2月17日晚8点，马年春晚准时开场，火山引擎数据中心进入了最为紧张的时刻。 与往年不同，这一次服务器面临的挑战不仅仅是视频流的分发，根据春晚后公布的数据，除夕当天，豆包AI互动总数高达19亿次，直播的某一秒，当舞台 上的水墨画随风而动时，或许正有一个观众在豆包上输入了生成一张赛博风格全家福的指令。 算力服务器集群必须在毫秒级内完成数千万亿次浮点运算，确保用户手机上的图片在数秒内生成。 所以，在2026年马年春晚的舞台上，"豆包们"都在做什么？又是怎么做到的？ 在节目《驭风歌》中，导演组要求背景中的水墨画必须动态化，画中的马匹不仅要跑起来，且数量、形态不能有任何偏差。 这创意难就难在审美，市面上的海外视频生成模型大多基于写实风格训练，无法理解中国水墨画的留白与晕染逻辑。 直接生成的视频往往像生硬的3D贴图，破坏了原作的二维质感。 豆包Seedance 2.0团队在训练阶段输入了大量中国传统文化语料，模型因此建立了对水墨笔触的理解。 最终呈现在舞台背景上的画面，马匹在奔跑时保留了墨色晕染的动态变化，画面没有出现常见的结构崩坏。 在视频生成之前，豆包图像创作模型Seedream 5.0 Lite还承担了关键帧的设计 ...

Core Viewpoint - The Chinese AI industry is entering a new "Qianwen Cycle," marked by significant advancements and innovations, particularly led by Alibaba's Qianwen AI family, which is set to redefine AI applications and capabilities in the market [1][6][12]. Group 1: Model Development - Alibaba's Qianwen AI family has launched the Qianwen-Image-2.0 model, positioning it alongside leading models like Google's Nano Banana Pro and GPT Image 1.5 [4]. - The newly open-sourced Qianwen 3.5-Plus model features a total of 397 billion parameters, outperforming larger models like Qianwen 3-Max, and offers competitive performance against top closed-source models such as GPT-5.2 and Gemini-3-Pro [9]. - Qianwen has released over 400 open-source models, making it the most widely adopted open-source model globally, with a download count exceeding 1 billion [7][9]. Group 2: Application and Market Impact - The Qianwen App has initiated a "Spring Festival 3 billion free order" campaign, achieving over 10 million orders in just 9 hours, indicating a strong market response and consumer engagement [4]. - The app's focus has shifted from merely being a conversational AI to a more functional assistant, emphasizing the ability to perform tasks and meet user needs effectively [11]. - AI shopping, as a key feature of the Qianwen App, aims to connect users with resources in a streamlined manner, enhancing user experience and creating new consumption opportunities [11][12]. Group 3: Strategic Vision - Alibaba is addressing fundamental questions about the capabilities and responsibilities of AI, aiming to guide the future direction of Chinese AI development [6][12]. - The "Qianwen Cycle" represents a commitment to making advanced AI accessible to small businesses and individuals, fostering a more inclusive technological landscape [14]. - The overarching goal is to integrate advanced technology with everyday applications, ensuring that AI serves practical human needs while driving innovation [14].

Core Insights - The 2026 CCTV Spring Festival Gala showcased a blend of cultural heritage and cutting-edge technology, emphasizing the integration of AI in real-time creative interactions [1][16] - The event marked the first time AI-driven real-time audience interaction was implemented, with a total of 1.9 billion interactions recorded on New Year's Eve [1][10] - The gala featured innovative performances, including the AI-generated program "Riding the Wind Song," which utilized the Seedance 2.0 model to create stunning visual effects [2][17] AI Integration and Performance - The "Riding the Wind Song" program combined Zhang Jie's singing with the visual of Xu Beihong's ink painting, demonstrating the capabilities of the Seedance 2.0 model in creating fluid animations [2][17] - The Seedance 2.0 model was developed in less than 50 days, showcasing its rapid deployment and effectiveness compared to other mainstream video generation models [2][17] - The model's ability to generalize and maintain the artistic style of ink painting while ensuring smooth motion was highlighted as a significant achievement [3][18] Creative Production Efficiency - The use of Seedance 2.0 significantly reduced the production cycle and complexity of creative programs, allowing for quick adaptation of dance movements from reference videos [4][19] - The "Flower God" program featured twelve performers, with each scene generated by Seedance 2.0, creating a unique visual experience for each character [5][21] - The gala's visual content met high standards, requiring 8K resolution and 50 FPS, achieved through enhancements from the Volcano Engine's video services [6][21] Robotics and AI Interaction - The gala featured a "domestic robot team," showcasing advancements in robotics with interactive performances, including a humorous skit with a humanoid robot [7][22] - Robots demonstrated enhanced capabilities, including visual perception and semantic understanding, allowing for natural interactions with human performers [8][22] - The integration of the Volcano Engine's language model provided robots with advanced dialogue capabilities, enhancing their emotional expressiveness [9][23] Technical Infrastructure and Challenges - The AI interaction during the gala represented a significant technical challenge, requiring robust computational resources to handle the massive volume of requests [10][24] - The Volcano Engine's Ark platform successfully managed the high demand for AI interactions, demonstrating its capability to support large-scale AI applications [12][26] - The system's optimization allowed it to handle peak traffic of 63.3 billion tokens per minute, marking a new phase in AI interaction for the gala [13][27] Conclusion - The 2026 CCTV Spring Festival Gala not only highlighted the integration of technology in cultural events but also showcased China's advancements in AI and robotics, setting a precedent for future innovations [14][28]

Group 1 - The Indian Artificial Intelligence Impact Summit is the largest of its kind in India, expecting 250,000 attendees from around the world [1] - Major tech companies like Google, Microsoft, and Amazon plan to invest a total of $68 billion in AI and cloud infrastructure in India by 2030 [1] - Key speakers at the summit include leaders from Google, OpenAI, Anthropic, and DeepMind, highlighting India's ambition in the AI sector [3] Group 2 - India aims to leverage AI for economic growth and attract investments, focusing on the deployment of advanced AI models rather than developing foundational models [3] - The IT industry in India, valued at nearly $300 billion, faces challenges with potential revenue losses of 50% in call centers due to AI adoption [4] - Over 60% of Global Capability Centers (GCCs) established in India in the past two years are focused on AI, data, digital engineering, or product development [4] Group 3 - The Indian government is actively seeking to establish domestic supply chains to attract investments from major tech companies, including a recent $18 billion semiconductor investment project [4] - The support from the government is seen as a guarantee for multinational companies to diversify their operations in India [4] - The summit is expected to lead to significant announcements of investments in AI data centers and large-scale infrastructure agreements [4]

Core Viewpoint - The article emphasizes the impressive capabilities of the Doubao model, showcasing its significant role in the production of the Spring Festival Gala, highlighting its ability to generate high-quality visual and audio content in real-time, and its potential to revolutionize the entertainment industry [3][31]. Group 1: Doubao Model Performance - Doubao generated over 50 million new year avatars and 100 million new year greetings, with a total interaction of 1.9 billion times during the Spring Festival Gala [3]. - The model was involved in the production of various segments of the gala, including the visual effects in performances like "He Hua Shen" and "Yufeng Ge" [5][9]. - Doubao's ability to understand and execute complex artistic styles, such as Chinese ink painting, sets it apart from other models, which often fail to grasp the nuances of cultural aesthetics [10][12]. Group 2: Technical Challenges and Innovations - The Spring Festival Gala demands zero tolerance for errors, making it a significant challenge for AI models, as any mistake could lead to viral criticism [17]. - Doubao's performance is characterized by its speed and precision, capable of generating high-quality content that meets the stringent requirements of live broadcasts [15][18]. - The model's architecture, powered by the Volcano Engine, allows for efficient resource allocation and high throughput, enabling it to handle massive simultaneous requests without crashing [27][28]. Group 3: Accessibility and Inclusivity - The gala featured real-time subtitles for the hearing impaired, showcasing Doubao's advanced speech recognition capabilities, which can understand various dialects and accents [25]. - This inclusivity reflects a broader trend in technology to make content accessible to all audiences, enhancing the overall viewing experience [25]. Group 4: Future Implications - The capabilities demonstrated by Doubao during the gala suggest a transformative potential for AI in the entertainment industry, indicating that future productions may increasingly rely on such technology for creative processes [31]. - The success of Doubao could lead to further advancements in AI, pushing the boundaries of what is possible in real-time content generation and interactive experiences [31].

2026 年央视马年春晚落下帷幕，除了明星表演，幕后的"AI 视觉魔法师" Seedance 2.0 彻底火出了圈。 作为字节跳动旗下火山引擎的最新视频生成模型，春晚成为了它正式发布前的全球首个重量级应用场景。这款依托豆包大模型打造的技术，深度参与了三个节目的 视觉定制，让观众直呼"大开眼界"。 最令人印象深刻的莫过于张杰演唱的《驭风歌》。 Seedance 2.0 精准读懂了中国画的精髓，让徐悲鸿的《六骏图》在舞台上"活"了起来 ，水墨骏马奔腾时的鬃毛 律动与肌肉线条，既保留了笔墨质感，又充满了动态张力。 此外， 青年歌手演唱的《快乐小马》，其青春活泼的舞台视觉也得益于它的加持。 火山引擎总裁谭待表示，模型因训练了大量东方美学素材，能完美契合春晚的视觉创意。这场首秀，不仅让观众看到了科技与传统文化的绝美融合，更标志着国产 大模型在影视级内容制作领域迈出了关键一步。 （参考：微博、央视春晚） | 张杰 奇 十关注 | | --- | | 26-2-17 13:42 发布于 北京 来自 微博视频号 已编辑 | | 当导演组第一次和我聊《驭风歌》的核心策划,说要借AI技术,让徐悲鸿先生的 | | 《六骏图》化作一 ...

2025年底，行业正驶入快车道，上市潮起之际，月之暗面却明确表示不急于叩响资本市场大门。杨植麟 在信中称："相比于二级市场，我们判断还可以从一级市场募集更大量资金。事实上，我们B/C轮融资 金额就超过绝大部分IPO募资及上市公司的定向增发。所以我们短期不着急上市，也不以上市为目 的。" 对月之暗面旗下的Kimi来说，2025年是技术持续演进的一年。杨植麟在内部信开头即祝贺团队取 得"SOTA成绩"（业界顶尖水准）。他写道，K2和K2 Thinking的发布标志着公司在AGI道路上走出重要 一步，并列举了"中国首个万亿参数基座模型""第一个开源的agentic model"等技术标签。 在月之暗面的布局中，技术突破与产品迭代正为商业化铺平道路。自5月起，Kimi高频推出包括 Researcher、OK Computer、PPT、Kimi Code在内的多项智能体（Agent）功能。杨植麟在信中透露，9 月至11月，海内外付费用户数平均月环比增长超过170%。同时，K2 Thinking的发布也显著带动了API 收入，同期海外API收入增长达4倍。 然而，百亿元的现金储备只是下一场更昂贵竞赛的燃料。杨植麟明确，C ...

来源：国民经略 新一轮科技革命与产业变革，再次加速。 01 "机器人全面入侵春晚"，释放什么信号？ 2026年央视马年春晚，无疑是人形机器人密度最高的一届。 多家具身智能企业亮相，大秀武术、跑酷、共舞、陪伴等多元技能，人机同台，硅基智能与碳基生命同 场"炫技"，未来智能社会的新图景徐徐拉开。 不只是机器人，马年春晚也是含"科"量最高的一届。 图像及视频生成大模型、数字人、VR、AR、AI眼镜、无人驾驶、量子科技等硬核技术纷纷"登场"，让 春晚成为文化、科技共融的盛宴，也成为大国科技飞速突破的重要注脚。 无独有偶，就在春节前夕，字节跳动发布的Seedance 2.0模型引爆全球，DeepSeek也被曝或将在春节期 间推出最新大模型。 这一场景，与去年同一时期的"DeepSeek时刻"遥相呼应，成为中国AI从"追赶"到"领跑"的又一见证。 剑指人工智能先锋城市，深圳凭什么？ 早在多年前，作为"深圳机器人十三太保"之一的优必选，就已连续4年登上春晚。 种种信号表明，以人工智能为核心的第四次科技革命，正在成为国际竞争的关键焦点、大国博弈的重要 变量、经济发展的强大引擎。 人工智能、低空经济、新能源等新兴产业和未来产业 ...

Core Viewpoint - The 2026 Spring Festival Gala showcased a significant presence of technology and innovation, particularly in robotics and AI, marking a shift in China's industrial landscape towards hard technology innovation [4][22]. Group 1: Robotics and AI - The gala featured over 20 partners, with a notable focus on robotics, including companies like Yushutech, Magic Atom, and Galaxy General, highlighting the evolution of embodied intelligence in China [3][4]. - Yushutech's performance demonstrated advanced capabilities, contributing to its valuation exceeding 10 billion RMB and accelerating its IPO process [10]. - Galaxy General, a newly established company, achieved a valuation of 3 billion USD after raising 300 million USD, showcasing the rapid growth in the robotics sector [10]. - The event served as a platform for these companies to compete not just in performance but also in intelligence and practical applications, indicating a maturation of the robotics industry in China [11]. Group 2: New Media and Consumer Engagement - The integration of platforms like Bilibili, Xiaohongshu, and Douyin into the gala reflects a shift in how younger audiences engage with traditional media, emphasizing real-time interaction and content creation [16]. - Douyin's role as the exclusive partner for "Vertical Screen Viewing" and Bilibili's position as the exclusive bullet screen video platform illustrate the diversification of media consumption among younger generations [16]. - The rise of emotional consumption brands like Miniso and KAYOU at the gala indicates a cultural shift towards valuing emotional and cultural connections over traditional material consumption [17][18]. Group 3: Industry Transformation - The gala serves as a microcosm of China's industrial transformation, moving from traditional consumer goods to advanced technology and cultural products, reflecting changing consumer preferences [20][22]. - The reduction in alcohol brand presence at the gala signifies a broader trend away from traditional industries towards tech-driven sectors, highlighting the evolving landscape of consumer engagement [20]. - The emphasis on technology and innovation at the gala marks a pivotal moment in China's industrial evolution, suggesting a future focused on hard tech and sustainable growth [22][23].

Group 1 - The AI Impact Summit in India is the largest of its kind in the country, expecting 250,000 attendees from around the world [1] - Major tech companies like Google, Microsoft, and Amazon plan to invest a total of $68 billion in AI and cloud infrastructure in India by 2030 [1] - Key speakers at the summit include leaders from Google, OpenAI, Anthropic, and DeepMind, highlighting India's commitment to leveraging AI for human-centric missions [3] Group 2 - India has not yet developed a globally dominant AI model, with the US and China leading in large model technology [3] - The Indian government is encouraged to focus on "innovative applications" rather than investing heavily in developing new large models [3] - India has a significant potential consumer market for AI, with over 72 million daily ChatGPT users expected by the end of 2025, making it one of OpenAI's largest user markets [3] Group 3 - The Indian IT industry, valued at nearly $300 billion, faces challenges from AI adoption, with call centers potentially experiencing a 50% revenue loss by 2030 [4] - The rise of Global Capability Centers (GCCs) in India is shifting focus towards AI, data, digital engineering, and product development, with over 60% of new GCCs established in these areas [4] - It is projected that over 80% of GCCs will be AI-driven within the next 6 to 8 months [4] Group 4 - India is actively seeking to establish domestic supply chains to attract investments from major tech companies, recently approving a $18 billion semiconductor investment project [5] - Government support for technology is seen as a guarantee for multinational companies to diversify their operations in India [5] - The summit is expected to lead to significant announcements of investments in AI data centers and large-scale infrastructure agreements [5]