机器之心发布 机器之心编辑部 SAIS Talk 是上智院主办的前沿技术分享会，迄今已成功举行 15 期，讲者背景多元，既有曾深度参与诺贝尔 奖评选的顶尖学者，也有活跃在科研一线的在读博士，以此激发灵感、共建生态。 9 月 26 日晚 ，五位来自共性技术、物质科学、生命科学、地球科学等方向的青年研究员将接连登场，分享 核心工作和创新思辨，内容涵盖 表征学习、催化反应预测、生物分子动态模拟、单细胞图谱、全球天气预 测 等多个领域。 当 AI 与科学深度融合，当年轻力量碰撞前沿课题，我们或将在不远的未来见证 "AI 爱因斯坦" 的诞生。诚 邀您与最具活力的 AI 力量同行，迎接科学发现的黄金时代的来临。 （赴 上海徐汇西岸的上智院 参与活动请发送姓名、机构、手机号至 sais@sais.org.cn ，报名截止 9 月 25 日 18:00，报名确认会以邮件在 21:00 之前回复） 活动议程 每个分享环节含 5-10 分钟交流 在全球人工智能浪潮奔涌向前的当下，创新的核心驱动力正越来越多地来自年轻一代。他们敢于挑战前 沿、不惧失败，正以跨界融合之姿重新定义科学发现的范式。 作为长期关注科学智能（AI for S ...

Core Insights - The global AI landscape is transitioning from a phase of technological exploration to one focused on creating tangible value through practical applications of AI technology [2] - There is a significant issue of homogeneity among current large model products, leading to market saturation [2] - The founder of Infinite Light Year, Qi Yuan, emphasizes that while the foundational large model market appears to be converging, industry applications are on the verge of an explosion, with unpredictable technological breakthroughs still possible [2] Industry Applications - Infinite Light Year has developed four major solutions for the financial sector, significantly expanding the coverage of index component stocks from 600 to 2600 and reducing the rebalancing cycle from quarterly to real-time responses in minutes [4][5] - The AI investment research assistant can complete a comprehensive analysis of a financial report within 5 minutes, improving efficiency by over 90% compared to manual analysis [10] Technological Innovations - The "Gray Box Large Model" concept proposed by Infinite Light Year aims to combine the probabilistic predictions of large language models with the logical reasoning of symbolic inference to address the issue of AI "hallucinations" [2] - The dual-engine technology system integrates neural-symbolic computing with large models, enabling precise handling of complex logical relationships and accurate predictions based on extensive data [9] Trust and Compliance - Trustworthiness is identified as a key factor for the successful implementation of AI in industries, particularly in finance where compliance with regulations is critical [8] - Infinite Light Year has introduced a "transparent reasoning mechanism" to enhance user trust by making the AI decision-making process clear and understandable [8] Future Outlook - The company is focusing on a dual-domain strategy for 2025, with horizontal development of a reusable AI infrastructure and vertical deepening in the financial and scientific intelligence sectors [3] - The future of AI competition is expected to shift from a focus on computational power to the ability to create value, with a strong emphasis on practical applications that address real-world problems [12]

AI Industry Trends - ByteDance has open-sourced its AI Agent "Coze," which supports commercial use and has over 6,000 stars on GitHub, providing a platform for developing intelligent agents without coding[14] - The "Step 3" model by Jieyue features 321 billion total parameters and 38 billion activated parameters, achieving a 300% inference efficiency compared to DeepSeek-R1, with expected revenue of nearly $1 billion in 2025[11] - Ant Group released the financial reasoning model "Agentar-Fin-R1," which outperforms similar models in multiple financial evaluations and is based on a comprehensive financial dataset[16] AI Applications and Platforms - SenseTime launched the "Wuneng" embodied intelligence platform, featuring a multimodal reasoning model that improves cross-modal reasoning accuracy by 5 times compared to Gemini 2.5 Pro[8] - Huawei introduced the AI-Box platform, designed for lightweight edge deployment, supporting local execution of multimodal large models with low power consumption[9] - Tencent's Tairos platform offers modular services for multimodal perception and planning, focusing on enhancing robotic software capabilities[10] AI Model Developments - Zhiyuan released the GLM-4.5 model, which integrates reasoning, programming, and agent capabilities, achieving top performance in global open-source model benchmarks[17] - JD Cloud announced the open-source enterprise-level intelligent agent "JoyAgent," which supports multi-agent collaboration and has been tested in over 20,000 internal applications[18] - ByteDance and Nanjing University developed the CriticLean framework, improving the accuracy of mathematical formalization from 38% to 84%[19] Market Risks - AI software sales are below expectations, leading to adjustments in capital expenditure plans and slower iteration speeds for core AI products[34]

Core Viewpoint - The World Artificial Intelligence Conference (WAIC) highlighted the strategic importance of AI for Science (AI4S), marking it as one of the ten core directions with dedicated forums and discussions, indicating its transformative role in reshaping scientific foundations [3][4]. Group 1: AI for Science (AI4S) Development - AI for Science has gained significant attention, especially after AlphaFold's success in solving long-standing biological challenges, demonstrating its real-world impact [3]. - The "Starry River Enlightenment" forum, co-hosted by Fudan University and the Shanghai Institute of Intelligent Science, served as a platform for discussing the trends and innovations in AI for Science [4][5]. - The forum gathered global experts, including Turing and Nobel Prize winners, to explore collaborative innovation and industrial practices in the AI4S 2.0 era [5]. Group 2: Open Collaboration and Ecosystem Building - Fudan University emphasized the need for an open scientific ecosystem, moving beyond the "tool mindset" to a collaborative "ecological mindset" involving human scientists and AI [7]. - The "Open Science Global Academic Cooperation Initiative" was launched to address the challenges of data disparity and promote a collaborative global scientific ecosystem [31][34]. - The initiative proposes four core actions: building open infrastructure, initiating large-scale scientific projects, fostering talent development, and creating a new era of human science [34]. Group 3: Educational and Research Paradigms - The dialogue among university leaders focused on how universities will be reshaped in the AI4S 2.0 era, emphasizing the transition from a "tool mindset" to an "ecological mindset" [39][40]. - The importance of foundational research in AI was highlighted, with calls for strengthening education in mathematics and physics to cultivate top AI talent [40]. - The need for a transformation in university structures and evaluation systems was recognized to adapt to the evolving landscape of scientific intelligence [40]. Group 4: Industry and Academic Collaboration - The forum discussions revealed a consensus on the necessity for collaboration among industry, academia, and new research institutions to foster a thriving ecosystem for AI4S [44]. - Industry representatives pointed out the mismatch between AI model generation and experimental validation, advocating for automated laboratories to bridge this gap [45]. - Academic perspectives focused on enhancing model learning capabilities and addressing ethical concerns related to AI applications in sensitive fields like life sciences [47]. Group 5: Practical Applications and Ethical Governance - The "Starry River Enlightenment" platform was introduced as a comprehensive system to empower scientists by providing open data, shared models, and automated experimental capabilities [53]. - Specific applications showcased the potential of AI in various fields, including life sciences and humanities, demonstrating its broad impact [55][56]. - Ethical governance was emphasized as crucial for the sustainable development of the ecosystem, with initiatives to enhance the efficiency and professionalism of ethical reviews in research [66][68].

Core Insights - The third World Science Intelligence Competition successfully concluded in Shanghai, featuring 30 teams competing for awards in various categories [1][3] - The event was supported by multiple Shanghai municipal committees and aimed to promote collaboration between academia and industry [1][3] Industry Trends - The competition focused on high-value industrial scenarios, addressing key scientific issues in fields such as aviation safety, materials design, synthetic biology, innovative pharmaceuticals, and new energy [3][4] - The event attracted nearly 16,000 participants from around 30 countries and regions, showcasing a trend towards greater openness and youth engagement in the AI for Science domain [3][4] Technological Developments - The competition emphasized the importance of open-source collaboration, providing access to real-world data and computational resources for participants [4][5] - A new "Physical AI track" was launched to focus on core technological challenges in spatial intelligence and reasoning models, further promoting the application of AI technologies [3][4] Youth Engagement - A dedicated middle school competition was introduced, involving 331 teams from 146 schools in Shanghai, with an average participant age of around 14 years [5][6] - This initiative aims to enhance the youth training system in AI, reflecting the industry's recognition of the importance of young talent in driving innovation [5][6] Future Outlook - The organizing committee plans to continue leveraging the competition platform to host more events focused on scientific intelligence, fostering a sustainable ecosystem for innovation in AI [7]

Core Insights - The third World Scientific Intelligence Competition was held in Shanghai, featuring 30 teams competing for awards in five major categories, with a total of 5 first prizes, 10 second prizes, and 15 third prizes awarded [1][3] - The competition aimed to select global talent in the field of AI for Science, with no restrictions on nationality or region, and attracted nearly 16,000 participants from around 30 countries and regions [1][4] Group 1: Competition Overview - The event was co-hosted by the Shanghai Institute of Scientific Intelligence and Fudan University, with support from various institutions including Alibaba Cloud and Shanghai Fosun Pharmaceutical [1] - The competition focused on high-value industrial scenarios, with real-world data sets provided for the challenges, such as aviation safety and renewable energy power forecasting [4][5] - A new "Physical AI track" was launched to address core technological challenges in space intelligence and reasoning models, promoting the application of AI technology [4] Group 2: Open Collaboration and Platform Development - The competition emphasized open-source principles, providing access to real data from industrial scenarios and offering computational resources and toolchain support for participants [5] - Outstanding models from the competition will be deployed on the newly launched Xinghe Qizhi Scientific Intelligence Open Platform, which aims to facilitate collaboration among scientists, AI researchers, and engineers [5] - The platform currently hosts over 200 scientific models across 12 disciplines and has accumulated more than 12PB of scientific data, attracting over 120 research teams [5] Group 3: Youth Engagement - The competition introduced a middle school category, attracting 331 teams from 146 schools in Shanghai, with an average participant age of around 14 years [7] - This initiative aims to enhance the youth training system and showcase the innovative potential of young participants in the field of scientific intelligence [7] Group 4: Future Directions - The organizing committee plans to continue leveraging the competition platform to launch more cutting-edge events focused on scientific intelligence, fostering a sustainable ecosystem for innovation and talent development [10]

Group 1 - GLM-4.5 is an open-source model designed for agents, excelling in reasoning, coding, and agent tasks, with leading performance in domestic tests [1] - The model employs a mixed expert architecture, offering two modes with high parameter efficiency, achieving performance comparable to larger competitors [1] - It features low cost (0.8 yuan per million tokens) and high speed (up to 100 tokens per second), supporting full-stack development tasks [1] Group 2 - Yuntian Lifa is focusing entirely on AI inference chips, aiming to enhance single-chip computing power to thousands of TOPS by 2028 to support trillion-parameter large models [2] - The company utilizes an innovative "computing power building block" architecture with fully domestic technology, compatible with mainstream open-source models and the HarmonyOS [2] - The strategy includes a triad layout of edge, cloud, and intelligent machines, forming four major business segments targeting edge computing, cloud-based large model inference, and intelligent machines [2] Group 3 - Coze has open-sourced two core products (Coze Studio and Coze Loop) under the Apache 2.0 license, receiving 9.5K stars on GitHub [3] - Coze Studio offers a no-code development platform allowing users to create agents through drag-and-drop operations, supporting multi-platform deployment; Coze Loop provides a full lifecycle management toolchain [3] - The open-source strategy aims to establish a new paradigm for agent development, providing a complete toolchain and flexible customization capabilities [3] Group 4 - Kuaishou's Keling AI has released significant updates, including a "spiritual canvas" supporting five-person collaborative creation and a greatly enhanced "multi-image reference" feature [4][5] - The new multi-image reference function addresses consistency issues in AI video generation, showing a 102% improvement in blind tests regarding character representation, dynamic quality, and artistic style stability [5] - A new local reference feature allows users to precisely define reference areas, making video generation results more controllable and significantly lowering the barrier for daily creative video production [5] Group 5 - Lovart, the world's first design agent, has officially launched, utilizing Tencent's Mix Yuan 3D model API for ultra-high-definition detail modeling [6] - The Mix Yuan 3D v2.5 version employs a sparse 3D native architecture, achieving a tenfold increase in geometric model accuracy compared to previous generations, supporting 4K PBR texture mapping [6] - The Mix Yuan strategy remains open-source, with plans for multiple upgrades by 2025, and has surpassed 2.3 million downloads on the Hugging Face platform, having also open-sourced the Mix Yuan 3D World Model 1.0 [6] Group 6 - Alibaba has open-sourced the Tongyi Wanshang Wan2.2 video generation model, the first in the industry to use the MoE architecture, with a total of 27 billion parameters, saving 50% in computing resources [7] - The new model introduces a cinematic aesthetic control system, offering over 60 parameters to adjust lighting, composition, and color [7] - The 5 billion version of the unified video generation model supports both text-to-video and image-to-video generation, deployable on consumer-grade graphics cards [7] Group 7 - SenseTime has launched the Wuneng Embodied Intelligence Platform, providing robots with perception, navigation, and multimodal interaction capabilities based on world models, addressing data bottlenecks [8] - The Wuneng platform can generate high-quality simulation data that adheres to physical rules and offers first and third-person perspectives, enhancing robot training efficiency [8] - This platform empowers robots with intelligent interaction capabilities, demonstrated by a robot that can present PowerPoint slides, showcasing global memory capabilities and transitioning from a tool to a partner in interaction [8] Group 8 - The Shanghai Institute of Science Intelligence, Fudan University, and Infinite Light Year have jointly launched the "Galaxy Enlightenment Scientific Intelligence Open Platform," providing AI-enabled full-link research tools for scientists [10] - The platform is designed with a "scientist-centered" approach, integrating over 200 scientific models across 12 disciplines and 12PB of high-value scientific data, attracting over 120 research teams [10] - It offers six core capabilities: native intelligent agent scientific exploration engine, universal scientific model repository, efficient scientific computing, wet and dry experiment closed-loop, high-value scientific data, and a multidisciplinary collaborative research community, marking the entry into the 2.0 era of scientific intelligence [10] Group 9 - Shopify announced its "All in AI" strategy, sharing successful implementation experiences three months post-announcement, emphasizing universal AI usage without cost limits and default legal team support [11] - The company has built a unified AI entry point, connecting all internal tools via an MCP server, allowing employees to freely construct workflows, significantly enhancing departmental efficiency [11] - Shopify employs a counterintuitive strategy by encouraging AI to demonstrate its thought process rather than hiding it, hiring more junior talent as "AI natives," increasing prototype creation, and linking AI usage to employee performance [11] Group 10 - OpenAI's board chair Bret Taylor believes the SaaS applications of 2010 will evolve into intelligent agent companies by 2030, indicating we are in an "accelerated internet bubble era" [12] - The AI market is divided into three main areas: frontier large models (high competition, difficult entry), AI tools (challenging but with opportunities), and application-layer AI (the greatest opportunity) [12] - Entrepreneurship requires a core "argument" rather than blindly "failing fast," with true customer value for B2B companies needing market validation, as the market explores the "LAMP" technology stack in the AI era, with future intelligent marginal costs approaching zero [12]

Core Insights - The World Artificial Intelligence Conference in Shanghai highlighted the launch of the "Xinghe Qizhi Scientific Intelligence Open Platform" and the "Early Chinese Civilization Multimodal Model," showcasing advancements in scientific intelligence and AI integration [1][2]. Group 1: Scientific Intelligence 2.0 - The transition to Scientific Intelligence 2.0 emphasizes a scientist-centric approach, moving from AI empowering science to a model where scientists lead the integration of AI in research [1][2]. - The "Xinghe Qizhi" platform serves as a bridge connecting scientists with AI, facilitating their participation in scientific advancements [2]. Group 2: Platform Features - The platform offers a comprehensive solution encompassing data, models, computing power, experimental validation, and collaborative research communities, aggregating over 40,000 datasets totaling 12PB of high-value scientific data [2]. - It features a "Universal Model Warehouse" with over 200 advanced models, significantly lowering the barriers for scientists to utilize AI [2]. Group 3: Computational Efficiency - The platform's intelligent computing system optimizes the integration of GPU and CPU resources, enhancing computational efficiency and providing acceleration of several times to up to a thousand times for specific research scenarios [2]. Group 4: Data Ecosystem and Collaboration - Utilizing blockchain technology, the platform establishes a trusted scientific data ecosystem, addressing data ownership and contribution calculation issues, and incentivizing data sharing through a points system [4]. - The platform aims to integrate dry and wet experiments, facilitating seamless connections between AI simulations and real-world experiments, promoting collaboration among scientists, laboratories, and enterprises [4]. Group 5: Cultural and Historical Research - The "Early Chinese Civilization Multimodal Model" is the first AI model focused on early Chinese civilization, developed collaboratively by Fudan University and other institutions, integrating vast historical materials [4]. - This model incorporates humanities research methodologies, enabling comprehensive analysis and traceable evidence for complex and controversial questions, serving as a powerful tool for researchers [4][5]. Group 6: Vision for the Future - The conference illustrated Shanghai's unique advantages in combining top-tier university resources with AI engineering capabilities to tackle significant scientific and cultural challenges [7]. - The ongoing development of scientific intelligence aims to create inclusive and collaborative global scientific knowledge ecosystems, making scientific exploration accessible to a broader audience [7].

Core Insights - The WAIC2025 Star River Intelligent Open Cooperation Forum was held, focusing on building an open and collaborative scientific intelligence ecosystem [1] - Multiple innovative achievements were announced, including the Early Chinese Civilization Multimodal Large Model and the Global Academic Cooperation Initiative [1][3] Group 1: Early Chinese Civilization Multimodal Large Model - The Early Chinese Civilization Multimodal Large Model was officially released, developed by Fudan University, Shanghai Intelligent Research Institute, and Shanghai Chuangzhi Academy [3] - The model encompasses 100TB of specialized corpus, SFT data, and evaluation sets, pioneering cross-modal intelligent alignment of civilization spatiotemporal data [3] - It supports the Chinese Civilization AI Agent platform, enabling multi-step reasoning and complex task planning, benefiting education, research, and cultural industries [3] Group 2: Global Academic Cooperation Initiative - A global initiative was launched by top international scientists, including Nobel Prize winners, aiming to break the "data divide" and ensure AI benefits reach every corner of the globe [3] - The initiative outlines four core objectives: building open-source scientific infrastructure, initiating multinational interdisciplinary scientific programs, cultivating international scientific talent, and establishing a fair value-sharing mechanism [3][4] Group 3: Star River Intelligent Open Platform - The Star River Intelligent Open Platform was launched, designed to accelerate scientific discovery and provide comprehensive infrastructure for scientists and AI engineers [6] - It aims to enhance cross-disciplinary collaboration and address key scientific challenges, significantly speeding up scientific discoveries [6] - The platform includes the "Guanxin" large model, which formalizes complex clinical diagnosis processes into a multi-agent collaborative system for cardiovascular specialties [6] Group 4: Ethical Review AI Agent "Yijian" - The ethical review AI agent "Yijian" was introduced, capable of automatic rule review and risk labeling, enhancing review efficiency and compliance [7] - It has been trialed at Fudan University and its affiliated Zhongshan Hospital, ensuring data security and supporting the generation of review reports [7]

Core Viewpoint - The 2025 World Artificial Intelligence Conference (WAIC) in Shanghai showcased significant advancements in AI technology, emphasizing the importance of global governance and application of AI in various sectors [1][3][4]. Group 1: Event Overview - The WAIC 2025 featured a record exhibition area exceeding 70,000 square meters, with over 800 participating companies and more than 3,000 cutting-edge exhibits, including 40 large models and 60 intelligent robots [1][5]. - The event attracted over 1,200 guests from 30+ countries and regions, including 12 top international award winners and over 80 academicians [3][5]. Group 2: AI Safety and Governance - Geoffrey Hinton, a Nobel laureate, highlighted concerns regarding AI safety, urging international consensus to prevent AI from surpassing human control [3][4]. - Various forums during WAIC will address AI governance, including the release of the "Global AI Governance Rules Map" and the "China AI Safety Commitment Framework" [7]. Group 3: Industry Growth and Trends - The scale of WAIC has expanded significantly, with the number of exhibiting companies increasing from over 400 in 2023 to 800 in 2025, and the number of showcased AI models growing from over 30 to nearly 100 [5][6]. - The event serves as a platform for industry collaboration, with significant investment agreements signed in previous years, including 288 billion yuan in 2023 and an expected 400 billion yuan in 2024 [5]. Group 4: Technological Advancements - The focus of WAIC 2025 includes advancements in humanoid robots, with several companies moving towards mass production, significantly increasing output from last year [9][10]. - New AI hardware products were introduced, such as AI electronic pets, showcasing the diversity of AI applications [14]. Group 5: Commercialization of AI - The commercialization of AI is accelerating, with companies collaborating with mobile and automotive manufacturers to integrate AI models into devices [14][15]. - Several AI firms announced new products aimed at practical applications in sectors like finance and research, indicating a shift towards tangible AI solutions [16].