" 具身智能爆发第三年，世界模型凝聚了哪些共识？ " 作者丨 张进 吴彤 梁丙鉴 刘欣 齐铖湧 编辑丨 林觉民 马晓宁 13 日，第八届 GAIR 全球人工智能与机器人大会世界模型分论坛圆满成功。 这场的演讲嘉宾是在世界模型领域，研究不同方向的五位青年学者，他们带来了五场围绕世界模型的精彩 演讲，话题聚焦通用感知、三维技术、物理模型、世界模型、数字人重建。通过他们的演讲、我们得以窥 见当下围绕着世界模型的研究是多么广泛与丰富。 目前，世界模型的研究尚处于起步阶段，共识尚未形成，有关该领域的研究形成了无数支流，而这股潮流 中，今天到场的几位嘉宾，用他们的智慧和力量给世界模型领域研究带来了不同的启发。 浙江大学研究员彭思达：面向具身智能的通用空间感知技术 在"世界模型"分论坛上，首位演讲者是浙江大学研究员彭思达。他是浙江大学软件学院"百人计划"研究 员、博士生导师，研究方向为三维计算机视觉和计算机图形学。此次他带来的主题演讲是《面向具身智能 的通用空间感知技术》，介绍了其团队近期在赋予机器人通用感知能力方面的多项工作。 团队主要聚焦于赋予机器人三项基础能力：一是相机定位（Camera Pose Estimatio ...

Core Viewpoint - The article discusses the importance of understanding the underlying principles of world models, emphasizing that relying solely on data-driven approaches ("炼丹") is insufficient for creating effective AI systems. It advocates for the integration of human-understandable structures and logic into AI models to enhance their interpretability and reliability [2][63]. Group 1: Development of 3D Generation - The evolution of 3D generation has transitioned from early attempts at creating 3D models from single images to the current era of large models capable of generating high-quality 3D content from textual descriptions [7][16]. - The emergence of "open world" 3D generation began around 2023 with the Dreamfusion project, which allowed for the generation of 3D models without category restrictions, marking a significant shift in the field [11][12]. - Current trends in 3D generation focus on achieving finer details, structured outputs for easier editing, and better alignment between generated models and input images [19][20]. Group 2: Challenges and Opportunities in 3D Generation - The article highlights a dilemma faced by the 3D generation field, particularly in light of advancements in video generation technologies that can produce content without the complex 3D modeling processes [24][28]. - Despite the rise of video generation, 3D content creation retains its value due to its ability to provide physical realism, spatial consistency, and detailed control over content [29][34]. - The potential crisis for 3D generation lies in the increasing capabilities of video generation models, which are beginning to exhibit controllable features, raising questions about the necessity of 3D in future content creation [34][38]. Group 3: The Role of 3D in World Models - The article categorizes world models into three types: macro models for societal understanding, personal experience models for exploration, and embodied models for machine intelligence, with 3D being essential for interactive virtual environments [43][44][45]. - For embodied intelligence, understanding human interaction with the physical world necessitates 3D modeling to accurately capture and simulate these interactions [48][50]. - The transition from digital to physical manufacturing processes, such as 3D printing, underscores the foundational role of 3D data in creating tangible products [52]. Group 4: Technical Approaches in AI - The article contrasts explicit and implicit approaches in AI development, with explicit methods relying on clear geometric and physical modeling, while implicit methods depend on data-driven neural networks [56][57]. - The need for explainability in AI systems is emphasized, suggesting that a balance between performance and interpretability is crucial for user trust and safety [58][63]. - The discussion concludes that 3D and 4D modeling are vital for providing a comprehensible framework for understanding complex AI systems, thereby enhancing user confidence [59][63].

Core Insights - The GAIR conference aims to explore the transformative power of AI technology beyond technical discussions, focusing on its impact on education, industry, and civilization [1] Group 1: Conference Overview - The 8th GAIR Global Artificial Intelligence and Robotics Conference took place in Shenzhen, featuring prominent scholars and industry leaders [2] - The conference has been a platform for academic exchange and a repository of China's AI development over the past 40 years since its inception in 2016 [2][3] - The main forum included discussions on redefining education and reconstructing paradigms in various fields, showcasing cutting-edge insights from top scholars [3] Group 2: Educational Transformation - Zhao Wei, a prominent academic, highlighted the profound impact of AI on higher education, emphasizing the need to redefine student training and educational management [6][7] - The "add-substitute-replace" model was proposed for student training, focusing on practical skills and reducing ineffective course content [6] - The traditional educational management systems need to evolve into intelligent systems that can provide real-time responses and decision-making capabilities [7] Group 3: AI in Education - Guo Yike discussed the shift in education from knowledge transmission to fostering curiosity, creativity, and collaborative awareness among students [9][10] - He emphasized the importance of integrating values and self-reflection into education, alongside knowledge acquisition [10] - The roundtable forum addressed the core contradictions and transformation paths in education due to AI, highlighting the need for a new educational philosophy [11][13] Group 4: Industry Insights - Kazuhiro Kosuge presented on the potential of AI-powered robotics to revolutionize the garment production process, noting the industry's significant market size and current low automation levels [22][23] - The global garment market is projected to reach $2.3 trillion by 2030, yet automation in textile industries remains minimal [23] - The need for automation in the garment sector is driven by high labor costs, particularly in Europe, where automation is becoming essential for competitiveness [25] Group 5: AI and Scientific Research - Jia Jiaya discussed the future of AI and large models, advocating for a shift towards "perceptual machines" and lifelong learning models [26][29] - The integration of AI into scientific research is seen as a pathway to enhance understanding across various scientific domains, including astronomy and life sciences [42][43] - The development of scientific foundational models aims to overcome language barriers and complex scientific data challenges [42][44] Group 6: Challenges and Opportunities in AI - The roundtable on AI industrialization highlighted the challenges of scaling AI applications and the need for a robust business model [48][49] - Experts noted the disparity between initial optimism in AI capabilities and the practical challenges faced in implementation [49][50] - Opportunities in AI lie in sectors with limited data, such as healthcare, where traditional models may still be necessary [51] Group 7: Future Directions - The conference concluded with discussions on the importance of continuous learning and the integration of AI with physical systems for enhanced capabilities [30][65] - The exploration of new modalities in perception, such as sound and millimeter-wave sensing, is expected to flourish in the coming years [67] - The emphasis on developing intelligent hardware that incorporates native memory and autonomous learning is seen as crucial for future advancements [63]

" 将 Key 跟 Value Cache 按照不同的方法压缩，可以让模型不掉 点。 " 作者丨张进 编辑丨 林觉民 目前，不同大模型厂商发布的大语言模型在处理超长上下文方面已经有显著突破，最高的已能支持数百万 Token 的输入，例如 MiniMax-M1、Qwen2.5-1M 系列模型，均支持百万Token（1M）级别的超长上 下文处理能力。 但是这场有关提升大模型上下文长度的"军备赛"依然不会停止，这是一项巨大的工程与效率之战。因为超 长下文为模型智能提供了最广阔的发挥空间——在处理如金融、法律、医疗等领域的长语境任务时表现更 好。所以谁能率先突破更长上下文处理能力，便有机会创造出更大的商业与技术价值。 胡侠团队便针对这一目标提出了一项最新研究方案——"通过有损计算（Lossy Computation）来提高大 语言模型的推理效率"。这项研究的基本思路是，利用大语言模型对来自低精度计算等"有损"操作产生的 噪声具有极强鲁棒性这一特点，主动引入可控的、不损害性能的信息损失，以换取显著的效率提升。 大模型中的"有损计算"是通过有选择地牺牲一部分精度来大幅降低计算或者存储成本，从而提升推理效 率，主要围绕模型 ...

Key Points - Xiaopeng Motors' founder He Xiaopeng made a bet with his team regarding the performance of their VLA2.0 compared to Tesla's FSD by 2026, indicating confidence in the advancement of autonomous driving technology [4][5] - Nvidia responded to allegations that Chinese startup DeepSeek used smuggled Blackwell chips for AI model training, stating they have seen no evidence of such activities [7] - ZTE Corporation announced its commitment to anti-corruption and is currently in communication with the U.S. Department of Justice regarding compliance investigations related to overseas bribery [9][10] - Zhu Xiaohu commented on Tencent's cautious investment strategy over the past 20 years, emphasizing that the company waits for market clarity before making significant moves [11] - The Chinese government is expected to continue its "national subsidy" policy for consumer goods in 2024, with a focus on optimizing implementation by 2026 [19][20] - MiniMax and Zhizhu, two domestic AI unicorns, are reportedly planning to conduct IPOs in Hong Kong soon, aiming to become the first publicly listed company in the large model sector [21] - JD Industrial, a subsidiary of JD Group, officially listed on the Hong Kong Stock Exchange, raising approximately HKD 2.827 billion [22][23] - Meitu's CEO announced an internal venture initiative, allowing employees to apply for funding to develop AI projects, aiming to enhance organizational efficiency [24] - Lantu Motors' chairman emphasized the need for a breakthrough in the luxury car market, which has been dominated by foreign brands [25] - Xiaomi launched its first self-produced central air conditioning unit at its Wuhan smart home appliance factory, showcasing advancements in its manufacturing capabilities [26][27]

" 在模型与算力的潮汐中，智能星火正在汇成产业巨浪，且看AI如 何重构产业生态的万千图景。 " 作者丨徐晓飞 编辑丨包永刚 12月12日的深圳，和世界万千城市一同蛰伏于智能产业爆发的黎明前夜，而一场汇聚前沿洞见的思想盛 会，正在此破土而出。 站在大模型技术深入"产业变革"的关键节点， 第八届 GAIR 全球人工智能与机器人大会 ，正式在深圳博 林天瑞喜来登酒店举办。 大会共开设四个主题论坛与两个闭门会议，聚焦 大模型、AI算力、世界模型、 数据&一脑多形、AI 硬件 等领域的创新脉搏。 这是GAIR大会走过的第八载，也是中国AI产学研投专家群体，对当前科技变革的又一次思想共振与方向校 准。 古有探骊得珠，需持炬而入深海，方可见骊龙颔下之至宝。 对眼下的AI大模型产业变革来说，亦是如此。 要知道，如今的AI大模型浪潮，已从几年前的"技术破壁"迈入了"价值深耕"阶段，愈发如深海骊龙的颔下 之宝，浮于浅水者必不可得。 而始于2016年的GAIR大会便如这枚探海之炬，八载深耕，薪火相传，汇聚前瞻学者与行业先锋的顶尖思 想，既照见了全球 AI 从业者的筚路蓝缕，也照彻了智能纪元从萌芽到勃发的浩荡征程。 GAIR大会至今 ...

Core Viewpoint - OPPO is intensifying its focus on AI by restructuring its AI center and consolidating key AI-related services into a single project called "Super Xiaobu" [2]. Group 1: Organizational Changes - OPPO has completed a new organizational structure adjustment for its AI center, merging three core services: Xiaobu Memory, Xiaobu Assistant, and Xiaobu Suggestions into "Super Xiaobu" [2]. - The newly formed Smart Product R&D Department will oversee the development of "Super Xiaobu," led by Jiang Yuchen, who was previously in charge of Xiaobu Memory [2]. - This restructuring reflects OPPO's strategic approach to unify AI capabilities, contrasting with other brands that are still competing for dominance between AI and operating systems [2]. Group 2: AI Center Development - The AI center was established in January 2024, integrating various AI-related fields from within the company, including digital engineering and software engineering [2]. - Over the past two years, the AI center has successfully delivered products such as the integration of Xiaobu Assistant with DeepSeek and the development of one-click memory capabilities [2]. Group 3: Future Prospects - Jiang Yuchen has recently founded a startup called Wave Intelligent, focusing on long-text generation for novel writing, which OPPO plans to acquire in October 2024 [3]. - In a recent interview, Jiang mentioned that the AI technology used in the Doubao phone serves as a fallback solution for OPPO to cover long-tail scenarios, indicating a preference for an Agent to Agent approach for ecosystem interconnectivity [3].

" 顶级AI产、学、研融合大会来到第八年，现场将会碰撞出怎样的 火花？ " 作者丨 徐晓飞 编辑丨 包永刚 2025年，是大模型从"技术破壁"迈入"价值深耕"的关键之年。这一年里，技术与产业纷繁变革，同频共 振，加速融合。那些对未来的颠覆，正在AI革命的黎明微光中酝酿、翻涌，向人们徐徐打开一个新世界。 在这时代的晨曦中，如何拨开迷雾，窥见先机，奋勇入局，拥抱这场庞大、多元、复杂的智能浪潮，正在 成为每一位行业先行者的叩问。 而始于2016年的"全球人工智能与机器人大会"（GAIR），在过往九年的时间里，数次站在了AI变革的潮 头浪尖，以前瞻性视野，精准而深刻地把握住了AI科技的时代脉搏，为产学研投各界搭建起了前沿交流的 核心桥梁。 作为人工智能领域的风向标，历届GAIR大会邀请了多位图灵奖、诺贝尔奖得主、50位院士、30位人工智 能国际顶会主席、100多位 Fellow，及500多位知名企业家、投资者和创新者共襄大会、论道AI。 这一次也不例外。 12月12日-13日，在深圳博林天瑞喜来登酒店三楼宴会厅， 第八届GAIR全球人工智能与机器人大会， 也 将汇聚上百位嘉宾和数千位专家，开设四个主题论坛与两个闭门会 ...

Core Viewpoint - The future of L4 autonomous driving companies is to become large integrators or major operators like Didi, as indicated by the founder of Sien Intelligent Driving, He Bei [1]. Group 1: Background and Development - He Bei graduated from Tsinghua University and joined Baidu in 2015, where he contributed to the development of autonomous driving technology [2][3]. - The early team at Baidu iV saw many members, including He Bei, leave to start their own ventures, driven by differing views on the future of autonomous driving technology [3][4]. - He Bei's initial focus was on a cost-effective, vision-based approach to autonomous driving, which contrasted with Baidu's emphasis on high-definition maps [4][10]. Group 2: Company Overview and Achievements - Sien Intelligent Driving, founded by He Bei in 2020, has completed eight rounds of financing and aims to achieve profitability by 2025 [4][5]. - The company has established strategic partnerships with two of the world's top three ports, Ningbo-Zhoushan and Qingdao, and has implemented solutions in key port areas [5][6]. - Sien Intelligent Driving's revenue is projected to reach around 500 million yuan in the coming year, with the company currently experiencing a slight loss [24][26]. Group 3: Market Challenges and Opportunities - The primary challenge for autonomous trucks is the difficulty in monetizing services, as clients are often reluctant to pay for autonomous vehicle operations despite the apparent market potential [6][35]. - The autonomous driving industry has experienced various waves of interest, with the current focus shifting towards logistics, ports, and mining [23]. - He Bei believes that the commercial value of autonomous vehicles is clearer in port operations compared to other sectors, as these environments have a more mature level of information technology [15][16]. Group 4: Future Outlook and Strategy - Sien Intelligent Driving plans to focus on delivery, customer acquisition, and expanding into international markets, with potential clients in Singapore, Brazil, and Abu Dhabi [36][37]. - The company aims to become a major player in the L4 commercial vehicle sector, emphasizing the importance of system integration capabilities and deployment efficiency [18][30]. - He Bei envisions Sien Intelligent Driving evolving into a large integrator, potentially diversifying through partnerships and acquisitions rather than solely expanding its own operations [39].

Core Viewpoint - The collaboration between Shopee and Dematech aims to innovate and implement smart logistics solutions to enhance sorting efficiency in response to the growing order volume and user experience demands in the fast-developing logistics sector [2][4]. Group 1: Collaboration and Partnership - Shopee and Dematech have established a deep partnership to address logistics pain points, with Dematech providing customized sorting solutions that ensure efficient and stable operations during peak order periods [2][4]. - The signing of an annual framework cooperation agreement marks a significant milestone in their relationship, highlighting mutual trust and Dematech's enhanced role in supporting Shopee's cross-regional logistics development [4][6]. Group 2: Technology and Solutions - Dematech's intelligent sorting system has been implemented in Shopee's sorting centers in Singapore and Brazil, achieving scalable applications across various business scenarios [4][6]. - The intelligent sorting system and digital management platform developed by Dematech offer flexible and efficient solutions, ensuring quick and accurate sorting processes, especially during peak order times [6][8]. Group 3: Future Outlook - Dematech plans to continue supporting Shopee's logistics network with innovative technologies and professional services, positioning smart sorting equipment and automation solutions as core elements for improving efficiency and ensuring fulfillment [8].