Workflow
世界模型
icon
Search documents
电动汽车百人会(2025)|博世吴永桥:中阶智驾将是标配,座舱才是车企主战场
Zhong Guo Jing Ji Wang· 2025-03-31 02:40
Core Viewpoint - The future of mid-level intelligent driving will become a standard feature in vehicles, similar to seat belts and airbags, leading manufacturers to rely on suppliers for development rather than investing heavily in in-house research and development [1][3]. Group 1: Industry Trends - Mid-level intelligent driving technology, exemplified by BYD's "Tianshen Eye," will be standard in vehicles priced between 80,000 to 150,000 yuan, featuring computing power of 80-150 TOPS and functionalities like highway navigation and automatic parking [3]. - Within three years, it is predicted that high-level intelligent driving will become widespread in vehicles over 150,000 yuan, with full navigation capabilities across various environments [3]. - The cost of developing mid-level intelligent driving systems is significant, with companies like Bosch investing nearly 2 billion yuan in the field last year [3]. Group 2: Supplier Dynamics - Tier 1 suppliers such as Bosch, Yuanrong Qihang, Huawei, and Momenta can support multiple clients and dozens of vehicle models, making it more cost-effective compared to manufacturers developing in-house with limited teams [3]. - Manufacturers typically require a team of 1,000 to 2,000 people to develop around 10 models, which is less efficient compared to the capabilities of suppliers [3]. Group 3: Future Focus Areas - For future L3 and L4 level autonomous driving, components can be managed by the supply chain, allowing manufacturers to focus on areas like user experience and after-sales service that suppliers may not adequately address [4].
中国银河证券:推理算力重要性提升 光模块等算力细分赛道发展再加速
Zhi Tong Cai Jing· 2025-03-24 08:58
Core Insights - The importance of inference computing power is increasing, with significant growth expected in related sectors such as optical modules and chips, driven by advancements in hardware and software from companies like NVIDIA [1][4] Group 1: Inference Computing Power Growth - Inference computing power is projected to continue growing, with NVIDIA's CEO stating that the demand for computing power will be 100 times greater than in the past to support advancements like AGI and embodied intelligent robots [1][2] - The number of tokens processed by models has increased to over 100 trillion, with inference models requiring 20 times more tokens and 150 times more computational power than before [2] - NVIDIA's Blackwell architecture shows a performance improvement of 68 times over the previous Hopper architecture, leading to an 87% reduction in costs [2] Group 2: Hardware and Software Developments - NVIDIA introduced the upgraded Blackwell Ultra architecture, emphasizing its potential to generate 50 times more revenue for data centers, with a clear development roadmap extending to 2026 and beyond [3] - The launch of the open AI engine stack, Nvidia Dynamo, aims to simplify inference deployment and scaling, potentially creating a new paradigm for efficiency in hardware and software [3] - The introduction of Nvidia Llama Nemotron is expected to serve as a foundational model for inference, facilitating exploration in related areas and forming an ecosystem [3] Group 3: Investment Recommendations - The current landscape indicates that the demand for computing power is not declining but is instead being stimulated by the growth of inference applications, suggesting substantial investment opportunities in the sector [4] - Recommended investment targets include telecom operators such as China Mobile, China Unicom, and China Telecom, as well as companies in optical modules and chips like Zhongji Xuchuang, Xinyi Guosheng, and others [4]
杨立昆“砸场”英伟达:不太认同黄仁勋,目前大模型的推理方式根本是错的,token 不是表示物理世界的正确方式|GTC 2025
AI科技大本营· 2025-03-21 06:35
责编 | 王启隆 出品丨AI 科技大本营(ID:rgznai100) 黄教主的演讲 感觉才没过几天,今年的 GTC 英伟达大会也即将迎来尾声了。 而今年比尔·达利则是对话"AI 教父" 杨立昆 (Yann LeCun),很有前后呼应的感觉。 但 GTC 并不只有黄仁勋和杨立昆,还有许多精彩的演讲与对话,比方说: ………… 接下来的一段时间, CSDN AI 科技大本营 将会在「 GTC 2025 大师谈 」栏目持续更新这些精华内容的全文整理,尽情期待。 比尔·达利 自己就在采访杨立昆之后进行了一场 演讲 ,系统性地讲解了英伟达 2024 一整年的四大项目进展,内容干货很多; OpenAI o1 作者 诺姆·布朗 (Noam Brown)和英伟达的 AI 科学家来了一场 对话 ,他认为现在 AI 圈最需要来一场革命的,就是这些五花八 门的 基准测试 (Benchmark),而且改这个东西还不需要花太多算力资源; 2018 年诺贝尔化学奖得主 弗朗西斯·阿诺德 (Frances Arnold)围绕 AI for Sciense 还有蛋白质工程进行了一场相当硬核的 圆桌对话 ; UC 伯克利教授 彼得·阿比尔 (P ...
范式集团创始人戴文渊:以“Agent +世界模型”赋能更多产业,加速通向AGI
IPO早知道· 2025-03-18 05:55
范式集团未来将剑指更多业务领域。 本文为IPO早知道原创 作者|Stone Jin 微信公众号|ipozaozhidao 据IPO早知道消息,在3月18日举行的范式集团发布会上,范式集团创始人戴文渊发布全新的集团 Logo,并分享了范式集团的 AGI 愿景与实现路径。 戴文渊强调:"范式集团未来将用'AI Agent+世界模型'的技术路径去赋能更多的产业,最终将大量 垂直的智能能力汇聚成为 AGI。" 因此, 通往 AGI 的路,就是不断提升 AI Agent 沟通能力,和不断丰富世界模型对各垂直细分领域 的认知,提升 AI 解决垂直领域的能力。 需要强调的一点是,我们所认为的世界模型和很多学者正在做的世界模型不同。当前绝大多数的世界 模型,在我们看来仅仅是3D空间模型,服务于具身智能。而 真正世界模型,远超具身智能的范畴, 包括认知从宏观到微观的世界,也包括认知人的内在需求,甚至包括所有的生命,即物理科学和生命 科学。 此前,我们更专注于企业市场,专注于企业软件中的垂直模型。我们虽然做出了一些成绩,但也意识 到智能的天花板也受限于企业市场。未来,我们要把格局放到整个市场,面向改造所有可能的业务, 当这些能力 ...
人形机器人的“iPhone时刻”快到了?
日经中文网· 2025-03-15 01:59
英伟达CEO黄仁勋在主题演讲中介绍人形机器人(1月6日,美国拉斯维加斯,摄影:积田檀) 大约15年前,iPhone成为新的技术平台,APP经济圈因此繁荣起来。随着生成式AI的发展,有观点 认为人形机器人也将迎来"iPhone时刻"。中美竞争激烈,中国有小鹏鹏行、宇树科技;美国有 Apptronik、Figure AI…… 奥平和行: 以美国和中国为中心,人形机器人的开发竞争火热。随着生成式AI的迅速发展,人形机器 人的实用化时期日益临近,有观点认为将迎来人形机器人渗透至社会的"iPhone时刻"。针对人形机器人 的乐观预测认为,到2050年全球市场规模将超过6亿台。在这种情况下,作为"机器人大国"显示出存在 感的日本也将被迫做出应对。 1月6日,美国拉斯维加斯,在科技展会CES(国际消费电子展)现场发表主题演讲的美国英伟达CEO黄 仁勋展示了14台人形机器人,将现场气氛推向高潮。黄仁勋表示,"它们是我的朋友。借助我一直介绍 的技术,未来几年将会实现飞跃发展"。 人形机器人的历史始于1920年代,大约20年前本田的"ASIMO"和索尼的"QRIO"曾引发热门话题。 当时由于用途有限且价格昂贵,这些机器人未能普及 ...
体验向上价格向下,端到端加速落地
HTSC· 2025-03-02 07:30
Investment Rating - The report maintains a rating of "Buy" for several companies in the automotive sector, including XPeng Motors, Li Auto, BYD, SAIC Motor, Great Wall Motors, and Leap Motor [10]. Core Viewpoints - The report emphasizes that by 2025, advanced intelligent driving (high-level AD) will see improved user experience and reduced prices, transitioning from a trial phase to widespread adoption among consumers [14][20]. - The penetration rates for L2.5 and L2.9 intelligent driving are projected to reach 3.5% and 10.1% respectively by November 2024, with expectations of further growth to 16% for highway NOA and 14% for urban NOA by 2025 [14][24]. - The report highlights the shift towards end-to-end architecture in intelligent driving systems, which allows for higher performance limits and seamless data transmission, enhancing the overall driving experience [30][31]. Summary by Sections Investment Recommendations - The report suggests focusing on companies with strong engineering capabilities and advantages in data, computing power, and funding, such as XPeng Motors, Li Auto, and BYD, as well as third-party suppliers like Desay SV and Kobot [5][10]. Market Trends - The report notes that the intelligent driving market is evolving, with a focus on enhancing user experience through features like "human-like" driving capabilities and the implementation of end-to-end architectures [14][20]. - The price of high-level intelligent driving systems is expected to decrease significantly, with current models priced below 100,000 and 150,000 yuan for highway and urban NOA respectively [24][28]. Technological Developments - The report discusses the advancements in end-to-end architecture, which is gaining traction among automotive manufacturers, allowing for improved data processing and decision-making capabilities [30][31]. - It also mentions the importance of AI-driven models and the need for automotive companies to adapt their organizational structures to support these technological shifts [15][41]. Competitive Landscape - The report outlines the competitive dynamics among leading automotive companies, highlighting their respective advancements in intelligent driving technologies and the rapid iteration of their systems [41][45]. - Companies like Tesla, Li Auto, and XPeng Motors are noted for their significant investments in R&D and their ability to push updates and improvements quickly [42][46].
对话小马智行楼天城:只靠端到端无法通向 L4,模仿优秀司机令人绝望
晚点LatePost· 2025-01-09 10:41
以下文章来源于晚点Auto ,作者晚点团队 晚点Auto . 从制造到创造,从不可能到可能。《晚点LatePost》旗下汽车品牌。 五年时间重构技术路线,"我们没有放弃在黎明前的那一刻"。 文丨程曼祺 编辑丨宋玮 Waymo,百度,再到 8 年前开始的创业——小马智行联合创始人兼 CTO 楼天城,可能是中国最懂 L4(全 无人级别)自动驾驶的人。 今年,百度萝卜快跑在武汉出圈,Google 旗下 Waymo 在旧金山的日均单量超过当地出租车,Robotaxi (无人出租车)和 L4 都在低谷反弹。 在小马智行于美股 IPO 后,我们与楼天城聊了他思考和实践着的 L4 技术:他几乎没提今年自动驾驶领域 最火的 "端到端"。 在他看来,描述车端模型架构的端到端本就不是关于 L4 的 "对的问题"。 真正的关键,是训练车端模型的虚拟环境,楼天城称之为 "世界模型。 楼天城说,世界模型是车端模型的 factory(工厂),L4 自动驾驶技术的真正差别在 "工厂" 的精度,因为 它会决定它所生产的产品,即车端模型的上限。 这套与流行的 "端到端" 相去甚远的方法背后,是小马智行过去 5 年经历的技术路线切换,楼天城总结 ...
AI 月报:10 亿美元训练不出 GPT-5;低成本中国开源大模型走红;AI 幻觉不全是坏处
晚点LatePost· 2025-01-07 14:59
2024 年 12 月的全球 AI 大事记。 文丨贺乾明 编辑丨程曼祺 2024 年 12 月的 AI 月报,你会看到: OpenAI、Google 发布新模型,中国的 DeepSeek 也抢到了风头 GPT-5 训练遇阻的更多细节 强化学习的重要性持续提升 至少有三个团队推出了世界模型 Google 霸占大模型竞技场前三 中国公司在开源社区存在感大涨 博通帮大公司自研 AI 芯片,市值破万亿美元 OpenAI 正式启动转型营利公司 20+ AI 公司获 5000 万美元以上投资,有 2 家中国公司 大模型的幻觉并不是一无是处 以下是我们第 2 期 AI 月报,欢迎大家在留言区补充我们没有提到的重要进展。 技术|10 亿美元没训出 GPT-5,新版 Scaling Laws 初步证明可行,多款世界模型亮相 GPT-5 训练遇阻的更多细节 OpenAI 训练 GPT-5(代号 Orion)遇阻,是大模型能力提升放缓的重要证据。12 月,多家媒体提供了更多的细 节: 2023 年 4 月推出 GPT-4 后,OpenAI 一直在开发 GPT-5,已经持续 20 个月。OpenAI 看到过乐观信号:24 年 4 月 ...
小马智行 IPO,8 年坚持 L4 的少数派
晚点LatePost· 2024-11-28 14:57
以下文章来源于晚点Auto ,作者晚点团队 晚点Auto . 从制造到创造,从不可能到可能。《晚点LatePost》旗下汽车品牌。 "看得到山顶,也要知道自己在接近山顶。" 11 月 27 日,小马智行正式在纳斯达克挂牌上市,股票代码为 "PONY",发行价为 13 美元,募资 4.52 亿 美元 ,比原计划超募 33%。 我们与小马智行的两位联创——CEO 彭军和 CTO 楼天城,聊了这 8 年的创业过程和小马智行独特的技术 判断。 7 年前的 2017 年 6 月,小马智行刚成立半年,组装出了自己的第一辆自动驾驶车,开始在硅谷道路上测 试。不久后,一位来访的投资人无预兆地冲到行驶的车辆前,想向同事证明自动驾驶技术的厉害。所有人 都吓了一跳,好在车成功自动刹停。 IPO 前,小马智行共获 7 轮融资,累计金额超 13 亿美元,投资方包括红杉中国、IDG 资本、五源资本、 丰田汽车、沙特新未来城等。 技术创业大致有两种,一种是有了技术,但需要找需求和场景;还有一类是,目标和愿景都很清晰,但实 现难度非常大。从业者一开始往往会低估困难,而随时间演进,又开始高估困难。 自动驾驶毫无疑问属于后者。彭军对能够带领小马 ...