编辑｜杜伟 多年前，苹果用「There's an app for that」开启了移动互联网的黄金时代。从那之后，一个个应用图标统治了我们的数字生活。 而如今，随着 LLM、Agent 的快速发展，这一切正在发生变化。 就在昨天，AI 大神 Karpathy「现身说法」，并抛出了一个激进的观点： 未来的应用不应该是被「下载」的，而应该是被「即兴创作」的 。 他以自己正在进行的有氧运动为例，没有选择去应用商店搜索任何一款「心率管理工具」，而是直接命令 AI 逆向工程了跑步机的云端 API，为自己量身定制了一 个为期八周的、极其私密的实验仪表盘。 这释放出了一个明显的信号：软件的本质正在从现成的商品降维成瞬时的服务。 那么问题来了： 当应用可以随用随建，我们还需要那个臃肿的应用商店吗 ？ | RHR 50 -> 45 Experiment | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 8-week zone 2 + HIIT plan · Feb 18 - Apr 14, 2026 · Dashboard | ...

上证指数+0.05%科创50+1.78% 上证50-0.28%创业板指+1.32% 沪深300+0.12%中证500+1.17% 总成交额(万亿人民币)2.16环比增长8% 今日中国股市基本平盘震荡，AI供应链表现出色。 交易台 – 高盛中国收盘 综述 上证指数+0.05%科创50+1.78% 上证50-0.28%创业板指+1.32% 沪深300+0.12%中证500+1.17% 总成交额(万亿人民币)2.16环比增长8% 今日中国股市基本平盘震荡，AI供应链表现出色。 午后成交量有所回升。 临近春节假期，本周板块轮动速度加快。 市场焦点重回 AI基础设 交易台 – 高盛中国收盘 综述 此外，液冷技术的强势支撑了数据中心相关个股的表现。 英维克(002837.SZ)涨停(+10%)。 美国冷却解决方案供应商维谛技术(Vertiv)乐观的第四季度订单数据(增长252%)提振了情绪，激发了对本土液冷供 应商的乐观预期。 非科技板块方面，在官方计划提高电力交易比例并统一电力市场后，电力电网/电力设备板块走势强劲。 有色金属势头依然强劲，稀土领涨。 反之，尽管官方出台了支持春节消费的补贴计划，消费股仍遭遇持续的获利回 ...

Group 1 - The core update in iOS 26.4 will introduce a new version of Siri that significantly changes user interaction and capabilities [1][3] - Siri will utilize a new feature based on LLM (Large Language Model), moving beyond simple voice-to-text and keyword searches to genuinely understand user queries and apply logical reasoning [3] - Although the update does not provide full chatbot interaction capabilities, it represents a substantial improvement over the current version, which has been overdue [3]

Financial Data and Key Metrics Changes - Cerence reported revenue of $115.1 million for Q1 2026, a 126% increase from $50.9 million in the prior year period [17] - Adjusted EBITDA was $44.6 million, representing a 39% margin compared to $1.4 million or 3% in the prior year [22] - The company generated record quarterly free cash flow of $35.6 million, marking a significant achievement [4][23] - GAAP net loss for the quarter was $5.2 million, an improvement from a net loss of $24.3 million in the same quarter last year [22] Business Line Data and Key Metrics Changes - Variable license revenue was $30.5 million, up 34% year-over-year, driven by steady customer utilization [18] - Fixed license revenue was $7.8 million, with expectations for it to be comparable to the prior year [18] - Connected services revenue was $14.5 million, up 6% year-over-year, with a potential increase of over 20% without prior year true-up [19] Market Data and Key Metrics Changes - Approximately 11.9 million cars produced included Cerence technology, flat from the prior year [23] - The number of connected cars shipped grew by 14% on a trailing 12-month basis [24] - 51% of worldwide auto production included Cerence technology, consistent with historical penetration [24] Company Strategy and Development Direction - Cerence's key priorities for 2026 include advancing technology through the xUI platform, maintaining cost diligence, and driving top-line growth [4][27] - The company showcased advancements in AI technology at CES, including the introduction of new AI agents and partnerships with major OEMs [5][7] - Cerence aims to expand its reach beyond automotive, with plans to operationalize strategies in new industries [14] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in the company's technology and customer momentum, anticipating continued growth in revenue and profitability [15][26] - The company reaffirmed its full-year guidance for fiscal 2026, expecting revenue between $300 million and $320 million [26] - Management highlighted the importance of the patent license revenue from Samsung as a validation of their IP strategy [15] Other Important Information - The company resolved its patent litigation with Samsung, resulting in a one-time payment of $49.5 million recorded in Q1 [15][19] - Total non-GAAP operating expenses were $57.3 million, primarily driven by legal costs associated with the patent license outcome [21] Q&A Session Summary Question: Interest in the mobile work agent and its impact on ARPU - Management indicated strong demand for the mobile work agent, which can be implemented in existing vehicles, potentially increasing ARPU [30][32] Question: Impact of new signings on TTM billings and backlog - Management confirmed that new signings would be reflected in the five-year backlog and TTM billings, with revenue expected to ramp up as production starts [34][38] Question: Usage trends of existing in-car connected systems - Management noted that usage of newer systems is increasing, especially with added functionalities like Microsoft 365 integration [40] Question: Competitive process for new wins and future win rates - Management highlighted that technology capability and team confidence are key differentiators in winning contracts, with a strong win rate expected moving forward [54][56]

Core Insights - Google DeepMind is at the forefront of AI research, focusing on breakthroughs that impact science, business, and society, particularly in the context of the AGI race [1][3][4] - The company has made significant advancements, including the development of Gemini, which is now competitive with ChatGPT, and has roots in technologies originally developed by Google [3][4][28] - The investment made by Google in DeepMind in 2014, approximately £400 million (around $540 million), has potentially grown to hundreds of billions, highlighting the strategic importance of this acquisition [4][28] Company Overview - Google DeepMind was founded in 2010 in London by Demis Hassabis, Shane Legg, and Mustafa Suleyman, with the latter now working at Microsoft [2][3] - The company has been pivotal in Google's AI advancements, particularly with consumer-facing products like Gemini, which leverage DeepMind's foundational technologies [4][28] Technological Developments - The AI landscape has evolved significantly since the emergence of ChatGPT, with Google facing internal restructuring to adapt to the competitive environment [3][4] - DeepMind's previous breakthroughs, such as AlphaGo and AlphaFold, have set the stage for its current innovations, emphasizing the company's commitment to solving fundamental scientific problems [4][5] AGI and Future Prospects - The pursuit of AGI is a long-term mission for DeepMind, with expectations of achieving significant milestones within the next 5 to 10 years [10][11] - Current AI systems, including LLMs, face limitations in achieving true AGI, particularly in areas like continuous learning and creative hypothesis generation [7][8][10] Energy and Efficiency Challenges - There are physical limitations in AI development, particularly concerning energy consumption and computational power, which need to be addressed as the field progresses [11][12] - Innovations in model efficiency, such as the use of Distillation, are expected to enhance performance significantly, with annual improvements projected at around 10 times [12][13] Competitive Landscape - The AI industry is experiencing intense competition, with many players, including startups and established tech giants, vying for leadership [28][29] - Concerns about potential financial bubbles in the AI sector are acknowledged, with some segments showing signs of unsustainable valuations [32][33] Global AI Dynamics - The competition between the U.S. and China in AI development is intensifying, with Chinese companies like DeepSeek and Alibaba making notable advancements [35][36] - Despite rapid progress, there are questions about whether Chinese firms can achieve significant innovations beyond existing technologies [36][38] Collaboration and Integration - Google DeepMind operates as a central hub for AI research within Google, integrating technologies across various products and ensuring rapid deployment of new capabilities [41][42] - The collaboration between DeepMind and Google is characterized by a close iterative process, allowing for swift adjustments to strategic goals and product development [42][43]

Core Insights - The number of active Agents is projected to surge from 28.6 million in 2025 to 2.216 billion by 2030, with a compound annual growth rate (CAGR) of 139% [1] - The total number of tasks executed annually is expected to explode from 44 billion in 2025 to 415 trillion by 2030, reflecting a CAGR of 524% [1] - The estimated annual Token consumption will increase dramatically from 0.0005 P in 2025 to 152,667 P by 2030, indicating a staggering CAGR of 3,418% [1] Group 1: Agent Development Trends - The trend is shifting from single LLMs to Agents, significantly boosting the demand for parallel processing [1] - Domestic and international models are accelerating Agent development, with notable advancements such as Kimi's new open-source model K2.5 and Anthropic's Claude in Excel plugin [1][2] - Agents enhance single LLMs by incorporating decision orchestration, enabling them to autonomously plan tasks and utilize external tools, thus addressing limitations in context awareness and real-time information retrieval [2] Group 2: Multi-Agent Systems (MAS) - Multi-Agent Systems are emerging as a new form of Agents, exemplified by Kimi K2.5, which can manage 100 sub-agents and execute 1,500 tool calls in parallel, reducing execution time by up to 4.5 times compared to single agents [2] Group 3: CPU as a Critical Support - CPUs are crucial for Agent performance, affecting latency, throughput, and power consumption, with CPU processing accounting for up to 90.6% of total latency [3] - In Agent operations, CPUs handle tasks that GPUs cannot, such as executing external tools and system-level task orchestration, thus becoming essential for efficient resource allocation [3] Group 4: Investment Recommendations - As the demand for Agents grows, CPUs are expected to become a key performance bottleneck, leading to increased demand for core supply chain companies such as Haiguang Information, Longxin Zhongke, Guanghe Technology, Tongfu Microelectronics, and Lanke Technology [4]

Group 1 - The core viewpoint of the article is that Yann LeCun, after leaving Meta, is launching a new company called Advanced Machine Intelligence (AMI), focusing on world models rather than large language models (LLMs) for achieving human-level intelligence [9][17][20] - LeCun criticizes Meta's product development decisions, stating that while research is acceptable, product execution has been poor, particularly under Mark Zuckerberg's leadership [2][3][15] - AMI aims to be an open-source platform, contrasting with the recent trend in Silicon Valley towards closed-source models, which LeCun believes is a misguided approach [11][13][16] Group 2 - The company will initially focus on research and development, specifically on world models, which LeCun argues are essential for building intelligent systems [17][19] - LeCun emphasizes that LLMs are not equivalent to AI and that understanding the real world is crucial for achieving human-like intelligence, which LLMs struggle to do [21][22][23] - AMI is seeking to raise €30 million (approximately 247 billion RMB) in funding, with an initial goal of €3.5 million for early financing, aiming for a total of €5 million in the first round [45][46][50] Group 3 - The company has already attracted interest from potential investors, including Cathay Innovation and Hiro Capital, indicating a shift in venture capital investment logic towards valuing founders over products [52][53][54] - LeCun is actively recruiting talent, including former Meta executives, to strengthen AMI's capabilities [40][42] - The ultimate goal of AMI is to become a leading supplier of intelligent systems, with a focus on practical applications of world models and planning capabilities [38][39]

Core Viewpoint - The article discusses the challenges faced by generative AI voice assistants, particularly in executing simple commands reliably, highlighting a gap between user expectations and actual performance [14][18]. Group 1: User Experience with AI Assistants - Users have reported frustrations with AI voice assistants like Alexa, which fail to execute basic commands such as brewing coffee or turning on lights, despite their advanced capabilities [4][8]. - The transition to generative AI has led to a situation where users experience inconsistent responses, with the AI providing creative but unhelpful reasons for not executing commands [7][16]. Group 2: Technical Limitations of Generative AI - Generative AI introduces a level of randomness that can lead to misunderstandings in command execution, making it unsuitable for tasks requiring precision and reliability [18][22]. - Traditional voice assistants operated on a template-matching basis, ensuring predictable outcomes, while generative models struggle to maintain consistency in system calls [19][23]. Group 3: Potential and Future Directions - Despite current limitations, there is recognition of the potential of generative AI to understand complex tasks and improve user interactions, suggesting a paradigm shift in capabilities [30][34]. - The article suggests that the chaos observed may not be a failure of generative AI but rather a misalignment of its application in contexts where deterministic execution is critical [44].

Summary of Key Points from the Conference Call Industry Overview - The conference call discusses the Internet sector in North America, focusing on key debates and investment opportunities for 2026, including advancements in LLM (Large Language Models), GenAI (Generative AI), hyperscaler growth, autonomous driving, and the impact of AI on various industries [1][2]. Core Themes and Arguments 1. **Thematic Debates for 2026**: - Key debates shaping the sector include LLM advances, GenAI productization, hyperscaler growth, the future of search, autonomous driving, and robotics [1][2]. - The market is expected to reward companies demonstrating positive ROIC (Return on Invested Capital) from GenAI or GPU-enabled technologies, while sectors facing disruption, such as rideshare and e-commerce, may trade at lower multiples [2]. 2. **Hyperscaler Growth**: - There is a bullish outlook on hyperscaler growth in 2026, particularly for AWS, GCP, and Azure, driven by increased AI tech adoption and diffusion across the economy [2]. 3. **Investment Opportunities**: - Key picks for 2026 include AMZN (Amazon), META (Meta Platforms), DASH (DoorDash), EBAY (eBay), and RBLX (Roblox) [1][2]. Company-Specific Insights 1. **Amazon (AMZN)**: - Target Price: $315, implying ~27% upside. - Expected to leverage both AWS and Retail to drive GenAI adoption, with a projected durable growth rate of over 20% for AWS in 2026 [3]. - Investments in AI-driven shopping assistants and logistics are expected to enhance retail growth and profitability [3]. 2. **Meta Platforms (META)**: - Target Price: $750, with ~15% upside. - Anticipated improvements in core engagement and monetization, with a focus on productizing new LLMs to drive revenue growth [7]. 3. **DoorDash (DASH)**: - Target Price: $300, indicating ~39% upside. - Investments in a unified tech infrastructure and autonomous delivery are expected to enhance ROIC and overall business performance [9]. 4. **eBay (EBAY)**: - Target Price: $112, suggesting a 23% upside. - Market skepticism about sustainability and profitability of growth is noted, but durable tailwinds in collectibles and new initiatives are expected to drive GMV growth [13]. 5. **Roblox (RBLX)**: - Target Price: $155, with ~100% upside. - Strong user-generated content platform performance is expected to drive bookings and engagement growth, despite short-term risks [14]. Additional Important Insights - **GenAI and AI Adoption**: - The call emphasizes the importance of demonstrating ROI from AI investments, particularly as companies face rising model training costs [40][41]. - The anticipated rollout of new AI models and tools is expected to enhance product monetization and engagement across platforms [49]. - **Agentic Commerce**: - The rise of agentic offerings is expected to impact e-commerce significantly, with personalized shopping experiences likely to drive consumer spending [67][69]. - **Autonomous Driving**: - 2026 is projected to be an inflection year for autonomous driving, with significant advancements expected in service availability and technology [78][90]. - **Physical AI**: - Companies are increasingly focusing on real-world data capture and physical AI, with Amazon's robotics-enabled warehouses expected to have a substantial impact on efficiency and cost savings [98][100]. This summary encapsulates the key discussions and insights from the conference call, highlighting the strategic outlook for the Internet sector and specific companies within it for 2026.

Core Insights - The article discusses the challenges and innovations in evaluating student performance in the context of AI advancements, particularly focusing on the shift from traditional assessments to AI-driven oral examinations [1][2][3]. Group 1: AI in Education - The traditional method of assessing students through written assignments has become ineffective due to the availability of AI tools that can assist students in completing their work [2][3]. - The introduction of AI-driven oral exams aims to evaluate students' real understanding and reasoning abilities, as it requires them to think on their feet without AI assistance [3][4]. Group 2: Implementation Challenges - Scaling oral exams presents logistical challenges, especially with larger class sizes, making coordination of exam schedules difficult [4][5]. - The use of AI to facilitate oral exams can streamline the process, allowing for personalized questioning and structured workflows [5][6][7]. Group 3: AI Oral Exam Structure - The AI oral exam consists of two main parts: discussing the student's project and analyzing a randomly selected case study, which tests their knowledge retention and application [9][10]. - A structured workflow with multiple AI agents is employed to ensure a smooth examination process, including identity verification and targeted questioning based on project details [11][12]. Group 4: Cost and Efficiency - The implementation of the AI oral exam system resulted in a total cost of $15 for 36 students, significantly lower than the estimated $750 for traditional human-led assessments [13][14]. - The average duration of the oral exams was 25 minutes, with a notable finding that shorter exams did not correlate with lower scores, indicating efficiency in understanding [32]. Group 5: Feedback and Assessment Quality - The AI system provides detailed feedback on students' performance, highlighting strengths and areas for improvement, which is more comprehensive than typical human feedback [29][30]. - The AI scoring system showed a high degree of consistency among different models, with a notable improvement in scoring accuracy after models reviewed each other's assessments [22][24]. Group 6: Student Reception - Student feedback indicated a preference for traditional assessments, with many feeling that AI oral exams increased pressure, yet a majority acknowledged that these exams better assessed their understanding [33][35]. - The article concludes that while the core idea of AI-driven assessments is promising, further refinement of execution details is necessary to enhance the student experience [35][36].