科创人工智能ETF华夏（589010）紧密跟踪上证科创板人工智能指数，覆盖全产业链优质企业，兼具高 研发投入与政策红利支持，20%涨跌幅与中小盘弹性助力捕捉AI产业"奇点时刻"。 （文章来源：每日经济新闻） 截至13点43分，科创人工智能ETF（589010）放量大涨 4.33%，盘中持续运行于均线之上，展现出强劲 的进攻姿态。持仓股方面掀起"涨停潮"，新点软件强势斩获20CM涨停，海天瑞声飙升近19.5%，中科 星图、福昕软件、星环科技-U 等多只核心成分股联袂大涨超15%，赚钱效应极佳。流动性方面，ETF成 交额突破 2.52亿元，换手率超 8%，交投情绪极度火热，显示出市场资金对科创板人工智能方向的长期 价值高度认可，配置窗口正如火如荼开启中。 消息方面，科技媒体The Information援引两位知情人士透露，DeepSeek计划在2月中旬农历新年期间发 布V4版本，同时表示发布时间仍可能变动。这款模型代号DeepSeek V4，是去年DeepSeek V3的迭代 版。基于公司内部基准测试的初步评估显示，该模型在编程能力上超越了现有模型，如Anthropic的 Claude和OpenAI的GPT系列 ...

Core Viewpoint - The successful listing of Zhipu marks a significant milestone for China's AGI industry, indicating a key transition from laboratory research to commercialization, and highlights the symbiotic relationship between Zhipu and its nurturing environment in Haidian District [3][6]. Group 1: Company Overview - Zhipu officially listed on the Hong Kong Stock Exchange on January 8, becoming the "first stock of global large models," with its stock price increasing nearly 77% and a market capitalization reaching 100 billion HKD [3]. - The company was founded in 2019 by Tsinghua University alumni, focusing on AGI and has achieved several milestones, including the launch of China's first hundred-billion model and the first open-source trillion model [5][6]. - Zhipu's revenue is projected to grow from 57.4 million RMB in 2022 to 312.4 million RMB in 2024, reflecting a compound annual growth rate of 130% [6]. Group 2: Ecosystem and Support - Haidian District provides a rich ecosystem for AI talent, housing approximately 43% of China's top AI talent, with over 80% located in Haidian [8]. - The district has established a diverse incubation system with 193 incubators, offering comprehensive support through mentorship, resources, and capital [9]. - Haidian has set up a technology growth fund totaling 20 billion RMB to accelerate the development of key industries and projects [9]. Group 3: Policy and Infrastructure - In June 2023, Haidian issued measures to support AI model innovation, including subsidies of up to 2 million RMB for innovative models [10]. - The district launched a public computing power platform in 2024, enhancing its AI infrastructure and supporting the development of AI applications across various sectors [10]. - Haidian's AI ecosystem is characterized by a high density of research institutions and companies, facilitating collaboration and innovation [14][15]. Group 4: Feedback Loop and Future Prospects - Zhipu has initiated the "Z Plan" to support entrepreneurs in the large model field, providing funding, technology, and resources [19]. - The company has invested in numerous startups, with over half of them establishing their headquarters or core R&D in Haidian [19]. - Haidian aims to continue nurturing future industries such as embodied intelligence and quantum information, suggesting that more leading companies will emerge from this innovative environment [20].

1月12日，"全球大模型第一股"智谱(2513.HK)今日为上市第三个交易日，股价继续上涨，盘中一度飙升45.34%至230.6港 元，再创上市新高，市值升至1014亿港元。 据最新消息，智谱与滴滴宣布达成战略合作，双方将围绕通用人工智能(AGI)关键技术及其在出行领域的智能体应用开展前 瞻性协同探索。双方将共同推进Agent场景落地和大模型领域人才培养，深化出行场景的意图对齐与推理能力建设，推动 Agent在更复杂业务场景中的验证与落地。 智谱是中国最早投身大模型研发的厂商，原创提出了基于自回归填空的通用预训练范式GLM，率先发布了中国首个百亿模 型、首个开源千亿模型、首个对话模型、首个多模态模型，以及全球首个设备操控智能体(Agent)，形成了全面的模型体 系，是国内罕有在原创技术路线上与全球顶尖水平保持同步的厂商，因此也被誉为"中国OpenAI"。 智谱2025年收入超1亿美元。其中，智谱本地部署收入占比85%，云端占比在过去两年从0提升到15%，预计今年云端收入占 比将持续提升，高性价比代码工具将对智谱云端收入产生更大影响。 港股频道更多独家策划、专家专栏，免费查阅>> 责任编辑：栎树 ...

大模型不赚钱，这事不多说，但简单回顾： OpenAI据说得2029年盈利，每年亏140亿刀。 Anthropic据说2028年能盈利，也亏损。 刚上市的MiniMax每年亏5个亿刀。 智谱和MiniMax差不多，也是每年亏这么多。 据说DeepSeek能赚钱，但这行当基础模型基本亏损是确定的。 不能赚钱就主要靠VC，国内VC不给力所以就导致了基模发展的一系列问题。 但基模属于非典型产品，不多说了，下面重点说智能体。（包括做基础模型也做应用的） 智能体也先说结果： 就拿我比较熟悉的Glean举例子，ARR最新报道过了2亿美金，我们就假设每年都是这么多收入，那现在 有1000个人，所以每个人对应收入是20万美金。这在AI行当基本会亏损。对应的事件呢，我觉得接下 来大概率会继续融资。 类似的，其他明星智能体公司也基本这样，大额融资的包括并购的... 如果就这么多，那就和上波AI有点像，还好不全是，比如： Midjourney一年赚了5亿美金的时候只有40个人，这就怎么都赚钱的。因为它流量基本还是自然流。 你比如Base44，一年350万美金，还就1个人，这怎么也是赚钱的，不太可能都是投流投出来的。 这样题目从结果看 ...

Core Viewpoint - The article emphasizes that AI has transitioned from being a mere functional plugin to a fundamental system capability, marking a significant shift in the hardware innovation landscape as large models become the default capability in smart devices [2][9]. Group 1: AI Hardware Landscape - The AI hardware landscape is witnessing a resurgence, with over 200 mainstream hardware manufacturers participating in events like the Alibaba Cloud Tongyi Intelligent Hardware Expo and CES, showcasing a variety of AI applications [3][4]. - The integration of large models, such as Qianwen, into over 1,000 smart devices across 76 categories indicates a strong industry signal that AI is moving beyond screens into the physical world [4][12]. Group 2: Historical Challenges - The AI hardware sector has faced significant challenges over the past decade, including uncontrollable model capabilities, difficult commercial logic, and high engineering complexity, which have hindered the successful implementation of AI in hardware [8][9]. Group 3: Turning Point in AI Hardware - Recent exhibitions signal a turning point for AI hardware, shifting the focus from showcasing model capabilities to system engineering, indicating that AI is now a reliable and efficient foundational capability [9][10]. - Alibaba Cloud's development of a multimodal interaction development kit allows manufacturers to integrate core model capabilities with various interaction technologies, significantly lowering the barriers for hardware selection and adaptation [10][11]. Group 4: Qianwen as a Universal Foundation - Qianwen has emerged as a universal foundation for AI hardware, with Alibaba Cloud collaborating with all AI hardware categories, indicating a shift from single-function AI to comprehensive system capabilities [14][15]. - The comprehensive AI capabilities provided by Qianwen address key pain points in AI hardware implementation, creating a complete ecosystem from models to tools and platforms [16][21]. Group 5: Market Impact and Future Outlook - The AI hardware market in China has surpassed 1.1 trillion yuan, marking a transition from conceptual exploration to large-scale implementation [12]. - The integration of over 1,000 smart devices with Qianwen signifies a clear signal that AI is deeply connecting with the physical world, transforming hardware into intelligent systems that continuously learn and evolve [23][24].

Group 1 - The core idea of the conference is that in the era of AI, human creativity, judgment, and responsibility remain irreplaceable by technology [1] - The conference highlighted the importance of integrating human insight with technological advancements in finance, emphasizing that AI should be seen as a creator and reshaper rather than a mere replacement [1] - Liu Xiaochun from Shanghai New Financial Research Institute stressed that financial innovation should focus on the essence of finance rather than technology itself, categorizing technology into three levels: financial technology, institutional technology, and scientific technology [2] Group 2 - Yuan Yue, chairman of Zero Point Data, outlined the transition from financial technology (FT) to financial intelligence (FI), indicating a shift towards intelligent decision-making in finance [3] - He introduced a framework for understanding the core technologies supporting risk control and service optimization, emphasizing the limitations of large language models in high-sensitivity fields like finance [3] - The conference also explored the mutual empowerment between financial technology and content creation, discussing how both sectors can benefit from each other [3][6] Group 3 - Zhang Wenyu from Zhejiang University of Finance and Economics highlighted the fundamental impact of AI on various industries, particularly finance, marking the emergence of a new era of AI capabilities since the launch of ChatGPT [4][5] - He emphasized that while 99% of routine tasks may be automated, the unique human qualities of creativity and insight are essential for navigating complex scenarios [5] - Zhu Guangye, a financial investment entrepreneur, acknowledged the reality of AI replacing many repetitive tasks in finance but noted that certain roles still require human judgment and experience, particularly in nuanced situations [5][6]

Core Insights - The article discusses the contrasting market responses to the IPOs of two AI companies, Zhiyun and MiniMax, highlighting their different commercialization strategies and market perceptions [1][3]. Group 1: Market Performance - Zhiyun, referred to as the "first global large model stock," saw its share price rise by 13.2% to 131.5 HKD on its first trading day after an initial dip [1]. - MiniMax experienced a strong performance, closing at 345 HKD, which represents a 109.1% increase from its issue price [1]. Group 2: Commercialization Strategies - Zhiyun focuses on delivering AI capabilities as "deployable engineering" within B2B services, while MiniMax emphasizes creating "consumable products" for the C-end market [3][4]. - The financial models of both companies show significant differences, with Zhiyun adopting a long-term technology approach and MiniMax taking a more aggressive, efficiency-driven strategy [3]. Group 3: Revenue Structure - For Zhiyun, localized deployment revenue is projected to account for over 80% of total revenue in 2024 and the first half of 2025, indicating a strong B-end orientation [4]. - MiniMax's revenue structure reflects a product and platform model, with over 71% of its revenue coming from AI-native applications like Hai Luo AI and Xing Ye [8][10]. Group 4: Market Sentiment and Future Outlook - The market's perception of Zhiyun is mixed, with some investors viewing it through a familiar lens of high margins and project-based revenue, while others see it as a platform-based business model in disguise [7]. - MiniMax's approach aligns more closely with consumer internet companies, focusing on user growth and cost efficiency, which has garnered positive market sentiment [10][11]. Group 5: Cost Structure and Cloud Dependency - MiniMax's reliance on cloud services is significant, with projected expenditures on Alibaba Cloud reaching up to 135 million USD annually by 2028 [12]. - Both companies' commercialization paths highlight a shared dependency on cloud infrastructure, with MiniMax exposing its costs directly while Zhiyun integrates them into its overall solutions [12][13].

Core Insights - The consensus among industry experts indicates that 2026 will be a pivotal year for the development of end-to-end (E2E) and VLA (Vision-Language Alignment) technologies in autonomous driving, with a focus on optimizing production processes rather than making significant algorithmic changes [1] - The industry is actively recruiting experienced algorithm engineers and developing talent to tackle the complex challenges ahead, particularly in areas such as BEV perception, large models, diffusion models, and reinforcement learning [1] Course Overview - The course on E2E and VLA autonomous driving is designed to provide a comprehensive learning path from principles to practical applications, developed in collaboration with industry leaders [3] - The course covers various aspects of E2E algorithms, including their historical development, advantages and disadvantages of different paradigms, and current trends in both academia and industry [6][7] - Key technical keywords that are expected to be frequently encountered in job interviews over the next two years are emphasized in the course content [7] Course Structure - Chapter 1 introduces the concept of E2E algorithms, discussing their evolution from modular approaches to current paradigms like VLA [6] - Chapter 2 focuses on the background knowledge necessary for understanding E2E technologies, including VLA, large language models, diffusion models, and reinforcement learning [11] - Chapter 3 delves into two-stage E2E algorithms, exploring their emergence and comparing them with one-stage approaches [7] - Chapter 4 presents one-stage E2E algorithms and VLA, highlighting various subfields and their contributions to achieving the ultimate goals of E2E systems [8] - Chapter 5 involves a practical assignment on RLHF (Reinforcement Learning from Human Feedback) fine-tuning, demonstrating how to build and experiment with pre-training and reinforcement learning modules [9] Learning Outcomes - The course aims to elevate participants to the level of an E2E autonomous driving algorithm engineer within approximately one year, covering a wide range of methodologies including one-stage, two-stage, world models, and diffusion models [15] - Participants will gain a deeper understanding of key technologies such as BEV perception, multimodal large models, reinforcement learning, and diffusion models, enabling them to apply their knowledge in real-world projects [15]

Group 1 - The release of GPT-4 has caused a global sensation, highlighting the capabilities of large models to surpass human rational thinking [1] - OpenAI's recent lack of significant achievements has led to a perception that open-source models have substantially outperformed closed-source models [1] - The development trajectory of OpenAI is seen as a double-edged sword, with its leadership being both a source of success and potential failure [1] Group 2 - The emergence of AI has transformed models into essential production factors, marking a shift from the information age to a new era where "dialogue equals production" [7] - The ability to produce models and code will become crucial for organizations and individuals, similar to the importance of understanding scientific experiments in the 20th century [6] - Knowledge post-2023 is characterized as being influenced by AI, leading to a significant change in how information is generated and perceived [13][14] Group 3 - The AI revolution has made certain professions, such as teachers and therapists, less susceptible to replacement due to their inherent complexity and human interaction requirements [9][10] - The traditional approach to hiring for various roles is shifting towards leveraging AI models to perform tasks that were previously done by multiple specialists [12] - The next three years are expected to see an explosive growth in AI-based software and research outcomes, fundamentally altering societal structures [13] Group 4 - The half-life of technical knowledge has decreased significantly, now averaging between 18 months to 3 years, increasing the demand for learning and cognitive abilities [19][20] - The importance of psychology is expected to rise in the AI era, focusing on how humans can better interact with AI systems [22][23] - Companies like Anthropic are employing psychologists to define concepts that relate closely to human knowledge structures, indicating a growing intersection between AI and psychology [24] Group 5 - The AI coding field has experienced a significant productivity leap, with expectations of 10x to 100x increases in efficiency across various knowledge work sectors [25][26] - The development of agents in AI is evolving, with future iterations expected to incorporate more complex functionalities and optimizations [28][29] - The cognitive gap is becoming a new divide, as AI enhances work efficiency, making it crucial for individuals to adapt to new workflows and technologies [30][31]

Core Viewpoint - Tencent is aggressively recruiting talent in the AI field, particularly for its large language model (LLM) project, "混元" (Hunyuan), aiming to compete with top global models. The company is experiencing a significant shift in its organizational structure and talent acquisition strategy to enhance its capabilities in AI development [10][20][23]. Group 1: Recruitment and Talent Acquisition - Tencent's "青云计划" (Qingyun Plan) targets top graduates for AI roles, directly competing with ByteDance's "Top Seed" program [10]. - The company is offering substantial salary increases, with some candidates seeing their compensation double upon joining Tencent from ByteDance [10][13]. - Key hires from Microsoft and other leading AI teams have been made to bolster Tencent's LLM capabilities, with a focus on candidates from specific high-profile companies [12][18]. Group 2: Leadership Changes and Organizational Structure - The appointment of Yao Shunyu as the chief AI scientist marks a pivotal change in Tencent's approach to its LLM project, granting him direct reporting lines to the company's president [20][21]. - Yao's leadership is expected to streamline decision-making and resource allocation, contrasting with the previous complex management structure [21][46]. - Organizational adjustments have been made to align with the demands of large model development, including the establishment of new departments focused on AI infrastructure and data [45][46]. Group 3: Competitive Landscape and Market Position - Tencent's late entry into the large model space has raised concerns about its competitive position, as it trails behind companies like OpenAI, Baidu, and ByteDance in model performance [23][24]. - The company is under pressure to deliver competitive models quickly, with industry insiders noting that its self-developed models have not been featured prominently in benchmark comparisons [23][24]. - The shift in focus towards LLMs is seen as a response to the urgent need for Tencent to catch up in the rapidly evolving AI landscape [23][47]. Group 4: Model Development Strategy - Yao Shunyu emphasizes a shift towards post-training and a more methodical approach to model updates, contrasting with the previous rapid release cycle [18]. - The upcoming "混元2.0" model, with 406 billion parameters, is anticipated to reflect Yao's influence, although it is unlikely to be entirely his work due to the typical training timelines [52]. - The strategy moving forward will likely involve leveraging proven methodologies from successful models in the industry to accelerate development [47][49].