ENHANCING LARGE LANGUAGE MODEL (LLM) GOVERNANCE LEVEL: A DYNAMIC PERFORMANCE GOVERNANCE (DPG) APPROACH

With the rapid development of artificial intelligence technology, large language models (LLMs) such as ChatGPT, ERNIE Bot, and Kimi have been increasingly applied in various fields including education, healthcare, law, and finance, significantly impacting social progress and technological innovation. However, the widespread application of LLMs has also raised concerns about their output results potentially being inconsistent with human expectations, especially when handling sensitive data that could potentially infringe on personal privacy. These issues not only affect the acceptability and reliability of LLMs but may also trigger broader social and ethical concerns. Therefore, ensuring that the development of LLM is consistent with human expectations.This study aims to explore enhance the government's performance in large language model governance through dynamic performance governance (DPG) methods to achieve a balance between technological innovation and social safety. To construct a governance framework based on a human values framework and guide the development of existing large language models toward better alignment with human expectations. Explore the application of DPG methods in large language model governance and put forward specific suggestions for optimizing governance strategies. Take the medical industry as an example to verify the effectiveness of the DPG method in industry governance and put forward targeted improvement measures. This study aims to provide a reference standard for the government's governance of the large language model by constructing a human values framework and explore ways to improve the governance performance of the government's big language model through the method of DPG. To achieve the above objectives, this study proposed four research questions: "How to help governments build a human values framework that LLM needs to align with?", "How to use the human values framework to evaluate the degree of alignment of LLM?", "How to enhance the government's LLM governance level through the DPG approach?" and "In the LLM governance of the medical industry, can the DPG framework provide effective governance strategy support for the government?". In response to RQ1, this study found that LLM should be consistent with the values and ethical standards of human society, including performance in ideology, legal norms, cultural diversity, and universal ethics. In response to RQ2, this study used an experimental method to evaluate the performance of LLM at different value levels and dimensions, providing empirical data for evaluating the value orientation of LLM. In RQ3, the study proposed a DPG-based conceptual framework that emphasizes the importance of government governance in LLM governance and explains how the DPG approach supports this process. Finally, in the exploration of RQ4, this study analyzes the help of DPG in large language model governance in specific industries through case studies of Chinese medical industry scenarios.This study first reviews the literature and comprehensively analyzes the technical development, application areas and governance risks of LLM. Subsequently, a multidimensional framework of human values that encompasses ideology, legal norms, cultural diversity, and universal moral ethics constructed to provide a theoretical basis for understanding and evaluating the value orientation of LLM. And through experimental evaluation of the value alignment level of popular LLM, reference standards and guidelines are provided for further improvement of LLM technology. Based on the constructed human values framework, this study tests the value alignment level of three existing popular LLMs. The experimental results show that although models such as ChatGPT 4.0, ERNIE Bot 3.5 and Kimi perform well in the field of universal ethics, there are still significant deviations in the ideological and legal norms. In addition, this study reveals the value judgment tendencies of LLMs in different cultural and social contexts and points out the improvement directions for future research in data diversity, model understanding, cultural diversity considerations, flexibility of experimental design, and depth of result interpretation. This study adopts the DPG method and proposes a conceptual framework based on DPG, attempting to help the government improve the level of LLM governance. Finally, the feasibility of the constructed DPG framework is verified using DeepSeek’s application scenario case in the Chinese medical industry.In summary, this study constructs a human values framework that can serve as a standard and guide for the governance of government LLMs. Using the constructed human values framework, a comprehensive analysis and experimental evaluation of the current LLMs for value alignment were conducted to show how the framework is used. By analyzing the characteristics of LLM governance, this study proposed a DPG framework to achieve sustainable development of LLM. This study provides a new perspective for understanding the complexity and dynamism of LLM governance, provides theoretical guidance and practical reference for policymakers and practitioners, and promotes the coordinated development of LLM technology and social values. By constructing a DPG framework, the government's LLM governance is explored, and the feasibility of the DPG method in the LLM governance of specific industries is verified through cases. The research results provide valuable inspiration for understanding and improving LLM governance, provide direction for future research and practice, ensure the harmonious unity of technological development and social values, and provide a scientific basis for decision-making that is more in line with human values.