Core Viewpoint - The A-share market is experiencing a fluctuating upward trend, with specific sectors such as CRO, weight loss drugs, and innovative pharmaceuticals showing significant gains. The rise of self-controllable technology and the acceleration of AI commercialization are driving the technology sector's performance [1][2]. Group 1: Market Performance - The A-share market maintained a fluctuating upward trend, with a slight pullback in the afternoon [1] - The CRO, weight loss drugs, and innovative pharmaceuticals sectors saw the highest gains in the afternoon [1] - The 信创 ETF (562570) rose nearly 1% at the close, with top holdings including 用友网络 (600588) and 泛微网络 (603039) [1] Group 2: Industry Trends - The computer industry has entered a new AI Agent market phase, accelerating the commercialization of technology [1] - The 信创 industry chain is undergoing mergers and acquisitions, evolving towards "technology complementary integration" and "ecosystem closure construction" [1] - Current mergers and acquisitions exhibit three main characteristics: 1) Technology collaboration orientation, where companies acquire to fill technological gaps [2] 2) Policy-driven concentration, enhancing industry concentration and encouraging mergers for resource optimization [2] 3) International ecological layout, allowing companies to acquire core technologies through cross-border mergers [2] Group 3: Related ETFs - 信创 ETF (562570) is closely related to the strategic restructuring of 海光信息 and 中科曙光, which are significant components of the index [3] - 云计算50 ETF (516630) tracks an index with a high AI computing content, covering various popular computing concepts [3] - As of May 30, 中科曙光 is the sixth largest component of the 云计算50 ETF, with a weight of 4.32% [3]

作者 | 孟旭 编辑 | 李忠良 策划 | AICon 全球人工智能开发与应用大会 在近期举办的 AICon 全球人工智能开发与应用大会·上海站（2025） 现场，网易有道词典笔产品负责人孟旭以一款全新的 AI 原生硬件 【有道 AI 答疑笔】 为例，分享了智能学习硬件在大模型技术催化下的变革逻辑——从解决单一需求的"学习工具"，进化为陪伴学习的"智能伙伴"。 孟旭指出，从多年的经验和认知出发，有道智能学习硬件的进化本质是 用户需求、硬件创新与 AI 技术三者的螺旋推进，三者像齿轮一样咬合转动，推 动产品进化 。即使是在大模型爆发的当下，纯软件升级或者纯硬件创新都更像是炫技，唯有软硬结合才能让技术润物无声地渗入场景，去解决用户的真 问题 ，这也是垂类硬件在技术爆发时代的生存法则。 以下根据演讲实录整理（部分内容有删改），供大家深入了解： 大家好，我来自网易有道硬件产品团队，我叫孟旭。 现在 AI 可以说无处不在了，作为智能学习硬件的产品团队，我们也一直在思考：当 AI 教育碰撞，如何让这项前沿技术真正成为孩子学习成长路上的"智 慧引路人"? 如何突破传统学习工具的局限，解决孩子在学习过程中遇到的实际痛点，为他 ...

Investment Rating - The report maintains a "Recommended" investment rating for the AI industry, highlighting the continuous evolution of domestic agents and the promotion of universal protocols for systematic applications [2]. Core Insights - The AI agent field is experiencing rapid advancements, with capabilities doubling approximately every seven months since the release of ChatGPT in 2022, leading to an exponential increase in the tasks that AI agents can complete [5]. - Major internet companies are competing to capture the AI agent market, which is expected to surpass the existing app ecosystem by providing more versatile and service-oriented solutions [9]. - The introduction of the Agent2Agent (A2A) protocol by Google aims to enhance communication and collaboration between different AI agents, significantly improving operational efficiency across platforms [69][72]. Summary by Sections Section 1: Domestic Intelligent Agent Technology Innovation - The report discusses the launch of ByteDance's Coze Space, a versatile agent product designed to facilitate efficient collaboration between users and AI agents, enabling the completion of complex tasks [13][14]. - Coze Space features two collaboration modes: an exploration mode for quick task completion and a planning mode for in-depth user engagement [14][17]. - The report also mentions the introduction of specialized agents within Coze Space, such as a user research expert and a stock observation assistant, which enhance the platform's functionality [19][20]. Section 2: MCP Open Protocol Continues to Expand - The A2A protocol allows different AI agents to communicate and collaborate on complex tasks, enhancing the overall efficiency and innovation potential of AI systems [69][72]. - Microsoft has announced support for the A2A protocol through its Azure AI Foundry and Microsoft Copilot Studio, indicating a significant shift towards collaborative AI systems [69][72]. - The report highlights the integration of various tools and applications with the MCP protocol, enabling seamless task execution across platforms [76][80]. Section 3: Related Companies - The report identifies several companies involved in the AI agent space, including Zhuoyi Information, which is developing low-code IDE tools integrated with AI capabilities to enhance software development efficiency [84][85]. - Puyuan Information is also mentioned for its low-code platform that incorporates advanced model capabilities to assist developers in code generation and data modeling [89]. - Hehe Information has launched an MCP server service for document processing, aimed at facilitating the use of intelligent document processing agents in various industries [90].

Group 1 - The core viewpoint highlights the recent rebound in technology sectors, particularly in military, pharmaceutical, and TMT industries, with the Sci-Tech Innovation Board ETF (589630) rising nearly 1.5% [1] - The A-share merger and acquisition market has been active this year, with significant participation from Sci-Tech Innovation Board companies, especially in sectors like biomedicine, semiconductors, and new-generation information technology [1] - Data indicates that out of 86 major restructuring events in 2025, 18 involved Sci-Tech Innovation Board companies, showing a significant increase compared to the same period last year [1] Group 2 - Dongwu Securities emphasizes the importance of self-reliance in technology, focusing on sectors such as artificial intelligence, autonomous control, new energy technology, aerospace information technology, and data elements [1] - In aerospace information technology, areas like low-altitude economy, satellites, and commercial space are highlighted as key focus points [1] - The report suggests that the AI sector should concentrate on AI agents, AI applications (like standalone software and smart terminals), humanoid robots, and autonomous driving [1] Group 3 - Western Securities notes that the level of merger and acquisition activity reflects the direction of the industrial cycle, with the highest number of mergers occurring in the automotive, electronics, and machinery sectors in 2024 [2] - The report indicates that the trends in automotive intelligence, semiconductor self-control, and high-end manufacturing are accelerating technological upgrades and industry evolution [2] - It is suggested that if AI commercialization leads to performance improvements through mergers and acquisitions, it could create a mainline market trend; otherwise, it may only represent a short-term thematic opportunity [2] Group 4 - The Sci-Tech Innovation Index ETF from Guotai (code: 589630) tracks the Sci-Tech Innovation Index (code: 000680), which includes representative stocks from the Sci-Tech Innovation Board, with an average market capitalization of approximately 11 billion [2] - The index focuses on technology innovation companies, covering more early-stage innovative firms and emphasizing hard technology sectors [2] - Investors without stock accounts can consider Guotai's linked ETFs for the Sci-Tech Innovation Board Comprehensive ETF [2]

Important Market News - The Ministry of Commerce announced export control measures for rare earths, emphasizing their dual-use nature and compliance with international practices. China will continue to review export license applications and has approved a certain number of compliant requests [1] - As of May 2025, China's foreign exchange reserves reached $328.53 billion, an increase of $3.6 billion from April, while gold reserves rose to 7.383 million ounces (approximately 2,296.37 tons), marking a continuous increase for seven months [1] Industry Insights - Deutsche Bank is exploring stablecoins and tokenized deposits, considering options for issuing its own tokens or joining industry initiatives. The bank recognizes the potential for stablecoins, especially in the U.S. regulatory environment, and sees various roles it could play in the stablecoin sector [2] - The storage industry is experiencing a shift due to anticipated production cuts of DDR4 memory, leading to price increases and supply shortages. The demand for DDR4 remains stable in niche markets, and the overall market is expected to improve by the second half of 2025, driven by AI device demand and inventory adjustments [3] - The Ministry of Industry and Information Technology is promoting the integration of artificial intelligence and manufacturing, focusing on enhancing software and hardware capabilities. The initiative aims to accelerate the intelligent upgrade of key industries and foster innovation in industrial data and models [4][5] Company Updates - Ruoyu Chen announced a share reduction plan by its shareholder, Langzi Co., intending to reduce up to 4.768 million shares, accounting for 3% of the total share capital. The stock has seen a significant increase of 535% over the past year [5] - Silk Road Vision disclosed plans for share reductions by its executives, with specific limits on the number of shares to be sold [5] - Mindong Electric Power announced a plan to reduce 4,579,514 shares, representing 1% of its total share capital, between June 30 and September 26, 2025 [5] - *ST Haiyue received a decision from the Shanghai Stock Exchange for stock delisting, with the last trading date expected to be July 4, 2025 [6]

Core Insights - The "Wujie" series of open-source large models was officially launched by the Zhiyuan Institute, covering various applications in embodied intelligence, brain science, and micro-life models [1][3] - Wang Zhongyuan, the director of the Zhiyuan Institute, emphasized that the development of large model technology is far from reaching its limits, despite the slowdown in performance improvement of large language models [1][2] Group 1: Large Model Development - The industry is exploring three main paths to overcome the performance bottleneck of large language models: reinforcement learning, data synthesis, and multi-modal data utilization [2] - The Zhiyuan Institute focuses on research and layout around the native multi-modal world model to enable AI to perceive and interact with the physical world [2][3] Group 2: Embodied Intelligence and Robotics - The "Wujie" series includes models like Emu3, Brainμ, RoboOS 2.0, and OpenComplex2, which are designed to adapt to various types of robots [3] - The rise of humanoid robots is seen as a significant direction in embodied intelligence, with expectations that humanoid robots will learn to walk by 2024 and run by 2025 [5][6] Group 3: Industry Collaboration and Ecosystem - The Zhiyuan Institute has established a strategic cooperation framework with Hong Kong Investment Management Company to build a world-class AI ecosystem [4] - Different industry players are exploring various solutions, including reducing hardware costs and utilizing simulation data for effective model training [7]

Investment Rating - The investment rating for the industry is "Positive" [5] Core Insights - AI Agents represent the third stage of AI development, transitioning from simple Q&A and content generation to becoming true "executors" capable of completing actual work tasks independently by 2025 [1][15] - The Model Context Protocol (MCP) is redefining the paradigm for AI Agents, serving as a crucial infrastructure that enhances the interaction between AI models and external services, making it more natural and precise [2][20] - Major tech companies are actively investing in AI Agent products, indicating a shift from technical competition to ecological value reconstruction in the AI Agent industry [2][34] Summary by Sections MCP Protocol Restructuring AI Agent Paradigm - AI Agents are identified as the third stage of AI development, with capabilities to represent users in actions [1][8] - The MCP protocol standardizes tool interfaces, allowing for seamless data interaction and decision execution across platforms [17][20] Acceleration of AI Agent Applications - Tech giants are rapidly deploying AI Agent products, with a noticeable shift towards ecological value reconstruction [34] - The market shows a strong preference for general-purpose AI Agents, with significant funding differences compared to vertical industry-focused agents [37] Investment Recommendations - The MCP protocol is likened to the "HTTP protocol" of the AI era, marking a transition to a standardized era of AI development [3][44] - Recommended companies to focus on include: Yonyou Network (commercial platform), Kingsoft Office (office solutions), iFlytek, and Wankong Technology (AIGC) [3][44] Industry Key Company Profit Forecasts - Profit forecasts for key companies indicate a positive outlook, with expected net profits for Yonyou Network, Kingsoft Office, iFlytek, and Wankong Technology showing growth from 2025 to 2027 [45]

Core Viewpoint - The article discusses the launch of AutoMat, an AI agent developed by Tsinghua University and other institutions, which automates the process of converting atomic-level STEM images into standard CIF structures, significantly reducing the time required for material analysis and discovery [1][2][5]. Group 1: AI Agent Functionality - AutoMat acts as a precise "map translator," transforming atomic-level STEM images into standard CIF structures and providing key physical properties like formation energy in one step [2]. - The traditional manual process of analyzing images, which could take hours or days, has been reduced to a few minutes, effectively bridging the gap between "microscopic imaging" and "structural reconstruction" [3][5]. - The team created a dataset called STEM2Mat-Bench, consisting of over 450 samples of two-dimensional materials, to validate AutoMat's performance [3][7]. Group 2: Methodology and Performance - AutoMat operates in four main steps: noise reduction, structure prior retrieval, atomic reconstruction, and property prediction, allowing for a closed-loop process from image to material insight [11][16]. - The performance evaluation of AutoMat against existing models shows it significantly outperforms multimodal models and specialized tools like AtomAI in both reconstruction accuracy and energy prediction [19][21]. - AutoMat achieved an average absolute error of 332 ± 12 meV/atom in formation energy prediction, which, while higher than the theoretical best of 48 meV, is substantially lower than the errors typically seen in visual-language models [20][21]. Group 3: Challenges and Future Directions - The article identifies two main challenges: template retrieval failures (39.3%) and downstream reconstruction failures (60.7%), which can lead to significant errors in atomic arrangement and structure output [22]. - Future improvements will focus on integrating experimental, simulation, and AI processes to enhance the accuracy of complex material structure and property predictions [23].

Core Insights - The article discusses the evolution and differentiation of AI Agents, emphasizing the need to distinguish between genuine innovative companies and those merely capitalizing on the concept of AI Agents [1][19][22]. Group 1: AI Agent Framework - The operation of AI Agents is broken down into three layers: perception, decision-making, and execution, highlighting the importance of each layer in the overall functionality of AI Agents [10][14][15]. - The "white horse is not a horse" concept is introduced to analyze the diversity of AI Agents in the market, categorizing them into pure, neutral, and free forms based on their operational characteristics [17][18]. Group 2: Technological Evolution - The article identifies the internalization of Agentic capabilities as a necessary evolution for LLMs, with examples from OpenAI's o4-mini and Anthropic's Claude 4 showcasing different design philosophies [30][38]. - Engineering integration is increasingly contributing to model capabilities, with tools like Prompt Engineering revealing significant potential for Agent products [2][31]. Group 3: Multi-Agent Systems - The limitations of Single-Agent systems are discussed, including memory constraints and the complexity of tool interactions, leading to the conclusion that Multi-Agent systems are becoming essential for overcoming these challenges [79][80]. - Multi-Agent architectures offer advantages in complexity, robustness, and scalability, allowing for parallel exploration of solutions and improved adaptability to human collaboration [82][83]. Group 4: Future Directions - The article suggests that the future of AI Agents will involve a competition between "experience universality" and "deep reliability," with hybrid architectures likely becoming a common choice [40][41]. - The emergence of protocols like MCP (Model Context Protocol) and A2A (Agent-to-Agent Protocol) is highlighted as a significant development in facilitating communication and tool integration among AI Agents [61][70].

Core Insights - Manus AI has launched a text-to-video generation service that transforms text prompts into fully structured, storyboarded video narratives, currently available for select subscription users and soon to be open to all [1] - This service directly competes with leading products in the industry, such as OpenAI's ChatGPT Pro Sora, which costs $200 per month, and other subscription-based or pay-per-use models from Runway, Synthesia, and Google [1] - The text-to-video sector is rapidly evolving, with Chinese tech giants Alibaba and Tencent pushing open-source solutions to challenge Western proprietary models [1] Group 1 - Manus AI gained significant recognition after launching its AI Agent tool earlier this year, which has led to investment from Benchmark Capital amid rising US-China AI competition [1] - The company has recently partnered with Microsoft Azure AI Foundry, enhancing its capabilities in the AI space [1] - Unlike traditional AI chatbots like ChatGPT, Manus can autonomously plan, execute, and complete complex multi-step tasks without continuous human intervention [2] Group 2 - Manus operates on a cloud-based asynchronous mechanism, allowing users to delegate tasks for offline processing, which can continue even during internet outages [2] - The system utilizes a multi-agent architecture, enabling specialized sub-agents to manage task planning, execution, and knowledge retrieval, thus supporting the breakdown and management of complex workflows [2]