说它脸萌，是因为它正脸长这样，没啥恐怖谷的感觉： 衡宇 发自 凹非寺 量子位 | 公众号 QbitAI 全球首款开卖的家务人形机器人来了！ 它就是 1X Technologies （以下简称1X） 公司推出的 NEO家用机器人 。 来看，下面这个萌脸翘臀的家伙，很有可能是你会拥有的第一台家庭具身智能机器人—— 当然，对人形机器人来说颜值不是正义。 能自主干各种各样的家务活，才是大家最盼望实现的： 家里有大别野的也别担心，NEO自己会平稳地上下楼梯： NEO今日开售，首发有米色、灰色与深棕色三种颜色可选，2026年发货。 早鸟价格20000美元 （约141978元） /台。 嫌贵的话还可以月租，500美元 （约3549.45元） 一个月。 1X的AI副总裁Eriic Jang表示： 这是一款超前时代的产品。 我们仍在积极开发和完善部分功能，可能会出现失误，我们会迅速汲取经验，并利用用户的早期反馈来改进NEO。 我不少亲爱的网友，发出了和我第一眼看到视频时一样的感叹： 里——面——有——人！ 而有的人已经手快下单了。 一买就是两台，希望它们承包家里所有的家务活，以及在闲暇时刻打打架来让自己看个乐子。 而且开箱即用 ...

Core Viewpoint - The article highlights the significant advancements and strategic initiatives by NVIDIA in the fields of AI computing, quantum computing, and 6G communication, emphasizing the competitive landscape and potential challenges from rivals like AMD and Qualcomm [1][49]. Group 1: NVIDIA's New Chip Developments - NVIDIA introduced the Vera Rubin superchip, which boasts a computing power of 100 PFLOPs, marking a 100-fold increase over its previous AI computing model, DGX-1 [5][6]. - The Vera Rubin platform is designed with a new architecture, integrating a Vera CPU and two Rubin GPUs, with the first samples produced by TSMC [10][12]. - The upcoming Vera Rubin NVL144 platform is expected to deliver 3.6 Exaflops of FP4 inference power and 1.2 Exaflops of FP8 training power, representing a 3.3-fold improvement over the previous GB300 model [19]. Group 2: Strategic Collaborations and Investments - NVIDIA plans to collaborate with the U.S. Department of Energy to build seven new supercomputing clusters, including two new supercomputers based on the Vera Rubin platform [22]. - The company has invested $1 billion in Nokia to develop AI-native 6G communication platforms, which has positively impacted Nokia's stock price [45]. Group 3: Quantum Computing Initiatives - NVIDIA announced NVQLink, a new interconnect architecture that enables seamless integration between quantum processors (QPUs) and NVIDIA GPUs, facilitating high-speed data transfer essential for quantum error correction [29][31]. - The CUDA-Q platform was introduced to extend CUDA capabilities to support quantum GPU computing, allowing for collaboration between classical and quantum computing [33][43]. Group 4: Competitive Landscape - AMD has secured two supercomputer contracts worth $1 billion, with its Lux supercomputer expected to outperform existing systems in AI performance [50]. - Qualcomm is entering the data center market with new AI chips, AI200 and AI250, focusing on cost efficiency and enhanced memory processing capabilities [52]. - The article notes that despite NVIDIA's advancements, it faces competition from various players in the quantum computing and 6G sectors, including significant developments from Chinese companies [54][60]. Group 5: Market Reaction - Following the announcements, NVIDIA's stock price rose by 4.98%, reaching $201.03 per share, with a post-market price of $204.43, resulting in a market value increase of $315.4 billion [65][66].

Core Insights - The article discusses the rapid development of image generation technology based on diffusion models, highlighting the limitations of the Variational Autoencoder (VAE) and introducing the EPG framework as a solution [1][19]. Training Efficiency and Generation Quality - EPG demonstrates significant improvements in training efficiency and generation quality, achieving a FID of 2.04 and 2.35 on ImageNet-256 and ImageNet-512 datasets, respectively, with only 75 model forward computations [3][19]. - Compared to the mainstream VAE-based models like DiT and SiT, EPG requires significantly less pre-training and fine-tuning time, with 57 hours for pre-training and 139 hours for fine-tuning, versus 160 hours and 506 hours for DiT [7]. Consistency Model Training - EPG successfully trains a consistency model in pixel space without relying on VAE or pre-trained diffusion model weights, achieving a FID of 8.82 on ImageNet-256 [5][19]. Training Complexity and Costs - The VAE's training complexity arises from the need to balance compression rate and reconstruction quality, making it challenging [6]. - Fine-tuning costs are high when adapting to new domains, as poor performance of the pre-trained VAE necessitates retraining the entire model, increasing development time and costs [6]. Two-Stage Training Method - EPG employs a two-stage training method: self-supervised pre-training (SSL Pre-training) and end-to-end fine-tuning, decoupling representation learning from pixel reconstruction [8][19]. - The first stage focuses on extracting high-quality visual features from noisy images using a contrastive loss and representation consistency loss [9][19]. - The second stage involves directly fine-tuning the pre-trained encoder with a randomly initialized decoder, simplifying the training process [13][19]. Performance and Scalability - EPG's framework is similar to classic image classification tasks, significantly lowering the barriers for developing and applying downstream generation tasks [14][19]. - The inference performance of EPG-trained diffusion models is efficient, requiring only 75 forward computations to achieve optimal results, showcasing excellent scalability [18]. Conclusion - The introduction of the EPG framework provides a new, efficient, and VAE-independent approach to training pixel space generative models, achieving superior training efficiency and generation quality [19]. - EPG's "de-VAE" paradigm is expected to drive further exploration and application in generative AI, lowering development barriers and fostering innovation [19].

Core Viewpoint - The article discusses the entrepreneurial journey of Liao Qian, the former VP of Product at Shengshu Technology, who has founded a new company named Apex Context, focusing on creating a multi-modal AI agent for marketing scenarios. The company has already secured millions of dollars in funding within a short period after its establishment. Group 1: Company Overview - Apex Context was founded by Liao Qian after leaving his previous job at the end of August [2][10]. - The company aims to develop a multi-modal agent specifically for marketing applications, which is seen as a productive and quantifiable area for AI implementation [11][12]. - The name "Apex Context" reflects the company's vision of AI deeply understanding and responding to user context, enhancing the precision and relevance of generated content [4][5]. Group 2: Product Focus - The primary goal of Apex Context is to create an AI Video Agent that assists brands in visual expression, providing end-to-end capabilities from creative ideation to video production [18]. - The agent is designed to be user-friendly, requiring minimal input from users, and aims to understand vague ideas or uncertain requests to generate appropriate content [15][16]. - The company plans to expand its capabilities beyond marketing to include education, lifestyle, and entertainment in the long term [22]. Group 3: Market Positioning - Liao Qian believes that the next phase of competition will revolve around who can help individuals and brands express themselves more effectively [21]. - The current technological landscape, marked by advancements in AI, presents a unique opportunity for startups to innovate while larger companies are preoccupied with defending their core businesses [38][40]. - The company emphasizes its understanding of user needs and scenarios as a potential competitive advantage in the market [40].

首次公开了内部研究目标的具体时间表，其中最引人注目的是 "在2028年3月实现完全自主的AI研究员" ，具体到月份。 这次发布会信息密度非常大，连奥特曼自己都说："鉴于这些内容的重要性，我们将以不同寻常的透明度分享我们的具体研究目标、基础设施 计划和产品战略。" 难道重组后的OpenAI，真的重新Open了？ 梦晨 发自 凹非寺 量子位 | 公众号 QbitAI OpenAI完成史上最重要的一次组织架构调整后，紧接着开了一场直播。 不过也有一些事故，本来OpenAI发帖征集大家的问题，结果抱怨GPT-4o对敏感对话强制路由机制的人太多，两人支支吾吾面面相觑了一 阵。 奥特曼最终还是承认"这次我们搞砸了"。 我们的目标是在保护脆弱用户的同时给成人用户更多自由。我们有义务保护未成年用户，保护那些不在合理心态下的成人用户。 直播一开始，奥特曼就承认自己的错误。 过去，我们把AGI想象成"天上的神谕"，超级智能会自动为人类创造美好的事物。 但现在我们意识到，真正重要的是创造工具，让人们用这些工具创造自己的未来。 这种思维转变并非偶然，人类历史上每一次技术革命都源于更好的工具，从石器到蒸汽机，从计算机到互联网。 Ope ...

Core Viewpoint - Qualcomm has officially entered the data center market with the launch of two new AI chips, AI200 and AI250, aiming to compete with Nvidia and AMD in the AI accelerator space [2][6][7]. Group 1: Product Launch and Features - Qualcomm's AI200 and AI250 are designed as rack-level inference accelerators and systems, focusing on the inference phase of AI models, with the lowest total cost of ownership (TCO), higher energy efficiency, and enhanced memory processing capabilities [8][11]. - The AI200 is expected to be commercially available by 2026 and can be sold as a standalone chip or as part of a complete rack server system [11]. - The AI250, planned for release in 2027, features a new near-memory computing architecture that claims to provide over 10 times effective memory bandwidth improvement while significantly reducing power consumption [13]. - Both products support enterprise-level features such as direct liquid cooling, PCIe and Ethernet expansion, and confidential computing, targeting high-density rack scenarios [13]. Group 2: Market Context and Competitive Landscape - Qualcomm's entry into the data center market comes after a six-year gap since its last data center product, the AI100, which was primarily aimed at edge and lightweight inference [5][15]. - The global data center investment is projected to reach $6.7 trillion by 2030, indicating a lucrative market opportunity [20]. - Currently, Nvidia dominates the market with over 90% share, while AMD holds a smaller portion, leaving room for competitors like Qualcomm to capture market share [21]. Group 3: Strategic Positioning and Future Plans - Qualcomm has a history of technology accumulation in mobile chips, which has been leveraged in the development of the AI200 and AI250, utilizing advancements in its Hexagon neural processing unit (NPU) [17]. - The company plans to advance its data center product roadmap at a pace of one generation per year, continuously improving AI inference performance, energy efficiency, and overall TCO competitiveness [14]. - Qualcomm has already secured an order from Saudi AI startup Humain for deploying rack-level computing systems based on AI200/AI250, with a total power of up to 200 megawatts starting in 2026 [23].

鱼羊 发自 凹非寺 量子位 | 公众号 QbitAI 刚刚，OpenAI宣布已完成资本结构重组。 这就意味着， OpenAI上市的道路已经铺平 ，而软银前几天刚批准的225亿美元投资，也将顺利到账。 具体来说，OpenAI重组后，非营利主体 （即原本的OpenAI Nonprofit） 改名为 OpenAI Foundation ，继续掌控营利实体—— 持有营利实体的26%股份，目前估值约1300亿美元。 拿到期权的员工们和投资者则持有47%的股份。 而大金主微软，将在OpenAI营利实体中持有32.5%的股份。同时，OpenAI已同意额外购买价值2500万美元的微软Azure云服务。 美股盘前交易中，微软股价一度上涨3.5%。 使命仍是"确保AGI造福人类" OpenAI官方通告中表示，OpenAI在2015年时作为一家非营利机构成立，使命是" 确保通用人工智能能造福全人类 "。 今天，OpenAI仍是一家致力于相同使命的非营利组织。 OpenAI作为一家公司越成功，非营利组织的股权价值就越高，非营利组织将利用这些资金来资助慈善工作。 资本结构重组之后，OpenAI营利实体达到估值里程碑后，基金会还将获得额 ...

Time-HD团队 投稿 量子位 | 公众号 QbitAI 从原来几百维的小数据集到万维系统，时序预测模型面对真实世界，终于不用再"一碰就脆"。 这是来自全华人团队最新研究——业界首个专为高维时间序列预测设计的大规模基准 Time-HD 。 该基准涵盖神经科学、云计算、气象、金融等十个领域的16个数据集，变量数量最高可达两万。 团队还围绕该基准发布了首个高维时序开源框架 Time-HD-Lib ，提供了标准化预处理、统一评估策略、自动化超参数搜索以及分布式训练， 系统地填补了高维时序预测评测的空白。 同时提出预测模型 U-Cast ，不仅可以在多个数据集中将误差降低 15% ，训练速度还能提升近一倍。 下面是有关该基准的更多细节内容。 时序预测领域向高维迈进 时间序列预测来到真正意义上的"高维赛场"！ 从金融市场的上千支股票，到智慧城市交通网络的上万个传感器，毫无疑问我们正全面进入一个由高维时间序列数据驱动的时代。 然而，当前主流的时间序列预测 （TSF） 模型，大多仍停留在仅包含几个或几百个变量的低维环境 （如ETT、Traffic） 。 | Datasets | ETT Weather Solar ECL ...

Jay 发自 凹非寺 量子位 | 公众号 QbitAI 又一位休学创业的名校少女登场。 19岁的华人女生 Christine Zhang 从哈佛休学，创办了一家名为Veil的公司，刚完成种子轮， 被塞了100万美元 。 拿到这笔钱后，Christine表示可能还会再休学一年。 不得不说，「100万美元已到账」的邮件通知，好像确实要比哈佛学位证书更有吸引力一点哈。 下面一起来看看，这位年仅19岁的华人少女，连哈佛的课都不上了也要做的公司，究竟是什么来头？ Veil：讨AI喜欢的电商优化方案 一句话概括，Veil是一个专为电商卖家打造的智能优化平台。 简单来说，它能帮你把商品介绍「翻译」成 AI 能看懂的语言，让产品更容易出现在ChatGPT、Gemini、Grok等AI的搜索结果中，从而吸引 更多潜在客户。 具体而言，Veil会先分析你的商品详情，然后运用GEO （生成式引擎优化） 和 AEO （问答引擎优化） 两种技术，找出那些让AI「看不 清」、「听不懂」的地方。 接着，Veil会根据痛点给出一份具体可行的修复方案，比如：增加结构化的常见问题解答页面 （FAQ） 、补充权威信息来源…… 打个比方，如果客户卖 ...

Core Viewpoint - The article emphasizes the transformative impact of artificial intelligence (AI) on various industries and society, marking the beginning of a new era where AI becomes an integral part of infrastructure and daily life [1][7]. Group 1: AI Integration and Evolution - Intelligent technology has deeply penetrated production and daily life, evolving from mere tools to intelligent partners that understand human needs [2]. - AI technology is no longer confined to specific fields but transcends industry, discipline, and scenario boundaries, creating new ecosystems and opportunities [3]. - Emerging technologies such as multi-modal, AR/VR, and spatial computing are blurring the lines between the digital and physical worlds [4]. Group 2: MEET2026 Conference Overview - The MEET2026 Intelligent Future Conference will focus on the theme "Coexistence Without Boundaries, Intelligence to Inspire the Future," inviting leaders from technology, industry, and academia to witness industry transformation [5][7]. - This year marks the seventh edition of the MEET Intelligent Future Conference, which attracts thousands of technology professionals and millions of online viewers, establishing itself as an annual barometer for the intelligent technology industry [9][12]. - The conference will feature prominent figures such as Dr. Kai-Fu Lee and Professor Zhang Yaqin, along with leaders from major tech companies like Baidu, Alibaba, Tencent, and Huawei [9]. Group 3: AI Trends and Awards - The "Artificial Intelligence Annual List" initiated by Quantum Bit has become one of the most influential lists in the AI industry, aiming to recognize those who lead change and explore new frontiers [16]. - This year's awards will evaluate companies, products, and individuals across three dimensions, with results to be announced at the MEET2026 conference [17][18]. - The "2025 Annual AI Top Ten Trends Report" will also be released at the conference, highlighting significant AI trends and their potential impact [23][24].