该工作由上海人工智能实验室钟志航团队联合四川大学、东京大学、上海交通大学、西北工业大学共同完成。 在李飞飞团队 WorldLabs 推出 Marble、引爆「世界模型（World Model）」热潮之后，一个现实问题逐渐浮出水面： 世界模型的可视化与交互，依然严 重受限于底层 Web 端渲染能力。 Marble 所依赖的 基于 WebGL 的 3D Gaussian Splatting (3DGS) 渲染器 SparkJS，让世界模型首次在浏览器中「跑起来」，但也暴露出明显瓶颈：大场 景以及复杂场景下，CPU 排序成为性能天花板，动态场景与生成模型难以接入。 相比 Genie3 等视频生成范式的世界模型，其对算力的依赖极为庞大，距离在 Web 端实现高质量、实时运行仍有不小差距。反观神经渲染路线，尤其是 3D Gaussian Splatting ，凭借其高效性，已经成为构建世界模型的重要表示形式。 3DGS 让高质量、实时的 3D 世界成为可能，但在实际落地中，仍存在明显断层： 近日，开源项目 Visionary 给出了一个截然不同的答案： 基于 WebGPU 与 ONNX，在浏览器中实现真正的动态 3DG ...

导语 投资人正在急切地寻找下一个百亿级估值的引爆点。如果说过去两年，有一家公司一个刺激大家神经，那无疑是Mercor，它重新定义了LLM时代的数据基 础设施。 硅谷过去十年的一个共性规律 ： 每一轮AI技术范式迁移，从计算机视觉（CV）到大型语言模型（LLM），最终都会在"数据层"沉淀出一次基础设施级的 巨大机会。 这一轮由大语言模型驱动的技术革命，其核心竞争要素已经清晰：模型层决定能力上限，而数据层是驱动突破的核心燃料。 除了模型层存在大机会，数据 层同样孕育着下一个基础设施级的平台机会 。 关键在于，谁能规模化地解决"高质量数据从哪里来"这一根本问题，谁就能掌握通向未来的钥匙。上一轮AI 范式的赢家，正是专注于此的Scale AI，它已然证明了数据基础设施的巨大价值。 这家成立不到三年、团队平均年龄仅22岁的初创公司，在新一轮融资中，估值一举突破100亿美元，成为AI基础设施赛道最年轻的百亿美金独角兽。这个数 字，是它转型前估值的五倍。 硅谷看重Mercor的，远不止一个高效的人才市场，其真正的价值，在于它大胆尝试重构了 AI研发的生产关系 。Mercor精准地找到了其利基市场，并将这个 时代最核心、最昂 ...

Core Insights - The integration of intelligent driving technology is reshaping lifestyles at an unprecedented pace, driven by advancements in artificial intelligence and a unique market environment in China [1][3] - The Chinese intelligent driving industry is transitioning from a phase of rapid growth to one of high-quality development, with regulatory frameworks being strengthened alongside pilot programs for higher-level autonomous driving [3][4] - The rapid adoption of electric vehicles is providing an optimal platform for intelligent driving technologies, creating a virtuous cycle between electrification and intelligence [4][6] Industry Trends - The emergence of cognitive intelligence technologies is transforming intelligent driving from a rule-based tool to a cognitive-driven entity, with new architectures like end-to-end and VLA opening new possibilities for high-level autonomous driving [3][5] - The intelligent driving sector is witnessing a clear focus on L4-level scenario-based applications, with significant investments directed towards areas like unmanned delivery and logistics [6][7] - Key components of the supply chain, such as sensor manufacturers and chip companies, are receiving substantial funding, highlighting their foundational role in the development of autonomous driving [7] Regulatory Environment - The regulatory landscape is evolving, with policies being introduced to facilitate the testing and commercialization of L3-level and above autonomous driving technologies in multiple cities [3][4] - The dual approach of relaxing pilot programs while simultaneously enhancing regulatory frameworks is creating clearer competitive advantages for companies with core competencies [3][4] Investment Landscape - Investment activities in the intelligent driving sector are increasingly concentrated in later-stage financing, indicating a shift from technology validation to large-scale commercial applications [7] - Traditional automotive companies are actively participating in investments to address technological gaps, while collaborations within the supply chain are emerging to build ecological advantages [7] Future Outlook - The competition in intelligent driving is entering a new phase where success will depend on the ability to integrate technology, compliance, and commercialization effectively [9] - The industry is at a historical turning point, with the potential for new industry giants to emerge from the convergence of technology, policy, and market dynamics [8][9]

Core Insights - Google has launched the Nano Banana Pro image generation tool, leveraging the capabilities of Gemini 3 Pro to set a new standard in the AI image generation industry [2][3] - Nano Banana Pro addresses long-standing challenges in the field, including consistency, understanding of the physical world, text rendering, deepfakes, and cost [4][5][8] Group 1: Key Features of Nano Banana Pro - The tool excels in detail control, semantic understanding, and cross-ecosystem collaboration, significantly improving the quality of generated images [3] - It can maintain high consistency and control, processing up to 14 reference images and accurately preserving facial features and clothing details across multiple images [9] - Nano Banana Pro integrates real-time information retrieval from Google's knowledge base, enhancing the accuracy of generated content [11] Group 2: Addressing Industry Challenges - The tool effectively resolves over 80% of the industry's major issues, including consistency and controllability, which have historically plagued AI image generation models [9] - It offers advanced text rendering capabilities, allowing for accurate integration of text into images, overcoming previous limitations [13] - To combat deepfake risks, Nano Banana Pro incorporates SynthID digital watermarks, ensuring traceability even after image modifications [15] Group 3: Market Position and Pricing - Nano Banana Pro is positioned as a premium product, with higher costs for generating images compared to standard versions, catering to professional commercial use [18] - The pricing strategy differentiates user groups, with the Pro version designed for low-tolerance error scenarios in professional settings [18] - Despite its advanced features, the tool still faces challenges related to high operational costs, which may limit accessibility for individual developers and researchers [8][18] Group 4: Integration and Ecosystem - The tool is deeply integrated with Google's ecosystem, enabling seamless collaboration with platforms like Adobe and Figma, thus expanding its application in creative fields [18] - The rapid increase in monthly active users of Gemini, from 450 million to 650 million, highlights the tool's impact on user engagement [18]

Core Insights - Yann LeCun, a Turing Award winner and a key figure in deep learning, announced his departure from Meta to start a new company focused on Advanced Machine Intelligence (AMI) research, aiming to revolutionize AI by creating systems that understand the physical world, possess persistent memory, reason, and plan complex actions [2][4][11]. Departure Reasons & Timeline - LeCun's departure from Meta was confirmed after rumors circulated, with the initial report coming from the Financial Times on November 11, indicating his plans to start a new venture [10][11]. - Following the announcement, Meta's market value dropped approximately 1.5% in pre-market trading, equating to a loss of about $44.97 billion (approximately 320.03 billion RMB) [11]. - The decision to leave was influenced by long-standing conflicts over AI development strategies within Meta, particularly as the focus shifted towards generative AI (GenAI) products, sidelining LeCun's foundational research efforts [11][12]. Research Philosophy & Future Vision - LeCun emphasized the importance of long-term foundational research, which he felt was being undermined by Meta's shift towards rapid product development under the leadership of younger executives like Alexandr Wang [12][13]. - He expressed skepticism towards large language models (LLMs), viewing them as nearing the end of their innovative potential and advocating for a focus on world models and self-supervised learning to achieve true artificial general intelligence (AGI) [14][15]. - LeCun's vision for AMI includes four key capabilities: understanding the physical world, possessing persistent memory, true reasoning ability, and the capacity to plan actions rather than merely predicting sequences [16][15]. Industry Context & Future Outlook - The article suggests a growing recognition in the industry that larger models are not always better, with a potential shift towards smaller, more specialized models that can effectively address specific tasks [18]. - Delangue, co-founder of Hugging Face, echoed LeCun's sentiments, indicating that the current focus on massive models may lead to a bubble, while the true potential of AI remains largely untapped [18][15]. - Meta acknowledged LeCun's contributions over the past 12 years and expressed a desire to continue benefiting from his research through a partnership with his new company [22].

Core Insights - Yann LeCun, a prominent figure in AI and Turing Award winner, announced his departure from Meta to establish a startup focused on advanced machine intelligence research [2][3] - Meta confirmed LeCun's departure and expressed gratitude for his contributions over the past 12 years, while also indicating a partnership with his new venture [2] Group 1: Departure and New Venture - LeCun plans to create a startup aimed at developing systems that can understand the physical world, possess long-term memory, reason, and plan complex actions [2] - Prior to the official announcement, LeCun's startup project had already attracted interest from several major companies [2] - Meta's spokesperson acknowledged LeCun's significant contributions to AI and expressed anticipation for future collaborations [2] Group 2: Disagreements and Internal Changes - LeCun had fundamental disagreements with Mark Zuckerberg regarding AI strategy and technology, particularly concerning the limitations of large language models (LLMs) [3] - He advocated for a "Joint Embedding Predictive Architecture" (JEPA) to build systems with long-term memory and reasoning capabilities, contrasting with Meta's focus on LLMs [3] - The acquisition of Scale AI by Meta for $14.3 billion and the appointment of new AI leadership diminished LeCun's control over key projects [3][5] Group 3: Impact on Meta and AI Landscape - The restructuring at Meta significantly affected the FAIR lab, leading to layoffs of core team members, including experts in reinforcement learning [4] - LeCun's departure may signify the end of the FAIR era at Meta and could resolve ongoing internal conflicts related to technology strategy [6] - LeCun's new company is expected to continue the "open-source ecosystem" approach, potentially competing directly with Meta's current closed-source strategy [6]

Core Insights - The article discusses the limitations of Visual-Language Models (VLMs) in complex visual tasks, highlighting their tendency to act impulsively rather than thoughtfully due to their perception of the world being limited and noisy [2][6]. - The VAGEN framework aims to enhance VLMs by teaching them to construct an internal world model before taking actions, thereby promoting a more structured thinking process [3][12]. Group 1: VAGEN Framework - VAGEN enforces a structured "thinking template" for VLMs, which includes two core steps: State Estimation (observing the current state) and Transition Modeling (predicting future outcomes) [7][11]. - The framework utilizes reinforcement learning (RL) to reward this structured thinking process, demonstrating that the "World Modeling" strategy significantly outperforms both "No Think" and "Free Think" approaches [12][32]. Group 2: Internal Monologue and Reward Mechanism - The research explores the best format for the internal monologue of the agent, finding that the optimal representation depends on the nature of the task [13][14]. - VAGEN introduces two key components in its reward mechanism: World Modeling Reward, which provides immediate feedback after each thought process, and Bi-Level GAE for efficient reward distribution [18][20]. Group 3: Performance Results - The VAGEN-Full model, based on a 3B VLM, achieved an impressive overall score of 0.82 across five diverse tasks, outperforming various other models including GPT-5 [27][30]. - The results indicate that VAGEN-Full not only surpasses untrained models but also exceeds the performance of several proprietary models, showcasing its effectiveness in enhancing VLM capabilities [30][32].

Core Insights - The exploration towards Artificial General Intelligence (AGI) highlights embodied intelligence as a key direction, focusing on the interaction and adaptation of intelligent agents within physical environments [1][3] - The development of embodied intelligence technology has evolved through various stages, from low-level perception to high-level task understanding and generalization [6][14] Industry Analysis - In the past two years, numerous star teams in the field of embodied intelligence have emerged, establishing valuable companies such as Xinghaitu, Galaxy General, and Zhujidongli, transitioning from laboratories to commercial and industrial applications [3] - Major domestic companies like Huawei, JD, Tencent, Ant Group, and Xiaomi are actively investing and collaborating to build a comprehensive ecosystem for embodied intelligence, while international players like Tesla and investment firms support advancements in autonomous driving and warehouse robotics [5] Technological Evolution - The evolution of embodied intelligence technology has progressed through several phases: - The first phase focused on grasp pose detection, which lacked the ability to model task context and action sequences [6] - The second phase introduced behavior cloning, allowing robots to learn from expert demonstrations but revealing weaknesses in generalization and performance in multi-target scenarios [6] - The third phase, emerging in 2023, utilized Diffusion Policy methods to enhance stability and generalization by modeling action trajectories [6][7] - The fourth phase, starting in 2025, explores the integration of VLA models with reinforcement learning and tactile sensing to overcome current limitations [9][11][12] Educational Initiatives - The demand for engineering and system capabilities in embodied intelligence is increasing as the industry shifts from research to deployment, necessitating higher engineering skills [17] - A comprehensive curriculum has been developed to cover various aspects of embodied intelligence, including practical applications and advanced topics, aimed at both beginners and advanced learners [14][20]

Core Viewpoint - The exploration of Artificial General Intelligence (AGI) is increasingly focusing on embodied intelligence, which emphasizes the interaction and adaptation of intelligent agents within physical environments, enabling them to perceive, understand tasks, execute actions, and learn from feedback [1]. Industry Analysis - In the past two years, numerous star teams in the field of embodied intelligence have emerged, establishing valuable companies such as Xinghaitu, Galaxy General, and Zhujidongli, which are advancing the technology of embodied intelligence [3]. - Major domestic companies like Huawei, JD, Tencent, Ant Group, and Xiaomi are actively investing and collaborating to build a robust ecosystem for embodied intelligence, while international firms like Tesla and investment institutions are supporting companies like Wayve and Apptronik in the development of autonomous driving and warehouse robots [5]. Technological Evolution - The development of embodied intelligence has progressed through several stages: - The first stage focused on grasp pose detection, which struggled with complex tasks due to a lack of context modeling [6]. - The second stage involved behavior cloning, allowing robots to learn from expert demonstrations but revealing weaknesses in generalization and performance in multi-target scenarios [6]. - The third stage introduced Diffusion Policy methods, enhancing stability and generalization by modeling action sequences, followed by the Vision-Language-Action (VLA) model phase, which integrates visual perception, language understanding, and action generation [7][8]. - The fourth stage, starting in 2025, aims to integrate VLA models with reinforcement learning, world models, and tactile sensing to overcome current limitations [8]. Product and Market Development - The evolution of embodied intelligence technologies has led to the emergence of various products, including humanoid robots, robotic arms, and quadrupedal robots, serving industries such as manufacturing, home services, dining, and medical rehabilitation [9]. - The demand for engineering and system capabilities is increasing as the industry shifts from research to deployment, necessitating higher engineering skills for training and simulating strategies on platforms like Mujoco, IsaacGym, and Pybullet [23]. Educational Initiatives - A comprehensive curriculum has been developed to cover the entire technology route of embodied "brain + cerebellum," including practical applications and real-world projects, aimed at both beginners and advanced learners [10][20].

Core Viewpoint - The article discusses the significant advancements brought by Google's DeepMind with the release of Genie 3, which showcases a new path towards Artificial General Intelligence (AGI) through the concept of "World Models" [4][5][6]. Group 1: Introduction of Genie 3 - On August 5, Google DeepMind launched Genie 3, a model capable of generating interactive 3D virtual environments based on user prompts, demonstrating enhanced real-time interaction capabilities compared to previous AI models [5]. - Genie 3 features a "Promptable World Events" function, allowing users to dynamically alter the generated environment through text commands, showcasing its advanced interactivity [5]. Group 2: Concept of World Models - World Models are inspired by the human brain's ability to create and utilize an "inner world" to simulate future scenarios, which is crucial for decision-making and action [8][9]. - The development of World Models has evolved from early attempts to mimic human cognitive functions to more sophisticated models that can predict and simulate real-world dynamics [10][11]. Group 3: Technical Implementation of World Models - The implementation of World Models involves several key stages: Representation Learning, Dynamic Modelling, Control and Planning, and Result Output, each contributing to the AI's ability to understand and interact with the world [15][16][17][18]. - Representation Learning allows AI to compress external data into an internal language, while Dynamic Modelling enables the simulation of future scenarios based on actions taken [15][16]. Group 4: Applications of World Models - World Models can significantly enhance "embodied intelligence," allowing AI agents to learn through simulated experiences in a safe environment, reducing costs and risks associated with real-world trials [20][21]. - In the realm of digital twins, World Models can create proactive simulations that predict changes and optimize processes in real-time, enhancing automation and decision-making [21][22]. - The education and research sectors can benefit from World Models by creating virtual laboratories for precise predictions and interactive learning environments [22]. Group 5: Potential and Challenges of World Models - While World Models present vast potential for various applications, they also raise ethical and governance concerns, such as the blurring of lines between reality and virtuality, and the potential for behavioral manipulation [24][25][26]. - The debate surrounding World Models as a pathway to AGI highlights differing opinions within the AI community, with some experts advocating for their necessity while others question their effectiveness compared to model-free approaches [28][29][30].