Domestic News - The State-owned Assets Supervision and Administration Commission (SASAC) is promoting the integration of technological innovation and industrial innovation, guiding central enterprises to accelerate the industrialization of scientific and technological achievements, and facilitating deep connections between the industrial chain and innovation chain [1] - SASAC supports central enterprises in establishing 97 original technology sources and has set up several joint enterprise funds, achieving a number of landmark results [1] - Over 1,000 outstanding achievements have been included in the recommended directory of scientific and technological achievements of central enterprises, which have been promoted and implemented in more than 6,000 enterprises [1] International News - The unemployment rate in the UK for the three months ending December 2025 reached 5.2%, the highest since January 2021, matching expectations and up from a previous value of 5.1% [4]

作者：是地理空间 AI 公司 Niantic Spatial 创始人兼 CEO，在科技与地图交叉领域深耕 25 年，主导开发 了《宝可梦 GO》、谷歌地图、谷歌地球及 Keyhole 地球查看器等产品。 四年前，元宇宙概念正流行时，我曾发文称它是一场反乌托邦噩梦。我当时主张，与其沉浸虚拟世界， 我们更应多与身边的现实世界联结。 时间会证明元宇宙最终能否成真，但 2024 和 2025 年，全球商务与休闲旅行、线下活动参与度均已超过 疫情前水平，这很明显说明：人类本就不该活在头显设备里。 转眼到今天，有人正试图打造这样一个世界：内容越来越不来自朋友，甚至不来自人类网红，而是直接 由 AI 模型生成 —— 一种经过工程化制造的像素与波形，只为抓住你的注意力，直到下一条广告出 现。 短短几年前，先进 AI 刚崭露头角时，人们对未来充满热情：一个人类繁荣的未来，医疗、材料、制 造、机器人技术的突破性进步将让世界变得更好，提升地球上所有人的生活质量。 我认为现在有理由问一句：我们真的在朝这个方向走吗？我们对 AI 的巨额投资中，有多少会真正用于 改善人类生存状况，又有多少会被用来制造更精巧的娱乐与分心手段？ 为什么这件 ...

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 ...