编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位均为全职，工作地点：北京中关村。 岗位面向： 加入我们，你可以获得： 以下是岗位详情： 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 AI产业方向 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 AI财经商业方向 岗位职责： 任职要求： AI产品方向 岗位职责： 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用 ...

Core Viewpoint - The Galbot S1, a fully autonomous heavy-duty embodied intelligent robot, has been deployed in the production line of CATL, marking a significant advancement in the industrial application of embodied intelligence [1][8][56] Group 1: Product Features and Capabilities - Galbot S1 has a maximum continuous operational load capacity of 50 kilograms, breaking industry limits and addressing the heavy load requirements in industrial settings [2][9] - The robot features a pioneering fully autonomous, zero-remote operation "embodied handling model," which utilizes pure visual perception to operate in complex factory environments without relying on static paths or pre-set routes [5][17] - The design of Galbot S1 aligns with industrial standards, featuring IP54 protection against dust and water, and a maximum working height of 2.3 meters, making it suitable for various industrial environments [15] Group 2: Industry Impact and Significance - The deployment of Galbot S1 in leading manufacturing companies like CATL signifies a milestone in integrating embodied intelligence into core production processes, moving beyond auxiliary roles [8][26][27] - This development reflects a shift from merely demonstrating technological capabilities to establishing a replicable and scalable productivity solution within the manufacturing sector [46][51] - The successful operation of Galbot S1 in real production lines validates its reliability and long-term value, indicating that embodied intelligence is now a critical component of industrial upgrades [28][29] Group 3: Market and Future Prospects - The recent completion of a 2.1 billion yuan financing round and a valuation exceeding 20 billion yuan positions the company as a leader in the embodied intelligence sector in China, showcasing confidence in the long-term prospects of intelligent manufacturing [53] - The ongoing collaboration with major manufacturers and the accumulation of operational data and engineering experience are paving the way for large-scale applications of embodied intelligence [34][51] - The emergence of Galbot S1 represents a crucial step in the evolution of embodied intelligence, as it begins to play a key role in enhancing productivity within the industrial core [56][57]

Core Viewpoint - The launch of Colossus 2, the world's first 1GW supercomputing cluster, marks a significant advancement in AI infrastructure, with plans to upgrade to 1.5GW by April and potentially reach 2GW, which could match the power consumption of major U.S. cities [2][12]. Group 1: Colossus 2 Overview - Colossus 2 is equipped with approximately 200,000 NVIDIA H100/H200 GPUs and around 30,000 NVIDIA GB200 NVL72 GPUs, significantly enhancing its computational power compared to its predecessor, Colossus 1, which was built in just 122 days [9][10]. - The cluster's 1GW capacity can power about 750,000 households, equivalent to the peak power demand of San Francisco [11]. - Once fully operational, Colossus 2 will house 555,000 GPUs, surpassing the GPU counts of Meta, Microsoft, and Google [13][14]. Group 2: Implications for AI Development - The advancements in Colossus 2 are expected to facilitate the development of Grok 5, which is projected to have parameters around 6 trillion, more than double that of Grok 4 [15][18]. - With the recent $20 billion funding round for xAI, the scaling capabilities for Grok 5 are increasing, leading to larger model parameters and faster training and deployment speeds [18][19]. - The rapid development of AI models is seen as a competitive advantage in the industry, emphasizing that speed is a crucial factor in the AI era [20]. Group 3: Energy Supply Concerns - The construction of large data centers like Colossus 2 is contributing to a projected annual electricity demand growth of 4.8% over the next decade, which is unprecedented for the U.S. energy system [27]. - The imbalance between rapidly increasing demand and slow supply growth is causing concerns about the stability of the power grid, leading to potential rolling blackouts for 67 million residents in 13 states during extreme weather [5][22][23]. - PJM, the regional transmission organization, is struggling to maintain supply-demand balance and has proposed measures to reduce peak demand from data centers, which have faced opposition from major tech companies [32][34].

Choice Policy团队 投稿 量子位 | 公众号 QbitAI 在家庭厨房自主使用洗碗机，在办公室边移动边擦拭白板——这些人类习以为常的场景，对人形机器人来说，却是需要调动全身关节协同运 作才能完成的 "高难度挑战" 。 近日，UC Berkeley加州大学伯克利分校团队在arXiv平台发表了题为《Coordinated Humanoid Manipulation with Choice Policies》的 研究论文，通过"模块化教学+智能选动作"的创新方案，成功破解了人形机器人全身协同的核心难题，为其走进真实人类环境铺平了道路。 阻碍人形机器人走进日常生活的"两大困境" 人形机器人一直被寄予厚望，有望在家庭、办公等非结构化环境中帮助人类完成日常工作，但长期以来，两个关键难题让它始终无法突 破"实验室边界"，难以真正落地应用： 难题1. 全身协同难，"教学数据"获取贵且难 像使用洗碗机、移动擦黑板这类"长时连续任务"，需要机器人同时协调头部 （定位目标） 、双手 （抓握操作） 、腿部 （移动平衡） ，实 现类似人类"眼到手到、脚步稳健"的状态。 但传统的"遥操作"模式，需要操作员同时控制机器人几十个 ...

henry 发自 凹非寺 量子位 | 公众号 QbitAI 已经是全球市值第一了，还怎么继续往上走？ 英伟达给出的答案很简单： 挖人，挖更多的人。 过去的2025年，黄仁勋一边扩编管理层，一边掏钱收团队—— 从挖角市场、政策、人力资源高管，到收购初创公司"打包"引入技术负责人，一套典型的"黄氏挖人+黄氏收购"正在成型。 不止芯片，用挖人重塑"第二增长曲线" 2025财年，英伟达营收 1305亿 美元，较前一财年增长逾一倍，成为科技史上的增长奇迹。 与此同时，英伟达正在用挖人重塑自己的"第二增长曲线"： 一方面系统性"挖人"，补齐市场、政策、研究与组织管理等关键能力。 另一方面则通过收购初创公司，直接将核心技术负责人和软件骨干纳入体系。 在今年最新的人事动作中，英伟达把"挖人"的铲子伸向了谷歌。 据悉，英伟达将聘请谷歌云老将 Alison Wagonfeld 出任公司 首位首席营销官 （Chief Marketing Officer，CMO） 。 Wagonfeld于今年2月正式履新，将此前分散在多位高管手中的相关职责，统一整合，全面负责英伟达的市场与传播工作。 （注：英伟达此前从未设立过专职的首席营销官（CM ...

编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位均为全职，工作地点：北京中关村。 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 AI产业方向 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用于工作，提升工作效率和创造力。 打造个人影响力 ：通过撰写独家原创内容，建立个人知名度，成为AI领域的意见领袖。 拓展行业 ...

Core Viewpoint - The ongoing lawsuit between Elon Musk and OpenAI has revealed significant and controversial details, particularly regarding the leadership's intentions and actions within OpenAI, which may impact the company's future and its relationship with Musk [2][23]. Group 1: Lawsuit Developments - The court has unsealed over 100 witness statements, leading to surprising revelations that have heightened public interest in the case [2][3]. - Musk expressed eagerness for the trial, suggesting that the outcomes and testimonies will be shocking [2] Group 2: OpenAI's Response - OpenAI has created a dedicated page on its website to counter Musk's claims, indicating a proactive approach to managing public perception [3]. - The organization argues that Musk is misrepresenting facts and that he had previously agreed to a profit-oriented structure for OpenAI's future [26]. Group 3: Leadership Controversies - Sam Altman, OpenAI's CEO, was found to have concealed his indirect ownership of OpenAI shares through the YC Fund, contradicting his earlier statements about not holding any shares [4][12]. - Greg Brockman's private diary entries from 2017 reveal intentions to remove Musk from the organization and shift towards a profit-driven model, despite publicly maintaining a non-profit stance [15][20]. Group 4: Musk's Allegations - Musk allegedly sought significant control over OpenAI, including a majority stake and CEO position, which was rejected by the board [27]. - OpenAI claims that Musk's ongoing litigation is a strategy to delay their progress and benefit his own company, xAI [29]. Group 5: Trial Timeline - The trial is scheduled to begin on April 27, 2026, and is expected to last approximately four weeks, with the judge noting the presence of numerous disputed evidences suitable for jury deliberation [31][32].

编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位面向： 加入我们，你可以获得： 以下是岗位详情： 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 AI产业方向 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用于工作，提升工作效率和创造力。 打造个人影响力 ：通过撰写独家原创内容，建立个人知名度，成为AI领域的意见 ...

Core Viewpoint - The article discusses the introduction of OVSeg3R, a new paradigm for open-set 3D instance segmentation that significantly reduces training costs and improves performance by leveraging mature 2D instance segmentation data and 3D reconstruction techniques [2][3][10]. Group 1: Challenges in 3D Perception - 3D instance segmentation is crucial for applications like autonomous driving and robotics, as it enables machines to understand and delineate objects in 3D space [4]. - The high cost and difficulty of acquiring and annotating 3D data have been major bottlenecks in the development of 3D perception models [5][6]. - Existing methods to integrate 2D models into 3D tasks often fail to enhance the 3D model's ability to recognize new objects, limiting their effectiveness [7][8][9]. Group 2: OVSeg3R's Technical Principles - OVSeg3R connects 3D models with 2D models through 3D reconstruction, allowing the 3D model to learn from the rich data available in 2D segmentation [10]. - The learning process involves three stages: data preparation, model input and annotation preparation, and model learning [12][19]. - Key innovations include the use of Instance-Boundary-aware Superpoints (IBSp) to improve training stability and the generation of high-quality semantic labels from raw video data [16][19]. Group 3: Performance Metrics - OVSeg3R achieved a significant performance leap in the ScanNet200 benchmark, reducing the performance gap between long-tail and head classes from 11.3 mAP to 1.9 mAP [21][22]. - In open-set settings, OVSeg3R outperformed previous methods, achieving a mAP of 24.6 and a notable improvement of 7.7 mAP in novel categories [23]. Group 4: Application Scenarios - OVSeg3R's capabilities are expected to drive advancements in open-set 3D instance segmentation, particularly in embodied intelligence by reducing reliance on expensive manual 3D annotations [27]. - The model's open-set recognition ability allows for precise identification of previously unseen "long-tail" objects, enhancing safety in robotic navigation and operation [28]. - OVSeg3R also compensates for the limitations of 3D geometry in recognizing non-rigid objects, providing a solid foundation for robotic grasping and navigation applications [29]. Group 5: Industry Progress - The technology is being advanced towards industrial application by IDEA's incubated company, Vision Future, indicating a move towards practical deployment [30].

Core Insights - The article discusses a groundbreaking experiment by Cursor, where hundreds of AI agents collaboratively developed a usable web browser from scratch, producing over 3 million lines of code [2][3]. Group 1: Experiment Overview - The project, codenamed FastRender, resulted in a browser with a rendering engine written in Rust and a custom JavaScript virtual machine [2]. - The browser is described as "barely usable," with performance significantly lagging behind established browsers like Chrome, but it can render Google's homepage correctly [3][4]. Group 2: AI Model Utilization - The success of the experiment relied on OpenAI's GPT-5.2-Codex, which is designed for complex software engineering tasks and can autonomously plan and execute coding tasks [5][6]. - GPT-5.2-Codex incorporates a technique called "Context Compaction," enhancing its ability to maintain logical consistency while handling large codebases [8]. Group 3: Multi-Agent Collaboration - Cursor developed a multi-agent collaboration architecture to enable hundreds of AI agents to work simultaneously without conflicts [12][18]. - Initial attempts at a flat collaboration model led to significant inefficiencies, prompting a shift to a hierarchical structure with planners, workers, and judges to streamline the process [15][18]. Group 4: Insights and Challenges - The experiment revealed that the general GPT-5.2 model outperformed the specialized GPT-5.1-Codex in long-term autonomous tasks, while other models like Claude Opus 4.5 were better suited for interactive scenarios [21]. - The design of prompts was found to be more critical than the model itself, emphasizing the need for extensive trial and error to guide AI agents effectively [22]. Group 5: Future Implications - The experiment sparked significant industry discussion, with predictions that the marginal cost of software development could approach zero as token costs decline [25]. - Despite existing challenges, such as planning responsiveness and agent overactivity, the experiment demonstrated the feasibility of scaling autonomous coding capabilities through increased agent numbers [29].