微软前AI副总裁、目前在OpenAI研究AGI的Sebastien Bubeck激动分享了这一消息，并表示： 30年悬而未决的数学难题就这样被AI证明了？！ 此时此刻， （前推特） 正在刮起一股讨论之风—— 来自Harmonic的数学AI模型独立证明了 Erdős问题 #124 ，而这个问题已经被数学家无奈搁置了近30年。 一水 发自 凹非寺 量子位 | 公众号 QbitAI 该解决方案100%由AI生成，总计耗时6小时。 甚至连陶哲轩这样的顶尖数学家也跑来围观讨论，他在对比了Gemini和ChatGPT的深度研究工具后发现，Harmonic模型对该问题的证明表 现更佳。 所以这到底是一个怎样的问题？Harmonic模型又是如何"大显神功"？ 咱接着瞧—— AI证明了Erdős问题 #124简易版 首先需要提醒，在听完各路大神讨论后，我们才意识到—— 原来Harmonic模型所证明的并非原版Erdős问题 #124 ，而是一个简易版本 。 Erdős问题 #124需要提供的证明如下 ： $$\sum_{1\leq i\leq k}{\frac{1}{d_{i}-1}}\geq1.$$ 通俗理解即为： 假设你有 ...

Core Viewpoint - The article discusses the advancements in Vidu Q2, a product from Shengshu Technology, highlighting its superior consistency and new features in AI-generated images and videos, positioning it as a competitive alternative to established players like OpenAI and Google [8][9][57]. Group 1: Product Features - Vidu Q2 has upgraded its reference image generation capabilities, claiming to have the industry's strongest consistency, allowing for repeated edits while maintaining character and object integrity [8]. - The new features include text-to-image generation and image editing, enabling users to create images with simple prompts, comparable to advanced editing software [9][35]. - Vidu Q2's image editing function allows users to change image proportions and details without complex processes, making it user-friendly and efficient [37][46]. Group 2: Performance Comparison - In a performance comparison, Vidu Q2 ranked fourth in the latest AA leaderboard, surpassing OpenAI and competing closely with major companies like Google and ByteDance [9]. - The article emphasizes that Vidu Q2 maintains high consistency in image generation, outperforming competitors like Nano Banana Pro in preserving background and structural details [20][29]. Group 3: User Experience and Accessibility - Vidu Q2 offers a one-month free membership for its new features, making it accessible for users to explore its capabilities [11]. - The platform provides a streamlined workflow for creators, allowing seamless transitions between image and video generation, which reduces the trial-and-error costs associated with content creation [52][57].

Core Insights - The article discusses the advancements in diffusion models, particularly focusing on the ShortDF and LeMiCa papers, which represent significant breakthroughs in the field of image and video generation [1][2][4]. Group 1: Technical Evolution - ShortDF serves as a theoretical pioneer in optimizing diffusion models through online training, while LeMiCa expands this theory into offline mapping for higher-dimensional tasks [4]. - The core challenge in diffusion models is the expensive inference costs, which hinder real-time applications [8]. - The non-linear denoising trajectory of diffusion models is identified as a primary reason for slow progress in the field [9]. Group 2: ShortDF's Mechanisms - ShortDF introduces a "shortest path optimization" approach to directly straighten the denoising trajectory during training, aiming to break the trade-off between speed and quality [12]. - The model's core insight is that the denoising process is fundamentally a correction of the initial error, which can be minimized to improve overall performance [13][14]. - ShortDF employs a three-pronged strategy: 1. Locking the "error upper bound" to optimize from the source [14][15]. 2. Utilizing graph theory to relax and compress paths, thereby minimizing the error upper bound [20][21]. 3. Implementing multi-state optimization to ensure training stability amidst random noise [28][29]. Group 3: Performance Metrics - ShortDF demonstrates superior performance in speed and quality, achieving a 5.0 times speed increase over DDIM while improving image quality (FID score of 9.08 compared to DDIM's 11.14) [36]. - The model shows robustness in complex scenarios, effectively restoring object contours faster than competing methods [37]. - In various datasets, ShortDF maintains a balance between performance and speed, showcasing its potential for real-world applications [40]. Group 4: Industry Implications - The advancements in ShortDF and LeMiCa highlight the importance of refined mathematical modeling over mere computational power in enhancing diffusion model speeds [41]. - These developments are crucial for the application of AIGC technology in resource-constrained environments, such as mobile devices and real-time interactive designs [42].

梦晨 发自 凹非寺 量子位 | 公众号 QbitAI ChatGPT广告代码泄露，就在发布三周年之际，终于要开始变现了？ 在ChatGPT安卓APP测试版的代码中，发现了多个与广告相关的引用。 就在2024年5月奥特曼还称ChatGPT加广告是 "最后的手段"和"令人不安的" 。 到2025年10月，他已经改口 "觉得广告有点令人反感，但并非完全不可取" 。 到现在从技术细节来看，这次的广告系统已经相当成熟，距离正式上线不远了。 OpenAI的运营模式面临着巨大的财务压力，探索广告变现终于还是提上了日程。 汇丰银行最近给OpenAI算了一笔账，运营ChatGPT的开销中仅维持其算力基础设施每年就可能需要数千亿美元。 目前ChatGPT Plus订阅服务每月收费20美元，加上API授权收入，远远无法覆盖这些成本。 OpenAI在2029年之前将持续处于亏损状态，累计亏损可能超过1000亿美元。 ChatGPT变现，奥特曼变脸 开发者Tibor首次曝光了这个重磅消息。他在ChatGPT安卓应用1.2025.329测试版的代码中，发现了多个与广告相关的引用。 代码中明确出现了"ads feature"字样，还包含了 ...

Core Viewpoint - The article discusses the launch and performance of Alibaba's new image generation model, Z-Image, which has quickly gained popularity and recognition in the AI community due to its impressive capabilities and efficiency [1][3]. Group 1: Model Overview - Z-Image is a 6 billion parameter image generation model that has achieved significant success, including 500,000 downloads on its first day and topping two charts on Hugging Face within two days of launch [1][3]. - The model is available in three versions: Z-Image-Turbo (open-source), Z-Image-Edit (not open-source), and Z-Image-Base (not open-source) [8]. Group 2: Performance and Features - Z-Image demonstrates state-of-the-art (SOTA) performance in image quality, text rendering, and semantic understanding, comparable to contemporaneous models like FLUX.2 [3][8]. - The model excels in generating realistic images and handling complex text rendering, including mixed-language content and mathematical formulas [6][15]. - Users have reported high-quality outputs, including detailed portraits and creative visual interpretations, showcasing the model's versatility [11][14][32]. Group 3: Technical Innovations - Z-Image's speed and efficiency are attributed to its architecture optimization and model distillation techniques, which reduce computational load without sacrificing quality [34][39]. - The model employs a single-stream architecture (S3-DiT) that integrates text and image processing, streamlining the workflow and enhancing performance [35]. - The distillation process allows Z-Image to generate high-quality images with only eight function evaluations, significantly improving generation speed [40][42]. Group 4: Market Position and Future Prospects - The timing of Z-Image's release is strategic, coinciding with the launch of FLUX.2, indicating a competitive landscape in the AI image generation market [44]. - The model's open-source availability on platforms like Hugging Face and ModelScope positions it favorably for further adoption and experimentation within the AI community [45].

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 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 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 ...

Core Insights - The article discusses the emergence of Temvideo, an AI video agent developed by Boolvector, aimed at addressing the marketing challenges faced by cross-border e-commerce businesses. The product enhances video production efficiency and reduces costs while maintaining high ROI for users [6][11]. Group 1: Product Overview - Temvideo is the world's first AI video agent designed specifically for marketing scenarios, targeting the pain points of low efficiency and high costs in video production for cross-border e-commerce [11]. - The core functionality of Temvideo includes batch video generation using verified high ROI templates, significantly reducing production time while achieving quality comparable to human-made videos [11][12]. - The product is designed to cater to e-commerce users with annual revenues between 10 million and 100 million, focusing on their advertising needs and ensuring high click-through and conversion rates [12][22]. Group 2: Unique Features and Advantages - Temvideo's design incorporates industry know-how, allowing it to generate effective marketing videos based on successful past campaigns, thus enhancing the quality of output [12][36]. - The product utilizes a combination of large models and industry-specific algorithms to improve video content understanding and production accuracy, addressing the limitations of generic AI models [30][32]. - Temvideo's ability to automatically segment video clips and match background music enhances the overall video quality, meeting the high detail requirements of merchants [29][30]. Group 3: Market Context and Trends - The article highlights that only about 10% of e-commerce businesses currently utilize AI video and image generation technologies, indicating significant room for growth in this sector [71]. - The demand for high-quality video content on social media platforms is increasing, with platforms like TikTok and Meta requiring more engaging and effective video advertisements [75][76]. - The potential market for AI-generated video content is substantial, with two primary business models: charging per video produced or sharing revenue based on performance metrics [78][79]. Group 4: Challenges and Future Directions - The article notes that many AI products face challenges in user retention due to high expectations and the complexity of AI capabilities, which can lead to unsatisfactory results [86]. - Boolvector aims to balance result delivery and cost control, focusing on optimizing the video generation process to ensure user satisfaction and retention [92][93]. - The future vision for Temvideo includes transitioning from a pay-per-video model to a performance-based payment system, fostering a sustainable business model that aligns with user success [95][98].

Core Insights - The article discusses a significant paradigm shift in AI, indicating that the development of AI is not slowing down but rather transitioning to a new phase of growth [1][7][12]. Group 1: AI Development Trends - There are two contrasting views on AI development: one claims that AI growth is slowing down, while the other highlights continuous advancements with new models like GPT-5.1 and Gemini 3 being released [3][12]. - Łukasz Kaiser argues that the perception of slowing growth is incorrect, stating that AI's capability growth follows a smooth exponential curve, akin to Moore's Law [15][16]. - The shift from pre-training to reasoning models is a key factor in this transition, with pre-training being in a later stage of its S-curve while reasoning models are still in their early stages [18][19]. Group 2: Reasoning Models and Their Impact - The industry is focusing on smaller, cost-effective models that maintain quality, leading to the misconception that pre-training has stalled [21]. - Reasoning models, which allow for more complex thought processes and the use of tools during inference, are expected to progress rapidly due to their emerging nature [22][27]. - The evolution of models like ChatGPT demonstrates a qualitative leap in performance, with newer versions incorporating reasoning and external tool usage for more accurate responses [23][24]. Group 3: GPT-5.1 Insights - GPT-5.1 is not merely a minor update but represents a significant stability iteration, enhancing reasoning capabilities through reinforcement learning and synthetic data [34][35]. - The naming convention for versions has shifted to focus on user experience rather than technical details, allowing for greater flexibility in development [38]. - Despite improvements, GPT-5.1 still has limitations, particularly in multi-modal reasoning, as illustrated by its struggles with basic tasks that require contextual understanding [41][42]. Group 4: Future of AI and Robotics - AI is expected to change the nature of work without eliminating jobs, as human expertise will still be needed in high-stakes scenarios [62][66]. - Home robots are anticipated to be the next visible AI revolution, driven by advancements in multi-modal capabilities and general reinforcement learning [67][69]. - The integration of these technologies is expected to lead to a significant leap in the capabilities of home robots, making them more intuitive and perceptible compared to current AI models like ChatGPT [69].

衡宇 发自 凹非寺 量子位 | 公众号 QbitAI ICLR 2026，居然有21%的评审是纯纯由AI生成的？！ 上面这个相当扎心的答案，来自Pangram实验室的分析报告。 这件事被发现的起因颇具戏剧色彩 ：CMU的AI研究员Graham Neubig，感觉自己收到的同行评审AI味超级重。 他之所以起疑心，是因为这些评审内容"非常冗长，且包含大量符号"，并且所要求的分析方式并非"审稿人通常在AI或ML论文中所要求的那种 标准统计分析方式"。 做事嘛，不能光靠直觉，要真凭实据啦。 Graham Neubig自己干不了这个事儿，就 在上发布了一个悬赏令，希望有人能做一轮系统性的检测 ，看ICLR的论文和审稿中到底夹杂了 多少AI文本。 我愿意悬赏50美元，给第一个做了这件事的人～ Pangram实验室就是那个接黄榜的。 这个实验室的业务之一，正好是开发检测AI生成文本的工具。 结论简单粗暴： 一个顶级AI学术会议，审稿和投稿两头都出现大规模AI代写…… 是怎么测出"AI味"的？ Pangram这次对ICLR的全部提交论文和所有评审做了系统分析，并且在博客中公开了全过程。 他们 先在OpenReview上，把I ...

Core Insights - The article discusses the launch of GELab-Zero, an open-source GUI Agent model that allows for easy deployment and aims to enhance the scalability of mobile agents in various applications [1][8]. Group 1: Model Performance and Capabilities - The 4B version of the GUI Agent model has achieved state-of-the-art (SOTA) performance across multiple GUI benchmarks on both mobile and desktop platforms [2][11]. - GELab-Zero-4B-preview outperforms other mainstream models, including larger parameter models like GUI-Owl-32B, demonstrating superior performance and easier deployment [13][11]. - The model is designed to handle complex tasks and vague instructions effectively, showcasing its versatility in various applications [19][24]. Group 2: Development and Deployment - The article emphasizes the need to lower development and usage barriers for mobile agents, allowing developers to focus on value creation rather than infrastructure setup [7][30]. - GELab-Zero includes a complete technical architecture that enables one-click deployment, facilitating a seamless experience for developers [25][26]. - The model supports lightweight local inference, enabling it to run on consumer-grade hardware while maintaining low latency and privacy [26]. Group 3: Evaluation Standards - The research team has established a new evaluation standard called AndroidDaily, which focuses on real-world applications and user scenarios, moving beyond traditional productivity benchmarks [5][31]. - AndroidDaily assesses the model across six core dimensions of modern life, including dining, travel, shopping, housing, information consumption, and entertainment [33]. - The evaluation framework includes both static and end-to-end testing methodologies to ensure comprehensive assessment of the model's capabilities [35][38]. Group 4: Future Directions - The research team aims to continue optimizing model performance, expanding cross-platform support, and enriching the ecosystem of tools while adhering to principles of openness, control, and privacy [41].