Group 1 - The conference highlighted five major trends in artificial intelligence, including accelerated iteration of foundational large models, a shift in focus towards post-training and inference stages, deep collaboration between hardware and software, the rise of intelligent agents and the intelligent agent economy, the promotion of open-source ecosystems, and increasing demands for AI safety governance [1] - Beijing Economic-Technological Development Area is committed to building a comprehensive AI city, with plans to establish a national AI data training base, the largest public computing power platform in the city, and to implement special policies and funding exceeding 1 billion yuan to support major projects in various AI-related fields [1] - By the end of 2025, the development goals include opening 100 landmark application scenarios, gathering 600 core enterprises, and achieving an industry scale target of 80 billion yuan [1] Group 2 - The AI hardware and software testing and verification center was officially launched, aiming to provide key testing and verification capabilities for AI hardware and software, with four core capabilities established [2] - The center has partnered with major companies to create innovation labs and testing facilities to accelerate the innovation and prosperity of intelligent computing technologies [2] - Five major achievements in AI hardware and software collaborative innovation were announced, showcasing significant breakthroughs across the entire technology chain from foundational computing power to framework software [2] Group 3 - The center completed the first batch of testing and evaluation for the adaptation of large models and domestic hardware and software, with several companies successfully passing the evaluation [3] - The conference awarded certificates to institutions that passed the unified benchmark testing, marking a new stage in the standardized and quantifiable development of AI collaborative innovation ecosystems [3] - The AI safety governance initiative was highlighted, with 18 companies disclosing their safety practices, contributing to the establishment of a solid foundation for responsible AI development [3] Group 4 - The vice president of the China Academy of Information and Communications Technology emphasized the urgent need to address challenges in hardware and software collaboration for building an open intelligent computing ecosystem [4] - The AISHPerf 2.0 benchmark system was officially released, featuring upgrades to support multiple inference engines and domestic open-source model loads, addressing various evaluation needs [4] - The academy has initiated a series of collaborative testing and verification efforts based on AISHPerf, focusing on large model adaptation and key collaborative technologies [4]

Core Insights - OpenAI has launched ChatGPT Agent, which integrates Operator and Deep Research capabilities, allowing it to perform complex multi-step tasks and interact with various tools [1][2][9] - Despite improvements, ChatGPT Agent scored 45.5% in spreadsheet editing tasks, significantly lower than the human score of 71.3% [6] Group 1: ChatGPT Agent Features - ChatGPT Agent can perform tasks such as checking calendars, analyzing competitors, and converting screenshots to editable formats [1] - The system combines capabilities of visual browsing, text processing, code execution, and API access [2] Group 2: Performance Metrics - In various benchmark tests, ChatGPT Agent achieved an accuracy of 41.6% in interdisciplinary expert tests, outperforming other models [3] - In data science tasks, ChatGPT demonstrated high accuracy with 89.9% in analysis and 85.5% in modeling [3] Group 3: Future Developments - OpenAI plans to continue iterating on the Agent, with a focus on releasing GPT-5, which is anticipated to enhance the foundational model's capabilities [9] - Developers expect the Agent to reach 90% accuracy in complex tool usage by the end of the year, indicating a move towards commercial viability [9]

Core Insights - OpenAI has launched ChatGPT Agent, an AI system capable of autonomously executing complex tasks, moving beyond simple dialogue to perform a range of functions including searching, filtering, and executing tasks [1][19] - The ChatGPT Agent integrates multiple tools, including a terminal, text browser, and visual browser, creating a comprehensive intelligent system that can operate like a controlled remote virtual operating system [1][2] Functionality and Capabilities - The ChatGPT Agent features three core components: a text browser for information processing, a visual browser for interface interaction, and a terminal for executing code and generating complex files [2][4] - The collaborative capabilities of these components enable a complete "perception-decision-execution" workflow, significantly enhancing task efficiency compared to human processing [6][8] Performance Metrics - ChatGPT Agent has demonstrated superior execution capabilities, achieving a score of 41.6% in the "Humanities Last Exam," nearly double that of models without tools [11] - In the WebArena assessment, the Agent's scores are approaching human levels, and it achieved a score of 45.5% in the SpreadsheetBench evaluation, indicating a twofold improvement over GPT-4o [14] - The Agent outperformed all previous state-of-the-art models in the DSBench test, showcasing its strength in real-world data analysis tasks [16][17] Integration of Previous Products - ChatGPT Agent represents the integration of OpenAI's Operator and Deep Research products, combining execution and content analysis capabilities into a unified model [17][18] - The development process involved reinforcement learning to teach the model effective tool usage, enhancing its operational intelligence [18] Future Implications - The introduction of ChatGPT Agent signifies a shift in AI's role from an assistant to an agent capable of executing tasks autonomously, marking a potential evolution in human-AI collaboration [19] - OpenAI plans to extend these capabilities to various service tiers, indicating a strategy to democratize access to advanced AI functionalities and expand its influence in the large model landscape [19]

又一位离职OpenAI的核心研究员发声！ 刚刚被曝加入Meta的 Hyung Won Chung ，分享了他对AI未来的深刻思考：人工智能正在成为有史以来最强大的杠杆机制。 Hyung Won Chung和一同离开OpenAI 的Jason Wei 是长期搭档，他们的合作可以追溯到谷歌大脑时期，两人曾共同作为第一作者发表了 关于模型微调的重要论文《Scaling Instruction-Finetuned Language Models》。 梦晨 量子位 | 公众号 QbitAI Jason Wei曾称赞他： Hyung Won Chung识别新范式并完全舍弃任何沉没成本的能力给我留下了深刻的印象。2022年底，他意识到了强化学习的力量，并从 那时起就一直在宣扬它。 Hyung Won Chung在OpenAI期间，是o1、o1-preview和Deep Research等核心项目的贡献者。 不过演讲中他没有聊具体的技术细节，而是从一个更宏大的视角解读AI——它如何从个人层面到人类文明层面，彻底改变我们创造价值的方 式。 演讲发生在两个月前，他却在被曝出离开OpenAI后这个时间点分享这些观点，特别是 AI如 ...

Group 1 - The CoDesign 2025 International Symposium was successfully held in Osaka, Japan, focusing on the challenges of large-scale computing and big data, emphasizing the importance of hardware-software co-design for the development of high-performance computing (HPC) [1] - The conference highlighted four core areas: algorithms, application systems, system software and middleware, and hardware-software co-design architecture, covering key fields of high-performance and scalable computing [2] - Keynote speeches and technical presentations showcased cutting-edge research and developments, including the challenges of system fragmentation and the need for collaborative design between hardware and software [3] Group 2 - Roundtable discussions addressed the integration of HPC and AI, with experts sharing differing views on the future direction of computing architectures and the role of AI in scientific programming [4] - The pursuit of Zeta Scale computing was discussed, with experts identifying system reliability and power consumption as core obstacles to scaling [4] - The symposium provided a platform for global experts to share insights and reach consensus, which will significantly advance the integration of HPC and AI, addressing future challenges and opportunities in the computing field [4]

近日，山东省网信办正式公布生成式人工智能服务备案名录，全省累计完成17款相关服务备案。其中，浪潮科技研发 的"焱宇大模型"顺利通过国家网信部门审核，正式获得国家级备案认证。 焱宇大模型以优质语料为核心支撑，通过数据源管理、语料提质、合成评测等全链路手段，为模型训练输送高质量资 源。其构建的LLMOps体系覆盖从数据管理到应用上线的全流程，支持在线平台快速调用，显著缩短应用交付周期。 在安全方面，模型集成数据分类分级、敏感信息过滤、提示词改写等工具，结合隐私计算与分布式可信训练框架，形 成多层次安全防护体系，可满足高合规场景需求。同时支持行业定制化开发，精准匹配业务需求。 场景落地：覆盖十余行业领域，打造丰富应用场景 基于技术突破，焱宇大模型已在基层治理、应急、法治、自然资源、科研、税务、工会、文物、文博、文化等10多个 垂直领域，开发了超百个智能体应用，涵盖辅助办公、话务助理、知识库、报告审查、智能图谱、智能运维、问数等 近20种实用服务场景，打造出20多个典型应用案例，实现了场景级大模型的定制化可视化构建与高效交付。 浪潮科技始终以技术创新驱动产业发展，此前焱宇大模型已成功获批4项深度学习算法备案。此次再次 ...

WAIC 2025"大模型智塑全球产业新秩序"论坛，将于7月28日14:00-18:00，在上海徐汇西岸拉开帷幕！ 论坛由中国信通院承办，华东分院协办，徐汇区新型工业化推进办公室、上海市先导产业促进中心、联 合国工业发展组织投资与技术促进办公室（中国·上海）、上海工创中心等单位支持，聚焦大模型的技 术演进与垂类领域的应用创新，探讨其在宏观发展和垂直领域的变革影响，搭建国际化思想交流平台， 全面展示大模型产业的最新进展及对各行业的变革力量。 剧透一：全球重量级嘉宾云集，权威观点对话 论坛汇聚了知名院士，BSI英国标准协会、北京前瞻人工智能安全与治理研究院等机构的顶尖学者；中 国信通院、上海人工智能实验室、上海大学、复旦大学附属中山医院等单位的权威专家；以及蚂蚁集 团、昆仑万维、印象笔记、中数睿智、上海联通、阿里云、第一财经等企业科技领军者。本次论坛将立 足全球视野，纵览大模型产业的发展态势、标准构建与治理框架，从尖端技术突破、垂类场景落地到安 全体系搭建，全方位、多维度勾勒大模型产业的发展新蓝图，探索未来演进新路径。 会院晖 中国信通院院长 Harold Pradal BSI英国标准协会 全球认证总裁 何宝宏 ...

科创人工智能ETF华夏（589010）紧密跟踪上证科创板人工智能指数，覆盖全产业链优质企业，兼具高 研发投入与政策红利支持，20%涨跌幅与中小盘弹性助力捕捉AI产业"奇点时刻"。 中金公司研报表示，自2023年ChatGPT兴起至今已近三年，随着大模型的持续迭代进步，构建AI Agent 智能体的技术基础和产品路线正在逐步成熟，产业也在愈发期待Agent的落地能够带来在AI应用端的拐 点，使得AI大模型产业能够形成完整的商业闭环。在2025年这一AI Agent"元年"，看好AI Agent产业浪 潮持续向前。 今日截至10:16，科创人工智能ETF华夏（589010）上涨0.39%，盘初最大涨幅达1.45%，有望冲击"六连 阳"。持仓股方面，合合信息领涨6.26%，金山办公上涨4.17%、福昕软件上涨3.18%；石头科技、奥普 特、威胜信息等跌幅超1%。流动性方面，盘中换手3.06%，成交总额达230万，市场持有稳健，长期价 值凸显。 消息方面，北京时间7月18日凌晨，OpenAI发布ChatGPT Agent。ChatGPT Agent一个重要功能模块是其 多工具集成能力，将Operator的网站交互能力 ...

Group 1 - AI applications are experiencing significant gains in the market, with companies like Kingsoft Cloud rising by 12.08% and Innovusion increasing by 11.2% [1] - The release of the K2 model by Kimi, which has over a trillion parameters, is generating excitement in the tech community due to its advanced capabilities in programming and reasoning [1] - Nvidia's CEO emphasizes the importance of the Chinese market, noting that it has cultivated 50% of the world's AI talent and highlighting the rapid development of AI models and applications in China [1] Group 2 - Tianfeng Securities expresses optimism about the AI industry and the AIDC supply chain, predicting that 2025 could be a pivotal year for AI infrastructure and application development in China [2] - CITIC Securities anticipates that the unlocking of H20 will boost capital expenditures among leading firms, creating multiple catalysts for investment opportunities in AI in the second half of the year [2]

一个共识是，AI 今年的大主题，是 Agent。 如果说过去我们习惯了 AI 的「动嘴」，那么接下来，就是 AI「动手」的时代。毕竟，既然 AI 已经能听懂、又足够聪明，那为什么不能直接把 事儿干完？ 2025 年下半场刚开局，一直在定义 AI 的 OpenAI，突然端出了他们的 Agent 答卷。不过有趣的是，乍一看，它和几个月前爆火的 Manus 模 式，惊人地相似。 7 月 18 日凌晨，Sam Altman 和四位 OpenAI 研究员在直播中介绍了 OpenAI 即将推出的 Agent 模式。 简单来说，在 Agent 模式下，你可以直接对 ChatGPT 提要求：婚礼缺双鞋，去电商平台帮我买了；或者，给我设计个宠物周边，直接下单打 印；查找信息，直接生成 PPT。然后，ChatGPT 会自己打开虚拟机，一步步操作。 在演示中，一个复杂任务大概需要 10 分钟完成。但从结果看，完成度很高。ChatGPT 可以在虚拟环境里可以调用文本浏览器、可视化浏览器和 终端。而基于终端，还可以进一步调用云服务 API、图片生成器，运行代码等。 更关键的，这次 OpenAI 不再 Pro 用户优先，Plus、Tea ...