Group 1 - The core theme of the event was "Work Methods in the AI Era," focusing on the evolution of human-machine collaboration [1] - DingTalk's CTO, Zhu Hong, emphasized that AI will evolve into an "intelligent agent" with cognitive and decision-making capabilities, enhancing efficiency and decision-making processes [1] - DingTalk aims to become a factory for industry models and enterprise intelligent agents, enabling businesses to build and apply their own exclusive large models at low cost and high efficiency [1] Group 2 - DingTalk introduced a pioneering "pay-per-performance" AI model, allowing enterprises to use AI with reduced financial risk, thus promoting wider adoption [2] - The company has successfully trained China's first vertical medical large model, achieving a "Chief Physician" professional standard, showcasing its capabilities in specialized fields [2] - DingTalk is committed to creating an open, intelligent, and human-centered collaborative work platform, facilitating deep integration of AI into various industries [2]

Core Insights - DingTalk showcased its AI products at the 2025 Yunqi Conference, introducing the next-generation office application DingTalk One, the first AI hardware DingTalk A1, and five other AI innovations [1] - DingTalk's CTO emphasized that the essence of AI transformation is the evolution of human-machine collaboration, where AI will act as an intelligent agent capable of cognitive and decision-making abilities [1] - The company aims to become a factory for industry models and enterprise intelligent agents, enabling businesses to build and apply their own exclusive large models efficiently and cost-effectively [1] Group 1 - DingTalk has established an AI productivity platform that assists enterprises in training exclusive models from data labeling to API deployment [1] - The company introduced a "pay-per-performance" model, allowing businesses to use AI first and pay based on results, making AI more accessible [1] - DingTalk has successfully trained the first vertical medical large model in China, achieving a "Chief Physician" professional level [2] Group 2 - Collaborations with partners like Gujia Home and Tongyi Laboratory have led to the development of industry-specific voice models that enhance AI sales assistance and service quality inspection [2] - The company is committed to creating an open, intelligent, and human-centered collaborative work platform, aiming to connect AI with the physical world across various industries [2]

Core Viewpoint - The year 2025 is seen as a critical juncture for the practical application of enterprise-level AI, transitioning from experimental tools to essential components of business operations, despite challenges in scaling and execution [1][5]. Group 1: AI Implementation Challenges - Companies face significant gaps between AI technology awareness and practical application, with discrepancies in understanding and goals between management and execution teams [8]. - A majority of AI projects (90%) fail to meet expectations, with 70% of executives reporting unsatisfactory results, primarily due to viewing AI merely as a tool rather than a collaborative partner [16][18]. Group 2: Data Quality and Management - Data quality issues span the entire data lifecycle, affecting AI implementation outcomes, with many CIOs questioning the value of accumulated data [31][33]. - The Hong Kong Hospital Authority has accumulated nearly 6 billion high-quality medical data points over 30 years, emphasizing the importance of structured data for effective AI application [36]. Group 3: AI Reliability and Interpretability - As AI becomes more widely adopted, ensuring the reliability and interpretability of AI technologies is crucial, particularly in high-stakes environments like finance [21][24]. - The "model hallucination" issue, where AI generates incorrect information, poses significant challenges for trust and compliance in sectors requiring high accuracy [23][28]. Group 4: Scene Selection for AI Projects - Companies often struggle with selecting appropriate AI application scenarios, caught between the allure of technology and practical business needs [44]. - The case of Yixin demonstrates how AI can transform financial services by providing tailored solutions to underserved markets, highlighting the importance of aligning technology with user needs [46][48]. Group 5: Knowledge Base Development - A dynamic and continuously updated knowledge base is essential for maximizing the value of AI applications, moving from static information storage to knowledge-driven processes [78][80]. - The Eastern Airlines' approach to knowledge management illustrates the shift towards integrating AI into operational processes, enhancing efficiency and service quality [83]. Group 6: Human-Machine Collaboration - The evolution of AI agents from simple task executors to collaborative participants in complex business scenarios is critical for digital transformation [87]. - Companies like Midea are leveraging AI to enhance production efficiency and redefine operational models, demonstrating the potential of AI in driving business innovation [89][91]. Group 7: Talent Acquisition and Development - The competition for AI talent is intensifying, with a significant mismatch between the demand for skilled professionals and the available talent pool, highlighting the need for strategic talent management [97][99].

Core Insights - DingTalk has established a new business line focused on industry-specific models, reporting directly to CTO Zhu Hong, marking a significant move in its AI strategy following the return of founder Wu Zhao [1][2] - The company has engaged with multiple industry clients and is advancing several industry-specific models, emphasizing the need for tailored AI solutions for vertical industries [1][2] - DingTalk's AI capabilities have been enhanced since April, with the integration of large model foundational capabilities and the launch of an AI assistant in January 2024 [1][2] Group 1 - The establishment of the industry-specific model team reflects the ongoing implementation of large models in enterprise settings, addressing the challenges faced by businesses, particularly SMEs, in adopting AI [2] - DingTalk provides comprehensive model training and data engineering services for SMEs lacking AI talent, ensuring that the models are tailored to specific business scenarios [2] - The first successful deployment of a vertical model, the Doukou Gynecology model, improved diagnostic accuracy for six major gynecological symptoms from 77.1% to 90.2% [2] Group 2 - In addition to developing industry-specific models, DingTalk is revamping its application market to create a closed AI ecosystem, with plans to launch an AI Agent Store in April 2024 [3] - The company aims to enhance the Agent market by opening capabilities to more ISVs and enterprises, facilitating the development of Agent applications and commercializing them through DingTalk [3]

Core Insights - The article discusses the limitations of general large language models in clinical scenarios, particularly in providing accurate medical diagnoses, highlighting the need for specialized training methods like supervised fine-tuning (SFT) [1][2][3] - The performance of the Doukou Gynecology model improved significantly from an initial accuracy of 77.1% to 90.2% through targeted SFT processes [1][3] Data Quality Control - The training dataset underwent a rigorous selection process involving systematic data cleaning, ensuring consistency between reasoning and results, and verifying the logical integrity of the data [2][5] - Low-quality data, such as those with clear medical inconsistencies, were excluded to maintain high standards [2] Model Training Phases - The first phase involved building a foundational SFT model using 1,300 meticulously labeled gynecological consultation data, achieving an initial accuracy of 77.1% [3] - The second phase focused on synthesizing symptom data and refining the model, resulting in a final diagnostic accuracy of 90.2% for six major gynecological symptoms [3][6] Iterative Optimization - Continuous iterative optimization was implemented, where high-quality samples scoring above 8 were added to the training set for further SFT, creating a cycle of training, evaluation, and retraining [10][18] - Key performance indicators were monitored throughout the process to ensure comprehensive model improvement [10] Evaluation System - A dual evaluation system was established, combining automated assessments with manual reviews by medical experts to ensure diagnostic accuracy [11][13] - The automated evaluation system utilized a high-performance language model to objectively score outputs based on a structured framework [11] Challenges and Lessons Learned - Initial reliance on manual labeling slowed data accumulation and increased costs, prompting a shift to a more efficient "machine distillation → expert review → post-training evaluation" system [14][15] - The model's ability to recognize rare diseases was enhanced through balanced sampling strategies [15] Future Directions - The company plans to explore a collaborative training paradigm combining SFT and reinforcement learning (RL) to enhance clinical reasoning capabilities [18]

作者 | 褚杏娟 近日，钉钉企业专属 AI 平台上成功训练出了首个高准确度、高实用性的专业领域大模型——由壹生 检康 (杭州) 生命科技有限公司研发的"豆蔻妇科大模型"，其在专业测试中准确率达 90.2%。 钉钉方面表示，妇科大模型的落地，意味着钉钉生态已经从 SaaS 生态、服务商生态、咨询生态、 交付生态，拓展到 AI 创业者。 与专业医生诊断吻合度达 90.2% 当前，各行各业都在努力将大模型与自身业务场景深度融合，打造行业或专业大模型，实现运营管理 的降本增效。 壹生检康是一家深耕女性精准检测及健康服务的生命科技公司，创业团队大多来自知名互联网企业、 妇产科医疗机构、生物医药公司。基于技术趋势和行业判断，王强宇团队认为，通过训练妇科专业大 模型打造 AI 医生，将有效缓解专业妇科医生、医疗服务不足的难题，对医美机构和女性用户都会带 来巨大的行业和社会价值。 专业性强的"妇科 AI 医生"并不是采用通用大模型就能简单训练出来。启动豆蔻妇科大模型研发以 来，壹生检康团队以开源大模型为基础，通过行业数据训练，第一个版本将模型诊断准确率做到 77.1% 左右。"77.1% 的准确率虽达到行业基础标准，但对于直 ...

Core Insights - The article discusses the "resource misalignment battle" in the AI sector, where large companies focus on parameter upgrades while smaller startups target niche markets that larger firms overlook [1] - The medical industry is highlighted as a high-risk area with stringent accuracy requirements, making it difficult for general models to meet specific needs [1] - There is a growing recognition among AI companies of the importance of specialized models in vertical fields, particularly in healthcare [1] Industry Analysis - The medical field requires vertical models to achieve higher accuracy, with general models only reaching a passing score [1][2] - The relationship between general and vertical models is likened to that of a medical student and a specialized doctor, emphasizing the need for extensive practical experience [2] - Companies like 壹生检康 are focusing on developing specialized models to address the limitations of general models in specific medical scenarios [4][5] Model Development - 壹生检康 has been developing a gynecological vertical model, selecting a 32B parameter model as the optimal balance between computational resources and response effectiveness [5][7] - The training process involved multiple rounds, with the first round yielding a 50% accuracy rate, which improved to 77.1% after addressing data imbalance issues [6][13] - The final model demonstrated superior performance compared to existing models, particularly in diagnosing specific gynecological conditions [13][14] Application and Impact - The gynecological model aims to provide precise and professional services to end-users, addressing common health issues faced by young women [18] - The model is also designed to empower healthcare providers in resource-limited settings, enabling them to offer reliable gynecological consultations [18] - The use of reinforcement learning is suggested as a future direction to enhance the model's capabilities and expand its application to other medical fields [19]