Core Insights - The 2025 Bund Conference gathered 550 guests from 16 countries to discuss the future of AI and innovation, featuring prominent figures like Richard Sutton and Wang Jian [1] Group 1: AI Development and Trends - Richard Sutton emphasized that AI is entering an "experience era" focused on continuous learning, with potential far exceeding previous capabilities [2] - Sutton also noted that fears surrounding AI, such as bias and job loss, are exaggerated and often fueled by those who profit from such narratives [2] - Wang Jian highlighted the shift from code open-source to resource open-source as a revolutionary change in AI, making the choice between open and closed models a key competitive factor [4] Group 2: Infrastructure and Economic Impact - Zhang Hongjiang pointed out that AI is driving large-scale infrastructure expansion, with significant capital expenditures expected, such as over $300 billion in AI-related spending by major tech companies in the U.S. by 2025 [6] - He also mentioned that the AI data center industry has seen a construction boom, which will positively impact the power ecosystem and economic growth [6] Group 3: AI in Healthcare - Ant Group's CEO, Han Xinyi, stated that AI will not replace doctors but will serve as a valuable assistant, enhancing the capabilities of specialists and supporting grassroots healthcare [9][11] - Han identified three core challenges for AI in healthcare: high-quality data, mitigating hallucinations, and addressing ethical concerns [11] Group 4: Challenges in AI Implementation - Wang Xingxing from Yushutech expressed optimism about the AI landscape but acknowledged that practical applications of AI still face significant challenges, particularly in aligning video generation with robotic control [13] - He noted that the barriers to innovation have lowered, creating a favorable environment for young entrepreneurs to leverage AI tools for new ideas [14]

9月11日，在2025Inclusion·外滩大会开幕主论坛上，香港大学计算与数据科学学院院长马毅指出，当前人工智能技术蓬勃发展，但仍缺乏对智能本质的 科学理解。他强调，必须将AI从依赖试错、不可解释的"黑箱"系统，转变为基于数学原理与闭环反馈的"白箱"模型，才能真正实现机器智能。 马毅回顾了智能演化的四个阶段：从DNA所代表的种系遗传智能，到生物个体出现大脑与感知系统形成的个体发育智能，再到借助语言实现的群体智 能，最后才是真正意义上的人工智能。他指出，生命进化本质是智能机制的启动，而当前以大模型为代表的AI仍处于最初级的"种系智能"阶段，依赖海量参 数与预训练数据，不仅资源消耗高、效率低，且缺乏个体记忆与自我意识。 马毅表示，智能的核心在于"自我验证与自我纠错"的能力，即能够批判性地审视既有知识，发现错误、修正并完善它。而当前的大模型仅是静态知识的 存储库，无法理解其内容，因而才会出现基础逻辑混乱和"幻觉"问题。"虽然拥有海量'知识'，但不具备真正的'智能'。"马毅表示。 展望未来，马毅认为，必须将智能作为一个严谨的科学与数学课题来研究，聚焦于构建具备个体记忆与闭环自治能力的系统，在可解释的理论框架下， ...

Core Insights - The evolution of intelligence is categorized into four stages: genetic intelligence represented by DNA, individual developmental intelligence formed by brains and perception systems, collective intelligence achieved through language, and finally, true artificial intelligence. The current AI models, represented by large models, are still in the primitive "genetic intelligence" stage, relying heavily on vast parameters and pre-training data, which leads to high resource consumption and inefficiency [1] Group 1 - The essence of life evolution is the activation of intelligent mechanisms, and current AI models lack individual memory and self-awareness [1] - The core of intelligence lies in the ability of "self-verification and self-correction," which allows for critical examination of existing knowledge to identify and rectify errors. Current large models serve merely as static knowledge repositories and cannot comprehend their content, resulting in logical confusion and "hallucination" issues [1] - Despite possessing vast amounts of "knowledge," current AI models do not exhibit true "intelligence" [1] Group 2 - Looking ahead, it is essential to study intelligence as a rigorous scientific and mathematical subject, focusing on building systems with individual memory and closed-loop autonomy capabilities [1] - The advancement of machine intelligence towards true "autonomous intelligence" should be promoted within an explainable theoretical framework [1]

马毅回顾了智能演化的四个阶段：从DNA所代表的种系遗传智能，到生物个体出现大脑与感知系统形 成的个体发育智能，再到借助语言实现的群体智能，最后才是真正意义上的人工智能。他指出，生命进 化本质是智能机制的启动，而当前以大模型为代表的AI仍处于最初级的"种系智能"阶段，依赖海量参数 与预训练数据，不仅资源消耗高、效率低，且缺乏个体记忆与自我意识。 新浪科技讯 9月11日下午消息，在2025 Inclusion外滩大会开幕主论坛上，香港大学计算与数据科学学院 院长马毅指出，当前人工智能虽技术蓬勃发展，却仍缺乏对智能本质的科学理解。他强调，必须将AI 从依赖试错、不可解释的"黑箱"系统，转变为基于数学原理与闭环反馈的"白箱"模型，才能真正实现机 器智能。 专题：2025 Inclusion·外滩大会 马毅表示，智能的核心在于"自我验证与自我纠错"的能力，即能够批判性地审视既有知识，发现错误、 修正并完善它。而当前的大模型仅是静态知识的存储库，无法理解其内容，因而才会出现基础逻辑混乱 和"幻觉"问题。"虽然拥有海量'知识'，但不具备真正的'智能'。"他说道。 责任编辑：江钰涵 展望未来，马毅认为必须将智能作为一个严谨的 ...

Core Viewpoint - AUO's digital transformation services aim to empower various industries by leveraging its extensive experience in smart manufacturing and digitalization [1][2]. Group 1: Company Overview - AUO has established AUO Digital Technology Services (Suzhou) Co., Ltd. to provide integrated solutions combining AI with manufacturing elements [1]. - The company has served over 1,000 manufacturing enterprises across more than 10 countries, covering 34 industries including electronics, healthcare, and automotive [1]. Group 2: Digital Transformation Strategy - The concept of "minimum element digitalization" allows users to select digital components tailored to their needs, minimizing transformation costs [1][3]. - AUO aims to share its manufacturing expertise to help other companies achieve digital transformation, creating a reciprocal growth model [2][3]. Group 3: Future Factory Concept - AUO defines the future factory as one that integrates large-scale AI capabilities, evolving from advanced factories that focus on lean, automated, and digital processes [5][6]. - The future factory will feature three core elements: autonomous intelligence, embodied intelligence, and spatial intelligence, supported by knowledge, digital, and embedded models [6][7]. Group 4: Client Segmentation and Services - Clients are categorized based on revenue, with tailored services ranging from enterprise hosting for smaller firms to co-creation with top-tier global clients [5]. - The company emphasizes the importance of large-scale factories for maximizing efficiency and value through system reuse [6].

Group 1 - The core viewpoint of the article emphasizes the emergence of WebDancer as a significant advancement in autonomous information retrieval, addressing the challenges of data scarcity and training in open environments [5][10][19]. - The article discusses the increasing demand for intelligent agents capable of multi-step reasoning and decision-making across various fields, highlighting the limitations of existing systems [4][5]. - WebDancer's innovative data synthesis strategies, including CRAWLQA and E2HQA, have successfully generated high-quality training datasets to overcome the scarcity of effective data [12][16]. Group 2 - WebDancer employs a two-phase training strategy, consisting of supervised fine-tuning (SFT) and reinforcement learning (RL), to effectively train agents in dynamic open environments [21][22]. - The article details how WebDancer utilizes the DAPO algorithm for dynamic sampling, enhancing data efficiency and the robustness of the agent's strategies [24][25]. - WebDancer's performance is validated through rigorous testing on challenging datasets like GAIA and WebWalkerQA, demonstrating superior capabilities in complex information retrieval tasks [28][30]. Group 3 - Future developments for WebDancer include integrating more advanced tools and expanding its capabilities to handle complex tasks such as web browsing and API calls [41]. - The article outlines plans to broaden the scope of tasks to include long-text writing, which will require enhanced reasoning and generation capabilities [42]. - The focus on open-source models aims to foster a deeper understanding of agentic models and their scalability in dynamic environments [44][45].

Group 1 - ChatGPT is recognized as the first AI agent to pass the Turing test, marking a significant milestone in AI development [4][6][19] - The rapid user adoption of ChatGPT, reaching over 100 million users within two months of launch, highlights its popularity and impact in the tech industry [3][6][19] - The evolution from GPT-3 to ChatGPT demonstrates substantial improvements in AI capabilities, particularly in natural language processing and user interaction [2][7][19] Group 2 - The structure of the IT industry is being reshaped by large models like GPT, with a layered architecture that includes cloud infrastructure, foundational models, and vertical models [9][11] - Opportunities for competitors in the AI large model era are significant, especially in vertical foundational models and SaaS applications [11][12][19] - The emergence of AI operating systems is being pursued by both established companies and startups, indicating a competitive landscape in the AI sector [12][19] Group 3 - The Chinese AI industry is expected to develop its own large models and killer applications, similar to the evolution of cloud computing [15][19] - The training of Chinese large models can benefit from multilingual data, enhancing their performance and capabilities [16][19] - The focus on generative AI is leading to a surge of new startups and investment in the sector, indicating a vibrant market landscape [18][19] Group 4 - The future of AI large models is projected to include advancements in multimodal intelligence, autonomous agents, edge intelligence, physical intelligence, and biological intelligence [32][33][34] - The integration of foundational models with vertical and edge models is expected to create a new industrial ecosystem, significantly larger than previous technological eras [34][35] - New algorithmic frameworks are needed to improve efficiency and reduce energy consumption in AI systems, with potential breakthroughs anticipated in the next five years [35][34]