机器之心报道 编辑：泽南、Panda 越通用，就越World Models。 我们知道，大模型技术爆发的原点可能在谷歌一篇名为《Attention is All You Need》的论文上。 如今，在通用人工智能（AGI）初现曙光，Scaling Laws 却疑似接近极限的当口，又是谷歌对未来方向进行了一番思考，想来想去还是只有五个词： 论文标题：General agents need world models 论文链接：https://arxiv.org/abs/2506.01622 香港中文大学博士 Richard C. Suwandi 撰写了一篇博客文章，详细解读了这篇开创性的论文及其对 AGI 未来的意义。 | AUTHORS | AFFILIATIONS | PUBLISHED | | --- | --- | --- | | Richard Cornelius Suwandi | The Chinese University of Hong Kong, Shenzhen | June 11, 2025 | 想象一下，如果我们能够构建一个像人类一样思考和计划的人工智能，未来会是什么样子。大语言模型（L ...

Group 1 - The year 2025 is widely regarded as the "Year of AI Agents" in the global technology and investment markets, with AI agents and their ecosystems seen as significant investment opportunities in the computer sector [1] - The performance gap between different AI models is rapidly narrowing, and domestic computing power supply is maturing, driving demand from traditional industries for AI technology to enhance efficiency and revenue [1][5] - A wave of technological innovation and commercialization surrounding AI agents is sweeping through the global tech sector, with major companies competing to seize the technological and market advantages [1][5] Group 2 - AI agents have evolved from single-task "Chatbots" to "Copilots" and now to "semi-autonomous entities," gaining identity and intent expression capabilities [2] - The new architecture of the AI economy is expected to lead to a revolutionary scenario of "AI hiring AI," where task-leading AI delegates sub-tasks to other AIs, creating a dynamic self-organizing task network [2] - Future development teams may consist of "AI project managers" directing "AI programmers," "testers," and "analysts" to collaborate seamlessly, marking a new dimension of competition for enterprises [2] Group 3 - Major tech companies are launching new competitive initiatives centered around AI agents, with Microsoft and Google leading the charge in creating new application ecosystems [5] - In China, companies are shifting focus from showcasing models to demonstrating the capabilities of AI agents, with a surge of vertical AI products expected in the coming months [5][6] - The launch of the "Nano AI Super Search" by 360 marks a significant milestone, as it is the first domestic product based on super-agent technology, aiming to redefine "AI search" with eight core capabilities [6][12] Group 4 - The "Nano AI Super Search" has quickly gained traction since its launch, becoming one of the largest AI-native search engines globally, ranking fourth in monthly active users in the Chinese market [7][10] - It ranks sixth in the global AI product rankings, being the only domestic product besides DeepSeek to make it into the top ten [10] - The eight core capabilities of the Nano AI Super Search include deep search, complex problem-solving, task decomposition, and multi-model collaboration, enhancing its decision-making, execution, and collaboration abilities [13][14] Group 5 - The Nano AI Super Search aims to realize the vision of "super individuals" and "super companies," where one employee can manage multiple AI agents [12] - It can seamlessly integrate various input methods and emerging hardware, expanding the potential applications of AI agents [15] - The launch of the Nano AI Super Search signifies a shift from AI agent concepts to practical changes in user interaction, strengthening 360's competitive position in the AI product matrix [16]

人在去机场的路上，有感而发，随便写写。 我昨天下午终于完成了OKKI 2025的新品发布会。 这是我第一次以发布会顾问+主持人的身份，深度参与一场外贸行业ToB产品的发布。 两个月前，我还对外贸两个字没什么特别的理解。在我认知里，它就是义乌的小商品、深圳的耳机工 厂、福建的丑拖鞋、浙江的男装。我只知道那是一个离钱很近的行业。 后来因为机缘巧合，认识了刘世奇，写了一篇他的专访： 专访刘世奇 - 他用AI设计丑拖鞋，一年卖 了3000万。 也因为刘世奇，跟阿里国际站结识，然后小满科技（OKKI产品的公司、也是阿里国际站的子公司） 正好要开AI新品的发布会，想找一个对外贸感兴趣的AI领域的KOL一起来聊，于是我们就正好，一 拍即合。 但当我真正走进去，跟他们聊完，看到数据，听完几个老业务员、年轻创业者、老板们的真实反馈之 后，我突然意识到，这帮子外贸人嗅觉真的很敏锐，他们已经把AI玩出花了。 不只用来写报告，也不止用来画个图。而是上来就用AI谈单、筛客户、做背调、跟单、复盘、质 检、建图谱、盘老客户，甚至是自动出击。 我是真的服了。 所以我决定写下这篇文字，总结8条我这半个月跟OKKI筹备过程中，和一些AI外贸专家 ...

国内科技厂商 Agent 布局与落地难点 20250612 摘要 字节跳动计划在 12 月发布豆包通用大模型 2.0 版本，该版本将合并豆 包和豆包 Thinking，并实现自推理技术路径的全自研以及多模态理解模 型的端到端贯通，预计 10 月或 11 月开始上线供用户使用。 豆包大模型日均 tokens 使用量从 2024 年 12 月的 4 万亿增长到 2025 年 6 月的 16.4 万亿，其中约 12%-15%通过各种渠道调用并收费，主 要集中在服务、工具和教育行业，大部分仍由字节跳动内部平台调用。 火山引擎占据国内公有云市场超过 55%的份额（截至 2025 年 6 月）， 增长主要源于中小企业选择和价格战，并投放了大量 H20 芯片，与科大 讯飞合作提供 AI 解决方案，尤其在民营企业和创业公司中具有显著优势。 字节跳动在 AIGC 领域布局广泛，提供从大型集团客户的 agent 私有化 部署（起步价 80 多万元）到公有云方案，以及模块化组件收费模式， 并投入 SaaS 套件，如扣子空间和豆包移动端及 PC 端，覆盖不同规模的 企业和个人用户。 扣子平台已成为国内领先的智能体搭建平台，拥有约一百 ...

Core Viewpoint - The article discusses the evolution of search engines into "super search" intelligent agents by 2025, emphasizing their transition from traditional keyword-based searches to advanced task execution capabilities that understand user intent and deliver actionable solutions [2][8][16]. Group 1: Evolution of Search Engines - The shift from traditional search engines to intelligent agents is marked by the emergence of AI search 3.0, which integrates intent recognition and task execution into a seamless user experience [8][16]. - AI search 1.0 and 2.0 focused on information aggregation and answer provision, respectively, but lacked the ability to execute complex tasks directly [5][8]. - The future of search engines lies in their ability to function as task engines, providing users with direct solutions rather than just information [6][8]. Group 2: Capabilities of Super Search - Super search must possess five key capabilities: task planning, multi-model collaboration, high-dimensional information recognition, multi-modal output, and personalized search experiences [9][10][11][12][13]. - Current AI search engines are still in the early stages of development, with some like Nano AI and Google's AI Mode showing promise in covering these capabilities [14][18]. Group 3: Market Position and Competition - Nano AI has emerged as a leader in the AI search engine market, outperforming competitors in user engagement and functionality [19][21]. - The competition between established players like Google and emerging platforms like Nano AI is intensifying, with both focusing on transforming search engines into intelligent agents [22][33]. - The article highlights the importance of technological infrastructure and the ability to execute complex tasks as critical factors for success in the evolving search engine landscape [18][22]. Group 4: Practical Applications - Practical examples of super search capabilities include generating comprehensive reports and conducting in-depth research based on user queries, showcasing the potential for AI to enhance productivity [26][30]. - The article illustrates how Nano AI can autonomously break down complex tasks and deliver tailored solutions, emphasizing the shift from information retrieval to actionable insights [30][31].

AI搜索3.0时代， 什么是"超级搜索"？ 超级搜索智能体未来已来。 2025年，"智能体化"不再停留在高谈阔论，而是实实在在已经登陆的台风。 各行各业的AI应用，正在蜕变为同时拥有"意图理解力"与"任务执行力"的智能体。其中，搜索引擎作为互联网时代最强大的流量入口、大众最熟悉的应用， 也应该成为最先完成"智能体化"的应用场景。 当传统搜索引擎还在筛选信息碎片时，搜索智能体已经能够通过主动识别用户的搜索意图，将海量数据融合在专业级交付中：一份旅游攻略、一份市场分析 报告、一次购物决策，搜索不再只是信息的搬运工，而是触手可及的生产力。 智能体对于传统关键词检索模式的颠覆与重构，也让"超级搜索"从概念走向了现实。 搜索引擎过去一直是老态龙钟，旁人却又无可奈何的状态。大模型时代，AI搜索的潜力肉眼可见，老牌巨头一成不变会有真正的风险，因为跃跃欲试的屠龙 者，谋划周密，变革汹涌，以重造搜索在新时期意图快速篡权。 AI搜索可以看到三个明显的产品分界。在AI搜索1.0时期，网页排序依然保留，加入了AI Overview，Bing、谷歌、百度都先这么"迎合潮流"。大模型就像为 传统搜索装上了一层智能化的"滤网"，总结出精华 ...

出品 | 搜狐科技 作者 | 梁昌均 "我是恨铁不成钢，想让市场部养成用AI的习惯。所以我这次提出说，要干掉市场部，实际是让市场部升级成什么部门呢？让他们去编程，去设计打造 智能体。" 在6月11日晚的发布会上，360创始人、董事长周鸿祎揭开了此前"要干掉360市场部"的谜底——还没有干掉市场部。 这次发布会的五天前，周鸿祎在多个自家渠道发文称，准备干掉360市场部，从而一年给公司省下几千万。他表示，从当天起，要做一个挑战，一个人 完成一场完整的新产品发布会。 "听起来像天方夜谭，但这次我准备动真格的了。过去做一个产品发布会，需要市场部几十号人，忙活大半个月，费时费力费钱，还不满意，这次自己 一个人全包了。" 不过，在这次发布会上，360的市场部工作人员依然在现场做着辅助工作，甚至周鸿祎演讲所用的PPT依然还是离不开市场部员工的参与。 "目前全世界所有AI做的PPT，我觉得忽悠领导，或者拿给客户或部门内部讨论肯定是够的，但做发布会估计还差点。"周鸿祎在开场中提到。 周鸿祎所谓的"干掉市场部"更像是一场话题营销，尤其是在AI取代人的社会焦虑下，这样的极致表达无疑会带来更多关注。可以说，"红衣大叔"再一次 把AI ...

Core Insights - Ant Group's Ant Technology has launched an energy service intelligent system that can accurately predict power generation and market supply-demand changes [1] - The intelligent system generates analytical reports and strategies for investment, operations, trading, and financing through simple text commands [1] Company Developments - The launch of the energy service intelligent system marks a significant advancement in Ant Group's capabilities in the energy sector [1] - This innovation aims to enhance decision-making processes in various energy-related activities [1] Industry Impact - The introduction of such intelligent systems could transform the renewable energy landscape by improving efficiency and responsiveness to market dynamics [1] - The ability to predict market changes and generate actionable insights may provide a competitive edge to companies in the energy sector [1]

每经记者｜李宇彤 每经实习编辑｜余婷婷 黑暗的场馆中，大屏幕上播放的宣传片吸引了全场人的目光。一段堪比动作大片的赛车竞逐之后，画面通过蒙太奇手法巧妙切换，观众瞬间穿越到艺术世 界，名画中的人物"活"了过来，花草也随之摇曳生姿——这，是字节跳动旗下火山引擎Force原动力大会的现场演示。 Seedance 1.0 pro的性价比同样受到市场关注。《每日经济新闻》记者现场了解到，Seedance 1.0 pro的价格是0.015元/千tokens（语言生成模型在运作时的最小 工作单元），相当于每生成一条5秒的1080P视频只需3.67元。据称，这个价格已经为当前行业最低。 把价格打下来的不只有视频生成模型，最新发布的豆包大模型1.6也展现了火山引擎要降低大模型使用成本的决心。 火山引擎总裁谭待 图片来源：企业供图 据火山引擎总裁谭待现场表示，豆包1.6首创按"输入长度"区间定价，深度思考、多模态能力与基础语言模型统一价格。豆包1.6的综合成本只有豆包1.5深度 思考模型或DeepSeek R1的三分之一。 豆包此番"双箭齐发"，字节跳动摁下Agent（智能体）加速键。 6月11日，"2025火山引擎春季Force ...

作者介绍： 本文作者来自通义实验室 RAG 团队，致力于面向下一代 RAG 技术进行基础研究。该团队 WebWalker 工作近期也被 ACL 2025 main conference 录 用。 它得能看懂网页，能做多步决策； 它得能适应开放动态环境； 它得能自主提问、自主行动、自主修正…… 一、背景：信息检索的新需求与挑战 在当今信息爆炸的时代，解决复杂问题不再仅仅是简单的知识检索，而是需要深入的信息挖掘和多步推理。从医学研究到科技创新，从商业决策到学术探索，每 一个领域都呼唤着能够自主思考、自主决策的智能体。Deep Research 等系统已经为我们展示了自主多步研究的巨大潜力，但构建这样的智能体并非易事。它们需 要在复杂的网络环境中感知、决策、行动，还要面对任务复杂度高、泛化能力弱等诸多挑战。 但打造这样一个 Deep Research 类智能体智能体，并不简单！ 在这种背景下，WebDancer 的出现，走出了一条复现 Deep Research 类智能体的可行路径。 自主信息检索智能体的构建，或者如何复现 Deep Research 类的模型一直面临着两大棘手难题：高质量训练数据的稀缺与开放环境训 ...