新任腾讯CEO/总裁办公室首席AI科学家姚顺雨1月10日首次公开露面，以嘉宾身份出席清华大学北京重点实验室主办的AGI-Next前沿峰会，姚顺雨此前曾 任OpenAI研究员。他表示，在OpenAI之前，他在一个公司实习过，这是一个toB的公司，他觉得在toB公司工作过有很多收获，最大的收获是即使今天的 模型不再变好，所有的模型训练全部停止了。但是我们把这些模型部署到世界上各种各样的公司，已经能带来今天10倍或者100倍的收益，能应对GDP产 生5%-10%的影响，但是今天它对GDP的影响还不到1%。 另外他觉得教育非常重要，他观察现在人和人的差距非常大，更多时候不是说人类替代了人类工作，而是会使用这些工具的人在替代那些不会使用工具的 人，就像当年电脑出来，如果转身学习编程跟你还在持续计算尺、使用算法，差距是巨大的。今天中国能做到的最大的有意义的事情是更好的教育，教育 大家怎么更好的使用像Claude或者ChatGPT这样的产品，在国内我们可以用Kimi或者智谱这样的国产模型。 深圳最新引入的天才少年顶尖科学家 姚顺雨入职腾讯后首次公开发声！ 1月10日，在由清华大学基础模型北京市重点实验室、智谱AI发起的AG ...

Core Insights - The competition in large models has shifted from "Chat" to "Agent," focusing on executing complex tasks in real environments rather than just scoring on leaderboards. The industry anticipates 2026 as the year when commercial value will be realized, with a technological evolution towards verifiable reinforcement learning (RLVR) [2][4][5]. Group 1: Competition Landscape - The engineering challenges of the Chat era have largely been resolved, and future success will depend on the ability to complete complex, long-chain real tasks. The core value of AI is transitioning from "providing information" to "delivering productivity" [4]. - The bottleneck for Agents lies not in cognitive depth but in environmental feedback. Future training paradigms will shift from manual labeling to RLVR, enabling models to self-iterate in systems with clear right or wrong judgments [5][6]. - The industry consensus suggests that while China has a high chance of catching up in the old paradigm (engineering replication, local optimization, toC applications), its probability of leading in new paradigms (underlying architecture innovation, long-term memory) is likely below 20% due to significant differences in computational resource allocation [5][11]. Group 2: Strategic Opportunities - Opportunities for catching up lie in two variables: the global shift towards "intelligent efficiency" as scaling laws encounter diminishing returns, and the potential paradigm shift driven by academia around 2026 as computational conditions improve [5][19]. - The ultimate variable for success is not leaderboard scores but the tolerance for uncertainty. True advancement depends on the willingness to invest resources in uncertain but potentially transformative new paradigms rather than merely chasing scores in the old paradigm [5][10]. Group 3: Perspectives from Industry Leaders - Industry leaders express cautious optimism regarding China's potential to lead, with probabilities of success varying. For instance, Lin Junyang estimates a 20% chance of leading due to structural differences in computational resource allocation and usage [11][12]. - Tang Jie acknowledges the existing gap in enterprise AI lab research but bets on a paradigm shift occurring around 2026, driven by improved academic participation and the emergence of new algorithms and training paradigms [15][19]. - Yang Qiang believes that China may excel in toC applications first, drawing parallels to the internet history, while emphasizing the need for China to develop its own toB solutions to bridge existing gaps [20][24]. Group 4: Technological Innovations - The future of AI will require advancements in multi-modal capabilities, memory structures, and self-reflective abilities, which are essential for achieving higher levels of intelligence and functionality [68][70][73]. - The introduction of new optimization techniques, such as the MUON optimizer, aims to enhance token efficiency and long-context processing, which are critical for the performance of agent-based models [110][116]. - The development of linear attention mechanisms is expected to improve efficiency and performance in long-context tasks, addressing the limitations of traditional attention models [116]. Group 5: Future Directions - The industry is focused on distinguishing between scaling known paths through data and computational increases and exploring unknown paths to discover new paradigms [98][99]. - The potential for AI to participate in scientific research is anticipated to expand significantly, opening new possibilities for innovation and application [101].

Core Insights - The article discusses the evolving landscape of Artificial General Intelligence (AGI) and highlights three key trends shaping its future development in China and the U.S. [10] Group 1: Trends in AGI Development - Trend One: Beyond Scaling, a New Paradigm is Emerging - The discussion around Scaling has shifted from whether to continue expanding model sizes to questioning the value of such investments. Efficiency has become a critical concern as the marginal returns on increased computational power diminish [14][15]. - Trend Two: Token Efficiency is Becoming a Decisive Factor - Token efficiency has emerged as a crucial variable in determining the potential of large models. The ability to utilize tokens effectively is now seen as essential for achieving higher intelligence levels and completing complex tasks [20][22][24]. - Trend Three: Diverging Evolution Paths for Chinese and American Models - The development of large models in the U.S. is increasingly focused on productivity and enterprise applications, while in China, the emphasis is on cost sensitivity and stability. This divergence reflects different market demands and cultural approaches to research and development [26][28][29]. Group 2: Key Discussions and Insights - The AGI-Next summit gathered leading figures in AI to discuss the future of AGI, emphasizing a shift from application-level discussions to foundational questions about the direction of next-generation AGI [6][10]. - The consensus among researchers indicates that the next phase of AGI development will require a reevaluation of existing paradigms, with a focus on efficiency and the role of token utilization in model performance [10][11][20]. - The cultural differences between U.S. and Chinese AI research environments contribute to the distinct paths taken by their respective large model developments, with U.S. labs often pursuing high-risk, high-reward projects, while Chinese labs focus on practical applications and efficiency [29].

智通财经记者 | 陆柯言 佘晓晨 智通财经编辑 | 文姝琪 1月10日，一场关于AI的顶级对话在北京中关村悄然上演。 这场对话由清华大学基础模型北京市重点实验室、智谱AI共同发起。最重要的是，它罕见地集结了中 国大模型领域最受关注的几位"顶流"人物。他们分别是： 唐杰：清华大学教授、智谱创始人，三天前刚刚带领智谱在港股敲钟，造就"中国基模第一股"。 姚顺雨：前OpenAI研究员，27岁的他现为腾讯CEO办公室首席科学家、 AI Infra 部及大语言模型部负责 人。这也是入职腾讯后的公开首秀。 林俊旸：阿里巴巴通义千问大模型负责人，阿里最年轻P10级技术专家，掌舵着全球开源生态中下载量 第一的阿里通义系列。 几位平均年龄不到35岁的年轻人，几乎握着中国AGI赛道上最贵的一批筹码。 如此规格的同台向来罕 见，不仅因为他们背后代表着大厂与创业公司的生态博弈，更因为在"Chat（对话）范式"已成存量的今 天，他们对下一代 AGI 路线图的任何一次押注，都可能决定未来十年的行业座次。 从大模型的下一个"奇点时刻"，到对模型分化的观察；从AI Agent的未来，到中国AI的胜算，这场万字 交锋试图在现实算力与商业落地的 ...

Core Insights - The AGI-Next summit highlighted the challenges and opportunities for Chinese large model companies, featuring prominent figures in AI discussing new paradigms and advancements in technology [2][4]. Group 1: AI Market Dynamics - The Chinese large model market is showing significant differentiation between To C (consumer) and To B (business) segments, with distinct underlying logic for each [4]. - In the To C market, most users do not require high intelligence from models, leading to a trend of vertical integration where model and application layers are closely coupled for better user experience [4][5]. - Conversely, in the To B market, higher intelligence correlates with increased productivity and willingness to pay, creating a head effect where top models command higher subscription fees [5][6]. Group 2: Future AI Paradigms - The next generation of AI is expected to focus on context capture rather than just model parameter competition, emphasizing the importance of understanding user context for better responses [5]. - There is a belief that signals of autonomous learning will emerge by 2025, although current attempts lack the pre-training capabilities seen in leading companies like OpenAI [8]. - The potential for AI to evolve autonomously and act proactively is seen as a key feature of future paradigms, though it raises significant safety concerns [9]. Group 3: Technological Advancements - Memory technology is anticipated to develop linearly, with breakthroughs expected in the near future as algorithms and infrastructure improve [10]. - The gap between academia and industry in large model development is narrowing, with more academic institutions gaining access to computational resources, fostering innovation [11]. - The industry faces efficiency bottlenecks, with the need to achieve greater intelligence with less investment becoming a driving force for new paradigms [11]. Group 4: AI Agent Development - The evolution of AI Agents is seen as a critical change for the AI industry by 2026, moving from human-defined goals to AI autonomously defining objectives [13]. - The ability of AI Agents to address long-tail problems is highlighted as a significant value proposition for AGI [13]. - The commercialization of AI Agents faces challenges related to value, cost, and speed, necessitating a balance between solving real human issues and managing operational costs [14].

清华攒了个局，把AI圈大半边天聚到了一块。 基模四杰全员到场： 智谱唐杰、Kimi杨植麟、阿里林俊旸 ，还有…… 突然贴脸跳屏的 姚顺雨 。 这场由清华大学基础模型北京市重点实验室发起的AGI-Next前沿峰会，相当硬核。 各位大咖的演讲简直像是在做技术报告，信息密度极高，而且用词相当犀利。 以下附上演讲原文，为提升可读性，量子位在不改变原意的前提下做了适当调整。 清华论剑 刚毕业那会儿我去港科大，学校几乎所有空间都在一栋楼里：教室、实验室、会议室、咖啡厅都在一起。 唐杰：DeepSeek横空出世后，Chat已经基本结束了，下一步是走向做事。 杨植麟：做模型，本质上是在创造一种世界观。 林俊旸：中国想在AI赛道反超，很难。20%这个数字已经很乐观。 姚顺雨：toC的话，大部分人其实用不着那么强的智能。 2019年，我们在清华的支持下完成成果转化，成立了智谱。 同一时期，我们也持续推动开源，既有模型和工具层面的项目，也有面向开发者的大模型 API 体系。 我在清华待了将近二十年。 回头看，我做的事情其实很简单，主要就两件： 唐杰 一是早年做AMiner；二是大模型。 我的题目是「让机器像人一样思考」。 有一个对 ...

以下文章来源于数字生命卡兹克 ，作者数字生命卡兹克 数字生命卡兹克 . 希望能激发你对AI的好奇。 来源丨数字生命卡兹克 （ID： Rockhazix ） 作者丨 卡兹克 昨日受邀，参加了一个非常有趣的活动，现场人真的爆满了，很多人都是从外地特意赶过来的。 这个活动，叫AGI-NEXT。 主要是几个演讲的嘉宾，过于重磅了。 开源四巨头除了DeepSeek没来，智谱的唐杰老师、Kimi的杨植麟、Qwen的林俊旸，齐聚一堂。 甚至腾讯最近最有话题度的姚顺雨，都以远程"巨头"的方式，远程参加了这场会议。 这个巨头，真的是非常的AI巨头。 这场活动因为没有座位了，我站着听了3个小时，收货非常的多。 包括唐老师说，随着DeepSeek这类模型的横空出世，Chat聊天这种范式，其实已经没有仗可打了。 下一仗是什么？是 Action ，是 Doing things 。 杨植麟说，Agent的本质，其实是一个搜索问题。 还有，智能和电力不一样，它不是等价交换品。 你在深圳用的一度电，和在北京用的一度电，完全一样，但一个CEO产生的智能，和一个设计师产生的智能，截然不同。 未来的模型竞争，比的就是谁更有Taste，谁更有品味， ...

1月10日，在由清华大学基础模型北京市重点实验室、智谱AI发起的AGI-Next前沿峰会上，罕见集齐了腾讯"CEO/总裁办公 室"首席AI科学家姚顺雨、Kimi创始人杨植麟、智谱创始人唐杰、阿里巴巴Qwen技术负责人林俊旸以及加拿大皇家学院院士、 香港科技大学荣休教授杨强等AI界知名大咖，围绕AI新范式、Agent、中国大模型公司的挑战及机会等话题展开了讨论。 其中，杨植麟首次深度分享了Kimi的技术重点，他透露，2025年，月之暗面的两个技术进化主线是提升"TokenEfficiency"，以在 有限的数据下冲击更高的智能上限；以及扩展"长上下文"能力，以满足Agentic时代越来越长程的任务对模型的记忆能力需求。 履新腾讯后，姚顺雨首次公开亮相 值得注意的是，此次峰会是姚顺雨加入腾讯后，首次对外界分享其对AI产业的观察。现年27岁的姚顺雨毕业于清华大学姚班和 普林斯顿大学。他在2024年加入OpenAI后，迅速成为团队核心研究者之一，参与推动 AI Agent和任务执行系统方向的开发。 2025年12月17日，腾讯宣布升级大模型研发架构，新成立AI Infra部、AI Data部和数据计算平台部，旨在全面 ...

Core Insights - The AGI-Next conference highlighted the competitive landscape of AI in China, focusing on the importance of foundational models and their impact on future business strategies [4][5] - Key players in the AI industry, including Zhiyuan, Tencent, and Alibaba, are exploring different paradigms for AGI, emphasizing the need for new metrics to evaluate model intelligence [6][7] - The discussion revealed a consensus on the increasing differentiation between consumer (ToC) and business (ToB) applications of AI, with distinct strategies for each segment [11][12] Group 1 - The AGI-Next conference featured prominent figures in China's AI sector, including Zhiyuan's founder Tang Jie and Tencent's newly appointed chief scientist Yao Shunyu, indicating a significant gathering of industry leaders [4][5] - The conference underscored the belief that the capabilities of foundational models will determine the success of future AI ventures, with a focus on maintaining a leading position in model development [5] - Tang Jie expressed concerns that the gap between Chinese and American models may not be closing, as many American models remain closed-source [5][6] Group 2 - The participants discussed the evolution of AI paradigms, with Tang Jie suggesting that the exploration of conversational models has reached its peak, and future efforts should focus on coding and reasoning capabilities [6][7] - Yao Shunyu emphasized the importance of scaling not just in computational power but also in architecture and data optimization to enhance model performance [6][7] - The need for new standards to measure AI intelligence was highlighted, with concepts like Token Efficiency and Intelligence Efficiency being proposed as metrics [7][41] Group 3 - The differentiation between ToC and ToB applications was a key theme, with Yao Shunyu noting that while ToC requires strong integration of models and products, ToB focuses on enhancing productivity through the best models available [11][12] - Lin Junyang pointed out that the success of AI applications depends on understanding real user needs, suggesting that effective communication with enterprise clients is crucial for developing successful AI solutions [8][12] - The conversation also touched on the potential for AI to automate significant portions of human work, particularly in the ToB sector, where higher model intelligence correlates with increased revenue [43][44] Group 4 - The participants acknowledged the challenges of deploying AI models effectively, with a focus on the need for better education and training to maximize the benefits of AI tools [44][57] - The discussion included insights on the importance of collaboration between academia and industry to address unresolved questions in AI research, such as the limits of intelligence and resource allocation [20][21] - The potential for new paradigms in AI, such as continuous learning and memory integration, was identified as a critical area for future exploration [38][40]

2026.01. 10 本文字数：1458，阅读时长大约2分钟 因此，姚顺雨认为，自主学习这件事已经发生了，只是受效率等因素限制，还存在各种问题，他认为目前自主学 习的范式迭代更像是渐变，而非突变。 至于目前全球市场中哪一家企业最可能率先引领范式创新，姚顺雨表示，虽然OpenAI经历了商业化等各种变 化，创新基因被削弱，但仍是最有可能诞生新范式的地方。 林俊旸认为，目前的RL（强化学习）范式尚处早期，潜力远未被充分挖掘，全球范围内仍面临诸多共性挑战， 而下一代范式的核心在于"自主进化"与"主动性"。只是自主进化是否需要更新参数，见仁见智。 作者 | 第一财经 吕倩 当大模型陷入Scaling Law（缩放定律）的增长瓶颈，下一代技术范式将会是什么？ 1月10日，在由清华大学基础模型北京市重点实验室、智谱AI发起的AGI-Next前沿峰会上，腾讯控股"CEO/总 裁办公室"首席AI科学家姚顺雨、阿里巴巴Qwen技术负责人林俊旸、Kimi创始人杨植麟、智谱创始人唐杰等人 工智能行业人士齐聚，共话大模型下一代技术范式。 对下一代范式的猜测中，自主学习（Autonomous Learning）是个热门概念，是大模型摆 ...