罗曼股份 20251009 摘要 罗曼股份 2022 年受疫情和市政建设下滑影响业绩负增长，2024 年因 毛利率下降、管理费用增加及减值计提出现亏损，但此前营收复合增长 率达 39.5%，净利率保持在 15%以上。 公司收购英国霍洛维茨进军数字文娱，拟收购算力服务企业梧桐树布局 算力行业，后者 2025 上半年营收 1.3 亿元，毛利率 25%，净利率 15%，优于同行，且在教育赛道布局较早。 梧桐树科技具备服务器性能、集成服务（全液冷技术）、财务表现和教 育赛道布局四大核心优势，其全液冷技术可将 PUE 降至 1.06~1.15， 显著降低数据中心电力消耗。 梧桐树科技承诺 2025-2027 年累计扣非损益不低于 4 亿元，截至 2025 年 9 月 5 日已确认订单约 5.5 亿元，并中标芜湖 AIDC 项目 30 亿 元，与资兴市政府达成 15 亿元合作框架协议。 若梧桐树科技签约 45 亿元总包标段，在手订单将达 50.5 亿元，按 15.2%净利率测算，可贡献约 7.7 亿元净利润，完成业绩承诺确定性较 高。 Q&A 罗班股份的核心业务及其市场表现如何？ 罗班股份成立于 1999 年，主营业务为 ...

撰文丨王聪 编辑丨王多鱼 排版丨水成文 " AI for Science "的目标之一，是发现能够通过真实世界实验验证的定制材料。目前，我们在计算预测和材料合成自动化方面已经取得了开创性的进展。然而， 大多数材料实验仍局限于采用单模态主动学习 的 方法，依赖单一的数据流，这限制了 对材料设计和性能优化中固有的复杂性进行深入解析的能力。而 人工智能 （AI） 在解释 实验复杂性中的潜力，在很大程度上仍未得到开发。 2025 年 9 月 23 日，麻省理工学院 （MIT） 李巨 教授团队在国际顶尖学术期刊 Nature 上发表了 题为： A multimodal robotic platform for multi-element electrocatalyst discovery 的研究论文。 该研究推出了一个 多模态 AI 机器人平台 —— CRESt （ Copilot for Real-world Experimental Scientists，真实世界实验科学家副驾驶 ） 。该平台将 大型多 模态模型 （LMM，融合了化学成分、文本嵌入和微观结构图像） 与 知识辅助贝叶斯优化 （KABO） 以及 机器人 ...

Core Viewpoint - JingTai Holdings (02228.HK) has achieved significant growth, reporting a revenue of 517 million yuan for the first half of 2025, a year-on-year increase of 403.8%, and marking its first half-year profit of 142 million yuan, attracting market attention [1][2] Company Overview - JingTai Holdings, founded by three MIT PhDs in Shenzhen, focuses on AI-driven drug development and has established a partnership with DoveTree worth up to 5.9 billion USD, leading to a stock price increase of over 40% in ten trading days [1][4] - The company has also signed a memorandum of cooperation with South Korea's Dong-A Pharmaceutical, further expanding its pipeline in immunology and inflammatory diseases [1][2] Growth Trajectory - JingTai's growth trajectory mirrors that of Tencent, both originating from Shenzhen and achieving similar market valuations at IPO, with JingTai being the first profitable Hong Kong-listed company in the AI application era [1][3] - The company is transitioning from a pure AI medical focus to new materials and agricultural technology, establishing a "molecular research and development" infrastructure [2][3] Market Potential - The global drug development outsourcing market is projected to reach 363.2 billion USD by 2030, while the materials science R&D spending is expected to hit 177.9 billion USD, indicating a combined market potential of approximately 3.9 trillion yuan [5][6] - JingTai's AI technology can significantly reduce drug discovery timelines to one-third of traditional methods and lower R&D costs by 70% in the new materials sector [6] Business Model - JingTai has established a profitable business model through robotic R&D services, receiving milestone payments and future sales shares from drug discovery projects, which can lead to substantial profit increases upon successful drug launches [7][8] - The company has announced numerous collaborations with leading pharmaceutical companies, indicating a strong pipeline of potential high-revenue drugs [7][8] Competitive Position - JingTai is building a "molecular research infrastructure" similar to Tencent's information flow platform, aiming to create scalable effects in the AI-driven scientific discovery field [8][9] - The company is expanding into various sectors, including energy, chemicals, and agriculture, showcasing its potential for horizontal industry expansion [9]

Core Viewpoint - The article discusses the potential and challenges of Artificial General Intelligence (AGI) in the context of scientific research, emphasizing the need for a balanced perspective on its capabilities and limitations [1][2]. Group 1: AI for Science Developments - AI for Science (AI4S) is recognized for its value in scientific research, with recent achievements presented at the 2025 Pujiang Innovation Forum, including the Amix-Agent for protein design and the DeepPeptide model for peptide synthesis [2][5]. - The integration of AI in various scientific fields such as biomanufacturing, quantum technology, and climate energy is being promoted through the establishment of the "Scientific Intelligence Strategic Technology Alliance" [5]. Group 2: Six Questions on AGI for Science - The first question addresses the boundaries of AI, questioning whether all scientific problems can be solved by AI, highlighting the historical context of this debate [2]. - The second question examines the predictive capabilities of AGI, cautioning against overestimating current models' ability to predict scientific phenomena due to limitations in existing human knowledge [3]. - The third question focuses on the representation of scientific concepts, suggesting that AI should move beyond natural language to include symbolic languages for better expression [3]. - The fourth question explores the potential for AGI to foster interdisciplinary collaboration, emphasizing its role in revealing unseen connections between different scientific fields [3]. - The fifth question proposes a thought experiment to evaluate AGI's ability to make significant scientific discoveries, using the example of deriving general relativity from prior knowledge [4]. - The sixth question discusses the evolving relationship between researchers, research subjects, and tools, indicating AI's potential to identify valuable patterns in unstructured data [4].

Core Viewpoint - The article discusses the emergence of "LunLun," an AI question-and-answer platform that utilizes a database of academic papers, aiming to democratize access to scientific knowledge and transform the landscape of scientific research and communication [3][4][6]. Group 1: Introduction of LunLun - LunLun is introduced as a groundbreaking AI platform that leverages a vast database of academic papers to provide answers and insights, targeting a broader audience beyond just scientists [3][4]. - The platform aims to bridge the gap between scientific research and the general public, promoting the concept of "Science and Technology Civilization" (STC) [4][6]. Group 2: Features of LunLun - The platform employs an innovative proactive AI interaction model, moving from a traditional "question-and-answer" format to actively disseminating scientific knowledge and insights [7][8]. - LunLun emphasizes the importance of "trustworthy knowledge," ensuring that all responses are sourced from reputable academic papers, thus combating the issue of AI hallucination [11][12]. - The backing of scientists and researchers provides credibility to LunLun, reinforcing its mission to make scientific knowledge accessible to everyone [12][13]. Group 3: Impact on Knowledge Acquisition - LunLun's proactive approach allows users to receive relevant scientific information effortlessly, enhancing the efficiency of knowledge acquisition and fostering curiosity and critical thinking [10][23]. - The platform facilitates cross-disciplinary knowledge connections, enabling users to explore related fields and gain a more comprehensive understanding of complex topics [21][23]. Group 4: Transformation of the Scientific Landscape - The rise of LunLun signifies a shift towards a more inclusive and decentralized scientific ecosystem, where innovation is no longer confined to large institutions but can emerge from diverse communities [15][19]. - The platform's capabilities are expected to reshape the technology industry, promoting localized production and personalized manufacturing through advancements in AI and related technologies [19][20]. Group 5: Future Prospects - LunLun is positioned to create personalized knowledge graphs and real-time research alerts, allowing users to stay informed about significant developments in their areas of interest [24][25]. - The platform represents a significant step towards making scientific research accessible to the general public, potentially leading to a future where everyone can engage in scientific exploration [28][29].

Group 1 - The 2025 Pujiang Innovation Forum opened in Shanghai with the theme "Shared Innovation, Shaping the Future: Building an Open and Cooperative Global Technology Community" [1] - This year's forum has the largest scale and number of attendees in its history, with over 550 guests from more than 300 institutions across 45 countries and regions [1] - The focus of the forum includes the "China-Belarus Year of Technological Innovation" activities, which involve project signings and the establishment of collaborative institutions [1] Group 2 - The forum invites representatives from government departments, scientists, entrepreneurs, and financial institutions to contribute to the national science and technology strategy [2] - Over 30 specialized forums and closed-door meetings are planned, featuring top global scientific award winners to discuss trends in technological revolution and industrial transformation [2] - The InnoMatch technology transfer conference will globally release 10,000 technology demands and showcase over 80 cutting-edge products, facilitating effective connections between technology, talent, and capital [2] Group 3 - The forum has introduced an upgraded AI assistant, "Xiao Jiang Tun 2.0," which offers real-time live streaming, multilingual queries, and smart navigation for an enhanced attendee experience [3] - Since its inception in 2008, the Pujiang Innovation Forum has been held 18 times, aiming to inspire innovative thinking and serve as a platform for international scientific cooperation [3]

Core Viewpoint - The 2025 Pujiang Innovation Forum, themed "Shared Innovation to Shape the Future: Building an Open and Cooperative Global Technology Community," opened in Shanghai, highlighting international collaboration in technology and innovation [1] Group 1: Forum Overview - The forum is co-hosted by the Ministry of Science and Technology of the People's Republic of China and the Shanghai Municipal Government, with Belarus as the guest country and Liaoning Province as the guest province [1] - This year's forum has the largest scale and number of participants in its history, with over 550 guests from more than 300 institutions across 45 countries and regions [1] Group 2: Key Activities and Initiatives - The forum features over 30 specialized forums and closed-door meetings focusing on cutting-edge fields such as AI for Science, quantum intelligence, and controllable nuclear fusion, with participation from top global scientific award winners [2] - The InnoMatch technology transfer conference will globally release 10,000 technology demands, with over 20 billion yuan in corporate investment, and showcase 2,000 talent demand positions along with over 80 first-time products and experience scenarios [2] - The WeStart entrepreneurship investment conference will stimulate innovation and entrepreneurship through diverse formats, with 1,487 teams registered, including 71 international projects, and 100 quality projects selected for roadshows [2] Group 3: Forum's Mission and History - Since its inception in 2008, the Pujiang Innovation Forum has been held 18 times, aiming to inspire innovative thinking, disseminate innovative concepts, and motivate innovative spirit [3] - The forum serves as an important platform for international scientific cooperation, national technological innovation practices, and a gathering space for global scientists and entrepreneurs [3]

论坛上，由中国科学院自动化研究所与武汉人工智能研究院（以下简称"武智院"）联合打造的紫东太初 4.0多模态推理大模型在光谷正式发布。中国科学院自动化研究所副总工程师、武智院院长王金桥在接 受记者采访时介绍，从3.0原生的多模态统一框架到4.0多模态细腻度的复杂思考，紫东太初完成了国产 大模型从"纯文本思考""简单操作带图思考"到"细粒度多模态语义思考"的三重跃迁。 本报讯 （记者李万晨曦） 9月19日，2025东湖国际人工智能高峰论坛在湖北省武汉市中国光谷科技会展中心圆满召开。大会旨在 深入贯彻国家关于加快发展新质生产力的战略部署，积极响应国务院"人工智能+"行动规划，以人工智 能核心要素为创新牵引，汇聚产学研用各界力量，共同推动产业高质量发展。 湖北省、武汉市、东湖高新（600133）区相关单位负责人、业界专家学者、产学研各界代表齐聚一堂， 围绕大模型演进、具身智能、AI for Science及产业发展等前沿议题展开深度交流。 为深入探讨AI赋能科研的核心价值与前沿趋势，本次论坛特别设立"AI for Science"圆桌对话环节，围 绕"人工智能在科研中的代表性成果""AI4S对研究范式的革新影响""紫 ...

Group 1 - The State Council's recent opinion emphasizes the importance of "AI + Science and Technology" as a top priority, indicating a strategic shift to leverage AI for scientific breakthroughs and societal development [1][3] - The concept of "scientific paradigm" is introduced, highlighting how AI is becoming a new paradigm in scientific research, akin to previous shifts from experimental methods to data-driven approaches [4][10] - AI is seen as a powerful tool that enhances research capabilities, enabling scientists to tackle previously insurmountable challenges, such as protein structure prediction [4][6] Group 2 - AI's integration into scientific research is already widespread, particularly in literature review and knowledge synthesis, significantly improving efficiency [5][10] - The development of "scientific large models" is crucial, which are designed to understand and analyze complex scientific data, thus acting as advanced research tools [7][8] - The current stage of scientific large model development faces challenges related to data quality and accessibility, but there is a significant opportunity in leveraging the country's educational resources for data annotation [8][9] Group 3 - AI is breaking down traditional disciplinary barriers, allowing for interdisciplinary research that focuses on problem-solving rather than adhering to specific academic fields [10][11] - The rise of AI is prompting a reevaluation of philosophical and social science research methodologies, expanding the scope of inquiry to include the societal impacts of AI [11][12] - Future scientific research is expected to be transformed by AI, enabling young scientists to focus more on innovation rather than repetitive tasks, thus enhancing their productivity [13][14]

Core Insights - The article discusses the founding of Kepler by Ashton Teng and Quinn Leng, aiming to revolutionize scientific workflows through AI, particularly in the life sciences sector [2][3][5]. Group 1: Company Overview - Kepler is designed to address the inefficiencies in life sciences data analysis, where scientists often wait days or weeks for results, thus slowing down scientific iteration cycles [3][19]. - The company positions itself as the "central nervous system" for research organizations, facilitating literature searches, experimental ideas generation, and data analysis [3][27]. - Kepler's AI Agent can handle multiple queries simultaneously, enhancing the speed and breadth of scientific exploration [3][28]. Group 2: Market Opportunity - The life sciences sector is technologically underserved, with data analysis capabilities lagging behind other tech industries, creating a significant market opportunity for specialized AI solutions [3][19]. - Major pharmaceutical companies are increasingly seeking partnerships with AI startups like Kepler, marking a shift in the industry where collaboration with startups was previously uncommon [7][41]. - Kepler's first client was secured within a week of its founding, indicating a strong demand for specialized AI solutions in the biotech field [5][36]. Group 3: Competitive Landscape - Kepler differentiates itself from general AI providers like OpenAI by focusing on the specific needs of the life sciences sector, addressing the "last mile" problem of integrating AI into existing workflows [5][41]. - The company faces competition from other startups and established firms like Palantir, but its unique focus on life sciences and execution capabilities provide a competitive edge [41][42]. Group 4: Future Aspirations - Kepler aims to expand its technology beyond life sciences into other fields such as materials science, climate science, and agriculture, reflecting the limitless potential of scientific exploration [7][43]. - The company envisions becoming the "central nervous system" for every research organization, emphasizing the need for specialized AI agents tailored to research tasks [43][44]. Group 5: Challenges and Innovations - Kepler is tackling complex challenges in processing large-scale multimodal data, requiring innovative solutions for effective scientist-AI interaction [31][32]. - The company is focused on creating a user-friendly interface that allows scientists to interact with the AI Agent seamlessly, addressing the unique demands of scientific research [34][35].