Core Viewpoint - The article discusses the launch of Meituan's AI assistant, Xiaomei, which simplifies daily tasks such as ordering food and making restaurant reservations through natural language commands, eliminating the need for complex graphical interfaces [1][6][49]. Group 1: Functionality and User Experience - Xiaomei serves as a "small and beautiful" life secretary, efficiently handling daily needs and making life simpler [3][6]. - Users can interact with Xiaomei using voice commands, allowing for easy completion of tasks like ordering takeout and finding restaurants without navigating through multiple screens [7][9]. - The assistant can recommend food based on user preferences and past orders, acting as a "wish box" for meal suggestions [29][30]. Group 2: Technology and Data Integration - Xiaomei is powered by Meituan's LongCat model, which excels in natural language processing and can handle complex tasks due to its extensive training on real-world data [51][54]. - The integration of Xiaomei with Meituan's service system allows for seamless execution of tasks, ensuring that user requests are processed accurately and efficiently [58][60]. - The assistant is designed to learn from user interactions, adapting to individual habits and preferences over time, thus enhancing user experience [61][62]. Group 3: Comparison with Traditional Assistants - Unlike traditional AI assistants that require multiple clicks and operations, Xiaomei aims to create a more human-like interaction through natural dialogue [63][64]. - The assistant captures subtle changes in user habits and responds appropriately, fostering a sense of familiarity and understanding [65][66].

不圆 发自 凹非寺 量子位 | 公众号 QbitAI 网络搜索Agent效果不好，猛猛投喂一波数据，表现还那样，咋回事？ 港科大&MiniMax团队指出问题核心：不是模型参数不够多，而是缺乏足够有挑战性的训练数据。 换句话说，别死记硬背了，来做点"真题"吧。 他们提出了一种构建高质量QA对的方法 WebExplorer 。 用该方法构建的数据集去训练，即使是较小的模型，也可以在复杂、长程的搜索任务上超越更大的模型。 训练后的8B模型支持高达 128K的上下文长度 和 100次工具调用轮次 的长期推理，能在参数量低于10B的模型中取得顶尖结果。 网友评价：用模型驱动的方式做探索，确实比传统图谱方法更能让智能体的浏览行为变灵活。 模型及数据集均已开源，链接可见文末。 优质训练数据稀缺 随着大语言模型（LLM）的快速发展，智能体的能力边界不断扩展。 网络搜索智能体作为这一发展的重要组成部分，能够自主地从广泛的在线资源中检索信息；长视野（Long-Horizon）网络智能体更是需要在 多个网站间进行复杂的推理和搜索。 可是呢， 现有的开源网络智能体在处理复杂搜索任务时往往表现有限，更强大的商业模型又缺乏透明的训练细节 ...

iDLab团队 投稿 量子位 | 公众号 QbitAI 当大数据席卷各行各业，控制理论也迎来新的拐点：从依赖模型到 依赖数据 。 但是，在数据驱动控制领域，却缺乏一种 标准 化的数 据表示形式 。 针对这一问题， 清华大学李升波教授课题组（iDLab） 首次将现代控制理论中的标准型概念引入数据驱动控制（datatic control）范式， 提出了一种基于数据的系统描述新形式。 目前，该成果已发表于ACC2025。 从模型标准型到数据标准型 人工智能的蓬勃发展，离不开数据这一核心支柱。 近年来，随着人工智能技术的广泛应用，以数据为核心的系统表征方法迅速渗透到控制领域。 控制系统的设计方法正迎来一场从模型驱动向数据驱动的范式变革，即从传统的 模型驱动控制 （modelic control，即model-driven control）到 数据驱动控制 （datatic control，即data-driven control）。 每个标准形式的样本由必要的转移和可插拔的属性组成，分别用于描述系统变化规律和人为定义特征。 不仅如此，该数据标准型还可根据算法需求定制属性，显著加速控制器设计，为提高数据驱动算法效率提供 ...

鹭羽 发自 凹非寺 量子位 | 公众号 QbitAI AI生图再进化！图像分辨率直接卷到 2K 。 腾讯开源 混元图像2.1 （HunyuanImage2.1） ，画质直接拉满的同时，还能读懂千字长文本，甚至中英文混搭渲染。 或者搞个美洲驼的概念图，也是轻轻松松～ 新一代模型在技术上全面升级，不仅显著提升图文语义一致性和跨场景泛化能力，还能够精细控制场景、角色姿态甚至多物体描述，达成开源 生图模型中的SOTA。 模型开源之后，在Hugging Face趋势榜上一路飙升，目前已拿下第一名的宝座。 话不多说，先来看几个网友试玩感受一下。 首先康康真实场景下的表现，细腻的手部和脸部纹理，处理细节过关 海报制作上，文本渲染也相当干净。 还有每次必不可少的动漫风环节：魔女宅急便 （圆润猪咪版） 可以说，混元图像2.1更懂语义、更擅图文、更多风格、更高清画质…… 所以咱们这不赶紧上手体验一波。 四大亮点 打开官网，操作界面是酱紫的～选择需要生成的图像尺寸和数量，填写prompt （上限2048） ，就能秒获取超高分辨率图像。 我们体验了一下，总结下来这个模型有四大亮点。 亮点1：复杂语义生成能力强 得益于多样化的大规模图 ...

时令 发自 凹非寺 量子位 | 公众号 QbitAI Qwen下一代模型架构，抢先来袭！ Qwen3-Next 发布，Qwen团队负责人林俊旸说，这就是 Qwen3.5的抢先预览版 。 基于Qwen3-Next，团队先开源了Qwen3-Next-80B-A3B-Base。 模型参数80B，但训练成本连Qwen3-32B的 十分之一都不到 ，并且在32 k以上的上下文推理吞吐能达到后者的 十倍以上 。 基于这一模型，团队接连出手，同步开发并发布了两大新模型： Qwen3-Next-80B-A3B-Instruct ：在256K超长上下文处理任务中展现出显著优势。 Qwen3-Next-80B-A3B-Thinking ：在多项基准测试中超越闭源模型Gemini-2.5-Flash-Thinking。 同时，在保留的标准注意力层中，他们进一步引入了多项优化设计： 网友表示，这更新频率令人震惊。 混合注意力机制 混合注意力机制 高稀疏度MoE结构 稳定性优化 多token预测机制 线性注意力在长上下文处理中效率很高，但召回能力有限，而标准注意力计算开销大、推理效率低，单独使用均存在局限。 为此，Qwen团队引入Ga ...

Core Viewpoint - The article discusses the launch of Gaode's new feature, "Gaode Street Ranking," which aims to provide a more authentic ranking system for offline services based on real user behavior rather than manipulated ratings [1][3][4]. Summary by Sections Introduction of Gaode Street Ranking - Gaode has introduced a new credit system called "Gaode Street Ranking," which promises to bring 10.8 million consumers to offline service businesses daily [2]. - The ranking system is designed to combat the existing issues of fake reviews and manipulated ratings in the restaurant and service industry [3][4]. Unique Features of the Ranking System - The ranking is based on real user behaviors, such as navigation and visits, rather than artificial manipulation [3][4]. - Gaode's data includes 514.3 million users and 1.3 billion navigation instances over a year, covering a distance equivalent to 57 times around the Earth [12][68]. Data and AI Integration - The success of the Gaode Street Ranking is attributed to the integration of AI with Gaode's extensive historical data, which has been accumulated over 20 years [6][67]. - AI helps in processing user behavior data to ensure the authenticity of the rankings, making it difficult for users to manipulate the system [48][65]. Ranking Dynamics - The ranking system includes various categories such as "Top Performers," "Street Rankings," and "Popular Check-ins," which are updated daily to reflect real-time user preferences [14][15]. - The "Top Performers" list is updated annually, while the "Street Rankings" are refreshed daily, allowing for more dynamic and responsive rankings [15]. User Engagement and Experience - Users can only leave reviews after visiting a location, which helps maintain the integrity of the ratings [47][65]. - The system encourages genuine user engagement, allowing for a more accurate representation of popular venues [39][70]. Conclusion and Implications - The Gaode Street Ranking represents a significant shift in how offline services are evaluated, emphasizing authenticity and user experience [72]. - This initiative could set a precedent for other industries facing similar challenges with fake reviews and ratings [73].

Core Viewpoint - The article discusses the latest advancements and discussions in AI technology, highlighting key insights from industry leaders at the Inclusion Bund Conference, emphasizing the transition to an "experience era" in AI development and the importance of open-source strategies in AI competition [6][10][20]. Group 1: AI Development Insights - Richard Sutton, the Turing Award winner, states that the human data dividend is nearing its limit, and AI is entering an "experience era" focused on continuous learning, which will unlock greater potential [9][10][19]. - Sutton emphasizes the need for new data sources generated through direct interaction between intelligent agents and the world, moving beyond static knowledge transfer [12][14]. - The concept of "experience" in AI involves the exchange of observation, action, and reward signals, which is crucial for developing intelligence [16][17]. Group 2: Open Source and AI Competition - Wang Jian, founder of Alibaba Cloud, highlights that the choice between open-source and closed-source models has become a critical variable in AI competition, especially after the U.S. imposed export controls on closed-source models [20][25]. - He discusses the evolution of open-source from code sharing to resource sharing, which now includes data, computing power, and model weights, significantly lowering the barriers to entry in AI development [26][30]. - Wang Jian's recent project involves deploying a complete AI model into space, showcasing the potential of collaborative resource sharing in advancing AI capabilities [31][35]. Group 3: Future of AI Applications - Yuval Noah Harari, a historian and philosopher, warns that the pace of technological change should not overshadow the need for understanding and governance, emphasizing that true progress is measured by cooperation and empathy rather than speed [60][67][70]. - Zhu Xiaohu from Sinovation Ventures predicts a significant explosion in AI applications next year, particularly in low-code and no-code software, and encourages entrepreneurs to seize opportunities in the AI space [54][57]. - Wang Xingxing from Yushutech believes that the integration of AI and robotics is on the verge of a breakthrough, enabling robots to perform tasks autonomously, although challenges remain in data quality and model alignment [46][50].

Core Viewpoint - The article highlights the severe impact of funding cuts on scientific research in the U.S., particularly at UCLA, where even top mathematicians like Terence Tao are struggling to secure funding for their research and students [2][4][5]. Funding Cuts and Their Impact - On July 25, 2023, the U.S. government abruptly suspended funding from the National Science Foundation (NSF) and the National Institutes of Health (NIH) to UCLA, amounting to $500 million [4]. - Although a federal court restored some funding on August 12, the funds had not yet been disbursed by early September [5]. - The funding cuts have led to significant disruptions in ongoing research projects and have severely affected graduate students' opportunities for academic development [12][21]. Personal Experiences and Concerns - Terence Tao expressed greater concern for his students than for himself, emphasizing that the funding is crucial for their participation in academic conferences and career development [6][12]. - Tao has had to shift his focus from research to fundraising activities due to the funding crisis [7][19]. Broader Implications for the Research Ecosystem - The article discusses how the cuts have eroded the independence of the research ecosystem, with many ongoing projects being abruptly halted [12][15]. - Tao noted that the NSF's funding has historically facilitated collaboration between universities, laboratories, and industry, which is essential for scientific advancement [14]. Psychological Impact on Students - The funding instability creates an environment of uncertainty, negatively affecting students' mental health and their ability to focus on research [23]. - Tao highlighted that the loss of funding directly impacts students' scholarships and research assistant positions, which are vital for their livelihoods [21][22]. Future Considerations - In an interview, Tao expressed uncertainty about his future in the U.S. academic landscape, indicating a potential shift in his perspective on remaining in the country due to the funding issues [24].

henry 发自 凹非寺 量子位 | 公众号 QbitAI 究竟是啥论文？ 让模型在博弈中学习 总的来说，MSL这篇新论文的核心思想是通过一种 Language Self-Play （LSP)的方法，让大型语言模型 在没有额外训练数据的情况下实 现自我提升 。 这一方法旨在应对当前大语言模型高度依赖大规模、高质量训练数据，且训练数据有限所带来的困境。 为此，LSP将模型的学习过程设计成一个博弈框架，让同一个语言模型扮演两个角色进行对抗，从而实现无数据训练。 Meta超级智能实验室（MSL）又被送上争议的风口浪尖了。 不过，这次不是人事风波，而是他们的 第二篇 论文《Language Self-Play For Data-Free Training》被质疑 忽视前人研究、缺乏创新 。 具体来说，这两个角色分别是： 在对抗过程中，挑战者不断生成越来越刁钻的问题或指令，以降低解决者的预期回报；而解决者则必须努力理解并回答这些指令，以最大化自 身回报——这其实就是我们熟悉的极小极大博弈（minimax game）。 通过这样的对抗训练，模型能够在不断博弈中持续改进，逐步提升能力。 此外，与传统对抗训练不同，LSP让 ...

Core Viewpoint - The article discusses the career transition of Shunyu Yao, a prominent researcher from OpenAI, as he embarks on a new phase in the AI field, focusing on personal AI and the evolving landscape of AI development, which is now entering its "second half" [2][47]. Group 1: Background and Achievements - Shunyu Yao, a 29-year-old researcher, has an impressive academic background, including graduating from Tsinghua University and obtaining a PhD from Princeton, where he focused on natural language processing and reinforcement learning [4][22]. - His notable contributions to AI include the development of frameworks like Tree of Thoughts, SWE-bench, and ReAct, which enhance the reasoning and decision-making capabilities of language models [6][36]. Group 2: Career Transition - Yao's departure from OpenAI has been confirmed through various channels, and he is rumored to be considering entrepreneurship or joining another tech giant [3][51]. - His recent work emphasizes the shift in AI development from model-centric approaches to defining meaningful tasks and evaluating AI systems' performance in real-world scenarios [47][48]. Group 3: Philosophical Insights - Yao's approach to research is characterized by a cross-disciplinary perspective, drawing inspiration from various fields, which he believes is essential for innovation in AI [9][20]. - He advocates for the importance of language as a medium for reasoning and decision-making in AI, highlighting its role in enabling agents to generalize across different contexts [28][30].