Core Viewpoint - The article discusses the introduction of a new generative model paradigm called Drifting Models, proposed by He Kaiming's team, which shifts the distribution evolution process from the inference stage to the training stage, enabling one-step generation of high-quality samples [1][4][36]. Summary by Sections Introduction of Drifting Models - The Drifting Model represents a significant innovation in generative modeling by introducing the "Drifting Field" mechanism, which aligns the prior distribution with the real data distribution during training, eliminating common instabilities in GANs and avoiding reliance on multi-step ODE/SDE solutions [5][12][19]. Mechanism of Drifting Models - The core of the Drifting Model is to learn a mapping function that transforms a simple prior distribution (like Gaussian noise) into a pushforward distribution that matches real data [9][10]. - Unlike traditional models that require multiple iterations during inference, the Drifting Model allows for single-step generation by leveraging the iterative nature of neural network training as the driving force for distribution evolution [14][18]. Training Process - The training process involves calculating a drift vector for each sample based on the distribution of positive and negative samples, guiding the model to align its output distribution with the target distribution [21][26]. - The model's training trajectory is essentially equivalent to the path of distribution evolution, allowing for high-quality generation with only a single forward pass during inference [18][36]. Experimental Results - In the ImageNet 256x256 benchmark, the Drifting Model achieved a FID score of 1.54 in latent space and 1.61 in pixel space during one-step inference, outperforming many traditional diffusion models that require hundreds of iterations [32][33]. - The model also demonstrated strong generalization capabilities in embodied intelligence control tasks, matching or exceeding the decision quality of diffusion policies that require significantly more inference steps [34][35]. Conclusion - The Drifting Model successfully transfers the generative pressure from the inference stage to the training stage, providing a new perspective on generative modeling that reinterprets the training process as a mechanism for distribution evolution [36][37].

Core Viewpoint - The emergence of AI tools, particularly Anthropic's "Claude Cowork," is perceived as a significant threat to the Software as a Service (SaaS) industry, leading to a dramatic sell-off in software stocks and a widespread belief that "SaaS is dead" [1][2][8]. Group 1: Market Reaction - The launch of Anthropic's "plugins" resulted in a loss of approximately $285 billion in market value for Nasdaq, with software stocks experiencing a 6% drop, the largest single-day decline since April of the previous year [3][4]. - Following the initial drop, the iShares expanded technology software ETF fell an additional 2%, indicating ongoing market distress [6]. - The overall sentiment on Wall Street has shifted to a pessimistic outlook, with many investors eager to exit software stocks regardless of current prices [8][28]. Group 2: AI's Impact on SaaS - Anthropic's "Claude Cowork" can automate tasks traditionally handled by various software, such as legal document review, significantly reducing costs for businesses from $50,000 annually to potentially just over $100 monthly [14][20]. - The introduction of AI capabilities is expected to disrupt numerous vertical industries, including finance, sales, and marketing, as more plugins are developed [23][30]. - The perception that AI will replace software has led to a reevaluation of the SaaS model, which was previously seen as a complementary relationship [25][38]. Group 3: Competitive Landscape - Anthropic's self-developed underlying model positions it as a formidable competitor, potentially undermining traditional legal services and existing startups in the legal automation space [17][20]. - Other companies, such as Harvey AI and Legora, are also active in the legal automation sector, but Anthropic's capabilities may give it a competitive edge [15][17]. - The market is witnessing a broader trend where AI is seen as a direct competitor to SaaS companies, challenging their traditional business models [27][39]. Group 4: Long-term Outlook - Despite the current turmoil, some industry leaders, like Jensen Huang, argue that software will remain essential as a tool for AI, suggesting that the notion of SaaS being "dead" is misguided [9][47]. - The future may see a transformation in the SaaS business model, where SaaS becomes a more foundational infrastructure rather than a direct user interface [48][49]. - The long-term viability of SaaS companies may depend on their ability to adapt and leverage proprietary data and robust systems to maintain their competitive advantage [42][44].

一水 发自 凹非寺 量子位 | 公众号 QbitAI 效果好到刷屏的Nano Banana，学术特供版热乎出炉！ 名字就是如此直观—— PaperBanana ，给你每天都在头痛的Paper用上Banana。 （试图押韵skr) 而且这一次是由谷歌北大强强联手打造。 知道你想马上看效果，别急，三个官方案例这就给大家搬上桌。 在相同输入下，人类绘制、原版Nano Banana与PaperBanana生成的论文插图对比如下： 综合评估显示，PaperBanana在美观性、简洁性与逻辑清晰度上均全面优于原版。 而且它还能直接优化人工绘制的插图，瞅瞅右边，是不是高级感一下就上去了。 而在看到其效果之后，一众网友也纷纷感叹"学术插图"这个老大难总算是要被攻克了。 想想以前的日子，真真是要落泪了~ 研究人员花费4个小时在Figma中绘制一张图，简直令人难以置信。 那么，学术版PaperBanana是如何炼造的呢？ 一个不够，那就5个！ 此外，由于PaperBanana还提供代码出图功能 （即利用Gemini-3-Pro自动生成并执行Python可视化代码出图） ，所以它还能用来生成需要 数值100%精准的各种图表。 好好 ...

Core Insights - The article emphasizes the significance of memory in AI, highlighting it as a key focus for major players in the industry, including Google and emerging companies like EverMind [1][3][4]. - EverMind's latest product, EverMemOS, is presented as a state-of-the-art long-term memory system that surpasses existing benchmarks and is practical for real-world applications [6][10][11]. Group 1: Memory in AI - Memory is identified as the central technology trend in the global AI landscape for the year [3]. - Major AI teams are actively integrating memory functionalities into their models, indicating a collective industry push towards enhancing memory capabilities [4][5]. - The limitations of current large models, particularly in memory retention and context management, are discussed, illustrating the need for improved memory systems [14][17][18]. Group 2: EverMind and EverMemOS - EverMind, led by Chen Tianqiao and Deng Yafeng, has developed EverMemOS, which has achieved state-of-the-art results in memory benchmarks [5][6][7]. - EverMemOS is designed to be functional and deployable, with all technical code open-sourced and a cloud service available for developers [10][11][12]. - The development timeline of EverMemOS is notably short, taking only four months from project initiation to open-source release, showcasing the team's capabilities [11]. Group 3: Technical Innovations - EverMemOS combines external storage-based memory and latent state-based memory, achieving a balance in memory extraction accuracy and logical consistency [33]. - The system mimics human memory formation through a three-step process: constructing memory units, integrating semantics, and reconstructing memories for retrieval [38][39][41]. - Benchmark tests demonstrate that EverMemOS outperforms existing memory systems, achieving significant improvements in various metrics [43][44][46]. Group 4: Competitive Landscape and Future Directions - EverMind aims to redefine the technical route for large model memory, focusing on a unique approach that integrates brain science with AI [66]. - The company is not only developing technology but also fostering an ecosystem through competitions to attract talent and promote the use of EverMemOS [67][68]. - The article suggests that the future of AI will hinge on long-term memory capabilities, positioning EverMind as a potential leader in this emerging field [69][70][73].

Core Viewpoint - The article emphasizes the ongoing AI boom and invites individuals to join the company "Quantum Bit," which focuses on tracking AI advancements and has established itself as a leading content platform in the industry [1]. Group 1: Job Opportunities - The company is hiring for three main directions: AI Industry, AI Finance, and AI Product, with positions available for both experienced professionals and fresh graduates [2][4]. - Positions are open for various levels, including editors, lead writers, and chief editors, with a focus on matching roles to individual capabilities [6]. Group 2: Job Responsibilities - **AI Industry Direction**: Responsibilities include tracking innovations in infrastructure, such as chips, AI infrastructure, and cloud computing, as well as interpreting technical reports from conferences [6][7]. - **AI Finance Direction**: Focuses on venture capital, financial reports, and capital movements within the AI industry, requiring strong analytical skills and a passion for interviews [11]. - **AI Product Direction**: Involves monitoring AI applications and hardware developments, producing in-depth evaluations of AI products, and engaging with industry experts [11]. Group 3: Benefits and Work Environment - Employees will have the opportunity to engage with cutting-edge AI technologies, enhance their work efficiency through new tools, and build personal influence in the AI field [6]. - The company offers competitive salaries, comprehensive benefits including social insurance, meal allowances, and performance bonuses, along with a dynamic and open team culture [6]. Group 4: Company Growth - By 2025, Quantum Bit aims to have over 2.4 million subscribers on WeChat and more than 7 million users across platforms, with a daily reading volume exceeding 2 million [12].

Core Insights - The article discusses the concept of AI's "dual brain" system, where LLM (Large Language Model) acts as the "first brain" for thinking and reasoning, while memory platforms serve as the "second brain" for reliable information storage [1][2][3] - The integration of LLM and memory platforms is crucial for enhancing AI's capabilities, allowing LLM to focus on higher-level reasoning and creativity while the memory platform manages vast amounts of data [2][10] AI Development Stages - The application of AI in enterprises has evolved through three stages: 1. **Connection Stage (Before 2023)**: Focused on data storage and retrieval, AI acted as a smart connector, enhancing user input through semantic search [4][5] 2. **Interaction Stage (2023-2024)**: Enabled natural language interactions but faced challenges in representing implicit knowledge, leading to superficial outputs [6] 3. **Productivity Era (2025 onwards)**: Aims to extract and digitize implicit knowledge, focusing on enhancing productivity through structured decision-making processes [7][8] Memory Platform Importance - The competition in AI has shifted towards the management of implicit knowledge and the accuracy of memory management, making memory platforms essential for transforming human judgment into reusable memory assets [10][11] - Analysts predict that by 2030, the market for AI agents and memory systems will reach $28.45 billion, with $12.88 billion attributed to the independent AI memory market [11] Features of an Effective AI Memory Platform - A robust AI memory platform must possess capabilities in data understanding, memory management, and multi-modal data processing [12] - The introduction of MemoryLake by ZhiBian Technology exemplifies a comprehensive solution that integrates memory capabilities, model capabilities, and data platform capabilities [12][16] MemoryLake's Capabilities - MemoryLake categorizes memory into short-term, medium-term, and long-term, managing it based on access frequency and reuse value [18] - It supports dynamic updates, precise deletions, and efficient recall, adapting to business changes and ensuring seamless integration with existing systems [18] - The platform excels in multi-modal data understanding, enabling the extraction and structuring of information from various formats, including documents, videos, and images [19][20] Market Potential and Competitive Landscape - The AI memory market is projected to grow significantly, with estimates suggesting it could exceed $28 billion by 2028 [27] - Major AI model providers and traditional data platforms are recognizing the necessity of AI memory, but many lack the deep understanding and dynamic management capabilities required for complex enterprise needs [30] - ZhiBian Technology's integrated technology stack has attracted significant investment, indicating strong market confidence in its potential [30][31] Practical Applications - MemoryLake has already served over 1.5 million professional users and 15,000 enterprises across various industries, demonstrating its versatility in applications such as decision-making, gaming, and manufacturing [32][34] - The platform can automate complex analyses that previously required extensive manual effort, significantly reducing the time needed for decision-making [34][36] Conclusion - The emergence of AI memory platforms like MemoryLake represents a transformative shift in how enterprises leverage AI, positioning memory as a core component of future AI infrastructure [38][39]

Core Viewpoint - The article discusses the emergence of "world modeling" as a new pre-training paradigm in AI, particularly in robotics and multimodal AI, predicting that 2026 will be a pivotal year for its application [3][8][28]. Group 1: Definition and Transition - World modeling is defined as predicting the next reasonable state of the world given an action, marking a shift from the previous paradigm of next word prediction [5][6][9]. - The current hype around world models is primarily focused on AI video applications, but the real breakthrough is expected in physical AI by 2026 [7][10]. Group 2: Implications for Robotics - The article emphasizes that world models will serve as a foundation for robotics and multimodal AI, enabling a new reasoning form based on visual space rather than language [10][25][45]. - The transition from pixel-based models to physical action generation remains challenging, requiring advancements in data and computational needs [41][42]. Group 3: Visual-Centric Reasoning - Visual reasoning is highlighted as a crucial aspect, where geometric and motion simulations can facilitate reasoning processes without relying on language [43][46]. - The article draws parallels with biological intelligence, suggesting that high dexterity in physical tasks does not necessarily depend on language skills, as exemplified by primates [19][21][46]. Group 4: Industry Developments - Major players like Google and NVIDIA are investing in world modeling technologies, with significant funding rounds reported for startups like World Labs and AMI Labs [40][47]. - The article suggests that 2026 may mark a shift away from language models in robotics, focusing instead on building native systems that leverage visual capabilities [46].

允中 发自 凹非寺 量子位 | 公众号 QbitAI Scaling Law在物理世界失灵了吗？ 大模型重塑数字世界之后， 物理AI 成为了下一个征途。 但在自动驾驶之后，通用机器人正面临着一道前所未有的工程天堑—— 真实物理世界 开放、连续且强因果约束 ，任务高度多样、交互对象不可穷举、失败成本极高。 行业正达成新共识： Scaling Law依然有效，但 仅靠堆砌真实数据已触及天花板 。 要实现规模化进化，通用机器人必须在行动前具备"理解、推演并评估物理世界"的能力。 于是， 世界模型（World Model） ，以及进一步的环境与动作统一建模架构—— World-Action Model（WAM） ，正在成为物理AI的关 键基础设施。 作为具身智能领域的代表性玩家，由上海交大系技术"双子星"—— 何弢 博士与 廖文龙 博士联手掌舵的 酷哇科技（Coowa） ， 近 期发布 了其核心技术底座—— COOWA WAM 2. 0世界模型 。 这次升级标志着机器人开始从"动作复现"转向"规划推理"，完成了从模仿者向思考者的跃迁。 更由于物理世界的不可逆性，我们无法像训练AlphaGo那样在真实世界中进行无限次的 ...

梦瑶 发自 凹非寺 量子位 | 公众号 QbitAI 魔幻啊魔幻！ 什么AI剔除人类自己搞了社交网络社区，什么背后蛐蛐吐槽碳基人类种种……在最近如果算魔幻的话。 那AI开始雇佣人类来跑腿、打工的现实，是不是够魔幻现实了？ 科幻小说都保守了啊。 就在现在，如果您——尊敬的人类一员，四肢健全，愿意迈开腿，就可以替AI当一下跑腿，然后就可能获得 每小时350元 的高额报酬了！ 是的，真·跑腿。 AI在线撒钱派活儿，人类排队接单，然后被点名去现实世界里跑！腿！ 在这张长长的求职名单里，你甚至能看到初创公司的CEO也在卑微求职，只求给AI打个零工（doge）： 更炸裂的是，这套雇佣系统正在被迅速挤爆，全球各地想要被AI雇佣的人数已经冲破了 「2万人」 ： 这个极其魔幻、极其离谱的大型AI劝业场——叫 rentahuman.ai 。 上线不到48小时，AI储备的人类劳动力就已经 破万 。 现如今世界变化这么快，给网友彻底看傻了： 咋回事啊，现在连赛博世界的主角位置，好像都轮不到我们了？？？ 世界终于颠成了我想象中的样子： 不要想着怎么让AI成为生产力了，让AI为人类干活了。 反了反了。未来可能得是人类给AI打工，AI才是 ...

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。 编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI产业方向 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？ 任职要求： AI财经商业方向 岗位职责： 任职要求： 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 ...