梦瑶 发自 凹非寺 量子位 | 公众号 QbitAI 当代打工人「酷刑」四件套，看看友友们有没有躺枪： 一点外卖就贼纠结还嫌麻烦、Excel一开人先宕机，攻略越做头越大、买东西还总能买贵…… （光想想都脑仁疼.jpg 但好消息是：可以不用疼了，因为现在AI，已经能《直接上手》替我们把这些糟心事儿给办了。 这！是AI帮我选购下单的27杯霸王茶姬，一键魂穿 「淘宝闪购」 ，优惠券自动加好，顺手帮我小薅一把～ 还有这！AI帮我制定了一份超详细的南京旅游攻略，自动直达 「飞猪」 页面，订机酒、订门票全都一把掐！ 不卖关子，这就是 阿里千问App 的新能力，一口气上线400多项新功能，把 淘宝、闪购、支付宝、飞猪 这些阿里自家生态全给安排进来 了。 四天前，谷歌宣布了与沃尔玛等零售商的AI购物合作计划，但目前尚未上线。 而阿里领先于谷歌，目前已成为了全球首个大规模开放"搜索-决策-支付-履约"全链路AI功能的科技公司。 不用在N个App间来回跳转，说一句指令，就能在手机里把点外卖、买东西、订机票、订酒店，甚至是办签证、查社保这些事儿轻松搞定。 Qwen模型+最全阿里生态强强联手，AI终于不只会聊天，也开始有模有样地替人 ...

Core Viewpoint - The company INMO has rapidly advanced in the smart glasses sector, achieving three rounds of financing within a year, indicating strong market demand and investor confidence [1][2][10]. Financing - INMO completed its B2 round financing of 150 million yuan in July 2025, followed by a B3 round shortly after, showcasing a fast-paced fundraising strategy [7][10]. - The total financing amount for the year reached nearly 500 million yuan, highlighting the company's ability to attract significant capital [10][12]. Product Development - INMO launched the GO3 smart glasses, which achieved over 20,000 pre-orders within three days of its release, indicating strong consumer interest [3][51]. - The GO3 is the first smart glasses to implement real-time translation and two-way dialogue, focusing on high-demand scenarios such as translation and meeting assistance [17][32]. Market Position - INMO has established itself as a leader in the lightweight, all-in-one AI+AR smart glasses category, with a valuation of 2 billion yuan and recognition as a category creator [19][60]. - The company has consistently ranked first in sales among startup companies in the all-in-one smart glasses sector for five consecutive years [52]. Technological Innovation - INMO's products utilize the self-developed IMOS operating system and the GLM large model, enhancing interaction capabilities and making AI more proactive [25][30][29]. - The GO3 and AIR3 models incorporate advanced features such as voice control for everyday tasks and real-time translation, showcasing the integration of AI capabilities into wearable technology [33][38]. Market Expansion - INMO is expanding its market presence through partnerships with major players like Tencent and Ant Group, and plans to invest 20 million yuan to foster AI application development [45][46]. - The company is also focusing on offline retail partnerships to enhance customer experience, addressing the challenge of online-to-offline transitions in smart glasses sales [49][51]. Industry Context - The global smart glasses market saw a shipment of 4.296 million units in Q3 of the previous year, but investment patience is tightening, making INMO's rapid financing pace stand out [8][7]. - As the industry shifts towards usability and long-term wearability, INMO's approach of creating lightweight, integrated AI+AR glasses is gaining traction [60][61].

一水 发自 凹非寺 量子位 | 公众号 QbitAI 就在刚刚，清华大学的一项 AI for Science研究 不仅登上Nature，而且还被Science深度报道了。 | Explore content | | --- | 这项来自清华大学李勇团队的研究通过分析全球 2.5亿篇 科学文献，揭示了 AI for Science领域存在的一个典型矛盾 —— AI在助力科学家"个体加速"的同时，却导致科学界的集体注意力窄化和趋同优化的"群体登山"现象。 就是说，虽然AI帮助科学家发表了更多论文、更早成为项目负责人，但却导致人们集体涌入少量适合AI研究的"热门山峰"，从而无形中削弱了 科学探索的广度。 而且进一步分析表明，这一矛盾绝非偶然，而是由当前科学智能AI模型 缺乏通用性 导致的系统性影响。 下面详细来看这到底是一项怎样的研究。 第一步： 寻觅 AI for Science的演化踪迹 回到起点，团队之所以进行这项研究，主要是发现AI for Science领域存在一个明显矛盾—— 在AI持续赋能科研的背景下，为何各学科的整体科学进展未见明显加速？ 在论文中，团队进行的首项工作是： 从浩如烟海的文献中找出那些 ...

henry 发自 凹非寺 量子位 | 公众号 QbitAI 最新消息：姚班大神陈立杰，加盟OpenAI了。 据"Top华人社消息"，OpenAI内部确认：清华姚班天才、UC伯克利EECS助理教授 陈立杰 已加盟 OpenAI ，负责数学推理！ 值得一提的是，OpenAI 在去年 9 月发表的出圈论文《Why Language Models Hallucinate》中，也引用了陈立杰参与的另一篇研究《Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations》。 与此同时，陈立杰近期参与的最新研究方向也十分"当下"，聚焦于 扩散语言模型（Diffusion Language Models） ，紧跟当前生成模型的 重要演进路线。 截至目前，陈立杰主页未有更新。 | C chen-lijie.github.io | 27 2 ﮧ | ← → | | --- | --- | --- | | Publications (by Years) | | | | Publications (by Categories) | | ...

Core Viewpoint - Meta is significantly scaling back its metaverse ambitions by laying off over 1,000 employees from its Reality Labs division, reallocating resources towards AI hardware and wearable devices instead [2][3][5]. Group 1: Layoffs and Strategic Shift - Meta is cutting approximately 10% of its Reality Labs workforce, which translates to over 1,000 job losses [5]. - The layoffs are part of a broader strategy to reduce investment in the metaverse, with resources redirected to emerging fields like AI and wearable technology [3][29]. - The company has closed three notable VR game studios, indicating a fundamental shift in its VR content strategy [11]. Group 2: Financial Context - Reality Labs has incurred cumulative losses exceeding $70 billion since the company's pivot to the metaverse in 2021, highlighting the unsustainable nature of its current business model [18][19]. - The financial strain has prompted management to take drastic measures to restore overall financial health [19][28]. Group 3: New Strategic Focus - Meta is transitioning from a metaverse-first approach to a comprehensive focus on AI, with AGI identified as a core future goal [29]. - A new department, "Meta Computing," has been established to oversee infrastructure development necessary for this AI-centric strategy [29]. - The company aims to integrate generative AI technology into its applications, enhancing advertising efficiency and providing stable cash flow for future investments [30]. Group 4: Hardware and User Interaction Changes - The positioning of hardware has fundamentally changed, with smart glasses being redefined as "sensors" for AI assistants [32]. - Meta is moving away from traditional VR interactions, adopting a new standard based on visual recognition and voice commands [33]. - The goal is to create a wearable AI assistant that users can interact with naturally, without manual input [34].

在这个背景下，腾讯混元团队借鉴其在视频生成大模型上的成功经验，提出了一套全新的、旨在突破当前瓶颈的文生动作解决方案，通过构建 一套严格的数据处理与标注管线，覆盖大规模预训练、高质量精调、强化学习对齐的全阶段训练流程，并将Diffusion Transformer （DiT） 模型扩展至10亿级别参数量，成功研发了 混元Motion 1.0 （HY-Motion 1.0） 这一业界领先的动作生成基础模型，并将该模型于2025年12 月30日对外开源 （见文末链接） 。 腾讯混元团队 投稿 量子位 | 公众号 QbitAI 在3D角色动画创作领域，高质量动作资产的匮乏长期制约着产出的上限。 游戏、动漫、影视与数字人等产业始终面临一个成本困局：从数万元起步的专业动捕采集，到动画师以"天"为单位的手工精修骨骼动画，每一 秒丝滑动作的背后，都是高昂的资源堆砌。 而在生成式AI领域，文生动作 （Text-to-Motion） 也因高质量数据的稀缺与计算范式的局限，长期处于"小模型"阶段，这类模型在面对复杂 的自然语言指令输入时，很难做出创作者希望得到的正确动作。 近年来，也有不少研究开始尝试通过大语言模型扩展词表的方式来 ...

Core Insights - The article discusses the emergence of numerous keywords in the AI product sector by 2025, highlighting transformative AI products that are leading the market [4] - The "AI 100" list by Quantum Bit Think Tank aims to evaluate and recognize the top AI products in China, reflecting the industry's evolution and future trends [4][12] Group 1: AI 100 List Overview - The "AI 100" list is divided into three main categories: "Flagship AI 100," "Innovative AI 100," and the top three products in ten popular sub-sectors [6] - The "Flagship AI 100" will focus on the strongest AI products of 2025, showcasing those that have achieved significant technological breakthroughs and practical application value [7] - The "Innovative AI 100" aims to identify products that are expected to emerge in 2026, representing cutting-edge AI technology and potential industry disruptors [8] Group 2: Sub-sector Focus - The ten hottest sub-sectors for the top three products include AI browsers, AI agents, AI smart assistants, AI workstations, AI creation, AI education, AI healthcare, AI entertainment, Vibe Coding, and AI consumer hardware [9] Group 3: Application and Evaluation - The evaluation of the "AI 100" list employs a dual assessment system combining quantitative and qualitative metrics, focusing on user data and long-term development potential [13] - Quantitative metrics include user scale, growth, activity, and retention, while qualitative metrics consider technology, market space, design, monetization potential, team background, and growth speed [13]

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 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。 拓展行业人脉 ：与AI领域大咖零距离接触，参与重要科技活动和发布会，拓展行业视野。 获得专业指导 ：应届新人会由主编级编辑出任mentor，提供一对一指 ...

金磊 发自 凹非寺 量子位 | 公众号 QbitAI 感谢AI！ 原生1个G的视频，现在只需要传200K数据就能看了—— 视频数据的压缩率干到了 0.02% ，但依旧能保持画面的高清、连贯和画面细节。 或许你会问，这又有什么用呢？ 想象一下，你身处于太平洋的一搜远洋货轮中，卫星信号只有一两格，刷个朋友圈，加载内容的圈圈都要转好久。 但正是因为有了这项AI技术，现在在如此极端的环境之下，你甚至可以直接看 高清的世界杯直播！ 而这项新研究，正是来自中国电信人工智能研究院（TeleAI）的技术—— 生成式视频压缩（GVC，Generative Video Compression） 作为国资央企、全球领先的综合智能信息服务运营商，中国电信不仅拥有覆盖海陆空天的通信网络基础设施，更具备将前沿AI技术与实际通 信场景深度融合的能力。 这种"云网融合+AI原生"的独特优势，使得GVC技术从实验室走向远洋船舶、应急现场等真实极端环境成为可能。 那么这项研究到底是如何做到的，以及又能给我们现实生活带来什么改变，我们继续往下看。 。 没错，视频传输的物理法则，算是被重写了。 用计算，换宽带 在介绍这项黑科技之前，我们需得先聊聊现 ...

Core Viewpoint - Google has officially entered the AI short video arena with the upgrade of Veo 3.1, enhancing video generation quality and introducing vertical and 4K formats [1][11][12]. Group 1: Features of Veo 3.1 - The upgraded Veo 3.1 allows users to generate videos from a single vertical image and a simple prompt, showcasing creative capabilities [3][14]. - It supports native 9:16 vertical video format optimized for mobile platforms like YouTube, and has increased resolution from 720p to 4K [15][12]. - The model has significantly improved consistency, ensuring characters maintain their appearance across different scenes [16][26]. - Element fusion capabilities have been enhanced, allowing for coherent video generation from simple descriptions of characters, objects, and backgrounds [20][21]. Group 2: Market Context and Competition - Google is not the first to pursue vertical AI video; competitors like OpenAI and Disney have also made strides in this area [33][40]. - OpenAI's Sora app, which mimics TikTok, faced challenges with user retention, highlighting operational difficulties in managing such platforms [36][37]. - Google benefits from its comprehensive operational capabilities, leveraging platforms like YouTube to create a closed-loop ecosystem for content creation and distribution [38][39]. Group 3: Industry Trends - The trend towards vertical AI video is becoming increasingly evident, with various players in the industry recognizing its importance [42][43]. - Domestic AI players in China are also exploring similar video generation applications, indicating a growing interest in this format [44][46].