scaling

Search documents
Kimi没有梦想
Hu Xiu· 2025-06-24 05:32
Core Viewpoint - The article discusses the rise and challenges faced by Kimi, an AI company, highlighting the impact of FOMO (Fear of Missing Out) on its growth and subsequent issues, including a shift in investor sentiment and operational strategies [10][22]. Group 1: Company Overview - Kimi has transitioned from a promising AI startup to facing significant challenges, including a decline in its competitive edge and user growth [7][22]. - The company was once valued at $30 billion, largely due to FOMO-driven investments, particularly from Alibaba, which invested nearly $800 million [14][15]. Group 2: Business Strategy and Challenges - Kimi's aggressive user acquisition strategy involved significant spending on marketing, reminiscent of past failed models like ofo bike-sharing [16][17]. - The reliance on the "Scaling Law" and "data flywheel" theories has been criticized, with experts suggesting that merely increasing data and computational power does not guarantee improved model performance [18][20]. Group 3: Market Dynamics and Future Outlook - The AI landscape is shifting, with new models challenging existing paradigms, indicating a need for Kimi to adapt its technological approach [21]. - Kimi's recent controversies, including arbitration cases and ethical concerns, have severely impacted its ability to secure further funding, particularly from state-owned enterprises [22][23].
OpenAI路线遭质疑,Meta研究员:根本无法构建超级智能
3 6 Ke· 2025-06-20 12:00
Core Insights - The pursuit of "superintelligence" represents a significant ambition among leading AI companies like Meta, OpenAI, and Google DeepMind, with substantial investments being made in this direction [1][3][4] - Sam Altman of OpenAI suggests that building superintelligence is primarily an engineering challenge, indicating a belief in a feasible path to achieve it [3][4] - Meta AI researcher Jack Morris argues that the current approach of using large language models (LLMs) and reinforcement learning (RL) may not be sufficient to construct superintelligence [1][2] Group 1: Current Approaches and Challenges - Morris outlines three potential methods for building superintelligence: purely supervised learning (SL), RL from human validators, and RL from automated validators [2] - The integration of non-text data into models is believed not to enhance overall performance, as human-written text carries intrinsic value that sensory inputs do not [2][6] - The concept of a "data wall" or "token crisis" is emerging, where the availability of text data for training LLMs is becoming a concern, leading to extensive efforts to scrape and transcribe data from various sources [8][19] Group 2: Learning Algorithms and Their Implications - The two primary learning methods identified for potential superintelligence are SL and RL, with SL being more stable and efficient for initial training [10][22] - The hypothesis that superintelligence could emerge from SL alone is challenged by the limitations of current models, which may not exhibit human-level general intelligence despite excelling in specific tasks [15][16] - The combination of SL and RL is proposed as a more viable path, leveraging human feedback or automated systems to refine model outputs [20][22][28] Group 3: Future Directions and Speculations - The potential for RL to effectively transfer learning across various tasks remains uncertain, raising questions about the scalability of this approach to achieve superintelligence [34] - The competitive landscape among AI companies is likely to intensify as they seek to develop the most effective training environments for LLMs, potentially leading to breakthroughs in superintelligence [34]
天工不止造物,也能修bug:Skywork-SWE给代码智能体补上软件工程课
机器之心· 2025-06-20 02:22
机器之心报道 编辑:Panda 400 多年前,宋应星著成《天工开物》。这是一部写给匠人、也写给未来的书。它让人相信:技术不是死物,而是人与世界持续互动的方式。 如今,代码系统早已成为现代文明的骨架。它们运行在日常软件、银行服务、交通调度等各式系统中,也支撑着我们所依赖的 AI 算法本身。但和古代器物一样, 再精妙的程序也难免出现 bug—— 有些是逻辑失误,有些是环境变迁,有些甚至源于协作失控。比如,就在前几天,AWS、谷歌云、Azure 和 Cloudflare 都发生了 中断,连带着 ChatGPT 和 Cursor 等热门 AI 应用也一并短暂失联;而这一事故的原因可能是一次错误的自动配额更新导致谷歌的 API 管理系统出现了故障。 同时,bug 修复也是软件工程中最基础,却也是最复杂、最消耗人力的任务之一。特别是在真实的 GitHub 项目中,修一个 bug 并不是「找到一行错字那么简 单」,它常常需要: 那么,我们能否使用 AI 智能体来完成这些任务呢? 当然可以!但我们需要的绝不是传统的用于解决单独编程任务的 AI 编程模型,而是需要像人类开发者一样能够理解历史上下文、进行多轮推理、在模糊与不确 ...
小鹏想要的,不止“留在牌桌上”
虎嗅APP· 2025-06-19 23:55
出品丨虎嗅汽车组 作者丨李赓 头图丨视觉中国 在所有造车新势力中,今年1-5月依旧保持高速增长的只有两家:小鹏和零跑。 两家车企的销量都保持了大幅的提升 (1-5月零跑相比去年同期增长161%,小鹏增长293%) ,今年 一季度的营收也实现了大幅增长 (零跑同比增幅187%,小鹏同比142%) ,净亏损则实现了大幅的 收窄 (零跑净亏损缩小87%,小鹏净亏损缩小52%) 。除去数据上的略微不同,更加不同的是两家 心态的外露。 零跑依旧保持了自己不怎么开发布会不怎么大力做营销的状态 (今年正式发布会也就两场,而且全 是车型更新) ,而去年刚"触底反弹"的小鹏显然更加"珍惜"市场给的又一次机会,在方方面面都选 择了投入到"极点",几乎每个车型都要按着"曝光、预热、预发布、实际发布、会后沟通"的充分流程 走下来,更是在一众车企中罕见地结合产品发布会搞了几次针对实际车主的品牌文化活动。 就拿4月中,上海车展开幕前夕的关键时刻,何小鹏就跑到了香港去,不仅豪横地再次定下了香港启 德邮轮码头的场地 (2021赴港上市,也是这块场地) ,请了近500家中外媒体看新款X9发布。在主 活动之外,小鹏还在香港独立地举办了两场媒体沟 ...
小鹏想要的,不止“留在牌桌上”
Hu Xiu· 2025-06-19 23:13
Core Insights - Both Leapmotor and Xpeng have significantly increased their sales, with Leapmotor growing 161% and Xpeng 293% year-on-year from January to May. Their Q1 revenues also saw substantial growth, with Leapmotor up 187% and Xpeng up 142%. Net losses were reduced significantly, with Leapmotor's loss shrinking by 87% and Xpeng's by 52% [2] - Xpeng's proactive marketing and product launch strategy contrasts with Leapmotor's more reserved approach, indicating a different mindset in responding to market opportunities [2] - Xpeng's recent product, the MONA M03, has been a key driver of its sales rebound, accounting for over 50% of monthly sales since its launch [7][12] Sales and Marketing Strategy - Xpeng's marketing strategy includes extensive media engagement and product launch events, such as the recent X9 launch in Hong Kong, which attracted nearly 500 media representatives [3][4] - The company has focused on creating a strong brand presence through various promotional activities, including events targeting actual car owners [2][3] - The MONA M03's competitive pricing and features, such as a 620 km range, have made it appealing to consumers, particularly in addressing range anxiety [9][8] Product Development and Features - The MONA M03 has been designed with a focus on user needs, balancing cost control with essential features, which has resonated well with consumers [8][12] - The vehicle includes enhancements like electric tailgates and smart parking, while also simplifying certain features to reduce costs [10][11] - Xpeng's product team demonstrated efficiency in refining the MONA model within a short timeframe after acquiring it from Didi [12] Consumer Demographics and Feedback - The MONA M03 has attracted a notably high percentage of female consumers, with 38.6% of users being women, which is significantly above the industry average [18][19] - Feedback from female users highlights the vehicle's aesthetics and practical features, contributing to its popularity among this demographic [20][21] - Xpeng has quickly adapted to market feedback by introducing new interior options that appeal to female consumers, further boosting sales [21][25] Technological Advancements - Xpeng is focusing on technological innovation, particularly with its self-developed "Turing AI chip," which will enhance the capabilities of its vehicles, including the upcoming G7 model [27][30] - The G7 will feature advanced computing power, significantly exceeding that of competitors, which is part of Xpeng's strategy to differentiate itself in the market [30][31] - The company is also exploring the application of scaling laws in AI to improve autonomous driving capabilities, indicating a commitment to ongoing technological development [40][42] Future Outlook - Xpeng's CEO has emphasized the importance of building a robust system rather than relying solely on individual product successes, indicating a long-term vision for the company [26][51] - The company aims to maintain its focus on technological advancements and market responsiveness to ensure its competitive position in the automotive industry [51]
推荐大模型来了?OneRec论文解读:端到端训练如何同时吃掉效果与成本
机器之心· 2025-06-19 09:30
机器之心报道 机器之心编辑部 人人都绕不开的推荐系统,如今正被注入新的 AI 动能。 随着 AI 领域掀起一场由大型语言模型(LLM)引领的生成式革命,它们凭借着强大的端到端学习能力、海量数据理解能力以及前所未有的内容生成潜力,开始重 塑各领域的传统技术栈。 作为互联网流量的核心引擎,推荐系统面临着级联架构导致的算力碎片化、优化目标割裂等问题,并逐渐制约其创新发展。实现从碎片化拼装到一体化整合的范 式跃迁,是推荐系统重焕生机的必由之路,而利用 LLM 技术重构架构以实现效果提升、成本降低成为关键。 近日,快手技术团队交出了他们的答卷,最新提出的「OneRec」首次以端到端生成式架构重构推荐系统全链路。 在效果与成本这场看似零和的博弈中,OneRec 让「既要又要」成为可能 : 目前,该系统已在快手 App / 快手极速版双端服务所有用户,承接约 25% 的QPS( 每秒请求数量 ) ,带动 App 停留时长提升 0.54%/1.24%,关键指标 7 日用户生 命周期(LT7)显著增长,为推荐系统从传统 Pipeline 迈向端到端生成式架构提供了首个工业级可行方案。 下图(左)展示了快手 / 快手极速版中 O ...
云载 AI·健行未来——火山引擎“AI+医药大健康”行业论坛圆满落幕
Cai Fu Zai Xian· 2025-06-19 09:13
Core Insights - The "AI + Healthcare" forum highlighted the transformative impact of AI in the healthcare sector, emphasizing the integration of cloud computing, big data, and AI technologies to enhance medical services and patient experiences [1][17] - The forum featured contributions from various experts, indicating a collaborative effort in advancing AI applications in healthcare, particularly in areas like disease prevention, diagnosis, and drug design [3][10] Group 1: AI Applications in Healthcare - AI is expected to address the increasing demands of life sciences and medicine due to rising life expectancy, with a focus on developing new AI technologies tailored for healthcare [3][10] - The collaboration between Volcano Engine and researchers has led to the development of Bio-OS-Co-Pilot, which significantly reduces research timelines from years to hours, enhancing efficiency in modeling and analysis [4] - Companies like Tianjin Pharmaceutical Group have reported a 14.3% increase in digital maturity through strategic digital transformation initiatives, showcasing the effectiveness of AI in optimizing workflows [6][8] Group 2: Future Directions and Challenges - The healthcare industry faces challenges such as high complexity and strict requirements for data governance, necessitating a shift towards sustainable iterative mechanisms for AI applications [12] - AI is positioned to enhance pre-consultation processes, patient education, and overall efficiency in healthcare delivery, while maintaining a supportive role rather than replacing human decision-making in high-risk scenarios [15] - Future efforts will focus on low-risk, high-value areas for AI implementation, such as research data analysis and logistics support, to ensure effective integration into healthcare systems [14]
电子行业2025年中期投资策略:算力需求仍将加大,端侧应用加速落地
Dongguan Securities· 2025-06-17 09:21
超配 (维持) 电子行业 2025 年中期投资策略 投 资 算力需求仍将加大,端侧应用加速落地 2025 年 6 月 17 日 罗炜斌 S0340521020001 电话:0769-22110619 邮箱: luoweibin@dgzq.com.cn 陈伟光 SAC 执业证书编号: S0340520060001 电话:0769-22119430 邮箱: chenweiguang@dgzq.com.cn 电子行业指数走势 究 资料来源:东莞证券研究所,iFind 动,2024 及 25Q1 业绩向好 投资要点: 本报告的风险等级为中高风险。 本报告的信息均来自已公开信息,关于信息的准确性与完整性,建议投资者谨慎判断,据此入市,风险自担。 请务必阅读末页声明。 电子行业 SAC 执业证书编号: 终端复苏及AI创新驱动,2024及25Q1业绩向好。受益于全球宏观经济改 善、国内补贴政策推出以及AI大模型导入等因素推动,智能终端需求呈 现复苏状态,同时大模型快速发展进一步加大对云端算力硬件需求,电 子行业整体业绩向好。2024年营收同比增长17.04%;归母净利润、扣非 后归母净利润同比分别增长24.10%和36.1 ...
Scaling Law首次在自动驾驶赛道被验证!小鹏汽车CVPR演讲详解:AI「吃」下6亿秒视频后,智能涌现
量子位· 2025-06-16 04:50
贾浩楠 发自 凹非寺 量子位 | 公众号 QbitAI CVPR 2025 ,自动驾驶传来重大进展: Scaling Law , 首次在这条赛道被验证! 来自中国的 小鹏汽车 ,完整拿出了技术方案和AI司机"智能涌现"的成果。 自动驾驶的"ChatGPT时刻",真的要来了吗? CVPR 2025,小鹏汽车拿出了什么成果 今年的CVPR线下会议在美国田纳西州纳什维尔举办,日期是6.11-6.15。观众老爷们看这篇推送的时候, CVPR才刚刚结束几个小时——新 鲜出炉 。 CVPR的自动驾驶分论坛 (Workshop on Autonomous Driving) ,历年都是业内极具影响力的技术风向标和盛会。比如2022年的WAD, Wayve首次披露了自己低传感器端到端路线方案,马上成为自动驾驶赛道炙手可热的明星公司;再比如,特斯拉最早在CVPR WAD上详细分 享了占用网络技术,随后成为业内悉数跟进的量产方案…… 今年的WAD,中国的 小鹏汽车是唯一一家受邀发表主题演讲的车企 。 小鹏在演讲前一天,刚刚开启了最新SUV G7 的预售,创造了 量产L3级AI算力第一车 的纪录,单车算力超过2200TOPS,何小鹏 ...
Scaling Law首次在自动驾驶赛道被验证!小鹏汽车CVPR演讲详解:AI「吃」下6亿秒视频后,智能涌现
量子位· 2025-06-16 04:49
CVPR 2025,小鹏汽车拿出了什么成果 今年的CVPR线下会议在美国田纳西州纳什维尔举办,日期是6.11-6.15。观众老爷们看这篇推送的时候, CVPR才刚刚结束几个小时——新 鲜出炉 。 CVPR的自动驾驶分论坛 (Workshop on Autonomous Driving) ,历年都是业内极具影响力的技术风向标和盛会。比如2022年的WAD, Wayve首次披露了自己低传感器端到端路线方案,马上成为自动驾驶赛道炙手可热的明星公司;再比如,特斯拉最早在CVPR WAD上详细分 享了占用网络技术,随后成为业内悉数跟进的量产方案…… 今年的WAD,中国的 小鹏汽车是唯一一家受邀发表主题演讲的车企 。 贾浩楠 发自 凹非寺 量子位 | 公众号 QbitAI CVPR 2025 ,自动驾驶传来重大进展: Scaling Law , 首次在这条赛道被验证! 来自中国的 小鹏汽车 ,完整拿出了技术方案和AI司机"智能涌现"的成果。 自动驾驶的"ChatGPT时刻",真的要来了吗? 小鹏在演讲前一天,刚刚开启了最新SUV G7 的预售,创造了 量产L3级AI算力第一车 的纪录,单车算力超过2200TOPS,何小鹏 ...