今已达 15.3 万。 注意力机制能让 AI 模型像人类一样，知道在处理信息时该 "重点关注" 什么、"忽略" 什么，抓住信息中最 关键的部分。 在大模型的训练阶段和使用（推理）阶段，注意力机制都会发挥作用。它的大致工作原理是 ...

DeepSeek 的大语言模型（LLM）因其在性能上接近 ChatGPT ，但成本却大幅降低而受到关注。市 场的即时反应褒贬不一。虽然数据训练成本，比如数据标注和归类等方面的成本可能没有体现出 来，而这部分成本由政府支持，但 DeepSeek 在训练效率和低成本方面的优势依然十分明显。 DeepSeek 能否助力中国芯片制造？ 中国的存储芯片或已具备成本竞争力 ： 假设中国政府对构成总成本很大一部分的固定成本提供支持 ，那么与同行相比，中国已能实现有竞争力的成本。例如，2024 年第一季度 DRAM （动态随机存 取存储器）每 Gb （ 千兆字节 ）的价格为 $0.34 ，此时高价的 HBM （高带宽存储器）对平均售价 的影响较小。而韩国 DRAM 的总成本大约为售价的 67% ，约为 $0.23 ，在不计固定成本的情况 下，中国的成本可能低至 $0.20 。（不过，中国的固定成本远高于韩国。） 高效的软件为低端硬件打开市场： 中国的策略是通过规模优势弥补与竞争对手在性能上的差距。华 为最新的 GPU —— Ascend 920 支持 HBM2 和 HBM2e ，而这些对于行业同行来说已是两年前的标 准，并未 ...