群核科技登陆港股 成为“杭州六小龙”首家完成IPO的企业 在我们《硅谷101》的专访中 创始人黄晓煌说 将押注空间智能的未来 我们所有训练的模型以及我们的工具 都紧贴着物理世界在做 漫画效果之类的 我们是不打算做的 李飞飞他们那个模型就啥都能做 特别在游戏行业等等的效果是特别好的 但那一块我们就完全不去介入 你觉得中国跟美国在空间智能上面的发展 有什么样不一样的路线的区别吗 我觉得美国更贴近于虚拟世界 中国更贴近于物理世界 2026年 AI的新风向 似乎已经越来越清晰了 那就是 世界模型 而在当前的世界模型所有技术分支中 很多研究者都把空间智能 Spatial Intelligence 视为了那个真正的“关键拼图” 对我来说没有空间智能是不可能去实现AGI的 而我想要去解决这个问题 那么究竟什么是空间智能呢 它跟世界模型 还有具身智能之间 究竟是什么样的关系 它的应用前景究竟在哪里 现在还存在着什么样的问题 Hello 大家好 欢迎收看《硅谷101》 我是陈茜 这个视频 我们和有着 “杭州六小龙”之一称号的群核科技 创始人黄晓煌一起来聊聊这个话题 群核科技的定位 是做空间智能服务的提供商 他们目前已经建立起了 ...

Core Viewpoint - The rise of "world models" is seen as a key to overcoming the limitations of current AI, enabling a deeper understanding of the physical world and paving the way for Artificial General Intelligence (AGI) [1][3][6]. Group 1: World Models and AGI - World models allow AI to understand the laws of the physical world, facilitating reasoning and interaction, which is essential for achieving AGI [1][3]. - The development of world models is still in its early stages, and the first company to leverage physical interaction data effectively will gain a competitive edge [1][3][6]. Group 2: Industry Trends and Investments - OpenAI's recent shift to focus on world model research indicates a strategic pivot in the industry towards understanding reality rather than generating it [3][6]. - Significant investments have been made in world model companies, with over $10 billion raised by notable firms this year alone, reflecting a growing consensus that the next battleground for AI lies in the physical world [6][7]. Group 3: Challenges and Opportunities - The current challenge for world models is the scarcity of high-quality physical world data, which limits their widespread adoption [11][13]. - Companies are exploring the integration of world models with existing AI frameworks to enhance capabilities, particularly in complex environments [12][13]. Group 4: Future Outlook - The year 2026 is anticipated to be pivotal for world models, potentially establishing a foundation for AGI and physical AI [11][12]. - The evolution of world models is expected to complement existing models, with a focus on physical intuition and decision-making, while other models handle semantic understanding [13].

Group 1 - Manycore Tech Inc. has successfully passed the Hong Kong Stock Exchange listing hearing, moving into the final stage of its IPO process, with Morgan Stanley and CCB International as joint sponsors [1] - If the IPO is completed, Manycore Tech will become the first company from the "Hangzhou Six Little Dragons" to go public and is expected to be the "first global stock in spatial intelligence" [1] - The company is projected to achieve a revenue of 820 million yuan in 2025, with a gross margin of 82.2%, marking a significant turnaround from losses to a net profit of 57.1 million yuan [4] Group 2 - Manycore Tech has established a strong position in the spatial intelligence sector, leveraging its extensive 3D data and foundational spatial capabilities, which align with the emerging "world model" concept in AI [2] - The company owns the largest spatial design platform globally, "CoolJiaLe," and the overseas version "Coohom," along with the new spatial intelligence solution "SpatialVerse," catering to both real and virtual environments [2] - Manycore Tech's business model has evolved into a "spatial editing tools - spatial data - spatial large model" system, transitioning from a 3D design software provider to a spatial intelligence service provider [2] Group 3 - The company has formed strategic partnerships with industry leaders such as Zhiyuan Robotics, Galaxy General, and PICO, supporting applications of spatial intelligence across various sectors including interior design, e-commerce, and XR [3] - Manycore Tech's AI-related vertical solutions have seen rapid growth, with the launch of the 3D AI design tool "CoolJiaLe E-commerce Studio" expected to increase revenue by 123% in 2025 [4] - The company plans to use the net proceeds from the IPO primarily for international expansion, enhancing existing products, and investing in core technologies and infrastructure [5]

Core Viewpoint - The article discusses the limitations of current AI models, particularly in the context of generative video technology, and proposes that the missing piece in AI development is a world model that can learn abstract representations and predict outcomes, with JEPA (Joint Embedding Predictive Architecture) being a potential solution [4][7][12]. Summary by Sections AI's Missing Component - Current AI lacks a significant component, which is a world model capable of learning abstract representations and supporting planning [8][9]. - The evolution of AI has seen two major revolutions: deep learning and large language models (LLMs), with the latter focusing on next-token prediction [9][10]. Limitations of Generative Models - The limitations of LLMs stem from their reliance on next-token prediction, which is not suitable for the unpredictable nature of the real world [7][14]. - Predicting every detail in real-world data, such as video, is fundamentally flawed; instead, the focus should be on learning abstract representations that can support predictions [12][13]. JEPA as a Solution - JEPA aims to find representations that retain input information while being predictive, contrasting with traditional methods that attempt to reconstruct all details [12][13]. - The approach emphasizes that effective modeling requires ignoring many details to retain sufficient structure for predictions [12][13]. Experience and Evidence - Historical experiments indicate that joint embedding methods consistently outperform reconstruction methods in learning representations [16][17]. - The article highlights that the best way to learn representations for natural signals is not through reconstruction but through methods that do not attempt to reconstruct every detail [17]. Transition to AMI Labs - The shift in focus at Meta towards short-term goals and LLMs led to the decision to leave and pursue JEPA at AMI Labs, where the application of these ideas can be explored in areas like industrial process control and robotics [21][22]. Future Directions - The potential for a hierarchical JEPA model is discussed, which would allow for predictions across different time and spatial scales, drawing parallels with concepts in physics [23]. - The article suggests that understanding complex systems, such as economic models, may benefit from a data-driven approach similar to JEPA, focusing on higher-level abstractions [26][27].

Industry Investment Rating - The industry investment rating is "Outperform the Market" and is maintained [1] Core Insights - The report highlights that domestic companies are accelerating their entry into the world model sector, which is expected to speed up the industrialization process. Previously, research in this area was mainly dominated by overseas institutions like Google and World Labs. Domestic firms, including ByteDance, Ant Group, Tencent, Huawei, and Alibaba, are now making significant strides in this field [5][6] - The application of AI technology in game development has reached an overall application rate of 86.36% as of 2025, with a focus on art design, automated testing, and sound generation. The penetration rate for core game asset generation is approximately 36.8%. The world model is anticipated to enhance AI's role in complex asset generation and scene construction, transitioning from a "point efficiency tool" to a "system-level productivity platform" [6] - The report suggests that game interaction is evolving from "character-level intelligence" to "scene-level generation + system-level interaction," which will transform content production from "pre-fabricated content supply" to "real-time generated supply" [6] Summary by Sections Industry Overview - The closing index is 784.68, with a 52-week high of 1021.75 and a low of 591.71 [1] Investment Highlights - Companies to watch include those with dual capabilities in world model development and scene application, such as Kunlun Wanwei, and large 3D game production companies like Perfect World and Giant Network [7]

Core Insights - The article highlights the significant achievement of GigaWorld-1, developed by the Chinese company 极佳视界, which has surpassed major competitors like Google and NVIDIA to become the top-ranked embodied world model globally [1][10]. Group 1: GigaWorld-1 Performance - GigaWorld-1 is the only embodied world model to score over 60 on the WorldArena leaderboard, achieving a score of 62.34% [2]. - It leads in several key dimensions, including Visual Quality (63.04), Motion Quality (39.16), Content Consistency (65.17), Physics Adherence (97.02), and 3D Accuracy (57.28) [2][6]. - The model shows a 16% improvement in Physics Adherence compared to the second-ranked model, Ctrl-World [6]. Group 2: Technical Advancements - GigaWorld-1 is designed as an Action-Conditioned World Model (AC-WM), integrating explicit action modeling and a differentiable physics engine for accurate physical interactions [11][14]. - The model has been trained using high-quality real robot operation video data, enhancing its generalization capabilities in open scenarios [14]. Group 3: Company Background and Funding - 极佳视界 is recognized as the first company in China to focus on world models, combining technology development with substantial financing [20]. - The company recently completed a nearly 1 billion yuan Pre-B round of financing, attracting investments from top firms in the semiconductor and automotive industries [21][22]. - Previous investments include a strategic investment from Huawei's Hubble Investment, indicating strong interest in the world model sector [24][25]. Group 4: Product Ecosystem - The company offers a product matrix that includes GigaWorld, a world model platform, GigaBrain, an embodied foundational model, and Maker, a general-purpose embodied ontology [28]. - GigaWorld serves as a digital sandbox for simulating physical world operations and generating high-fidelity synthetic data, achieving efficiency improvements of 10-100 times compared to traditional simulators [30][32]. Group 5: Team and Expertise - The core team of 极佳视界 includes experts with extensive experience in physical AI, robotics, and world models, led by founder and CEO Huang Guan, who has a strong background in automation and AI competitions [41][46]. - The team has a proven track record of achieving global recognition in AI competitions and has published numerous influential papers in the field [44][47].

Core Insights - The article emphasizes that the key limitation in advancing world models is not merely their representational capabilities but rather the dynamics modeling aspect, which is crucial for stable future predictions [2][9][78] - Recent research by Haoxin Lin and his team focuses on improving dynamics modeling to enable world models to support long-term predictions effectively [3][39][75] Dynamics Modeling - World models consist of two main components: the V model for state representation and the M model for dynamics modeling, which predicts future states based on actions [4][5] - Traditional dynamics models often rely on single-step predictions, which can lead to cumulative errors in long-term forecasts due to bootstrapping [13][14][78] - The Any-step Dynamics Model (ADM) proposes a method for direct multi-step predictions, reducing error propagation and enhancing stability in long-term rollouts [19][20][24] Research Contributions - The ADM model demonstrates that future states can be predicted without relying solely on the previous step's results, allowing for more robust long-term forecasting [19][39] - The ADM-v2 model further advances this by achieving full-horizon rollouts of over a thousand steps, marking a significant leap in the capabilities of world models [41][42][60] - The structure of ADM-v2 separates state initialization from action-driven evolution, improving flexibility and stability in multi-step predictions [43][44] Performance and Evaluation - Experimental results indicate that the ADM models lead to improvements in future prediction quality and overall policy performance, particularly in both online and offline settings [32][33][64] - The ADM-v2 model has shown superior performance in offline policy evaluation tasks, outperforming various existing methods [59][60][66] Future Implications - The advancements in dynamics modeling are crucial for the evolution of world models from short-term prediction tools to comprehensive simulation systems capable of long-term planning and evaluation [75][78] - The ongoing research highlights the importance of addressing dynamics modeling as a foundational aspect for the future development of larger and more capable world models [78][79]

Core Viewpoint - The article discusses the innovative "world model" technology introduced by Leap Motor, which aims to democratize advanced intelligent driving capabilities previously reserved for luxury vehicles, making them accessible in entry-level models priced under 100,000 yuan [1][6]. Group 1: World Model Technology - The world model is a new paradigm that connects AI models directly with the real physical world, showcasing potential for AGI [4]. - Leap Motor's world model technology allows for a more intuitive and effective driving experience, transitioning from merely functional to highly usable [3][5]. - The technology is designed to handle complex driving scenarios, such as narrow roads with mixed traffic, demonstrating superior decision-making capabilities compared to existing mass-produced systems [11][19]. Group 2: Performance and User Experience - Real-world testing of the Leap Motor world model in complex urban environments shows its ability to navigate challenging situations efficiently and safely, mimicking human driving behavior [9][14]. - The system's performance includes smooth lane changes and interactions with pedestrians, enhancing user confidence and comfort [16][19]. - Leap Motor's approach emphasizes a user experience that feels natural and familiar, avoiding abrupt or dangerous maneuvers typical of other systems [20][24]. Group 3: Technical Architecture - The world model architecture consists of three main components: a visual encoder, a dynamic core for state prediction, and a renderer for visual output [35]. - This architecture allows for real-time environmental recognition and decision-making, distinguishing it from traditional rule-based systems [40][41]. - Leap Motor's world model leverages a significant computational infrastructure, with capabilities to automatically identify and resolve data issues, enhancing overall system efficiency [42][46]. Group 4: Market Position and Strategy - Leap Motor has positioned itself as a leader in the intelligent vehicle market, achieving significant sales growth and aiming to redefine user experiences in smart driving [69]. - The company has successfully integrated advanced technologies into more affordable models, setting new benchmarks in the industry [63][68]. - Leap Motor's strategy focuses on making cutting-edge technology accessible, contrasting with competitors who may prioritize high-end models [61][66].

Core Viewpoint - The article emphasizes the transition from a focus on stronger models to the establishment of a comprehensive AI platform, highlighting the strategic direction of TianGong AI in the multi-modal AI landscape [1][2][8]. Group 1: Transition to AI Platform Economy - TianGong AI's CEO, Zhou Yahui, announced that the first leap from mobile internet to large model tools has been completed, and a second leap towards an AI platform economy is now underway [3][4]. - In this new era, models serve as engines, platforms act as factories, and creators are the bosses, collectively enhancing creativity [5][8]. - The choice to fully invest in AGI and AIGC indicates a long-term vision to build a complete AI platform rather than just stronger models [8][10]. Group 2: Model Releases and Ecosystem Development - TianGong AI launched three models at the recent conference, each positioned in the global first tier of their respective fields, contributing to a cohesive world model [12][13]. - The models include Matrix-Game 3.0 for gaming, SkyReels V4 for video, and Mureka V9 for music, each addressing specific industry challenges and enhancing interactivity and creativity [19][36][62]. - The integration of these models aims to create a comprehensive interactive world model, with each model reinforcing the others [14][77]. Group 3: Technical Innovations in Models - Matrix-Game 3.0 introduces long-term memory capabilities, allowing for consistent content generation even after extended interactions, achieving minute-level memory retention [27][32]. - SkyReels V4 addresses common issues in AI video generation, such as synchronization and narrative coherence, by employing advanced architectures and reinforcement learning techniques [43][51]. - Mureka V9 enhances music generation by optimizing for emotional expression and creative logic, marking a significant advancement in AI music capabilities [65][70]. Group 4: Strategic Framework and Future Directions - TianGong AI's "3+1 strategy" includes three major models and the Skywork Super Agents, forming a comprehensive ecosystem for content creation and distribution [82][84]. - The company aims to build a unified system that integrates multi-modal capabilities, facilitating scalable content production and interaction [105][106]. - The shift from single-modal capabilities to a platform approach reflects a broader industry trend where AI is becoming integral to production processes rather than merely serving as a tool [100][102].

Core Viewpoint - The autonomous driving industry is still in its early stages, and the notion of a "final conclusion" is premature. Continuous technological advancements are expected, and the competition will persist beyond 2026, contrary to some industry predictions [3][30][31]. Group 1: Company Developments - WeRide has emerged as a significant player in the autonomous driving sector, achieving a 90% revenue growth to 690 million yuan in the past year, driven by its Robotaxi business and the introduction of an end-to-end model [4][5]. - The company has developed a simulation platform called WeRide GENESIS, which generates high-quality training data for autonomous driving, addressing the industry's data bottleneck [5][9]. - WeRide GENESIS is designed to create realistic driving scenarios, enhancing the training of autonomous models by simulating complex environments and interactions [8][9]. Group 2: Technological Innovations - The end-to-end model developed by WeRide is seen as a breakthrough, allowing for direct decision-making from sensor data without predefined rules, which contrasts with traditional methods [6][7]. - The platform's ability to generate diverse and realistic scenarios is crucial for training autonomous systems, as it overcomes the limitations of real-world data collection [8][9]. - The company emphasizes that the quality of data is critical for the success of the end-to-end model, and WeRide GENESIS provides a solution to generate high-quality synthetic data [27][10]. Group 3: Industry Perspectives - The industry is characterized by ongoing debates about the future of autonomous driving technology, with some leaders expressing skepticism about the emergence of new paradigms [3][31]. - WeRide's leadership believes that the competition will remain fierce, and the notion of a few companies dominating the market by 2026 is overly optimistic [30][31]. - The company aims to balance its focus on both high-level autonomous driving (L4) and advanced driver-assistance systems (ADAS), indicating a strategic approach to market demands [35][36].