Domestic News - Alibaba Tongyi Qianwen announced the update of its flagship Qwen3 model, introducing the Qwen3-235B-A22B-FP8 non-thinking mode, with long text capability increased to 256K [2] - Youfang Technology plans to sign a server procurement contract worth up to 4 billion RMB to support its computing cloud service business, expecting a positive impact on its future financial status [3] - Shanghai Artificial Intelligence Laboratory released the DeepLink technology solution for large-scale cross-domain mixed training, enabling interconnection and mixed training of large models across multiple intelligent computing centers [4] - ByteDance's Volcano Engine is reportedly testing a new digital human platform named "Chimera," currently in an invitation-only phase [5] - Shanghai Artificial Intelligence Laboratory launched the multi-agent framework Intern·Agent, which enhances AI's autonomous exploration capabilities in scientific research [6][7] - Tongyi Qianwen's latest AI programming model Qwen3-Coder has been officially open-sourced, featuring multiple sizes and the most powerful version with 480 billion parameters [8] - Shanghai Data Exchange held a closed-door seminar on "New Opportunities in Capital Markets: RDA," introducing the RDA paradigm emphasizing data's role in verifying and enhancing the value of physical assets [9] - Lingyang announced the launch of Alibaba's first batch of enterprise-level Agents, providing multi-faceted services in various business scenarios [10] - 360 Group's founder Zhou Hongyi stated that the company is shifting its chip procurement towards domestic products, primarily purchasing from Huawei [11] - Alibaba Cloud's Qwen3-Coder API is now available on its platform, with competitive pricing compared to other models [12] - Weimob reported a 25% increase in consumer spending per transaction after deploying its AI recommendation engine [13] - Zhou Hongyi revealed plans for the first intelligent agent community in China, "Nano AI Intelligent Agent Community," to enhance the capabilities of digital experts [14] - ByteDance released the Seed LiveInterpret 2.0, a near-human-level simultaneous interpretation model for Chinese and English [15] - Xin'ao Paneng Network launched an autonomous driving architecture in the energy sector, integrating AI with energy expertise [16] - The World Internet Conference held a session on AI development and governance, highlighting China's commitment to AI safety and innovation [17] - Wisdom Eye Technology unveiled the Bian Stone AI Traditional Chinese Medicine Diagnostic Instrument, integrating traditional diagnostic methods with AI [18] - China Mobile's chairman emphasized the importance of the "AI+" initiative for industrial upgrades and innovation [19] - The large model industry is transitioning to a value realization phase, with significant market potential identified [20][21] - Luo Yonghao's digital human capabilities are expected to be mass-produced by October, leveraging advanced script generation technology [22] - Huawei Cloud launched a new AI computing service in Ulanqab, providing high-performance computing capabilities [23] - SenseTime released the new "Riri Xin" model and the "Wuneng" embodied intelligence platform, showcasing advancements in AI capabilities [24] - CITIC Group announced the "Rock" action plan for technological innovation, focusing on multiple key research areas [25] - JD.com upgraded its Yansai large model brand to JoyAI, offering a range of models to meet diverse industry needs [26] Overseas News - Neuralink successfully completed two surgeries in one day, with Elon Musk stating that brain-machine interfaces will change millions of lives [27] - Fidji Simo, the outgoing CEO of Instacart, will join OpenAI as the new application CEO, focusing on building and sharing AI technologies [28] - TRAE launched the SOLO mode AI development assistant, achieving over 1 million monthly active users [29] - Meta reportedly hired three top AI researchers from Google [30] - Alphabet's CEO announced that Google's AI model has over 100 million monthly active users in the US and India [31] - IBM reported a net profit of $2.2 billion for Q2 2025, with an 8% revenue increase [32] - Perplexity's CEO praised China's open-source large model Qwen3-Coder, highlighting its competitive performance and pricing [33] - The US government released an AI action plan aimed at easing regulations and promoting rapid data center construction [34] - Trump expressed his dislike for the term "artificial intelligence," suggesting a name change [35] - Intel reported a net loss of $2.92 billion for Q2 2025, with plans for significant layoffs [36][38] - Starlink experienced a 2.5-hour network outage due to software service failure, which has since been mostly resolved [37] - Jeff Bezos completed a large-scale stock sale of Amazon shares, raising $5.7 billion [39] Financing and Mergers - UBTECH announced a placement agreement for 30.16 million new H shares at a price of 82 HKD per share [40] - Hainan Da announced a plan to transfer its wholly-owned subsidiary for approximately 628 million RMB [41] - Musk's xAI is seeking to raise up to $12 billion for chip procurement and data center establishment [41] - KKR is in talks to acquire ST Telemedia Global Data Centres, with a valuation exceeding $5 billion [42] Policies and Trends - The National Data Bureau's deputy director announced the establishment of data labeling bases in seven cities [43] - The National Data Bureau reported that seven data labeling bases have built 524 datasets [44] - The National Data Bureau aims to promote market-oriented data element allocation and the "AI+" initiative [45] - The Ministry of Industry and Information Technology held a meeting to discuss the integration of information technology and industrialization [46] - IDC predicts that China's data warehouse software market will reach $2.09 billion by 2029 [46] - Over 40% of employees at leading AI companies are considering job changes [47][48] - Anhui Province aims to cultivate competitive data enterprises and industry clusters [49] - Shanghai's information transmission and software services sector grew by 14.6% in the first half of 2025, contributing significantly to GDP growth [50] - Shanghai Pudong plans to add 1,000 AI companies in the next three years, with a focus on infrastructure and application scenarios [51][52] - The Pudong AI Seed Fund was launched with a total scale of 2 billion RMB [52] - AI large language models have shown a high success rate in code writing, exceeding 90% for most software [53] - China leads the world in the number of large models, with 1,509 out of 3,755 globally [54] - Alibaba Cloud's VP noted that large models are experiencing rapid iterations [55]

Group 1 - The software ETF has a turnover rate of 4.32% and a transaction volume of 215 million yuan, ranking first among comparable funds in terms of average daily trading volume over the past week at 396 million yuan [3] - The latest scale of the software ETF reached 4.925 billion yuan, marking a one-year high and also ranking first among comparable funds [3] - Over the past five trading days, the software ETF experienced net inflows on three days, totaling 743 million yuan [3] Group 2 - The software ETF has seen a net value increase of 7.09% over the past three years, with the highest monthly return since inception being 39.35% and the longest consecutive monthly gain being three months with a total increase of 69.40% [3] - The average return during the months of increase is 9.98% [3] Group 3 - The intelligent software development industry focuses on AI technology to provide solutions that enhance business efficiency, optimize decision-making, and improve user experience [4] - The industry is characterized by strong technology dependence, short product iteration cycles, and diverse cross-domain applications, with China showing strong capabilities in this field [4] - The growth in smart hardware shipments is expected to drive demand for intelligent software, with opportunities arising from developments in smart homes and smart vehicles [4] Group 4 - As of June 30, 2025, the top ten weighted stocks in the CSI Software Service Index include iFlytek, Kingsoft Office, Tonghuashun, Hengsheng Electronics, Runhe Software, 360, Softcom Power, Tuo Wei Information, Compass, and China Software, collectively accounting for 60.56% of the index [4] - The stock performance of the top companies includes iFlytek with a 2.88% increase and a weight of 13.93%, while Kingsoft Office increased by 0.46% with a weight of 8.15% [6]

7月24日，360集团创始人周鸿祎在2025年世界互联网大会数字丝路发展论坛开幕式上发言。 中新 网记者 王东明 摄 周鸿祎在接受中新网采访时还说道，AI浪潮下，人工智能的安全性面临诸多挑战，比如内容安全、"胡 说八道"等问题。"大模型就像人一样，聪明且具有随机性，这就会带来安全问题。我们认为应'以模制 模'，必须理解大模型的认知问题，和大模型对话。" 中新网泉州7月25日电 (张钰惠 孙虹)"数字海上丝路孕育着广阔的发展机遇，也交织着网络安全的挑 战。"360集团创始人周鸿祎24日在2025年世界互联网大会数字丝路发展论坛上表示，AI(人工智能)时代 的网络安全更需要全球协作。 当天，2025年世界互联网大会数字丝路发展论坛在福建泉州开幕，论坛以"数智海丝 共迎未来——携手 构建网络空间命运共同体"为主题，吸引来自49个国家和地区及13个国际组织的600余名嘉宾参会。 7月24日，2025年世界互联网大会数字丝路发展论坛在福建泉州开幕。 中新网记者 王东明 摄 周鸿祎在论坛开幕式上发言并表示，数字化最鲜明的特征可以概括为，"一切皆可编程""万物均要互 联""数据驱动业务""AI提高生产力"，一句话就是软件在 ...

在展望未来应用及企业如何适应智能体带来的变革时，周鸿祎表示，中国拥有全球最丰富的工业门类和应用场景，从工业制造到政务服务，从内容创作到网 络安全，智能体正渗透到千行百业，推动效率革命。"不过，企业不必一开始就做复杂智能体，从会计、文员等简单场景入手，逐步积累经验，就能在智能 体时代占据先机。"他建议。 7月23日，在2025中国互联网大会开幕式期间，360集团创始人周鸿祎就智能体发展路径、安全挑战及应用前景等话题发表观点，他表示，2025年是"智能体 元年"，未来两年将是技术落地的关键期。他同时宣布，360将发布中国首个智能体社区"纳米AI智能体社区"，打造个人专属智能体服务。 周鸿祎表示，当前智能体正从概念走向实践，中国凭借丰富的应用场景，将在引领全球智能体技术革命及市场化推广等方面迎来重要发展机遇。在这个过程 中，各行各业都有机会被智能体重塑。周鸿祎认为，智能体在被广泛应用的过程中，个人对专属智能体服务的需求日益强烈，这块市场将大有可为。 从技术来看，周鸿祎认为，当前智能体发展存在"通用"与"专用"的路径之争，但从实践来看，专业垂直的智能体更具生命力。"通用智能体意味着对任何领 域都具备泛化分析能力，这在 ...

完整早报音频，请点击标题下方小耳机收听 【三菱汽车彻底退出中国汽车相关生产业务】 7月22日，三菱汽车宣布，终止和沈阳航天三菱汽车发动机制造有限公司（以下简称"沈阳航天三菱"）的合资合 作，以及合资公司的发动机业务运营。三菱汽车称，考虑到中国汽车产业迅速转向电动化，公司对中国地区的市场环境重新评估，不得不对区域战略进行调整， 决定终止在合资企业中的参与。这也就意味着，从整车生产到核心发动机业务，这家在华深耕四十余载的日系车企，彻底退出了中国汽车生产的舞台。（红星新 闻） 【微信公众号能回复点赞了】 7月23日，微信官方发文：在公众号，创作者收到用户的点赞或推荐后，可以在"公众号助手app"中表达感谢，用户会在微信"公众 号"中收到系统的通知。个人公众号可以直接通过手机端微信"公众号-右上角-我的公众号-互动"谢谢用户的点赞及推荐。在视频号，创作者"进入视频号→点击右 上角我的→打开视频号后台→点击视频号消息"能向点赞的用户表达感谢，还能谢谢用户的关注。（金融界） 【上海 AI 研究院解散了？亚马逊云科技回应：人员精简，将全力支持员工顺利过渡】 7 月 22 日， AWS 亚马逊云科技上海 AI 研究院的首席应 ...

中新网北京7月23日电(记者吴涛)在23日举行的2025中国互联网大会上，AI智能体成为众多与会者讨论 的热门话题。多位嘉宾认为，AI智能体将与互联网深度融合，影响人类的工作、生活。 （文章来源：中国新闻网） 中国工程院院士邬贺铨认为，未来，人工智能将和互联网深度融合，带来新的需求。"AI for Internet和 Internet for AI两个方向都需要着力，一个是互联网怎么适应大流量、高隐私、边缘计算等需求；另一个 角度，广大网友也需要在互联网上更多地应用到AI技术，来满足个性化需求。" 零一万物联合创始人马杰也指出，人工智能将带给创业者更多机会。人类可以基于AI做很多事情，原 来做一个新技术公司是很难的，也需要很多人，现在因为有了AI技术，一个小公司基于AI大模型或 Agent，也有可能成为独角兽公司。 中关村实验室主任、中国互联网协会副理事长吴建平认为，现在人工智能互联网还未实现，但人工智能 绝对会对互联网产生影响。比如改善和优化局部互联网，以前靠人工来纠正的地方，未来可能由"互联 网运维大模型"自动诊断和修复。(完) "未来每个人都将拥有一个由AI智能体组成的数字团队，你的工作就是管理这些数字 ...

y A age 出品 | 搜狐科技 作者 | 梁昌均 编辑 | 杨锦 "即使装上了DeepSeek，就是当聊天机器人在用，感觉还处在一个玩具的阶段，并没有真正变成大家所梦想的新质生产力的打造工具，智能体恰恰弥补了大 模型的不足。" 在7月23日举办的2025中国互联网大会期间，AI成为话题焦点。360创始人周鸿祎在与搜狐科技等媒体的沟通中，强调智能体才是大模型真正落地的关键。 他认为，智能体是大模型进化的一个新阶段，和大模型并不矛盾——大模型扮演大脑的角色，智能体扮演身体、手和脚的角色。 目前，业内做智能体主要有两种模式，一种是大模型厂商去做，如刚推出ChatGPT Agent的OpenAI；另一种则是产品应用公司基于现有大模型去做，典型如 Manus。 他还谈到了行业关注的DeepSeek流量下滑、黄仁勋中国行等话题。最近，Kimi、通义千问先后更新模型，冲到开源模型阵营前两名，市场也都在期待 DeepSeek-R2何时发布。 "DeepSeek可能在憋大招。"周鸿祎表示，DeepSeek并不在乎流量，但包括360等做智能体的不少公司都在用DeepSeek模型，其对中国大模型产业做出了两大 贡献。 具体来看 ...

本文字数：1357，阅读时长大约2分钟 作者 | 第一财经 刘佳 7月23日，身穿标志性红色T恤的360集团创始人周鸿祎，走进互联网大会的一间会议室，他说自己昨 晚没睡好，早上喝了一杯咖啡还是很困。但一谈起AI，便立刻打开了话匣子。 上周英伟达CEO黄仁勋访华后，周鸿祎发了条短视频，感慨国产芯片追赶的速度令人惊叹，强调中 国的开源模型和AI产业链迎来发展的新机会。 2025.07. 23 危险操作，例如诱导模型泄露企业机密文件——这种"注入攻击"已经发生。 "甚至我有时候开玩笑说，一个前台小姐姐可能不会编程，只要她对老板不满意，她都可以对公司的 大模型和智能体进行攻击。"周鸿祎说。 更严峻的是，黑客也在利用AI升级攻击手段。他还提到现在黑客智能体攻击方式，称现在黑客正尝 试把自己的能力和经验嵌入大模型，打造'黑客智能体'，"周鸿祎透露。一旦拥有足够算力，单个黑 客便能操控数十甚至数百个智能体进行攻击。这彻底颠覆了传统网络安全攻防格局——数字化的"机 器人黑客"无需休息，可全天候作战。 "网络安全就从人和人的对抗，变成了人和算法、人和机器、人和算力的对抗。"周鸿祎说，安全企业 也必须转型，打造自己的"智能体安全 ...

文|新浪科技 罗宁 7月23日，2025中国互联网大会在北京举行。360集团创始人、董事长周鸿祎在大会期间与新浪科技等进 行对话，围绕大模型与智能体的发展、国产芯片的应用、DeepSeek流量下降、网络安全风险、360 AI硬 件布局及智能体商业化等多个核心话题展开深入分享，以下为采访实录（不改变原意的情况下有微 调）。 谈大模型与智能体：进化了 谈国产芯片：近期采购都是华为产品 谈及近日英伟达H20供货以及企业的芯片选择，周鸿祎明确表示 "尽可能往国产转"。他承认国产芯片与 英伟达有差距，但强调："必须要用，不用差距永远存在。只有坚持用，才能在实践中改进。" 周鸿祎直言，当前大模型存在严重局限："它很'被动'，只能陪你聊天，更像一个聊天机器人，能力偏 文字功底，很多专业技能其实不会。" 他举例称，企业即便用上DeepSeek，也多当作聊天工具，仍处于 "玩具阶段"，尚未成为真正的生产力工具。 而智能体的出现恰好弥补了这一短板。"大模型像个聪明的大脑，能推理、规划，有渊博知识，但没有 手脚；智能体则扮演身体和手脚的角色。" 周鸿祎强调，智能体与大模型的本质区别体现在两点：一是 能使用各类工具，如同人类借助工具 ...

黄仁勋的"手术刀"割得很高明，割出了国产芯片的机会。 7月23日，身穿标志性红色T恤的360集团创始人周鸿祎，走进互联网大会的一间会议室，他说自己昨晚 没睡好，早上喝了一杯咖啡还是很困。但一谈起AI，便立刻打开了话匣子。 上周英伟达CEO黄仁勋访华后，周鸿祎发了条短视频，感慨国产芯片追赶的速度令人惊叹，强调中国的 开源模型和AI产业链迎来发展的新机会。 对于备受期待的DeepSeek-R2何时发布，周鸿祎表示并不清楚，"也可能在憋大招"，而且国内外竞争对 手的更新速度也不慢。但DeepSeek给中国大模型产业开了个好头，一是消灭了百模大战，行业不需 要"重复发明轮子"，二是证明了中国只有坚持开源路线和开放生态，才能建立国际竞争力。 对于如何用AI Agent重塑360自身，周鸿祎揭示了新的安全挑战。他指出，大模型降低了技术的使用门 槛，安全领域也不例外。如今，即使没有编程基础的人，也能通过精心设计的指令操控大模型执行危险 操作，例如诱导模型泄露企业机密文件——这种"注入攻击"已经发生。 "甚至我有时候开玩笑说，一个前台小姐姐可能不会编程，只要她对老板不满意，她都可以对公司的大 模型和智能体进行攻击。"周鸿祎说 ...