这项课题由上智院AIMath青年研究员、北京大学博士生马成栋于2024年发起。他希望在完成既有学术 成果后，挑战更小众、更高风险的问题。深耕强化学习的他，与数学出身的陶兆巍形成互补。 PackingStar项目是陶兆巍参加的第一项真正意义上的科研工作。从北京大学数学学院本科毕业后，陶兆 巍去法国留过学、当过国际学校的数学老师、在著名科普杂志做过编辑。GPT发布后，他开始自学计算 机专业知识并主动接触AI从业者。 牛顿和大卫·格雷戈里在1694年提出了三维空间的亲吻数问题：在一颗中心球周围，最多能紧贴放置多 少颗相同的球？对这个问题，数学界直到1952年才得到明确结论。过去50年，亲吻数构造仅有7次实质 性进展，而且每一次突破几乎都依赖完全不同的数学技巧，难以形成可复制的研究路径。 如今，这一数学界关心的经典难题迎来了新的突破。上海科学智能研究院（以下简称"上智院"）、北京 大学、复旦大学的联合研究团队通过设计PackingStar强化学习系统，将高维堆积问题转化为余弦矩阵上 的多智能体博弈学习问题，使AI能够探索远超人类直觉的复杂空间，在25-31维打破了人类已知的最佳 亲吻数结构，同时打破了长期保持不变的1 ...

过去数年里，有研究团队尝试使用AI进入亲吻数问题，但只产生过一次突破：DeepMind的AlphaEvolve 通过修补11维构型，将最优值从592提到了593，但其生成的构型较为混乱，缺乏内在的数学结构，也未 能产生新的数学研究对象，对该领域的推动作用有限，难以普适及进一步提升。 中青报·中青网记者 魏其濛 牛顿和大卫·格雷戈里在1694年提出了三维空间的亲吻数问题：在一颗中心球周围，最多能紧贴放置多 少颗相同的球？对这个问题，数学界直到1952年才得到明确结论。过去50年，亲吻数构造仅有7次实质 性进展，而且每一次突破几乎都依赖完全不同的数学技巧，难以形成可复制的研究路径。 如今，这一数学界关心的经典难题迎来了新的突破。上海科学智能研究院（以下简称上智院）、北京大 学、复旦大学的联合研究团队通过设计PackingStar强化学习系统，将高维堆积问题转化为余弦矩阵（刻 画球心之间几何关系的矩阵）上的多智能体博弈学习问题，使AI能够探索远超人类直觉的复杂空间， 在25-31维打破了人类已知的最佳亲吻数结构，同时打破了长期保持不变的14维与17维的"两球亲吻 数"以及12维，20维与21维的"三球亲吻数"。成果 ...

据了解，此次突破带来了该问题研究的方法论变革。此前DeepMind的AlphaEvolve仅实现11维单点优化，方法难以普适。PackingStar重新定义问 题，将高维几何难题转化为代数计算，形成跨维度迁移路径，突破了传统对称构造思路，发现多维度持平纪录的非对称构型。团队形成稳定人机 协作模式：人类提出研究边界，AI高速构造搜索，人类验证抽象结果，让高维几何探索从单点尝试走向系统推进。 1694年，牛顿与格雷戈里提出三维空间亲吻数问题：中心球周围最多可紧贴放置多少颗相同球体？牛顿认为是12，格雷戈里主张13，该猜想直至 1953年才被证实。作为希尔伯特第十八问题的局部形式，亲吻数问题关联格子理论、球面码等数学分支，且与卫星通信、量子编码、数据压缩等 工程技术紧密相关。2022年，数学家玛丽娜·维亚佐夫斯卡因8维与24维球体堆积最优解证明获菲尔兹奖。 高维空间中，亲吻数问题迅速进入研究"无人区"。过去50年，该领域仅7次实质性进展，方法难以迁移复用。 人类与AI结伴，共同探索科学的未至之境，并且在基础科学研究中实现新的发现与创新、推动社会发展，这可说是人类科学探索史上前所未有的 浪漫图景。上海科学智能研究院（ ...

亲吻数又叫牛顿数，是希尔伯特第十八问题（球体堆积）的局部形式，和通信技术中的"比特拥挤"问题是同一套底层逻辑。 闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 情人节到了… 那咱也来应应景，讲讲亲吻这件事—— AI的打开方式。 你或许知道，数学上有个正经问题叫做 亲吻数（Kissing Number Problem） ，卡了人类300多年，但就在最近，被 中国AI 狠狠推了一 把。 简单说，它研究的是：在n维空间中，一个球体周围，最多能有多少个和它大小相同的球体，刚好与它相切（kiss），不重叠的那种 。 它源自于1694年，牛顿和格雷戈里两位大佬的争吵： 在三维空间里，一个球周围到底能放12个，还是13个同款球？牛顿坚持12，格雷戈里不服，结果谁也没能当场辩过谁。 直到1953年，数学家用了 258年 时间才严格证明牛顿是对的。 就连2022年获得 菲尔兹奖 的玛丽娜·维亚佐夫斯卡， 正是凭借解决8维和24维空间的最密球体堆积问题，摘得桂冠。 但再往高维走，人类的直觉就崩了。在过去近50年里，亲吻数构造仅有7次实质性进展，而且每一次的方法都完全不同，在临近维度上难以迁 移与复用。 现在，僵局被打破了。 ...

（来源：上观新闻） 3维空间尚且如此，进入高维空间更是远超人类的想象，因此300多年来进展缓慢。玛丽娜·维亚佐夫斯 卡，正是凭借在8维和24维的突破性进展，获得2022年菲尔兹奖。 亲吻数问题实则深刻，且有着重要的应用价值。在信息编码中，如何用最少的比特数压缩最多的信息， 其底层逻辑和亲吻数是相通的。 300多年里，这道世界级数学难题，仿佛在等待着一场深刻的相遇。 2 牛顿可能想不到，他在1694年首次提出的亲吻数问题，会成为困扰数学界至今的一道世界级难题。 他一定想不到，300多年后，来自上海科学智能研究院（简称上智院）、北京大学和复旦大学的联合团 队，在一个名叫"人工智能"的助力下，让这一经典难题迎来系统性突破——在人类无法想象的多个高维 空间，打破已知的最优解。 他一定也想知道，这一切究竟是怎么发生的？ 1 亲吻数问题只是看上去简单——在N维空间中，一个球体周围最多能与几个相同的球体相切 （Kissing）。 3维空间就引发了牛顿和数学家大卫·格雷戈里的激烈争论，牛顿猜测说最多12个，格雷戈里说可能有13 个。直到258年后，数学家才严格证明牛顿是对的。 上智院AI Math青年研究员、北京大学博士生马 ...

Core Insights - The article discusses the integration of artificial intelligence (AI) in drug development, particularly focusing on siRNA (small interfering RNA) technology, which has shown significant potential in silencing disease-causing genes. AI models, such as the "Nüwa RNA model," are enhancing the efficiency of siRNA drug screening processes, moving from traditional trial-and-error methods to more precise selection techniques [1][2]. Group 1: AI and Drug Development - The application of AI models has improved in vitro screening efficiency by approximately 1.6 times compared to traditional methods [1]. - The Nüwa RNA model, developed by the Shanghai Institute of Intelligent Science in collaboration with Fudan University, aims to create a living scientific intelligence infrastructure that can be continuously evolved and utilized by scientists [2][3]. - The model integrates over 1 billion RNA sequences, structures, functions, and chemical modifications, achieving leading performance in RNA structure prediction and reverse folding tasks [2]. Group 2: Research and Development Process - The Nüwa RNA model allows for the rapid selection of around 200 high-potential candidates from thousands of sequences within hours, significantly enhancing the drug development process [3]. - The model has already validated siRNA design processes for over five targets, with preliminary experiments conducted for chronic diseases such as hyperlipidemia and hypertension [3][4]. - A closed-loop system has been established, where experimental data is continuously fed back into the AI model, facilitating iterative improvements in drug design [4]. Group 3: Star River Intelligence Platform - The "Star River Intelligence" platform consolidates over 400 scientific models and tools, aiming to lower research barriers and streamline the entire research process [5][6]. - The platform has built a repository of 40,000 high-value scientific datasets and covers nearly 500 million scientific papers, enabling intelligent search and report generation [6]. - The platform is designed to integrate various data, models, and methods into a unified research environment, enhancing the systematic advancement of scientific inquiries [6][7]. Group 4: Collaborative Research and Innovation - The platform promotes cross-disciplinary collaboration, allowing scientists from different backgrounds to work alongside AI algorithm experts, fostering innovation in life sciences [7]. - Successful outcomes from the platform include high-level research results, such as the "Sui Ren Catalytic Reaction Model," which have been published in top journals [7]. - The platform has seen significant engagement, with approximately 23,000 daily visits and active use among over 7,600 students and faculty from Fudan University and its affiliated hospitals [7].

Core Insights - The article emphasizes the role of the younger generation in driving innovation in the field of artificial intelligence, particularly in scientific research [2] - The Shanghai Institute of Scientific Intelligence (上智院) is highlighted as the world's first research institute focused on AI for Science, aiming to transform scientific research paradigms and empower various industries [2] - The SAIS Talk event showcases promising young researchers sharing their innovative work in scientific intelligence, indicating a vibrant future for AI in scientific discovery [3] Group 1: Event Overview - The SAIS Talk has successfully held 15 sessions, featuring speakers from diverse backgrounds, including top scholars and active researchers, to foster inspiration and collaboration [3] - The event on September 26 will feature five young researchers discussing topics such as representation learning, catalytic reaction prediction, and global weather forecasting [3] Group 2: Research Highlights - Research on hierarchical spatiotemporal representation and cross-scale implicit autoregressive modeling significantly improves long-term prediction accuracy in dynamic systems [5] - The RXNGraphormer framework unifies the prediction of chemical reaction performance and synthesis planning, achieving leading performance across multiple prediction tasks [10] - A 4D diffusion model framework for protein dynamics and conformational generation offers new computational paradigms for understanding protein functions and accelerating drug design [13] - The SCRIPT framework for single-cell gene regulatory relationship prediction shows over twofold improvement in long-range regulatory predictions, with implications for complex disease genetic diagnostics [17] - FuXi-Weather, a machine learning-based global weather forecasting system, demonstrates superior performance in sparse observation areas compared to traditional numerical weather prediction systems [21]

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].