DeepSeek又找到突破大模型推理瓶颈的新方法了！ 智东西2月27日报道，昨天，DeepSeek发布了一项名为DualPath的全新推理系统方案，直指当前大语言模型在智能体应用场景下遭遇的短板 ——KV缓存存储I/O瓶颈。该方案通过引入双路径加载机制，显著提升系统吞吐量，基本消除了KV缓存的I/O开销。 DualPath的核心创新在于开辟了一条从存储直通解码引擎的新通道。KV缓存不再仅由预填充引擎加载，而是可以加载至解码引擎，再通过计 算网络中的RDMA高效传输至预填充端。这一设计不仅缓解了存储端的压力，还避免了网络拥塞，确保延迟敏感型任务不受干扰。 与全局调度器协同后，DualPath实现了动态平衡两端负载，进一步提升资源利用率。在真实智能体工作负载测试中，DualPath将离线推理吞吐 量提升最高达1.87倍，在线服务吞吐量平均提升1.96倍。 在大规模可扩展性方面，DualPath系统在最多1152张GPU上进行了验证。离线推理从2P4D（2K智能体）扩展到48P96D（48K智能体）实现近 线性扩展，任务完成时间基本保持一致。 值得一提的是，与之前DeepSeek发表的许多研究论文类似，这篇论文的第 ...

Core Viewpoint - Kaos IoT Technology Co., Ltd. has submitted an application for listing on the Hong Kong Stock Exchange, aiming to enhance its global expansion strategy and strengthen its core platform capabilities [1][5]. Group 1: Company Overview - Kaos was established in April 2017 and is an industrial internet platform under Haier Group, having developed the COSMOPlat platform to assist enterprises in digital transformation [1]. - The company has achieved large-scale commercial application of its products and solutions across various verticals, serving over 9,500 paying customers [3]. Group 2: Financial Performance - Kaos's revenue is projected to increase from RMB 49.94 billion in 2023 to RMB 50.69 billion in 2024, with a revenue of RMB 44.21 billion reported for the first nine months of 2025 [3]. - The company has turned a net loss of RMB 82.72 million in 2023 into a profit of RMB 65.14 million in 2024, with a significant increase in net profit to RMB 17.6 million in the first three quarters of 2025, representing a 393% year-on-year growth [3][4]. - The net profit margin is expected to improve from -1.7% in 2023 to 4.0% by the end of September 2025, attributed to increased gross profit and operational efficiency [3]. Group 3: Market Strategy and Global Expansion - Kaos has shifted its listing strategy from A-shares to Hong Kong due to regulatory uncertainties in the A-share market and aims to leverage Haier Group's international network for global expansion [5]. - The company plans to enhance its market penetration in key industries and regions, focusing on emerging markets in the Americas, Southeast Asia, and the Middle East [5]. - As of September 2025, Kaos's solutions have been promoted in over 20 countries, serving more than 50 overseas enterprises [6]. Group 4: Shareholding Structure - Haier Group is the controlling shareholder of Kaos, holding approximately 78.04% of the voting rights, with direct ownership of 10.83% and additional stakes through subsidiaries [3]. Group 5: Industry Context - The listing of Kaos on the Hong Kong Stock Exchange is expected to expand the "Haier system," which currently includes multiple listed companies under Haier Group [7].

Core Viewpoint - The traditional "people-intensive" strategy of brokerage firms is outdated, and the future will see widespread application of intelligent agents to enhance service capabilities and efficiency [1][3]. Group 1: Intelligent Agents in Brokerage Firms - Brokerage firms should develop several intelligent agents based on their capabilities to consolidate stable service output [3]. - These intelligent agents can interface with wearable devices and internet platform agents to extend the brokerage's mature service capabilities to a broader range of investors [3]. - The company is cautious about investments but is willing to "boldly experiment" with intelligent systems [3]. Group 2: Efficiency Improvements through AI - AI has revolutionized the production of market analysis short videos, reducing the time required from several hours to just minutes [3]. - The new digital human system allows for the rapid generation of content, enabling multiple videos to be produced simultaneously, resulting in efficiency improvements of dozens to hundreds of times [3]. Group 3: Building Investor Trust - Gaining investor trust is likened to "making friends," emphasizing the importance of stable, continuous, and warm service [4]. - Service output must remain consistent, providing timely assistance when needed and maintaining appropriate distance when not [4]. - AI can be trained to improve emotional interaction and stability through ongoing optimization and feedback, akin to employee development [4].

Core Insights - The MCP protocol is evolving to include interactive user interface (UI) support through the MCP Apps proposal, enhancing the interaction capabilities of AI agents beyond text and structured data [2][10][11] Group 1: MCP Apps Proposal - The MCP Apps proposal (SEP-1865) aims to standardize support for interactive UIs, allowing MCP servers to provide visual interfaces directly to hosts [2][4] - This proposal has received positive feedback from the community, driven by contributions from key players like OpenAI and Anthropic [9][10] Group 2: Enhancements in User Interaction - The MCP Apps Extension introduces a standardized approach for declaring UI resources, linking them to tools, and enabling bidirectional communication between embedded interfaces and host applications [4][18] - The transition from a text-based interaction model to a graphical interface is likened to upgrading a customer service chatbot from text messaging to a smart assistant capable of providing visual dashboards and forms [6][11] Group 3: Standardization and Community Involvement - The current limitations of the MCP server in exchanging only text and structured data hinder the presentation of visual information and complex user input [13][18] - The MCP-UI project, supported by a vibrant community, has demonstrated the feasibility of integrating rich UIs into the MCP architecture, with backing from major companies [15][18] Group 4: Key Design Decisions - The MCP Apps Extension emphasizes security, backward compatibility, and the use of pre-declared resources to enhance performance and safety [20][23][24] - The initial extension specification supports rendering HTML content in a sandboxed iframe, with plans for future support of additional content types [22][24] Group 5: Community Engagement - The MCP community encourages participation in the development of the MCP Apps Extension, providing early access SDKs for developers to build applications [26][27]

Core Insights - The "2025 Digital Transformation Development Conference" highlighted ten key terms for digital transformation in enterprises, including AI+, intelligent applications, and data value realization [1][2] - The conference was organized by the China Communications Standards Association and supported by various industry associations, reflecting a collective focus on digital transformation strategies [1] Group 1: Key Terms for Digital Transformation - The ten key terms identified are: "14th Five-Year Plan" digitalization, architecture and process governance, enterprise-level AI+, data value realization, innovation ecology, AI application research and development, intelligent applications, government digital governance, comprehensive urban digital transformation, and digital maturity [1] - The emphasis on integrating technology with practical scenarios is seen as a main theme for the "14th Five-Year Plan" digital transformation [1] Group 2: Enterprise-Level AI+ - The implementation of "AI+" is viewed as a crucial step for industries to achieve revolutionary productivity leaps and explore new business models [2] - Enterprises are progressively realizing "AI+ scenarios" in areas such as intelligent Q&A, text generation, intelligent inspection, and code generation, marking a shift from basic applications to deeper integration of AI [2]

Core Insights - The article discusses a training session led by Wu Xiaonan, a senior lecturer at the Unknown AI Research Institute, focusing on "DeepSeek Prompting Techniques and News Writing" for over 120 participants from the Zhejiang Provincial Science and Technology Association [1]. Group 1: Training Overview - The training emphasized the characteristics of communication in the intelligent media era and systematically analyzed the core methodologies of AI-assisted writing [1]. - The course was structured into three main modules: optimizing prompt engineering, reconstructing scientific narrative logic, and generating promotional copy for various scenarios [1]. - Participants engaged in real-time operations to master practical skills for controlling AI output styles and quickly generating suitable content [1]. Group 2: Organizational Focus - The Unknown AI Research Institute is dedicated to AI frontier trends, commercial implementation, and talent development, aiming to become the "cognitive infrastructure of the AI era" [2]. - The institute actively develops practical training programs, including DeepSeek workplace applications and AI strategy workshops, to convert cutting-edge technologies into actionable training solutions [5]. - Future plans include deepening efforts in the AI field and promoting the integration of AI technology across various industries [5].

Market Overview - The Hong Kong stock market experienced a decline, with the Hang Seng Index down 1.19% to 25,215.1 points, the Hang Seng Tech Index down 1.08% to 5,515.77 points, and the National Enterprises Index down 1.26% to 9,013.58 points [1] - Major technology stocks saw widespread declines, with Alibaba down 2.63%, JD.com down 3.84%, and Meituan down 3.14%, while Tencent saw a slight increase of 0.76% [1] - Internet healthcare stocks surged, with Dingdang Health rising over 26%, while Chinese brokerage stocks strengthened, with Zhongzhou Securities up over 13% [1] Company News - Alibaba Group is launching a large-scale AI talent recruitment plan, aiming to hire nearly 1,000 people focusing on advanced technologies such as large language models and AI hardware, with positions available in major cities like Beijing and Shanghai [2] - China Telecom reported a revenue of 271.5 billion yuan for the first half of the year, a year-on-year increase of 1.3%, and a net profit of 23 billion yuan, up 5.5% year-on-year [3] - CK Hutchison Holdings reported a revenue of 240.66 billion HKD for the first half of the year, a year-on-year increase of 3.45%, but a significant net profit decline of 91.65% to 850 million HKD [4] - JD.com reported a second-quarter revenue of 356.7 billion yuan, a year-on-year increase of 22.4%, but a net profit decline of approximately 50.8% to 6.2 billion yuan [4] - NetEase reported a revenue of 56.72 billion yuan for the first half of the year, a year-on-year increase of 8.37%, and a net profit of 18.90 billion yuan, up 31.33% year-on-year [5] Institutional Insights - Analysts from Zhongtai International noted that the current valuation of Hong Kong stocks has significantly recovered, with the Hang Seng Index's forecast PE returning to the mid-level of 2018-2019, and the risk premium at a historical low [6] - Guotai Junan analysts indicated that the overall pressure from capital outflows in Hong Kong stocks may be relatively controllable, with an expected net inflow of over 1.2 trillion yuan for the year [6] - Ping An Securities highlighted that despite uncertainties from U.S. tariffs, China's export data in July was unexpectedly strong, supporting a bullish outlook for Hong Kong stocks [6] - Everbright Securities stated that the overall profitability of Hong Kong stocks remains strong, with relatively scarce assets in sectors like internet, new consumption, and innovative pharmaceuticals, suggesting a favorable long-term investment outlook [6]

Core Viewpoint - Baidu is testing an open entry for intelligent applications on its desktop search homepage, which will display previously used or potentially interesting intelligent applications for logged-in users [1] Group 1 - The intelligent applications integrated into Baidu's search homepage mainly come from the Wenxin intelligent application platform, external high-quality AI applications accessed through the search open platform, and some self-developed applications by Baidu [1] - This feature is currently in a gray testing phase and is expected to be fully opened soon [1] - As of the report's publication, Baidu has not officially responded to inquiries regarding the testing of the open intelligent application entry [1]

Core Viewpoint - The article discusses the launch of the new flagship model GLM-4.5 by Zhipu AI, which is designed for intelligent agent applications and has been released on HuggingFace and ModelScope platforms [2][3]. Group 1: Model Architecture and Performance - GLM-4.5 utilizes a mixture of experts (MoE) architecture with a total parameter count of 355 billion and 32 billion active parameters, while GLM-4.5-Air has 106 billion total parameters and 12 billion active parameters [4][6]. - The model integrates reasoning, coding, and intelligent agent capabilities, achieving a comprehensive performance ranking in the global top three, and is the leading domestic and open-source model [3][4]. - In comparative tests against models like Claude Code and Kimi-K2, GLM-4.5 demonstrated superior task completion and tool reliability, although it slightly lagged behind Claude-4-Sonnet in some dimensions [8]. Group 2: Cost and Efficiency - The API call pricing for GLM-4.5 is set at 0.8 yuan per million tokens for input and 2 yuan per million tokens for output, making it a cost-effective option [10]. - The high-speed version of the model supports a generation rate of up to 100 tokens per second, catering to high concurrency deployment needs [12]. Group 3: Training Data and Fine-tuning - The training data for GLM-4.5 encompasses 15 trillion tokens of general corpus, supplemented by 8 trillion tokens specifically fine-tuned for coding, reasoning, and agent tasks, enhanced through reinforcement learning [7]. Group 4: Agent Capabilities and Demonstrations - Zhipu AI has released multiple real-world scenario demos to showcase the agent capabilities of GLM-4.5, including a simulated search engine, a video platform simulator, a playable Flappy Bird game, and an automated PPT tool [14].

Summary of Conference Call Notes Industry or Company Involved - The discussion primarily revolves around the public fund industry in China and its impact on the broader financial market. Core Points and Arguments - **Impact of New Public Fund Regulations**: The new regulations aim to address the issue of active funds underperforming, potentially leading to a shift towards passive investment strategies and reducing market volatility. The main benchmarks are the CSI 300 and the CSI 800 indices [1][3][4]. - **Market Sentiment and Activity**: Positive market sentiment is noted, with an increase in total trading volume across all A-shares, a rise in the proportion of strong stocks, and an increase in margin trading. The micro-cap stock index has reached new highs, indicating a favorable market environment for structural opportunities [1][5]. - **Investment Recommendations**: Three main investment directions are highlighted: 1. Self-controlled sectors within the Sci-Tech 50, particularly in military aircraft supply chains and the Harmony operating system [2][7]. 2. Traditional core assets represented by the Shanghai Stock Exchange 50, benefiting from the return to performance benchmarks, including finance, insurance, and public utilities [2][3]. 3. New consumption sectors, focusing on platform economies, elderly care, and maternal and infant products [2]. - **Sectoral Fund Flows**: Following the new regulations, sectors such as banking, non-bank financials, and coal are expected to see significant short-term inflows, while high-tech companies are projected to gradually increase their market capitalization share [3][11]. Other Important but Possibly Overlooked Content - **Growth Themes**: Key growth themes include solid-state batteries, AI, and low-altitude economy, with upcoming technology exhibitions expected to catalyze market interest [8][9][10]. - **Long-term Market Trends**: The long-term outlook suggests a gradual increase in the market capitalization of high-tech companies, particularly in the TMT (Technology, Media, and Telecommunications) sector, which is worth exploring for growth potential [12]. - **Current Market Dynamics**: The current market environment is characterized by a recovery in risk appetite, with the U.S. delaying interest rate cuts and ongoing trade negotiations between China and the U.S. contributing to a stabilizing global equity market [5][6]. This summary encapsulates the key insights from the conference call, focusing on the implications of regulatory changes, market sentiment, investment strategies, and growth opportunities within the public fund industry and related sectors.