来源：环球网 【环球网科技综合报道】11月15日，2025人工智能+大会分论坛——"AI+影视"论坛在南中轴国际文化 科技园·云演艺中心举行。本次论坛由北京市丰台区南中轴地区管理委员会指导，中关村科技园区建设 投资有限公司、南中轴建设发展集团有限公司主办，南中轴（北京）国际文化科技发展有限公司、云戏 （北京）文化科技有限公司、故事接龙（北京）科技有限公司共同承办，以"技术赋能创意，AIGC重塑 影视创作新格局"为主题，汇聚了来自政府、产业、学界、研究、创作及投资等领域的百余位嘉宾，共 同探讨人工智能技术驱动下影视产业的新变革与新机遇。中关村发展集团副总经理伍发平、丰台区科信 局副局长薛明、南中轴管委会副主任安静等出席活动。 AI赋能，影视产业迎来全链条重塑 从2016年至2025年，人工智能在影视行业实现了从"概念验证"到"全链条协同"的深刻转型。当前，人工 智能技术已深度融入剧本生成、视觉预演、视频实时生成、智能剪辑及社群化分发等核心环节，不仅显 著提升生产效率，更系统重构产业底层逻辑。本次论坛全面呈现了人工智能从"辅助工具"升级为"全链 条赋能者"的产业图景，覆盖了从技术底层、内容创作到生态聚合的完整闭环 ...

Core Insights - The article discusses the evolution of algorithms in the chip design industry, particularly focusing on the advancements in attention mechanisms and their implications for future chip designs [2][4]. Group 1: Attention Mechanism Evolution - The Transformer architecture has dominated the large model field, but its self-attention mechanism poses significant computational challenges, especially in terms of power requirements during the prefill and decode phases [4]. - Various improvements to the Transformer structure have been proposed, such as Performer, Reformer, and lnformer, but none have achieved widespread application due to a lack of strong demand [4]. - The emergence of linear attention mechanisms aims to reduce computational complexity to linear levels, with models like RWKV and Mamba following this approach [5]. Group 2: Dynamic Sparsity and MoE Technology - Dynamic sparsity, particularly through Mixture of Experts (MoE) technology, has gained traction, allowing only a subset of experts to be activated during inference, which can lead to better performance and reduced computational costs [8]. - The trend towards increased sparsity in MoE models, such as Ant Group's recent models, indicates a significant shift in the industry, necessitating larger memory and bandwidth requirements [9]. Group 3: Low-Bit Quantization - The introduction of low-bit quantization techniques, such as FP8 training, has opened new avenues for model efficiency, with a focus on weight-only quantization to alleviate bandwidth bottlenecks [11]. - The article highlights the importance of fine-grained quantization and the potential for mixed quantization strategies to optimize model performance, especially in MoE models [12]. Group 4: Token Compression - Token compression has emerged as a critical area for reducing the computational burden of large models, particularly in visual token processing, which has shown high redundancy [14]. - The article notes a surge in research focused on token compression techniques, which could significantly impact chip design by lowering application barriers for large models [14]. Group 5: Future Implications for Chip Design - The advancements in attention mechanisms, dynamic sparsity, low-bit quantization, and token compression are expected to have substantial implications for the design of future edge chips, which have lagged behind the development of large models [14].

Group 1 - Sichuan Zigong Transportation Machinery Group Co., Ltd. (运机集团) signed a cooperation framework agreement with iFlytek (科大讯飞) and its subsidiary Zhejiang Tidal Power Technology Co., Ltd. (潮汐力) on September 4 [1] - The strategic cooperation focuses on emerging technologies such as acoustic large models, visual large models, predictive large models, and multi-modal large models, as well as intelligent products like optical fiber voiceprint diagnosis and industrial intelligent robots [1][2] - The collaboration aims to develop and share results in industries including mining, bulk material transportation ports, cement and building materials, metallurgy, electricity, intelligent equipment, and intelligent operations [1] Group 2 - The three parties will work together to seek technology projects, market resources, and business opportunities, establishing a joint development mechanism to address customer needs and industry pain points [1][2] - The partnership emphasizes principles of resource sharing, joint responsibility for results, and win-win cooperation, focusing on AI technology breakthroughs and project development [2] - Future efforts will center on AI large models, AI sensing instruments, and industrial intelligent robots, aiming to create a smarter, safer, and more efficient industrial production system [2]

Group 1 - The 2025 government work report has officially included "deep-sea technology" as a direction for cultivating new productive forces, emphasizing the promotion of its safe and healthy development alongside emerging industries like low-altitude economy [1] - In Zhoushan, a new low-altitude application integrating security, logistics, and tourism is rapidly unfolding, supported by a near-sea 5G communication real-time guarantee system developed by China Mobile [1] - The "Low-altitude Marine Intelligence Comprehensive Management Platform" utilizes 5G-A sensing base stations and a comprehensive perception network to ensure safe and precise drone operations, enhancing the quality of the island economy [1] Group 2 - Drones equipped with temperature-controlled preservation boxes can deliver freshly caught fish from breeding bases to tables in just 15 minutes, significantly reducing the time from the traditional 2 hours [3] - The launch of the "Zhoushan Police" drone flight reporting platform has facilitated the application of consumer drones, leading to a 45.6% month-on-month increase in recreational flying activities [3] - The deployment of a drone management platform based on digital twin technology has significantly expanded the low-altitude warning range in Zhoushan, reducing the occurrence of unauthorized flights [3]

Core Viewpoint - Hikvision collaborates with Zhenjiang Jin Gang Port Company to enhance the efficiency of belt conveyor inspections through large model and IoT perception technologies, significantly improving inspection efficiency and reducing safety risks for personnel and equipment [1] Group 1: Technology Implementation - The application of visual large models has improved target detection rates and reduced false alarm rates, enhancing intelligent recognition capabilities [2] - A self-cleaning camera installed above the belt can observe the belt's operation clearly in dusty environments, while an AI algorithm analyzes the belt's deviation status and provides timely warnings to prevent equipment damage, potentially saving over 300,000 yuan annually in cleaning costs due to deviations [2] - The drop accumulation detection model identifies when material accumulates to a certain level and sends alerts for timely cleaning, preventing disruptions to the belt's operation [2] Group 2: Smart Port Platform - Hikvision has developed a smart port platform for Zhenjiang Jin Gang Port Company, creating a dynamic digital twin of the port that integrates data from various sources into existing business processes [3] - The platform utilizes three-dimensional scene technology to replicate physical elements of the port, providing robust support for efficient management and decision-making [3] - The platform also incorporates DeepSeek, which uses natural language processing capabilities to facilitate interactions through simple text commands, acting as an "intelligent assistant" for management tasks [3] Group 3: Operational Efficiency - The project signifies a shift from on-site operations to centralized control, improving the working environment for personnel and reducing operational costs for the company [3] - This new operational model is expected to inject new intelligent momentum into the company's high-quality development [3]

以下文章来源于晚点Auto ，作者晚点团队 晚点Auto . 从制造到创造，从不可能到可能。《晚点LatePost》旗下汽车品牌。 比亚迪将推进大模型多线研发，这需要巨大的计算资源、数据量和训练量支持。 文丨司雯雯 这需要巨大的计算资源、数据量和训练量。目前，常见高阶智驾功能的车端大模型是十亿级参数，云端大模 型是百亿级参数，而基座模型、语言大模型等参数在千亿级别。 为快速补齐能力，比亚迪选择从大模型公司招聘有成熟经验的技术人才。聂鹏程 2013 年毕业，在 AI 计算领 域工作近 12 年，有管理大型算力资源、研发基座模型的经验，也参与研发 AI 应用。 编辑丨龚方毅 在偏好内部培优的比亚迪，外部人才的流向体现着一项业务的资源、级别和受重视程度 —— 2023 年起 智驾是行业的竞争焦点，也被比亚迪重视。2024 年下半年开始，以端到端大模型为代表的人工智能对汽 车公司越来越重要，车企的比拼从汽车的智驾、座舱延伸到人形机器人。 2024 年下半年起，比亚迪对人工智能业务的投入明显加大。 我们独家获悉，2024 年 11 月，零一万物前首席技术专家聂鹏程加入比亚迪先进技术研发中心，将主要负 责 AI 超算，也 ...