Core Insights - OpenAI has launched ChatGPT Agent, a new feature that allows its chatbot to perform multi-step tasks on behalf of users, such as scheduling meetings and creating presentations [2][3] - The feature is currently available only to paid users, with Pro subscribers allowed up to 400 agent prompts per month, while Plus and Team users are limited to 40 prompts [2] - This launch follows significant investments in AI agents by major tech companies like Google, Salesforce, and HubSpot over the past year [2] Functionality and Design - ChatGPT Agent integrates capabilities from previous breakthroughs, including interaction with websites, information synthesis, and conversational intelligence [3] - Users can issue commands for tasks like checking schedules or analyzing competitors, with the agent operating on a virtual machine that can switch between various tools [3] - The design allows for collaboration, enabling users to interrupt or reassign tasks without losing progress [3][4] Limitations and Safety Measures - The tool currently lacks real-time response capabilities, taking approximately 25 minutes to generate a presentation based on company financial reports [3] - Certain features, such as memory, are disabled for safety reasons, with the company taking a cautious approach to its release [4] - OpenAI has classified the underlying model as having "high biological and chemical capabilities," indicating potential risks associated with misuse [4][5] Risk Mitigation Strategies - To reduce the risk of real-world abuse, OpenAI has implemented measures such as requiring user confirmation for irreversible actions and monitoring high-risk tasks [5] - The company has introduced protections against prompt injection, which could lead to malicious instructions being executed by the agent [5] - OpenAI emphasizes that the release of ChatGPT Agent is just the beginning, with plans for ongoing improvements in efficiency and interaction quality [5]

近日，智象未来团队两项创新成果入选ICCV 2025，分别聚焦于图像生成与视频增强两个方向，全面展 示了其在生成式AI技术上的突破与领先。 这两项成果分别发表于ICCV 2025 论文《Denoising Token Prediction in Masked Autoregressive Models》 和《Aligning Global Semantics and Local Textures in Generative Video Enhancement》，为视觉生成技术开 辟新路径，推动高质量、高效率的多模态内容生成迈向实用化，未来有望在创意设计、影视制作等领域 广泛应用。 在视频增强方面，团队则推出了创新的生成式视频画质增强框架GenVE，攻克传统方法细节缺失难题。 其核心在于双重对齐机制：通过图像扩散模型生成语义参考，确保全局布局一致；借助局部感知交叉注 意力模块，精准迁移参考图像的纹理细节至视频。同时，多重增强策略提升模型鲁棒性：噪声增强平衡 质量与保真度，时间增强强化帧间连贯性，掩码策略提升特征利用率。在YouHQ40、VideoLQ等数据集 上，GenVE的MUSIQ、DOVER等指标全面领先 ...

Core Insights - The WAIC 2025 Large Model Forum will officially open on July 27 in Shanghai, aiming to drive innovation and accelerate the transformation of production power across various industries [1] - The forum will serve as a global platform for top researchers and technology experts to discuss the evolution of large models and their commercial applications [2] - Key topics will include the collaboration and ecosystem construction of domestic heterogeneous computing power, as well as breakthroughs in embodied world models [3] Group 1: Forum Objectives and Themes - The forum aims to gather global wisdom and leverage Shanghai's strategic position in large model technology to foster innovation [1] - It will focus on the construction logic, collaboration paradigms, and development bottlenecks of the large model industry ecosystem [2] - Discussions will also cover international collaboration opportunities in AI, including technology standards, ethical governance, and talent cultivation [4] Group 2: Key Participants and Contributions - Notable participants include representatives from leading AI companies such as SenseTime, NVIDIA, and various top universities [2][3] - The forum will feature keynote speeches from prominent figures, including the president of Nanyang Technological University and executives from major companies like Kingsoft and Xiaomi [6][7] - Insights from industry leaders will provide valuable, replicable blueprints for intelligent transformation across various sectors [4]

广西云－广西日报AI体验官在南京投石科技有限公司投石光影艺术馆体验AI漫画变脸透明屏，通 过AI大模型训练可以实现人脸头像的实时动漫变化。 "作为重要合作伙伴，讯飞看好广西积累的海量东盟语料资源和东盟语言人才资源，计划搭建多语 言数据平台。"科大讯飞董事长刘庆峰如是介绍。 "作为头部机器人企业，埃斯顿正积极评估在广西设立区域服务与展示中心，以强化本地化支持和 供应链能力，辐射我国西南及东盟市场。"南京埃斯顿学院副总监李明立透露。 广西云－广西日报AI体验官在科大讯飞采访拍摄。 "商汤科技将在广西开展智算中心及产业生态合作，共同打造面向东盟赋能千行百业应用创新的示 范展示窗口。"商汤大装置事业群解决方案总监刘波表示。 连日来，广西云－广西日报AI体验官走访长三角人工智能企业，在开阔眼界的同时，也有了更多 思考：硬核技术"扎堆"的背后，有哪些因素支撑？人工智能企业纷纷表示看好广西，仅仅是因为这里的 独特区位优势吗？共建人工智能国际合作高地，吸引人工智能企业落地，广西怎样提供更加丰富的应用 场景？ 既要留足"自主空间"，也要提供"安全感" "我们不需要政府宠企业，我们需要政府懂企业。"在杭州采访，浙江实在智能科技有 ...

截至2025年7月18日 09:35，中证人工智能主题指数(930713)强势上涨1.38%，成分股深信服(300454)上涨6.70%，中际旭创(300308)上涨6.24%，三六零 (601360)上涨3.96%，瑞芯微(603893)，用友网络(600588)等个股跟涨。AI人工智能ETF(512930)上涨1.44%， 冲击6连涨。最新价报1.48元。拉长时间看，截 至2025年7月17日，AI人工智能ETF近1周累计上涨7.22%。 流动性方面，AI人工智能ETF盘中换手0.5%，成交933.70万元。拉长时间看，截至7月17日，AI人工智能ETF近1周日均成交1.16亿元。 规模方面，AI人工智能ETF最新规模达18.43亿元。 数据显示，杠杆资金持续布局中。AI人工智能ETF最新融资买入额达558.70万元，最新融资余额达6661.93万元。 截至7月17日，AI人工智能ETF近5年净值上涨5.78%。从收益能力看，截至2025年7月17日，AI人工智能ETF自成立以来，最高单月回报为28.72%，最长连涨 月数为4个月，最长连涨涨幅为18.08%，涨跌月数比为36/35，上涨月份平均收益率为6 ...

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，在不远的未来将在更多的行业和场景实现规模化落地，成为真正意义上的智能生产力。 规模方面，科创AIETF最新规模达47.37亿元，创成立以来新高，位居可比基金1/7。 份额方面，科创AIETF近1周份额增长1.26亿份，实现显著增长，新增份额位居可比基金1/7。 资金流入方面，科创AIETF最新资金净流入2646.18万元。拉长时间看，近5个交易日内，合计"吸 金"4507.73万元。 数据显示，杠杆资金持续布局中。科创AIETF本月以来融资净买额达1888.04万元，最新融资余额达4.46 亿元。 截至2025年7月18日 09:38，上证科创板人工智能指数(950180)上涨0.96%，成分股金山办公(688111)上涨 4.10%，凌云光(688400)上涨3.82%，合合信 ...

南兴股份（002757）一字涨停，酷特智能（300840）、掌阅科技（603533）、立方控股、鼎捷数智、延 华智能（002178）涨超5%，万兴科技、值得买（300785）、汉得信息（300170）、通达海（301378） 等跟涨。消息面上，今天凌晨，OpenAI进行了技术直播发布了重磅产品ChatGPT Agent。ChatGPT Agent具备自主思考和行动的能力，能够主动从其技能库中选择合适的工具，包括Operator、Deep Research和ChatGPT来完成各种超复杂任务。 ...

目前Lexi支持Facebook、Instagram广告投放，不限国家、语言和行业。已服务超过 10 万家中小企业，覆盖北美市场与东南亚、非洲等多个新兴市场超90 多个国家，日注册用户稳定维持在四位数以上，服务对象涵盖初创企业主、线上商家、自雇创作者及本地服务商，体现出其强大的通用性与市场刚需。 Lexi采用SaaS订阅制，用户首周可低门槛试用，后续以每周计费的方式持续订阅，换算月均费用约为 199 美元。 锦秋基金表示："锦秋基金作为12 年期的 AI Fund，始终以长期主义为核心投资理念，积极寻找那些具有突破性技术和创新商业模式的通用人工智能初创 企业。" 投资界7月17日消息，AI创新科技公司「Sandwich Lab」宣布完成数百万美元首轮融资，由锦秋基金与汇量科技联合领投，澜松资本担任长期独家财务顾 问。本轮融资主要用于核心AI能力研发、全球市场拓展与国际化人才团队建设。 Sandwich Lab 的创始团队由来自阿里巴巴达摩院、苹果、微软、百度、Giorgio Armani 等全球顶尖公司成员组成，覆盖 AI 算法、产品设计、品牌创意与 海外市场多个维度。其创始人郭振宇博士本科毕业于浙江大学，博 ...

BigBear.ai (BBAI 16.01%) saw its stock close at $8.22 on Thursday, July 17, marking a significant 15.5% increase. The intraday trading showed notable volatility, ranging between a low of $7.25 and a high of $8.38.In the context of broader market movements, BigBear.ai's performance outstripped that of key indices. The S&P 500 saw a 0.54% increase, while the Nasdaq Composite rose by 0.74%, indicating that the stock's robust rise was primarily driven by company-specific excitement rather than macroeconomic fac ...

Core Insights - The article discusses the competitive landscape of AI large models, highlighting the recent release of xAI's Grok-4 and Kimi's K2 model, which have sparked a new wave of advancements in the field [1][4]. Model Performance Summary - Grok-4 achieved a significant score increase from 42.6 to 65.0 in the ScienceQA evaluation, marking a 50% improvement and surpassing OpenAI's o3 model to become the state-of-the-art (SOTA) model [4][8]. - Kimi K2, a non-thinking model, scored 49.6, placing it in the top ten, with a BoN (N=5) score of 73.0, indicating strong performance in multi-step reasoning tasks [11][24]. - OpenAI's o3-pro model scored 59.6, showing improvement over its predecessor, but with increased response time and API costs [11][25]. Cost and Efficiency Analysis - Grok-4 is noted for its competitive pricing at $15 per million tokens, significantly lower than o3-pro's $80, while maintaining high performance [15][21]. - Doubao-Seed-1.6 demonstrated a cost-effective model with a score of 56.6 and an output price of $1.1, making it one of the best value models [15][18]. - The analysis indicates a trend where longer reasoning times correlate with higher scores, with Grok-4 having the longest average response time of 227 seconds [17]. Model Innovations - Grok-4 incorporates advanced features such as real-time web retrieval and multi-agent collaboration for enhanced reasoning capabilities [23]. - Kimi K2 is recognized for its innovative training techniques, including the MuonClip optimizer and a comprehensive agent simulation pipeline, which contribute to its large parameter count and performance [24]. - OpenAI's o3-pro model has been optimized for scientific and programming tasks, showcasing improved reliability and reasoning capabilities [25]. Leaderboard Updates - The leaderboard reflects updates from 16 companies with 43 different model versions, maintaining a consistent ranking for major players like OpenAI, Google, and ByteDance [5][8]. - The leaderboard will continue to evolve with monthly updates, providing ongoing insights into model performance and capabilities [1][5].