量子位
Search documents
AI开始“动手”了,全世界第一个带头的是阿里千问
量子位· 2026-01-15 04:26
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 ...
一年拿下三轮融资!影目INMO正在鼻梁上“复刻”一个AI手机
量子位· 2026-01-15 02:30
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].
清华新研究,Nature+Science双杀!
量子位· 2026-01-15 01:23
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 ...
姚班传奇陈立杰入职OpenAI!16岁保送清华,30岁拿下UC伯克利助理教授
量子位· 2026-01-15 01:23
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) | | ...
Meta元宇宙部门狂裁千人:一醒来就收到邮件,刚入职也未能幸免
量子位· 2026-01-14 11:19
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].
让AI当「动作导演」:腾讯混元动作大模型开源,听懂模糊指令,生成高质量3D角色动画
量子位· 2026-01-14 11:19
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 ...
「AI 100」榜单启动招募,AI产品“年会”不能停丨量子位智库
量子位· 2026-01-14 08:10
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]
量子位编辑作者招聘
量子位· 2026-01-14 08:10
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 ...
不得了,这个新技术把视频压缩到了0.02%!
量子位· 2026-01-14 08:10
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 ...
谷歌也要「AI抖音」了!新Veo 3.1原生支持竖屏,4K分辨率高画质
量子位· 2026-01-14 08:10
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].