Core Insights - The article presents a counter-consensus viewpoint from Fei-Fei Li, emphasizing that large language models alone cannot lead to Artificial General Intelligence (AGI), and that spatial intelligence is a more foundational path [4][5][6]. Group 1: AGI Route Debate - Language is not the entirety of intelligence and is not its foundation; spatial intelligence, which has evolved over 500 million years, is crucial for AI development [5][6]. - If AI only possesses language capabilities, it will remain confined to the digital realm; true AGI requires understanding and interaction with the three-dimensional physical world [6]. Group 2: Redefining World Models - The newly introduced spatial intelligence model, Marble, can process multimodal inputs and create a navigable, interactive 3D world with physical consistency, differing from traditional video models [7][8]. - Marble has applications in various fields, including game development, visual effects, and even therapeutic settings for conditions like OCD [8]. Group 3: Scaling Law and Data Challenges - The slower development of physical world AI compared to language models is attributed to the noise in physical data and the difficulty in large-scale data acquisition [8][9]. - World Labs employs a hybrid data strategy, combining existing internet data with synthetic and real-world data to overcome these challenges [8][9]. Group 4: General Robotics vs. Autonomous Driving - General robotics is viewed as a higher-dimensional challenge compared to autonomous driving, which operates primarily in a 2D space [10][11]. - The core task of general robots involves interaction in 3D space, which presents significant technical challenges [10][11]. Group 5: AI as a Fundamental Infrastructure - AI is likened to electricity, with its success not measured by model size but by its ability to empower civilization and improve individual lives [11][12]. - The goal of World Labs is to integrate spatial intelligence into various industries, aiming for significant advancements by 2026 [12].

该原型率先向美国Google AI Ultra订阅用户开放。 如果人工智能领域的进步可以看作一部交响乐，那么过去几年，乐章的主题无疑是"生成"——生成文 字、图像、声音乃至视频。然而，在2026年初，一段崭新的旋律被奏响：它不仅生成，更能构建。 北京时间1月30日凌晨，谷歌DeepMind向外部开放了Project Genie，它被认为是目前最先进的世界模型 之一，可以算是世界模型Genie3的实验性研究原型，也是这套世界模型第一次以可交互形态对公众开 放。 "Genie"这个单词源于阿拉伯语 jinni（精灵），后经法语变形成 génie后成为一个英语词汇，最常见的含 义是指阿拉伯和伊斯兰神话传说中，一个能实现召唤者愿望的"精灵"或"神怪"。谷歌DeepMind将其世 界模型项目命名为"Project Genie"（精灵计划），正是在阐释该神话的内涵：这个AI模型能将你用文字 描述的任何场景（召唤者的愿望），瞬间生成一个可以进入并交互的虚拟世界。 当AI不仅能够描绘梦境，更能让人走进梦境并与之互动时，我们所讨论的"虚拟"与"现实"的边界，或许 已到了需要被重新思考的时刻。 目前，该原型率先向年满18岁的美国 ...

2026 年 01 月 27 日 Cowork 获得永久记忆，AI 协作迎来范式革新 推荐(维持) 投资要点 分析师：任春阳 S1050521110006 rency@cfsc.com.cn 行业相对表现 | 表现 | 1M | 3M | 12M | | --- | --- | --- | --- | | 计算机(申万) | 11.5 | 4.9 | 33.2 | | 沪深 300 | 1.0 | -0.2 | 23.3 | 市场表现 ▌AI 应用：Gemini 周访问量环比+3.43%，Cowork 获 得永久记忆 -20 -10 0 10 20 30 40 (%) 计算机 沪深300 资料来源：Wind，华鑫证券研究 相关研究 1、《计算机行业周报：DeepSeek 开 源含 Engram 模块，千问助理重塑人 机交互》2026-01-19 2 、 《 计 算 机 行 业 周 报 ： 英 伟 达 Rubin 架构重塑算力未来，MiroMind 发布 MiroThinker1.5》2026-01-13 3 、 《 计 算 机 行 业 周 报 ： 小 红 书 Video-Thinker 打 破 工 具 依 赖 ...

Core Insights - World Labs, founded by Fei-Fei Li, is seeking to raise up to $500 million at a valuation of approximately $5 billion, significantly increasing its valuation from $1 billion in just over a year [2][3]. Funding and Valuation - World Labs has previously raised a total of $230 million, achieving a valuation of $1 billion after its initial funding round in April 2024, which started at around $200 million [3][6]. - The first round of investors included Andreessen Horowitz and Radical Ventures, with subsequent funding rounds attracting major players like NVIDIA and Temasek [6][10]. Product Development - The company launched its first 3D world generation model, Marble, in November of the previous year, which allows users to create explorable 3D worlds based on text or image prompts [7][9]. - Marble utilizes 3D Gaussian Splatting technology to efficiently render scenes while also providing collision meshes for physical simulations [9]. Strategic Vision - Fei-Fei Li emphasizes that world models are crucial for achieving spatial intelligence and are considered the next core focus of AI after large language models [10][12]. - The world model is expected to have broad applications across various fields, including AIGC, robotics, and real-world task execution [12][13]. Competitive Landscape - Another venture, AMI Labs, founded by Yann LeCun, is also attracting investment, with a potential valuation of $3.5 billion, focusing on implicit world models [15][18]. - The landscape of world models is categorized into three layers, with LeCun's approach positioned at the highest abstract level, contrasting with Li's explicit and generative model [18].

Core Viewpoint - World Labs, founded by Fei-Fei Li, is seeking to raise up to $500 million at a valuation of approximately $5 billion, marking a significant increase from its previous valuation of $1 billion in 2024, indicating a 5x revaluation in just over a year [2][4]. Financing and Valuation - If the financing is successful, World Labs' valuation will jump from $1 billion to $5 billion, reflecting a rapid increase in investor confidence in its "world model" approach [2][4]. - World Labs has previously raised a total of $230 million, with initial funding rounds led by notable investors such as Andreessen Horowitz and Radical Ventures, and later rounds involving firms like NVIDIA and Temasek [5][6]. Product Development - World Labs is developing AI systems capable of navigation and decision-making in three-dimensional environments, focusing on creating "large world models" that understand the structure and evolution of the physical world [8][9]. - The company launched its first 3D world generation model, Marble, which can create explorable 3D environments based on text or image prompts, utilizing advanced techniques like 3D Gaussian Splatting for efficient rendering [10][14]. Strategic Importance - Fei-Fei Li emphasizes that world models are crucial for achieving spatial intelligence and are considered the next core focus for AI in the coming decade, following large language models [16][18]. - The world model is seen as a foundational capability that can influence multiple application areas, providing predictive representations of environments essential for effective decision-making and control [18][22]. Competitive Landscape - Another significant player in the world model space is AMI Labs, founded by Yann LeCun, which is pursuing a different approach focused on implicit world models. This indicates a broader investment interest in various technological paths within the world model domain [20][24]. - The world model landscape can be categorized into three layers, with LeCun's JEPA positioned at the highest abstract level, highlighting the diverse strategies being adopted by different companies in this field [24][27].

IT之家 1 月 23 日消息，据彭博社今日报道，"AI 教母"李飞飞正在就其创办的初创公司 World Labs 进行 新一轮融资洽谈，目标估值约为 50 亿美元（现汇率约合 349 亿元人民币）。 World Labs 于 2024 年结束隐身并完成 2.3 亿美元（IT之家注：现汇率约合 16.05 亿元人民币）融资，当 时估值约为 10 亿美元（现汇率约合 69.8 亿元人民币）。 World Labs 现有投资方包括 Andreessen Horowitz、NEA 以及 Radical Ventures（李飞飞本人在 Radical Ventures 担任科学合伙人），而英伟达旗下的风险投资部门也已参与投资。 随着投资者开始寻求超越大语言模型（LLM）的下一代 AI 技术突破，世界模型等相对尚处早期阶段的 方向正受到更多关注。当前主流的大语言模型主要为 ChatGPT 等聊天机器人提供支持。 彭博社本周早些时候还报道称，由前 Meta 研究员杨立昆（Yann LeCun）创办的世界模型初创公司 AMI Labs 正吸引包括凯辉创新（Cathay Innovation）在内的潜在投资者，其融资轮估值可 ...

Core Viewpoint - The collaboration between World Labs, led by Fei-Fei Li, and Guanglun Intelligent, a leading synthetic data company, aims to address the long-standing issue of "scalable evaluation" in the field of embodied intelligence, marking the entry into an evaluation-driven era for this technology [1][2][3]. Group 1: Companies Involved - World Labs is founded by Fei-Fei Li, a prominent figure in AI, known for her work on ImageNet and as a former chief AI scientist at Google Cloud [4][5]. - Guanglun Intelligent is recognized as a hot company in the embodied intelligence infrastructure sector, having established a strong partnership with NVIDIA and contributing to the development of simulation systems [54][55]. Group 2: Technological Innovations - World Labs is set to launch its first product, Marble, by the end of 2025, which can generate high-fidelity 3D worlds from minimal input [8][9]. - Marble aims to provide a visualized world model, allowing users to create and export 3D environments efficiently, thus serving as a productivity tool for visual effects and game developers [15][16]. Group 3: Challenges in Evaluation - The rapid advancement of models in embodied intelligence has outpaced existing benchmarks, creating a need for new evaluation methods [20][22]. - Traditional evaluation methods are inadequate for assessing the capabilities of embodied intelligence, necessitating the use of simulation as a scalable solution [29][30]. Group 4: Strategic Collaboration - The partnership between World Labs and Guanglun Intelligent is crucial for developing a comprehensive evaluation framework that integrates environment generation and physical interaction [37][49]. - Guanglun Intelligent's role is to provide the necessary physical assets and evaluation loops, ensuring that the simulated environments can support real physical interactions [49][50]. Group 5: Future Directions - The collaboration signifies a pivotal moment in the embodied intelligence sector, as it transitions into an evaluation-driven era, with the potential to shape research directions and identify technological bottlenecks [71][72][76]. - The establishment of robust evaluation standards, such as RoboFinals, highlights the industry's shift towards scalable and credible assessment frameworks for advanced robotic models [63][64].

Core Viewpoint - The article discusses the emergence of real-time world generation models, specifically highlighting PixVerse R1 as a significant advancement in this field, allowing users to interactively influence video narratives through prompts [2][4]. Group 1: Definition and Context of World Models - The term "world model" has become broad and somewhat ambiguous, referring to systems that can predict changes in a sustainable internal state and allow for interaction and validation [4][21]. - Current world model representatives can be categorized into three main directions: Google's Genie 3, Li Feifei's Marble, and NVIDIA's Cosmos, each serving different purposes such as video generation, 3D spatial intelligence, and physical AI applications [20][19]. Group 2: PixVerse R1 and Its Features - PixVerse R1 introduces a fourth direction in world models focused on real-time video generation, allowing for continuous and interactive storytelling [22][23]. - The platform offers a demo version that requires an invitation to access, indicating a controlled rollout to manage computational demands [26][30]. Group 3: User Experience and Interaction - Users report a highly engaging experience with PixVerse R1, describing it as one of the most enjoyable products they have encountered, emphasizing the joy of real-time interaction and narrative control [31][41]. - The platform allows for customizable prompts and templates, enhancing user creativity and engagement in generating unique storylines [46][57]. Group 4: Future Implications - The article suggests that the future of entertainment may evolve into dynamic, flowing narratives rather than fixed-duration content, where creators set the stage and audiences influence the direction of the story [56][58]. - This shift could redefine how content is created and consumed, fostering a deeper connection between creators and audiences through interactive experiences [60][62].

Core Viewpoint - The next decade of AI is defined by "spatial intelligence," which emphasizes the need for AI to understand depth, distance, occlusion, and gravity to achieve true embodiment [1][10]. Group 1: Li Fei Fei's Background and Career - Li Fei Fei, known as the "AI Mother," has over 20 years of experience in AI research, with a focus on spatial intelligence as her latest guiding principle [2]. - Her autobiography, "The World I See," details her journey from a challenging childhood in the U.S. to becoming a prominent figure in AI, reflecting on her struggles and achievements [2][5]. - Li Fei Fei's career spans the evolution of AI from laboratory research to industrial application, making her autobiography a significant account of AI's development [2]. Group 2: ImageNet and AI Development - ImageNet, a large-scale visual database created by Li Fei Fei, played a crucial role in the advancement of AI, marking the beginning of the AI golden age [6][9]. - The project faced initial skepticism and challenges, but the use of Amazon's crowdsourcing service was pivotal in its success, allowing for efficient image labeling [8]. - The introduction of deep learning models like AlexNet, which utilized ImageNet, significantly improved AI's performance in image recognition tasks, reducing error rates dramatically [9]. Group 3: Spatial Intelligence and Future Directions - Li Fei Fei believes that the next breakthrough in AI will come from developing spatial intelligence, which encompasses understanding and generating three-dimensional environments [10][11]. - The current state of technology in spatial intelligence is still in its early stages, but Li Fei Fei is confident that significant advancements will occur within the next one to two years [11]. - She views spatial intelligence as a critical component in the pursuit of Artificial General Intelligence (AGI), suggesting that it is one of many keys needed to unlock this complex field [12].

Core Insights - The article emphasizes the importance of human involvement in the development and application of AI, highlighting that AI is fundamentally a "civilizational technology" that significantly impacts society and culture [38][41]. Group 1: Background and Development of AI - Fei-Fei Li, known as the "godmother of AI," discusses her journey from a typical Chinese middle-class upbringing to becoming a leading figure in AI research, emphasizing the role of her parents in shaping her curiosity and resilience [13][15][16]. - The creation of ImageNet marked a pivotal moment in AI, representing a shift towards utilizing big data in the field, which had previously been stagnant during the so-called "AI winter" [20][21]. - ImageNet was developed between 2007 and 2009, becoming the largest dataset for computer vision training and evaluation at that time, which was crucial for advancing AI capabilities [20][21]. Group 2: Key Factors for ImageNet's Success - The success of ImageNet can be attributed to the timely recognition of the potential impact of big data, as well as the formulation of the correct scientific hypotheses regarding visual recognition [29][30]. - The project utilized Amazon Mechanical Turk for crowdsourcing image labeling, which allowed for the collection of millions of high-quality images necessary for training AI models [34][35]. - The careful consideration of data quality and the implementation of rigorous testing for labelers ensured the reliability of the dataset, which was essential for the project's success [36][37]. Group 3: Current AI Landscape and Future Directions - The article highlights the current cultural and economic significance of AI, noting that AI contributed to 50% of the GDP growth in the U.S. last year, indicating its transformative potential [38][39]. - There is a concern that discussions around AI often overlook the human element, which is crucial for ensuring that technology serves humanity and maintains individual dignity [41]. - The establishment of WorldLabs aims to develop spatial intelligence in AI, which is seen as a foundational capability for enhancing human creativity and interaction with the environment [45][46].