本周观点 投资建议 建议关注国内生成式大模型龙头科大讯飞；AI 硬件有望成为应用落地的新载体，建议关注海康威视、虹软科技、 禾赛等；AI 相关功能打磨能够带动付费率、Arpu 值提升，建议关注迈富时等。 风险提示 行业竞争加剧的风险；技术研发进度不及预期的风险；特定行业下游资本开支周期性波动的风险。 敬请参阅最后一页特别声明 1 7 月 28 日，智谱 AI 开源 GLM-4.5 系列（总参 3550 亿/激活 320 亿），首次在单模型中融合推理、编程与智能体 能力。该系列采用 MoE 架构与均衡路由优化，经 15T 通用语料+7T 代码特化训练后，在 12 项全球评测中综合 排名全球第三（开源/国产第一），生成速度达 100 tokens/秒。实测在工具调用成功率上超越 Claude、Kimi，已 在智谱清言与Z.ai平台开放体验并开源全部代码与技术报告。7月26日，阿里通义千问推出推理专用模型Qwen3- 235B-Thinking（MoE 架构 2350 亿/激活 220 亿），一周内第三次开源。模型在 256K 长文本与逻辑数学任务实 现跃升——高难度测试「人类最后的考试」得分较 4 月提升 54% ...

Core Insights - The World Artificial Intelligence Conference (WAIC 2025) showcased over 800 companies, highlighting the rapid development of AI technology across various sectors, including model applications, intelligent agents, embodied intelligence, and AI hardware [1] - RockAI's new generation model, Yan 2.0 Preview, features "native memory" and "offline intelligence," allowing devices to operate without internet connectivity and perform complex tasks autonomously [1][2] - The model's advancements include support for video modalities and a memory module that enables autonomous decision-making, moving beyond traditional AI limitations [2] Company Developments - RockAI is targeting a diverse customer base, including consumer electronics brands, ODM manufacturers, and companies in need of high-performance, low-power solutions [3] - A memorandum of understanding was signed with AMD to collaborate on AI PC technology, marking a significant step in the commercialization of the Yan architecture and positioning Chinese AI technology on the global stage [3] - RockAI aims for general artificial intelligence (AGI) by focusing on collective learning and collaborative evolution, allowing multiple intelligent units to solve complex problems through local interactions [3][4] Technological Impact - The technology showcased by RockAI at WAIC 2025 represents a shift from one-time delivery of smart hardware to continuous evolution throughout the product lifecycle, potentially paving the way for AGI [4]

Group 1: AI Model Developments - GPT-5, codenamed "Lobster," has been quietly launched on the WebDev Arena testing platform, showing performance significantly surpassing Grok-4 [1] - The new Step 3 foundational model by Jieyue Xingchen is a native multimodal reasoning model with a total parameter count of 321 billion and an active parameter count of 38 billion, achieving high inference efficiency [2] - RockAI showcased the Yan 2.0 Preview model, which operates offline and incorporates a "native memory module" for continuous learning and evolution [7] Group 2: AI Applications and Products - Tencent unveiled the "Hunyuan 3D World Model 1.0," the first open-source 3D world generation model, enabling quick generation of interactive 3D scenes [3] - Alibaba previewed its self-developed "Quark AI Glasses," which integrate various functionalities from the Alibaba ecosystem and are set to be released within the year [4][5] - Lovart launched the ChatCanvas feature, combining visual understanding and multimodal design, allowing users to perform advanced design operations on a smart canvas [6] Group 3: Marketing and Robotics Innovations - The Navos AI Agent by Taidong Technology can generate marketing materials in 5 minutes and execute cross-national campaigns within 72 hours, addressing localization cost challenges [8] - Unitree Technology introduced the humanoid robot Unitree R1, priced from 39,900 yuan, featuring 26 degrees of freedom and advanced capabilities [10] Group 4: AI Ethics and Future Perspectives - Geoffrey Hinton emphasized the potential for large models to achieve "immortality" while warning of the risks associated with AI surpassing human intelligence [11] - Hinton suggested separating the research on making AI "smarter" from making it "kinder," advocating for shared "kindness technology" to mitigate future AI risks [12]

Core Viewpoint - Google is initiating a transformation in AI architecture, moving beyond the limitations of the existing attention mechanisms with the introduction of the MoR architecture, indicating a consensus on the need for architectural innovation in the AI field [2][3]. Group 1: RockAI's Innovations - RockAI has developed the Yan architecture, which is a non-Transformer framework that significantly reduces computational complexity, allowing for offline operation on low-power devices like Raspberry Pi [5][10]. - The Yan 2.0 Preview model possesses native memory capabilities, enabling it to remember interactions over time, unlike traditional models that forget previous conversations [6][12]. - The architecture allows for end-to-end memory integration, making it easier for users as it does not require manual management of external knowledge bases [19][22]. Group 2: Challenges with Transformer Models - Transformer models face issues such as data scarcity and high computational requirements, making it difficult to deploy on low-power devices without significant modifications [9][10]. - The reliance on massive datasets for pre-training is becoming increasingly challenging due to the difficulty in acquiring high-value data [9]. - Current models often lack the ability to learn and update parameters during inference, limiting their adaptability [9][10]. Group 3: Vision for AI Development - RockAI aims to create intelligent devices that can learn and evolve independently, moving away from static models that require cloud connectivity [24][25]. - The concept of collective intelligence is emphasized, where individual devices with learning capabilities can share knowledge and evolve together, contributing to the development of AGI [26][29]. - The company’s mission is to ensure that every device possesses its own intelligence, promoting offline smart capabilities [27][28]. Group 4: Market Reception and Future Directions - RockAI's approach has gained recognition at industry events, with hardware manufacturers showing interest in their technology due to its unique memory capabilities [34][35]. - The company is committed to continuing its challenging yet correct technological path, focusing on enhancing autonomous learning capabilities [36][37]. - RockAI's long-term vision reflects a commitment to fundamental technological advancements, akin to the journeys of leading AI research labs [37][38].