每经记者｜李宇彤 每经编辑｜文 多 穿着一件黑色T恤、戴着标志性的框架眼镜，久未公开亮相的字节跳动创始人张一鸣，在10月9日上海徐汇知春创新中心的开业仪式 上，以一场技术味十足的发言引发行业震动。 张一鸣用模型中的"过拟合"（模型对已知数据预测得很好，但对未知数据预测得很差的现象）作为类比，直指当前AI（人工智能）人 才培养的硬伤："有的人才可能专业知识扎实，也有'高精准'的技能，但面对创新任务就不行了。" 这一犀利观点，在职场社交平台"脉脉"9月最新发布的《2025年AI人才流动报告》（以下简称《报告》）中或许能找到原因。 视觉中国图 《报告》显示，2025年前7个月，AI岗位需求同比上涨10倍，但"算法"相关人才持续紧缺，"搜索算法"人才甚至面临"5岗争2人"的极端 供需矛盾。而从"新发AI岗位量最多企业TOP20"榜来看，字节跳动的招聘指数为29.83，位列首位。 在这场被业界称为"字节跳动人才战略2.0"的活动上，张一鸣与上海交通大学ACM班（计算机科学班）创始人俞勇联手推出民办非营利 机构"上海徐汇知春创新中心"，计划招聘对泛计算机和人工智能感兴趣的年轻人，培育新一代的创新人才。 张一鸣此举，一方面 ...

图片来源：知春创新中心公众号文章截图 着一件黑色T恤、戴着标志性的框架眼镜，久未公开亮相的字节跳动创始人张一鸣，在10月9日上海徐 汇知春创新中心的开业仪式上，以一场技术味十足的发言引发行业震动。 张一鸣用模型中的"过拟合"（模型对已知数据预测得很好,但对未知数据预测得很差的现象）现象作为 类比，直指当前AI（人工智能）人才培养的硬伤："有的人才可能专业知识扎实，也有'高精准'的技 能，但面对创新任务就不行了。" 这一犀利观点，在职场社交平台"脉脉"9月最新发布的《2025年AI人才流动报告》（以下简称报告）中 或许能找到原因。据报告，2025年前7个月，AI岗位需求同比上涨10倍，但"算法"相关人才持续紧 缺，"搜索算法"人才甚至面临"5岗争2人"的极端供需矛盾。而从"新发AI岗位量最多企业TOP20"榜来 看，字节跳动的招聘指数为29.83，位列首位。 在这场被业界笑称为"字节跳动人才战略2.0"的活动上，张一鸣与上海交通大学ACM班（计算机科学 班）创始人俞勇联手推出民办非营利机构"上海徐汇知春创新中心"，计划招聘对泛计算机和人工智能感 兴趣的年轻人，培育新一代的创新人才。 张一鸣此举，一方面展示了字节 ...

文 | 窄播，作者 | 李威 国庆假期刚开始，Sora 2就引爆了整个AI圈子。 这个新的视频模型带来了对现实世界更精准的呈现，有更强的可控性，能创造出复杂的音频，并且能够 轻松地将现实世界中的人和物插入到AI生成的视频内容中，完成一场现实里很难实现的高难度客串 （Cameo）。 于是，我们在这两天看到了大量由OpenAI CEO Sam Altman参演的AI视频作品，他会与瑞克、莫蒂对 谈，和同事一起撞翻充满彩色泡泡球的浴缸，在一个播客中参与讨论。这些内容发布在与Sora 2同步推 出的Sora App上，被看作是「AI抖音」将要到来的体现。 很多人相信，人们会被客串激发出更多AI视频的创造行为。但我们认为，即便有完整的内容推荐功 能，当下的Sora App本质也还是一个工具，而不是平台，它与另一个近期大火的AI视频生成产品 Higgsfield是同类——都是在利用AI提供更高级的滤镜，激发出人们跟风和模仿的冲动。 Sora 2带来的模型能力的提升，更可能会加快To B方向的落地，推动整个视频大模型行业的技术更新， 让AI更好地服务于有创作冲动的人。我们不知道To C层面的AI抖音什么时候能够出现，以及A ...

以下文章来源于窄播 ，作者窄播 窄播 . 关注消费、互联网、AI，聚焦公司、产业、案例，提供信息、趋势、洞察。 来源丨 窄播（ID：exact-interaction） 作者 | 李威 图源丨Midjourney 国庆假期刚开始，Sora 2就引爆了整个AI圈子。 这个新的视频模型带来了对现实世界更精准的呈现，有更强的可控性，能创造出复杂的音频，并且能 够轻松地将现实世界中的人和物插入到AI生成的视频内容中，完成一场现实里很难实现的高难度客串 （Cameo）。 于是，我们在这两天看到了大量由OpenAI CEO Sam Altman参演的AI视频作品，他会与瑞克、莫蒂 对谈，和同事一起撞翻充满彩色泡泡球的浴缸，在一个播客中参与讨论。 这些内容发布在与Sora 2 同步推出的Sora App上，被看作是「AI抖音」将要到来的体现。 很多人相信，人们会被客串激发出更多AI视频的创造行为。但我们认为，即便有完整的内容推荐功 能， 当下的Sora App本质也还是一个工具，而不是平台 ，它与另一个近期大火的AI视频生成产品 Higgsfield是同类—— 都是在利用AI提供更高级的滤镜，激发出人们跟风和模仿的冲动。 ...

Industry Recognition - Video Arena Leaderboard showcases rankings of Text-to-Video and Image-to-Video models based on over 14,000 community votes [1] - Google DeepMind, Hailuo AI, Bytedance, Kling AI, Alibaba Wan, Pika Labs, and Genmo AI are recognized for their achievements in Text-to-Video technology [2] Text-to-Video Model Rankings - Veo3 (with audio) ranks 1 in Text-to-Video [2] - Hailuo 02 [Standard] and Seedance 1.0 pro rank 5 [2] - Kling 2.1 Master ranks 6 [2] - Wan 2.2 A14B ranks 9 [2] - Pika 2.2 and Mochi 1 rank 11 [2]

Investment Rating - The report maintains an "Outperform" rating for the communication industry [4][49]. Core Insights - NVIDIA is accelerating its expansion into the European market, planning to establish 20 AI factories, which will increase AI computing power in Europe by tenfold within two years, equipped with 10,000 GPUs [1][11]. - Oracle's cloud infrastructure revenue has shown significant growth, with a 52% year-on-year increase in the fourth quarter, and is expected to continue growing at a rate exceeding 70% in the next fiscal year [1][17]. - The report emphasizes the importance of AI cloud-side and edge-side developments while also highlighting the stable dividend value of major telecom operators [3][49]. Summary by Sections Industry News Tracking - NVIDIA's GTC Paris event announced plans for substantial AI infrastructure growth in Europe, while Oracle reported strong cloud revenue growth [1][11]. - ByteDance's Volcano Engine launched the Doubao model 1.6, showcasing improvements in various AI capabilities [2][23]. - The upcoming 2025 World Mobile Communication Conference (MWC) in Shanghai will focus on themes such as 5G integration and AI [2][33]. Market Performance - The communication index fell by 0.7835% this week, with the communication sector ranking third among major industries this month [2][42]. Investment Recommendations - The report suggests focusing on AI computing facilities and maintaining long-term investments in major telecom operators due to their stable performance and increasing dividends [3][49]. - Recommended stocks include China Mobile, Zhongji Xuchuang, Tianfu Communication, and Guanghetong [3][49]. Key Company Earnings Forecast and Investment Ratings - China Mobile, Zhongji Xuchuang, and ZTE Corporation are rated as "Outperform" with projected earnings per share (EPS) growth and favorable price-to-earnings (PE) ratios [4][50].

Investment Rating - The industry investment rating is "Positive" (maintained) [1] Core Insights - The report highlights the acceleration of AI applications and models, particularly from ByteDance and Oracle, indicating a strong growth trajectory in the AI computing and application sectors [10][20] - The report emphasizes the robust performance of Oracle's cloud business, with expectations for significant revenue growth in the upcoming fiscal year [18][19] - The report identifies seven key industry directions for investment, including AIDC infrastructure, IT equipment, network devices, cloud computing, AI applications, satellite internet, and 6G [20][21][22][27] Summary by Sections 1. Investment Outlook - ByteDance's Force Conference showcased significant advancements in AI models, with daily token usage exceeding 16.4 trillion, marking a 137-fold increase since its launch [13][15] - OpenAI's release of the o3-pro model enhances complex reasoning capabilities, further driving demand for AI computing [16][17] - Oracle's fourth fiscal quarter results exceeded expectations, with cloud infrastructure revenue projected to grow over 70% in the next fiscal year [18][19] 2. Communication Data Tracking - As of April 2025, China has 4.439 million 5G base stations, with a net increase of 188,000 from the end of 2024 [29] - The number of 5G mobile phone users reached 1.081 billion, reflecting a year-on-year growth of 21.6% [29] - 5G mobile phone shipments totaled 19.889 million units, accounting for 79.4% of total shipments, although this represents a 1.7% year-on-year decline [29] 3. Operator Performance - The report notes strong growth in the cloud revenue of major telecom operators, with China Mobile's cloud revenue reaching 100.4 billion yuan, a 20.4% increase year-on-year [43] - The average revenue per user (ARPU) for China Mobile remained stable at 48.5 yuan, while China Telecom's ARPU slightly increased to 45.6 yuan [43][46]

Core Insights - ByteDance's cloud business platform, Volcano Engine, launched the Doubao large model 1.6, introducing a tiered pricing strategy based on context length, resulting in a 63% reduction in comprehensive costs compared to the previous model [1][2] - The new pricing model aims to facilitate broader adoption of multi-modal deep thinking models and address the high cost pressures associated with enterprise-level AI agents [2][6] - Doubao large model 1.6 includes three versions, enhancing capabilities in deep thinking, multi-modal understanding, and real-time interaction [3][4] Pricing Strategy - The pricing for the Doubao large model 1.6 is structured into three tiers, with the most common input range (0-32K tokens) priced at 0.8 yuan per million tokens for input and 8 yuan for output, leading to a comprehensive cost of 2.6 yuan [1][2] - This pricing strategy is designed to lower the operational costs for enterprises, where a single AI agent can consume up to 20 dollars daily in token costs [2] Model Capabilities - Doubao large model 1.6 supports various functionalities, including deep thinking, multi-modal understanding, and GUI operations, making it suitable for applications in e-commerce and autonomous driving [3][4] - The model demonstrated strong performance in assessments, scoring 144 in a national math exam and achieving high scores in simulated exams [3] Market Penetration - Volcano Engine holds a 46.4% market share in China's public cloud large model service usage, significantly outpacing competitors like Baidu and Alibaba [6] - The daily token usage for Doubao large model surged from 4 trillion in December 2024 to 16.4 trillion by May 2025, marking a 137-fold increase since its launch [6] - The model has been widely adopted across various sectors, including consumer electronics, automotive, finance, and education, with partnerships established with major companies in these industries [6]

Core Insights - The core viewpoint of the article revolves around the aggressive growth and revenue targets set by Huoshan Engine, aiming for over 250 billion yuan in revenue by 2025, which represents a 100% growth from 2024's revenue of over 120 billion yuan [1][2]. Group 1: Company Growth and Market Position - Huoshan Engine has rapidly transformed from a minor player to a significant disruptor in the cloud market, largely due to its large model, Doubao, which has achieved a market share of 46.4% in 2024 [3][4]. - The competition between Huoshan Engine and Baidu Cloud is intensifying, with both companies engaging in a price war, impacting overall industry revenue [1][2]. - Huoshan Engine's president, Tan Dai, emphasizes the importance of focusing on innovation and core competencies rather than external factors [1][2]. Group 2: Pricing Strategy and Market Impact - The newly released Doubao 1.6 has a significantly lower cost structure, with input and output prices at 0.8 yuan and 8 yuan per million tokens, respectively, making it one-third the cost of its predecessors [4][9]. - The aggressive pricing strategy has led to a substantial increase in customer usage, with average daily token usage per customer growing by 20 to 30 times within three months of Doubao's launch [9][11]. - Despite the low pricing, Huoshan Engine faces challenges in converting its large user base into substantial revenue, as its 2024 revenue of 125 billion yuan still lags behind competitors like Alibaba Cloud and Baidu Cloud [12][16]. Group 3: Future Challenges and Technological Development - Moving forward, Huoshan Engine must focus on enhancing model performance and service quality to maintain competitiveness, as low pricing alone may not suffice [18][19]. - The company has established a research organization, "Seed Edge," to advance its AI technology and improve model capabilities, which is crucial for expanding its market presence [19][22]. - Recent developments show that Doubao 1.6 has surpassed competitors in certain performance metrics, indicating a shift towards prioritizing technological advancements alongside pricing strategies [19][22].

Group 1: OpenAI and Mistral AI Developments - OpenAI released the inference model o3-pro, which is marketed as having the strongest reasoning ability but the slowest speed, with input pricing at $20 per million tokens and output at $80 per million tokens [1] - User tests indicate that o3-pro excels in complex reasoning tasks and environmental awareness but is not suitable for simple problems due to its slow inference speed, targeting professional users [1] - Mistral AI launched the strong inference model Magistral, which includes an enterprise version Medium and an open-source version Small (24B parameters), showing excellent performance in multiple tests [2] - Magistral achieves a token throughput that is 10 times faster than competitors, with a pricing strategy of $2 per million tokens for input and $5 per million tokens for output [2] Group 2: Figma and Krea AI Innovations - Figma introduced the official MCP service, allowing direct import of design file variables, components, and layouts into IDEs, achieving a higher fidelity than third-party MCPs [3] - Krea AI launched its first native model Krea 1, focusing on solving issues of AI image "homogenization" and "plasticity," providing high aesthetic control and professional-grade output [4][5] - Krea 1 supports style reference and custom training, with native support for 1.5K resolution expandable to 4K, aimed at accelerating digital art creation processes [5] Group 3: ByteDance and Tolan AI Applications - ByteDance released the Doubao large model 1.6 series, which includes multiple versions supporting 256k context and multimodal reasoning, with a 63% reduction in comprehensive costs [6] - Tolan, an alien AI companion application, has achieved 5 million downloads and $4 million ARR, emphasizing a non-romantic, non-tool-like companionship experience [7] - Tolan's design integrates companionship with gamification, allowing users to customize their alien companion's appearance and develop unique planetary environments [7] Group 4: Li Auto and Figure Robotics Strategy - Li Auto established two new departments, "Space Robotics" and "Wearable Robotics," to enhance its AI strategy, focusing on creating a smart in-car experience [8] - Figure aims to provide a complete "labor force" system with humanoid robots, emphasizing fully autonomous operation and a production line capable of producing 12,000 units annually [9] - Figure plans to deliver 100,000 units over the next four years, targeting both commercial and home markets, while utilizing a shared neural network for collective learning [9] Group 5: Altman's Predictions and OpenAI Codex Insights - Altman predicts that by 2025, AI will be capable of cognitive work, with significant productivity boosts expected by 2030 as AI becomes more affordable [10] - OpenAI Codex is shifting software development from synchronous "pair programming" to asynchronous "task delegation," anticipating a transformation in developer roles by 2025 [11] - The team envisions a future where the interaction interface merges synchronous and asynchronous experiences, potentially evolving into a "TikTok"-like information flow for developers [11]