[author]WANG Lusheng

[content]

On the dynamic shaping of the legal large language model in the legal field

Wang Lusheng, 

Professor of Law School of Southeast University and researcher of the future rule of law and digital intelligence technology innovation laboratory

Abstract: Legal large language models (LLMs) are large technological systems that embody both technical and social attributes. The legal domain regulates the production, supply and allocation of data, knowledge, organizations, institutions and symbolic resources in the legal domain through multiple guidance of value pursuits, knowledge foundations, organizational preferences, institutional preparation and cognitive structures, and ultimately actively shapes the functional characteristics, application scenarios and diffusion paths of the legal LLMs. The resource endowment and constraints within China’s legal domain shape the technological style of Chinese legal LLMs through both technological impetus and technological constraints. Therefore the legal domain should not be considered merely as an application scenario or a transformation target by legal LLMs, but should be given the active role to break the resource dilemma in the legal domain. This requires taking the promotion of self-reliant innovation in China’s legal LLMs as the overarching goal, and coordinating multi-level intentions within the domain to form a consistent and proactive shaping action. On this basis, with organizational resources as the center, the complex interaction relationships among various resources are simplified, the optimization plan of domain resources is constructed, and the technological style of legal LLMs is built in line with the core requirements of Chinese modernization of the rule of law.
Key words: legal field legal big model generative artificial intelligence system theory dynamic shaping

1. Introduction

Since the end of 2022, the large language model has accelerated iteration and increasingly showed good generalization ability and emergence characteristics in general tasks. However, the characteristics of the universal large-scale model, which is broad but not specialized and comprehensive but not precise, will inevitably give birth to its deep adaptation to the needs of the times in the vertical fields such as law and medical treatment. It is generally believed that non-human agents need to have three prerequisites for legal people to feel the survival crisis: no less than legal people's professional skills; Self growth, even consciousness; Legal person and AI have a very broad and daily survival connection. In the long run, the big model of law may change the mode of production of legal knowledge and show its potential in many legal businesses; At the same time, based on the training of large parameters, the large legal model shows that the emergence and generalization ability of legal tasks has the primary level of "self growth"; Further, the promotion of the large legal model may establish a human-computer cooperation legal business model, thus forming a ubiquitous connection with the daily business of legal persons. Although many visions of the future imply the metaphor of AI replacing the legal person, the crisis has not been urgent before. The emergence of the legal model may mean that the prelude to the self digestion of the legal person's subjectivity has been played.
In general, the current academic research around the legal model is relatively limited, and focuses on one side of the coin, that is, the role of the legal model in shaping legal practice, legal organizations and legal profession. On the other side of the coin, there is even less research on "the big model of the influence of the legal field on the law". From the existing research, the impact of legal organizations and legal professionals on the legal model is negative or even hidden: it is either regarded as the applicator of the legal model, and may lead to ineffective technology promotion due to lack of technical literacy; Or it is regarded as the object of "transformation" by technology, and is treated negatively for fear of being replaced. In this narrative framework, actors in the legal field are passively involved in the tide of scientific and technological discipline, and "obedience" has become the natural extension of legal science and technology to shape legal modernity. The above perspective oversimplifies the development and application logic of the large legal model. In fact, the large legal model is not independent of the "independent existence" in the legal field, but is deeply embedded in the data, knowledge, organization, system and symbol resources in the legal field. The research and application results of the large legal model are jointly shaped by the technical field and the legal field. This is why although they have similar technical attributes, the promotion of large legal models in different countries is significantly different, some are in the ascendant, and some are superficial; Different subjects may use the same legal big model technology for different purposes, but its application patterns may be very different.

This paper attempts to explore the dynamic role of the legal field in shaping the legal model. Specifically, first of all, it reveals the essence of "large-scale technological system" with both technicality and sociality of the legal model at the level of theoretical basis, and the deep mechanism of how the legal field can shape it dynamically; Secondly, it analyzes how the resource endowment in China's legal field shapes the technical style of China's legal model, and on this basis, it combs the systematic constraints brought by the resource dilemma in China's legal field to China's legal model; Finally, based on the dynamic shaping theory, facing the independent innovation of China's legal model, this paper puts forward a systematic scheme to solve the resource dilemma in the legal field.

2. The theoretical basis of actively shaping the legal model in the legal field

The theory of actively shaping the legal model in the legal field regards the legal model as a large-scale technical system, and believes that the data, knowledge, organization, system, symbols and other resources in the legal field interact to jointly shape the functional representation and final form of the technology in the legal field. Under this theoretical framework, the big legal model is neither the independent force of essentialism nor the neutral object of instrumentalism that is divorced from the domain goal. Technological advancement is only a necessary and insufficient condition for the research, development, application and diffusion of the legal model. The legal field can influence the production, supply and allocation of resources in the legal field by purposeful regulation and guidance of value pursuit, knowledge base, organizational preference, institutional preparation, cognitive structure, and ultimately complete the dynamic shaping of the legal model.

2.1 Legal model as a large-scale technology system

Large technical system was proposed by Thomas Hughes, a scholar in the history of technology and sociology of science and technology. It refers to those systems with large scale and high complexity, involving many technical components (such as physical equipment) and social elements (such as organizational structure, technical standards, laws and regulations), such as power, railway, telecommunications, etc. Compared with general technology systems, the "technical momentum" of large-scale technology systems makes it increasingly difficult to change or stop due to the loss of plasticity and flexibility after the system expands and matures.
The legal (vertical) large model is a generative artificial intelligence system based on the large language model, which is specially designed for the legal field, aims to meet the strict requirements of the legal scene for accuracy and fairness, and has strong legal language understanding, professional task generalization and trusted knowledge generation ability. In essence, it is a large-scale technical system composed of the hardware components of the large language model (such as computing power and energy infrastructure) and the legal field as a social element, which has both technical and social attributes. First of all, as a new infrastructure in the legal field, the legal model includes front-end system applications that interact with users, computing infrastructure such as servers and cloud storage that need to be relied on for training and operation, and energy infrastructure that supports training and reasoning. Secondly, in the process of research, development and application of the large legal model, technological components and social elements are inseparable. It is not only subject to the innovation and development of artificial intelligence technology, but also subject to the intertwined influence of the value pursuit, organizational preference, institutional preparation and cognitive structure in the legal field around technological innovation and application. It can be seen that the hardware equipment to achieve a specific purpose, specific skills and knowledge, the huge physical structure and the bureaucratic system of the organization are the source of power for the development of the legal model, which determines the sustainability of the growth and evolution tendency of the legal model. "The problem of the unsmooth development of technology often lies not in the technology itself, nor even in a single social reason, but in complex social conditions." when technological innovation is not compatible with the legal field, the development direction and speed of the legal model will be substantially restricted. Finally, in the early stage of the development of the large legal model, due to the lack of "technical momentum", technology has a high degree of flexibility and adaptability, so it is more likely to be shaped by the initiative of the legal field. With the technology diffusion and deep application of the large legal model, its "technological momentum" is accumulating and forming an irreversible impact on the legal field.

2.2 Legal field as an action field and resource system

The legal field can be understood as a functional action field and complex resource system with unique value objectives, dynamic power structure, systematic normative system and clear code of conduct. Externally, the legal field is an irreplaceable social subsystem differentiated from the functions of modern society, which maintains its independent operation by referring to its relationship with the whole society, adjacent subsystems and itself; In this process, the legal field constructs itself by creating and maintaining the differences with the environment, and uses the system boundary to adjust the differences, so as to realize the functional operation of "self reference". Internally, the legal field is a field of independent action composed of specific rules, logic and power relations. Actors generate meaning and construct the reproduction structure of resource elements according to their own codes, outlines and procedural logic in the legal field. Through the production, supply and allocation of data, knowledge, organizations, systems, symbols and other resources, the legal field constantly adjusts the structure and function of the field, and shapes its complexity, diversity and dynamics. Furthermore, according to the characteristics of "closed operation" but "open cognition" of the social system, the legal field still needs to form resource exchange with other fields on the premise of ensuring its own independence, so as to maintain its own production and supply. It can be seen that there is a functional or structural "initiative" in the legal field, which is reflected in maintaining self reference, that is, responding to external environmental disturbances through internal communication and structural coupling. Compared with the "hard" components of the large legal model, the legal field provides irreplaceable resource support for the whole chain of research and development and application of the large legal model from the "soft" dimensions of technical basis, theoretical accumulation, cooperation mechanism, normative guarantee, and cultural identity.

Data resources in the legal field refer to the data resources represented by legal documents (including judgment documents, contract texts, etc.), laws and regulations, and legal papers, which are used to support the pre training, optimization, and operation of the large legal model. Data resources in the legal field are constantly generated by legal organizations and legal professional groups in the process of legal practice. The generation process embeds the value pursuit, professional knowledge and practical experience of legal professional groups, which is the direct embodiment of consensus in the legal field. Compared with general data resources, data resources in the legal field are highly professional, authoritative and normative.

Knowledge resources in the legal field refer to technical knowledge such as data annotation, value alignment, and instruction fine-tuning in the development process of the large legal model, as well as technical literacy such as legal tips, information identification, and risk aversion in the application process of the large legal model. Polanyi's epistemology contains a binary distinction between explicit knowledge and tacit knowledge. Explicit knowledge is mainly expressed in written words, charts and mathematical formulas; Tacit knowledge mainly refers to practical knowledge that is difficult to express and has individual characteristics and context dependence. As far as the legal model is concerned, the explicit knowledge with the accumulation of technology in the legal science and technology community as the core and the tacit knowledge with the technical literacy in the legal theory and practice circles as the core constitute the knowledge resources in the legal field. The formation of both stems from the common shaping of legal habits, values and culture. In terms of technical knowledge, they are embodied in the objective knowledge system embedded in legal professional knowledge and the needs of legal scenes. In terms of technical literacy, they are embodied in the practical experience of using technology in specific legal situations.

Organizational resources in the legal field refer to the resource network supporting the research, development, application and diffusion of the large legal model, with organizational preference, organizational structure and organizational ability as the core connotation. The research on technology under the framework of organization theory has experienced a paradigm shift from emphasizing the influence of technology on organization in the sense of determinism to emphasizing the interaction between technology and organization. In the above transformation process, the role of organizations in shaping technology has attracted increasing attention, and has gradually become a key resource affecting the evolution of technology. The organizational resources in the legal field have distinct characteristics. On the one hand, the preferences of various organizations in the legal field are deeply shaped by legal professionals at the individual and group levels. Specifically, the behavior preference of individual legal professionals has a micro impact on the preference of legal organizations through the daily accumulation of legal practice. On this basis, heterogeneous individual preferences rely on the organizational environment to achieve coordination and integration, and gradually converge to form the overall preferences of legal organizations. On the other hand, under the framework of structural functionalism, the structural elasticity of various organizations in the legal field is deeply affected by the positioning of organizational functions.

Institutional resources in the legal field are various formal and informal policy guidelines, industry standards, and ethical norms that support and regulate the development and application of the legal model. Institutional resources exert incentives and constraints on scientific and technological innovation and application in the legal field at all levels through the organic combination of formal rules, informal norms, material means and discourse framework. It is worth noting that the institutional resources in the legal field are not the institutional resources of all social phenomena, but the general name of various institutional norms for the research, development and application of legal science and technology produced by the legal field organization at the individual and overall levels.

Symbolic resources in the legal field refer to the recognition and acceptance of the legal model by professional groups in the legal field and the public receiving legal services. The symbolic resources in the field of law are shaped by cognitive mechanisms such as the thinking mode, knowledge framework and decision logic in the field of law. Bourdieu believes that symbolic capital is the capital that exists in the form of symbols, such as respect, honor, reputation and recognition that individuals or groups obtain in a specific social environment. Its essence is a resource based on recognition and reputation, which is intangible, symbolic and scarce. We can regard the capital related to honor and reputation as the symbolic resources of the legal model, that is, a kind of trust and expectation at the overall level of the legal field composed of non-material factors such as the recognition and trust of legal organizations and legal professional groups.

2.3 The dynamic shaping mechanism of the legal model in the legal field

The overall initiative of the legal field does not point to the "consciousness subject", but refers to its role as an action field and resource system to regulate the production and supply of resources in various fields through collective behavior, cultural norms, institutional arrangements and other ways, so as to shape the functional representation, application norms and social acceptance of the legal model. The big model of law is not an "independent existence" independent of social time and space, but a large-scale technical system deeply embedded in the legal field.

2.3.1 the dynamic shaping mechanism of data resources in the legal field on the legal model

The accessibility and availability of data resources in the legal field directly affect the comprehensiveness, practicability, timeliness and accuracy of the generated content of the legal model, and then deeply shape its legal language understanding, legal task generalization and credible knowledge generation ability. Among them, the pre training of laws and regulations data helps to strengthen the "understanding" of the large language model on the legal provisions' system structure, logical connection and normative content, and enhance its ability to solve legal problems by using legal rules. At the same time, the pre training of legal documents can help the big language model to strengthen the understanding and application of legal expression, strengthen the analysis of legal cases, and optimize the performance in the scenes of legal consultation and legal argumentation. In addition, if we can make full use of the huge amount of legal document resources such as electronic contracts, it will be conducive to strengthening the performance of the legal model in the tasks of contract generation, interpretation and analysis. Furthermore, the pre training of legal papers can help the legal model deepen its grasp of legal theory, improve its ability in legal epistemology and methodology, and optimize its performance in legal thinking and legal logic reasoning.

2.3.2 the dynamic shaping mechanism of knowledge resources in the legal field on the legal model

The explicit knowledge in the legal field of the development and application of the large legal model mainly includes: various algorithms designed and trained for specific tasks in the legal scene; Technical specifications for data acquisition, cleaning, labeling and management; Develop and deploy industry standards and engineering solutions for legal AI systems. Explicit knowledge in the legal field provides a reproducible, operable, verifiable and transmissible knowledge base and practice framework for the development and application of the legal model, which is conducive to the standardization and consistency of technology development and application. Correspondingly, the tacit knowledge in the legal field mainly includes the technical literacy of the legal person using the legal model at the level of experience and skills. Among them, technological cognitive literacy determines the perception and understanding of the legal model by the actors in the legal field, and profoundly shapes the way in which the technology is actually used; It suggests that literacy determines whether users in the legal field can interact with the legal model efficiently and accurately, and affects the accuracy and availability of generated content; Information identification literacy determines whether users in the legal field can distinguish the accuracy and reliability of the content generated by the legal model, so as to effectively avoid the illusion of knowledge and ensure the decision-making mode in the loop. When the technical efficiency is the same or similar, the technical literacy (knowledge resources) of legal professional groups will fundamentally affect their understanding and use of the legal model.

2.3.3. the dynamic shaping mechanism of organizational resources in the legal field on the legal model

Legal organizations limit and guide the adoption and implementation of the large legal model through technical preferences, structural flexibility, and practice scenarios. Firstly, organizational technology preference (Culture) affects the judgment of technological comparative advantage, compatibility and complexity, and determines the adoption of technological innovation by organizations. When the large legal model can meet the business needs of legal organizations, the technology is more likely to be widely used and accelerate the diffusion. Secondly, the structural characteristics of legal organizations, and the organizational flexibility between internal and external, horizontal and vertical, determine the possibility for legal organizations to respond, absorb and transform into actual effectiveness when facing the opportunities and challenges brought by the large legal model. Finally, the legal model can spread among organizations through isomorphic pressure and peer competition. When the R&D and application of the large legal model has become a general trend in the industry and evolved into the pressure and goal of legal organizations, legal organizations will imitate each other in the competitive self adjustment in order to increase their institutional legitimacy, and eventually tend to be "identical" in promoting the innovation and Application of the large legal model. With the diffusion of the legal model among legal organizations, legal organizations themselves become the supporting resources for its diffusion.

2.3.4. the dynamic shaping mechanism of institutional resources in the legal field on the legal model

The institutional resources in the legal field with different levels and attributes affect the development and application of the legal model through incentives or constraints. Among them, policy guidance refers to the general name of various guidance documents issued by political and legal organs at all levels to guide and regulate the research, development and application of legal science and technology. Policy guidance can guide the development direction of legal science and technology in different granularity, optimize the allocation of funds and resources in the legal field, and effectively control the risk of legal science and technology. Industry standards shape the technical effectiveness, legitimacy and legitimacy framework of the legal model through the setting of standardized processes and technical indicators. Ethical norms provide ethical constraints and bottom line standards for the development and use of the large legal model, and ensure that the large legal model develops at a high speed while paying attention to ethical legitimacy.

2.3.5. the dynamic shaping mechanism of symbolic resources in the legal field on the legal model

The essence of the supply of symbolic resources in the legal field is the process of convergence of the attitudes, values and beliefs of potential users of the legal model. This resource shapes the research, development, adoption and diffusion of the legal model by influencing the subjective cognition of the actors in the legal field. This is why some scholars believe that when deciding to adopt technology, the social forces formed around technology perception are more important than the physical attributes of technology. When the legal field has a high degree of trust in the legal model, it means that the technology conforms to the values and symbol system of the field, and the actors in the field will have more expectations for it, thus encouraging the relevant subjects to invest more resources to support the research and development of the technology. Therefore, symbol resources largely determine the initial adoption of technology. Furthermore, in the process of the development of the legal model, the common recognition and professional trust of the actors in the legal field will further reduce the resistance to promotion and improve the diffusion speed. It is worth noting that the accuracy of technology is not a sufficient and necessary condition for the subject's trust in the model, and the trust of social groups in the legal model is also related to the interpretability and transparency of the algorithm behind the legal model.

3. Resource endowment and resource dilemma in the legal field of China's legal model

The big model of law is not simply a "technological object" that exerts influence on the legal field, but a "social technological object" that is deeply influenced by the resources in the legal field. For this reason, while following the technical logic, the development and application of China's large legal model also need to pay full attention to the comprehensive impact of China's resource endowment and resource dilemma in the field of law on the two levels of technical power and technical constraints.

3.1 Resource endowment in the legal field of China's legal model

The resource endowment formed by the full accumulation of data resources, knowledge resources, organizational resources, institutional resources and symbol resources in China's legal field shapes the technology choice, path dependence and development mode of China's legal model, forming a distinctive policy oriented and application driven technical style.

3.1.1. data resource endowment in the legal field

The all-round digital transformation of China's legal field has accumulated a wealth of field data resources for the training of the legal model. This data resource determines the overall technical style of the localization and customization of the Chinese legal model, as well as the goal orientation of legal semantic understanding, legal reasoning and legal decision-making that adapt to the Chinese context. In terms of laws and regulations, the construction of various large-scale laws and regulations databases provides a natural training data set for the development of large legal models. For example, the national laws and regulations database, as an official database built by the national legislature, contains nearly 26500 legal texts including the constitution, existing effective laws, legal interpretations, and local regulations, covering the main content of the socialist legal system with Chinese characteristics. In terms of legal documents, the massive accumulation of electronic legal documents at the national and social levels can provide full support for the legal model. Although the online public judgment documents provide important support for various legal ai r&D subjects, the published part is only a very limited part of the data resource pool of legal documents in the whole society. For example, the number of signed electronic contracts in China reached 133.71 billion in 2023 alone. In terms of legal papers, the huge stock and considerable increment of electronic legal papers are also the important digital resource base of China's legal model. The digital library of the Supreme People's court alone has gathered 150.5 million documents with 1.929 trillion words.

3.1.2 knowledge resource endowment in the legal field

Under the background of the gradual improvement of the new national system in the field of scientific and technological innovation, relying on the project layout of the Ministry of science and technology's national key R&D plan in the fields of public security, smart justice and so on, China has produced a number of key algorithms and models around the fields of legal big data and legal artificial intelligence, forming an important explicit knowledge base for the research and development of China's legal big model. According to the data, as early as 2018, more than half (51%) of the legal science and technology patents applied worldwide came from China; The following year, the proportion of legal science and technology patents in China rose to 61.9%. In recent years, Tsinghua University, Peking University, Zhejiang University, Fudan University, Shandong University, Southeast University and other universities have successively released powerlawglm, chatlaw, Zhihai · Luwen DISC-LawLLM、 Master · insight, legal balance and other legal models. Technology companies are also actively developing large model products and applications for the legal industry, such as intelligent legal systems and legal intelligent assistants. It can be seen that the current in-depth exploration around legal AI in various fields has provided more explicit knowledge resources for the research and development of the large legal model.

In terms of tacit knowledge, the prosperity of "letting a hundred flowers blossom" and "general mobilization" in China's digital rule of law practice has boosted the knowledge enlightenment around legal science and technology in the legal field. As the social shapers of technology believe, actors tend to understand new technology through familiar patterns or frameworks. On the one hand, organizations and individuals in the legal field have formed a wealth of tacit knowledge around the deployment and application of cutting-edge technologies, logical mechanisms, and risk issues at different stages of digitization, networking, and intelligence; On the other hand, the public has also improved the technical literacy and practical knowledge of participating in digital judicial activities in an integrated and universal way in the transformation of the digital rule of law, and has generated awareness of its basic concepts, application methods and other aspects in the interaction with legal AI. All these provide sufficient practical experience support for the development and application of the large legal model.

3.1.3 organizational resource endowment in the legal field

The organizational needs and preferences of various organizations in the legal field, the strong flexibility and plasticity of the organizational structure, and the Inter Organizational pressure and competition around the application of technology jointly form the domain organization resource endowment of China's legal model. This resource endowment determines that the R&D and application of China's large legal model has the characteristics of global mobilization for national needs, and also determines that China's large legal model still continues the technology diffusion style of combining local pilot doctrine and peer competition to a certain extent.

First, in terms of organizational needs and preferences, the legal model has a wide range of application scenarios in the legal field. The application of intelligent technology has always carried positive, positive and eager expectations, and is regarded as a new road and new driving force for the modernization of the rule of law in China. Political and legal organs not only have realistic organizational needs for the legal model in dealing with the "litigation explosion", strengthening cooperation and restriction, optimizing the allocation of functions and powers, and improving legal services. As a political and legal organization community at the national level, it also connects the national system and field practice of the innovation of the legal model through the organizational action path of national promotionism and holism, so as to realize the effective alignment of organizational preferences.

Second, in terms of organizational structure and flexibility, the political and legal organs provide strong organizational support for the introduction of the legal model through the special adjustment of the internal structure. From the perspective of adaptive structure theory, legal organizations not only provide application scenarios for the large legal model, but also realize the adaptive use of the large legal model through self adjustment of their own internal structure. At the national level, the Supreme People's court and the Supreme People's Procuratorate have successively set up internal institutions or coordination bodies such as the information technology service center of the people's court and the leading group for digital procuratorial work, effectively improving the flexibility of the organizational structure to cope with the complexity of intelligent construction tasks. At the local level, the political and legal organs at all levels also optimize the institutional setting and authority allocation within the organization according to the local reality, and set up internal departments specialized in the management and application of legal science and technology, such as the technology department and the information department, to coordinate the informatization and intelligent construction. In general, on the one hand, these specialized agencies are flexible in scale. Under normal conditions, they are responsible for organizing daily work such as the evaluation and guarantee of internal technology applications, and can quickly organize a large number of human and material resources when major projects involving the application of legal science and technology are involved; On the other hand, it is flexible in terms of rights and responsibilities. The content it is responsible for is centered on technology, which can cover all business areas within the organization and effectively serve the complexity needs of the management of the large legal model.

Third, in terms of organizational pressure and competition, the Chinese legal model is expected to obtain multidimensional and multi-layer organizational resources support in the close interaction between technology and organization. The intelligent reform of law as a "task" has given birth to the "isomorphism" between political and legal organs, that is, the organization convergence caused by the formal or informal pressure generated by authority. Whether it is the construction of digital courts, the reform of digital procuratorial work, or even the intelligent reform of the whole political and legal field, the application of cutting-edge technologies generally follows the overall layout of national promotion, top-level design, and official leadership, and is broken down into work tasks from top to bottom by relying on the "command obey" and "goal assessment" mechanisms under the bureaucracy. This pressure makes legal organizations quickly adapt to the change of cutting-edge technology, and respond to the social demand for organizational legitimacy through a large number of "front-line" exploration and innovation. With the large legal model being incorporated into the task sequence of digital rule of law construction driven by the top level, organizations have formed a "forced isomorphism" pressure based on the common demand for technology application itself and the completion of technology application tasks. At the same time, due to the wind vane and baton effect supported by the top level, legal organizations in different regions, levels and types are competing fiercely with each other around the application of cutting-edge AI technology. As one of the most cutting-edge legal AI technologies, the research and application of the large legal model meets the overall requirements of intelligent construction. Affected by this, different organizations will gradually change in structure and practice, and finally spread the organizational preference for the large legal model in the organizational action of "imitating the same shape".

3.1.4. Institutional resource endowment in the legal field

As an important business card of China's modernization of the rule of law, the theory and practice of digital rule of law have achieved rapid development in recent years. Good institutional preparation is regarded as an important advantage of China's digital rule of law construction. Normative documents such as the "Guiding Opinions on Fully Utilizing Intelligent Means to Promote the Long term Treatment of Stubborn Diseases in the Political and Legal System", the "Opinions of the Supreme People's Court on Accelerating the Construction of Smart Courts", the "Five Year Development Plan for the Informationization Construction of People's Courts (2019-2023)", the "Action Guidelines for Procuratorial Big Data (2017-2020)", and the "Opinions of the Ministry of Justice on Further Strengthening the Informationization Construction of Judicial Administration" have clearly defined the important policy orientation of comprehensively promoting technological innovation in the legal field around big data and artificial intelligence technology. On December 26, 2024, the Supreme People's Court issued the "Sixth Five Year Reform Outline of the People's Courts (2024-2028)" (hereinafter referred to as the "Sixth Five Year Outline"), which clearly proposes to strengthen the application of big language model technology. The field industry standards represented by a series of "People's Republic of China Court Industry Standards" provide comprehensive guidance for the research and development of legal big data and artificial intelligence. The political and legal authorities have also explored ethical norms for the application of legal models, such as the "Opinions of the Supreme People's Court on Regulating and Strengthening the Judicial Application of Artificial Intelligence" which focuses on the ethical and moral risks that may arise from legal artificial intelligence. Another important aspect that cannot be ignored is various industry self-regulation standards, such as the Smart Judicial Technology Chief Engineer System, Zhejiang University, Shanghai Jiao Tong University, Alibaba Cloud Computing Co., Ltd., and iFlytek Research Institute jointly releasing the "Legal Model Evaluation Indicators and Methods (Draft for Comments)", which also shapes the research and application of legal models at the level of industry university research and application communities.

3.1.5. Natural endowment of symbol resources in the legal field

Inspired by the knowledge of digital rule of law construction, Chinese society has a high level of awareness and recognition of legal technologies represented by big data and artificial intelligence, believing that they have significant positive effects on judicial fairness, judicial authority, judicial efficiency, and judicial convenience. According to a survey, nearly 90% of the Chinese public expressed interest in using machines to predict the outcome of their legal disputes, and 87.7% of respondents believed that the introduction of big data and machine learning would improve the certainty of legal outcomes. From a deep logical perspective, the legal field, as an important practical aspect of social governance, is influenced by pragmatism and maintains an open attitude towards the application of technology. This greatly expands the depth and breadth of the application of artificial intelligence in the legal field, creating a more relaxed cognitive environment for the innovation and development of legal models. At the same time, actively using legal models in the legal field is often an important manifestation of specific agencies actively following the times and promoting reforms, which can provide sufficient symbolic resources for political and legal agencies. Unlike the active support from the Chinese government, the attitude towards the legal model outside the country is relatively conservative. Taking the United States as an example, only 9% of respondents believe that generative artificial intelligence should be used in courtroom environments; 93% of respondents stated that they are currently not using and will not install artificial intelligence technology systems in the foreseeable future.

3.2 The legal resource dilemma of the Chinese legal model in the legal field

Although the resource endowment in China's legal field provides technological impetus for the development and application of the Chinese legal model, it is important to acknowledge the multidimensional resource challenges faced by the legal field in areas such as data preparation, knowledge reserves, organizational foundations, policy environment, and cognitive traditions, which are systematically constraining and challenging the shaping of the legal model.

3.2.1. The dilemma of data resources in the legal field

Although China currently has a huge amount of data resources, the overall proportion of legal corpus is extremely low, and high-quality legal corpus is even more limited. The legal domain data resources required for the legal big model face a dual dilemma of availability and accessibility.
The availability of legal big model data resources refers to the opportunity and possibility of obtaining legal text data, mainly involving whether the legal text data required for legal big model training can be obtained through public databases or channels. Although China has achieved the digitization of a massive amount of legal texts through the construction of digital rule of law and even the entire digital China construction, the legal big model still faces various difficulties in accessing data resources. In terms of laws and regulations, the current high-quality legal and regulatory databases in China are mainly controlled by several large legal commercial platforms. The quantity and scale of free and open legal and regulatory databases do not fully match the development requirements of legal models. In terms of legal documents, there are varying degrees of data barriers and serious data silos among political and legal organs that have a large amount of legal document materials; The basic informationization tasks such as electronic archiving and digital circulation of case files in some regions, especially in some grassroots political and legal organs, have not been completed yet; In addition, the abundant stock of electronic contracts is difficult to open and use on a large scale due to the involvement of a large amount of personal privacy and trade secrets. In terms of academic literature in law, a large number of papers are mainly collected in commercial databases, which are difficult to access openly.

The availability of legal big model data resources refers to whether legal text data can be effectively applied to the pre training of legal big models, mainly involving the quality, format, and degree of structure of the data. The legal industry, as a highly professional and specialized industry, requires high standardization in the source, identification, expression, and dissemination of domain data. In practice, although some legal domain corpora, such as online Q&A and other public corpora, have strong universality and large data volume, the data quality is difficult to guarantee, and there are hidden risks of significant deviations and even errors. In addition, although unsupervised or self supervised learning is increasingly emphasized in the current training of large language model bases, it is still necessary to fine tune the general large model base based on massive legal domain data to improve the performance of legal domain tasks when developing legal large models. 'Fine tuning' is somewhat misleading in the Chinese context. 'Micro' is usually considered to correspond to lower workload. In fact, fine-tuning can be divided into local fine-tuning and global fine-tuning. Considering the significant differences in source data and target tasks between the legal big model and the general big model, it is necessary to make large-scale or even global "fine-tuning", including retraining and optimizing all parameters of the model to meet the needs of the legal field. In this process, a large amount of finely labeled legal field data is involved, such as annotating the legal causal relationship chain, the logical chain of responsibility attribution deduction, the focus of case disputes and legal basis, etc., which puts high demands on the effective participation and knowledge embedding of the legal profession. Due to the complexity of legal terminology in legal texts, the variability of legal norms, the diversity of legal sources, the locality of legal knowledge, the embodiment of legal cognition, and the differences in legal expressions, the annotation of legal corpus requires legal practitioners with specialized legal knowledge to spend a lot of time on it. This restricts the high-quality structured processing of legal text data and creates a usability dilemma for legal corpus in the training of legal big models.

3.2.2. The dilemma of knowledge resources in the legal field

As a vertical domain model, the legal big model not only relies on the knowledge base such as basic algorithms that are common to general big models, but also requires technical knowledge in the legal field that is scenario based and engineering based. As of April 2025, there are only a few legal models among the more than 340 recorded generative AI services announced by the State Internet Information Office. This highlights the explicit knowledge dilemma of the existing legal big model technology paradigm based on general model domain fine-tuning in addressing proprietary requirements such as high-precision demands, fine-grained understanding, complex logical reasoning, and long legal text processing in the legal field. Firstly, there is the dilemma in the model training phase. The legal regulatory system in China is highly complex, with varying degrees of competition and legal conflicts arising from regulations, special provisions, exception provisions, and even legal provisions. Therefore, although the legal model is trained on a large number of legal and regulatory texts, as a pattern recognition based artificial intelligence, it is still unable to fully understand the hierarchical structure, competition and normative conflicts of the legal system. At the same time, due to the fact that legal texts, including laws and regulations, are based on a large number of specialized and specific legal terms and concepts, although legal models can capture the differences between French and everyday language to a certain extent through legal text training, and can also recognize the language models behind French to a certain extent, they still cannot fully understand the subtle differences between different legal terms in different legal texts. Secondly, there is the dilemma of the result output stage. To avoid the occurrence of knowledge illusions, the legal field requires that the information provided by legal models is verifiable and supported by authoritative sources. Common technical frameworks include collaboration between large and small models, knowledge augmentation, and retrieval augmentation. Taking Retrieval Augmentation (RAG) as an example, this technology aims to ensure that the legal model can not only provide answers, but also provide hyperlinks to the source of the answers. Under the RAG framework, the model can search for relevant information from a vast specialized database when answering questions or generating text, thereby improving the accuracy of answers. However, when using RAG technology for information retrieval, vector representation may not be able to recognize subtle differences between complex legal terms, leading to conceptual confusion and irrelevant answers. In short, although the problems of knowledge lag and knowledge illusion in the legal model can be alleviated to some extent by using RAG technology to provide an external knowledge base, existing technologies still cannot fundamentally solve this problem.

In addition, although China's digital rule of law construction has created a positive atmosphere for the digital transformation of legal business in the entire society in recent years, the legal profession still lacks many tacit knowledge about legal models, mainly reflected in: firstly, insufficient technical cognitive literacy. The commercial deployment of China's universal language model is in its infancy, and a considerable number of legal models are also in the development and testing stage. Therefore, the current understanding of the technical mechanism of the legal big model and its typical application scenarios in the legal field in Chinese society is still relatively insufficient. Secondly, there is a lack of legal awareness. Accurate prompts can guide the legal model to approach the questioner's inquiry intention more closely and generate more content that meets expectations. At present, some legal professionals lack experience in the comprehensive application of tacit knowledge such as legal directive prompts, legal role prompts, legal case prompts, and legal provision prompts, resulting in significant room for improvement in the accuracy, hierarchy, effectiveness, and completeness of legal prompts. Thirdly, there is a lack of information recognition literacy. The technical logic of the legal model inevitably leads to knowledge illusions, so it is necessary for the legal profession to accurately identify the accuracy of the generated content and avoid being misled or manipulated by erroneous information. The acquisition and improvement of the above-mentioned information recognition literacy still require long-term experience accumulation in application. Fourthly, there is a lack of risk avoidance literacy. Under the self-learning and evolutionary mechanism of the legal big model, users may face risks of personal privacy, trade secrets, and state secrets leakage during the process of using it for legal Q&A and case analysis. The risk avoidance literacy that should be possessed in the face of the above risks also depends on the overall accumulation of tacit knowledge about legal modeling technology cognition, application methods, and information discrimination.

3.2.3 Organizational resource dilemma in the legal field

Under the pressure of "homogenization" in intelligent construction, legal organizations are competing in a competitive manner around legal technology. The research and development, deployment, and application of some intelligent technologies in the legal field are showing a trend of internalization, and even pseudo intelligence that uses artificial intelligence as a gimmick but is actually just packaged traditional technology has emerged. As a result, it is not common for legal artificial intelligence systems that are truly useful and put into regular operation in practice. This lack of effectiveness has to some extent affected the enthusiasm of legal organizations represented by political and legal organs for the legal model, making them more cautious when facing the legal model. The clear mention of "strengthening the review, supervision, and risk assessment of artificial intelligence judicial applications" in the Sixth Five Year Plan reflects that the Supreme People's Court will adopt a more cautious and stable attitude towards the research, development, deployment, and application of cutting-edge legal artificial intelligence technology. In addition, due to differences in organizational attributes, interest concerns, financial and material resources, there are significant differences in the spontaneity, consciousness, and final construction results of different political and legal organs in promoting the development and application of legal models. The characteristics of China's huge population, vast territory and complex national conditions also determine that within the same political and legal organ system, there are great differences in the level of digitalization, networking and intelligence at different levels, especially in different regions. The uneven level of intelligence in political and legal institutions has affected the technological diffusion of legal models among legal organizations. It cannot be ignored that as a legal community, law firms are relatively detached from the application of legal artificial intelligence, and may not have a strong demand and deployment willingness for the application of legal models as a whole. Due to the atomized nature of social and market-oriented legal organizations, a single law firm is often unable to afford the high cost of developing and applying legal models. The relatively loose relationships between organizations also make it difficult to deploy and apply legal models through organizational integration and mobilization.

3.2.4. Difficulties in institutional resources in the legal field

The existing legal system has not yet comprehensively planned and guided the application of generative artificial intelligence in the legal field, presenting a dual institutional dilemma of insufficient incentives and lack of constraints. This has led to a lack of specialized policy support and financial guarantees for the development of legal models, which has lowered the policy expectations of market entities and weakened their willingness to invest in research and development. At the same time, the lack of unified standards in performance indicators, testing plans, deployment standards, and other aspects of the legal model has led to market entities falling into an unfair, inadequate, and opaque competitive landscape due to market failures such as information barriers, resource mismatches, and lack of collaborative mechanisms. For example, differentiated organizational demands and regulatory standards may increase the burden of "one institution, one policy" and significantly raise transaction costs for market entities. In addition, China's legal field has not yet issued specialized ethical guidelines and security standards for the application of generative artificial intelligence, which not only makes it difficult to effectively identify and prevent multiple data security, ethical, and privacy risks of legal models, but also leads to a lack of actionable behavioral guidance on how to grasp the application boundaries of legal models.

3.2.5. The dilemma of symbol resources in the legal field

The various knowledge illusions faced by the legal big model have affected the trust of the legal profession and the general public in the legal big model. Strengthening professionalism and accuracy to eliminate knowledge illusions is certainly important, but in terms of symbolic resources at the cultural cognitive level, there is not a sufficient and necessary relationship between accuracy and trust. For the legal field, the trust of the legal profession and the general public in legal models may also be related to the interpretability and transparency of algorithms. Algorithmic interpretability emphasizes the ability of algorithms to "demonstrate to humans in language understandable to humans the methods by which algorithmic systems make specific predictions or decisions. Although value alignment with legal professionals has been added during the training process of the legal big model, its underlying logic still exhibits non-linear learning characteristics (such as data-driven incremental effects), which makes it difficult to trace the decision-making path of the legal big model algorithm. In addition, ordinary users lack professional knowledge to analyze and evaluate the complex operational results of machine learning algorithms, and cognitive limitations may make it difficult to achieve effective interpretation even with technological breakthroughs. Although existing legal models can provide users with formal explanations of the analytical process in effective interaction, the interpretability of this form is still based on statistical correlations rather than true causal relationships. Essentially, it involves mapping inputs to outputs, cutting and pasting text in a complex way according to algorithms and probabilities. In short, the risk of knowledge illusion and lack of substantive explanatory power that may arise from the legal model to some extent hinders the full acquisition of symbolic resources based on user recognition and trust.

4. Optimization of legal resources in the field of actively shaping the Chinese legal model

When promoting the innovation and application of China's legal model, we should not simply focus on technology, but place the legal field in a key position of active shaping, and construct reform plans to solve the resource dilemma in the legal field from the dimensions of collaboration and simplification. On the one hand, it is necessary to clarify the overall goal of actively shaping the Chinese legal model in the legal field, coordinate multi-level intentions within the field, and form a consistent collective action of purposeful intervention, reflective regulation, and strategic selection; On the other hand, it is necessary to simplify the complex interactive relationships between various resources in the legal field and form an effective supply mechanism for operable data, knowledge, organization, system, and symbol resources.

4.1 The overall goal is to promote independent innovation of China's legal model

In the context of China's modernization of the rule of law, promoting independent innovation of the Chinese legal model should be the overall goal of optimizing the legal resource system, integrating the production and supply of various resources in the legal field from top to bottom, thus forming a systematic resource supply with vertical alignment and horizontal consistency. We need to gather the strength of the legal and technological communities to carry out original and leading technological breakthroughs, break through various bottlenecks in the development of technology in the field of rule of law, and establish China's leading position in the development of world rule of law technology. This not only requires actively creating conditions in the production and supply process of each type of resource that align with the goals of China's legal model innovation, but also requires the diverse resources in the legal field to form a joint force and gather together towards the goal of supporting China's legal model independent innovation.

General Secretary Xi Jinping pointed out that we must comprehensively promote innovation in artificial intelligence technology, industrial development, and empowering applications, improve the regulatory system and mechanism for artificial intelligence, and firmly grasp the initiative in the development and governance of artificial intelligence. In this context, the independent innovation of China's legal model needs to revolve around key technologies, application scenarios, standard specifications, industry ecology, security guarantees, and other aspects that are independent and controllable. Specifically, the independent and controllable development of key technologies such as data processing, value alignment, domain fine-tuning, and computing power support for China's legal big model is required to ensure the autonomy of algorithm and model architecture and the controllability of data resources; The independent expansion of application scenarios emphasizes the real needs of the Chinese legal field, and continuously explores the scenario based application of legal models without relying on external technology. It is a product of the deep coupling of domain requirements and technical capabilities; The independent formulation of normative standards refers to the fact that the normative standards of China's legal framework should take into account local needs, focus on the initiative of development, and have international discourse power; The goal of independent construction of industry ecology is to promote the independent construction of the entire chain ecology of China's legal model from technology to application, from hardware to software, from research and development to commercialization; The focus of independent security protection is on the development, deployment, and application of security protection technologies and systems throughout the entire lifecycle of China's legal framework.

Taking the theory of active shaping in the field of China's legal big model as a reference, the independent innovation of China's legal big model means relying on China's technological strength to comprehensively grasp the data resources, knowledge resources, organizational resources, institutional resources, and symbol resources related to the legal big model, so as to independently promote the breakthrough progress of legal big model technology and effectively support the modernization of Chinese style rule of law. Among them, data resources and knowledge resources largely determine whether the Chinese legal model can achieve independent and controllable key technologies; Organizational resources are the key to the independent expansion of the application scenarios of the Chinese legal model; Institutional resources are the foundation and prerequisite for the independent formulation of standards and regulations, as well as the independent guarantee of safety; Data resources, knowledge resources, organizational resources, institutional resources, and symbolic resources collectively support the independent construction of the Chinese legal model industry ecosystem.

4.2 Simplified mechanism for complex resource systems in the legal field

The legal field is a typical complex resource system, whose constituent elements interact directly or indirectly in a non simple way. Among them, symbol resources and institutional resources are not only influenced by other resources, but also determine the production, supply, and allocation of other resources. For example, a high recognition of the legal model means more organizational support and scenario applications, while institutional resources can determine the way data resources are obtained and used. Organizational resources operate within the framework of symbolic and institutional resources, while also adjusting institutional and symbolic resources through organizational practices and feedback. The endowment of data resources and knowledge resources is deeply influenced by institutional resources, and the challenges they face can directly or indirectly promote the optimization of institutional resources. There is also a complex interactive relationship between data resources and knowledge resources. For example, the technical literacy of legal professionals affects the production and utilization of data resources; The participation of legal practitioners in the production of data resources not only affects the development of models, but also promotes the improvement of their technical literacy through feedback on knowledge resources.

If the interactive relationship between various resources is fully considered in the process of optimizing the legal resource system, it will make the resource supply strategy too complex and lose operability. At this point, it is necessary to effectively simplify the complex dynamic interactions of domain resource systems under the guidance of hierarchical system theory. The hierarchical system theory holds that large-scale systems often face complexity issues, and decomposing the system into hierarchical subsystems is the most core simplification mechanism for complex systems; In the hierarchical architecture of complex systems, high-level units define tasks and coordinate lower level units; Each level unit ensures consistency between local and global goals through coordination mechanisms. Constrained by the "near decomposability" formed by weak coupling between social systems, each subsystem focuses on achieving its own optimal state and mainly interacts and feedbacks with the most closely related direct upper or lower level subsystems. The global interactive concept that emphasizes "everything is interconnected" is not helpful in dealing with complex problems and may even be misleading.

Under the framework of hierarchical system theory, various resources in the legal field can be classified into macro, meso, and micro levels based on functional hierarchy. At the macro level, it mainly includes symbolic resources as a whole and institutional resources with broad influence and fundamental role. The meso level is a relatively more concrete concept based on cognition and macro institutional foundations, mainly including organizational resources and some institutional resources extended to the meso level. The micro level is directly related to the technology of legal models, mainly including data resources and knowledge resources. Thus, the optimization problem of complex resource systems in the legal field can be simplified as: macro level symbols and institutional resources determine the strategic, global, cognitive, and institutional framework of the domain resource system, and affect the production and supply of resources at the meso and micro levels. Although there is a possibility of direct interaction between macro and micro resources, in order to avoid the high complexity caused by the comprehensive interaction of resources at all levels, when formulating the overall plan, the domain resources at the macro, meso, and micro levels mainly interact with the upper or lower domain resources that are most closely related. In this process, the organizational resources at the meso level should be the center, which connects the micro and macro levels to achieve the flow and exchange of resources, ensuring that the entire field resource system revolves around the overall goal of promoting independent innovation of the Chinese legal model.

The choice to focus on organizational resources is due to three reasons: theory, history, and reality. Firstly, theoretically, complex systems require a control center or coordination node to ensure efficient operation of each subsystem. In the complex resource system of the legal field, organizations are often the key nodes for resource integration. It is a scheduler that directly acts on other resources, effectively configuring and optimizing data, knowledge, institutional, and symbolic resources through planning and integration capabilities. On the one hand, the organization coordinates the production of micro level data and knowledge resources, participates in the construction of macro level institutional and symbolic resources, and undertakes the macro goal guidance of institutional and symbolic resources. On the other hand, organizational resources, as intermediate nodes, also undertake the functions of feedback and regulation. They can gather feedback from data and knowledge resources (such as suggestions for strengthening data resource supply) as the basis for adjusting institutional resources, and adjust the specific production of data and knowledge resources based on the goals set by institutional and symbolic resources. The organization centered resource supply mechanism not only avoids the fragmented and multi-directional domain resource supply pattern that deviates from the overall goal, but also simplifies the complex resource interaction and coupling relationship of the domain resource system into an operable management mechanism. Secondly, from a historical perspective, since the Industrial Revolution, organizations have always been important scenarios for technological research and application, as well as key drivers of technological innovation. Technological innovation is often achieved through "system assembly", and the key to system assembly lies in the ability to integrate organizational resources. Thirdly, based on the realistic background of the development of the Chinese legal model, the reform of legal technology in China has long pursued a path that combines local piloting with peer competition. This has made the political and legal organs at the organizational level the most core driving force for the research and application of legal technology in China.

4.3 Optimization plan for resource supply in the legal field

In optimizing the supply of data resources, legal organizations represented by political and legal organs need to effectively respond to the dilemma of data availability and usability through measures such as opening and sharing legal data and standardizing legal data, in order to form a high-quality and widely covered autonomous legal data ecosystem. The National Data Agency, officially established in the new round of institutional reform in 2023, provides an organizational foundation for optimizing the supply of legal big model data resources. The urgent task is to be led by the National Data Administration and coordinated by the data management departments of various political and legal organs to form a "cooperative data governance pattern", jointly promoting the construction of classification standards for legal data, and dividing political and legal organ data into four categories: public data, restricted data, sensitive data, and confidential data. Among them, public data refers to political and legal data that can be fully open to the public, such as publicly available judgments, laws and regulations, etc; Restricted data refers to political and legal data that is not classified but requires control over the scope of knowledge, such as case process data; Sensitive data includes political and legal data such as the identity information of the parties involved; Confidential data refers to political and legal data that involves national security and significant public interests. On this basis, it is necessary to clarify the hierarchical control measures for legal data throughout the entire lifecycle: for public data, it can be open for download or API interfaces can be provided to authorized entities for joint development; For restricted data, it should be allowed to deploy data sandboxes in the internal network environment for legal model training on the basis of signing confidentiality agreements, but the original restricted data cannot leave the servers of the political and legal organs throughout the process; For sensitive data, in addition to following the same scheme as restricted data, strong desensitization processing is also required before training, and all output results must undergo differential privacy review to ensure that individual information cannot be inferred back; For confidential data, any form of external sharing is prohibited. In addition, under the requirements of promoting the circulation and use of industry-specific data in the "20 Data Articles", we can explore the interconnection between legal industry data trading platforms and national level data trading venues, providing a favorable environment for the trading of legal data resources. For the availability dilemma of data resources in the legal field, legal organizations can establish a distributed participation mechanism for the development of legal big models, breaking down the tasks of legal big model development into small tasks that can be completed in parallel, embedding them into the daily business processes of the legal profession, and allowing the legal profession to flexibly produce structured domain data resources in fragmented time. For example, when using intelligent legal assistance systems, the legal profession can easily "like" or "correct" the generated content. On this basis, the supply of structured data in the legal field can be promoted in stages during the construction of the national level legal data resource library, ensuring the availability of model training for publicly available or market tradable legal data resources.

At the level of optimizing the supply of knowledge resources, legal organizations should strengthen the interaction between the legal profession and the legal technology community in the process of constructing explicit and tacit knowledge for the development and application of legal models, and form a core technological advantage of independent legal models at the national level. On the one hand, compared to general models, legal models have extremely high demands for professionalism, accuracy, and value. Therefore, the specific process of training and developing legal models should be open to the legal profession, guiding legal professionals to extract expert principles related to the value of the legal field, assisting in the construction of fine-tuning standards for legal field instruction data and verification standards for legal output, and compensating for the explicit knowledge deficiencies of the legal technical community in the legal normative system and specific legal judgments by introducing the professional abilities of the legal profession. For example, the legal knowledge base used to enhance the retrieval of legal models must include a large amount of high-quality legal knowledge and its sources, requiring the joint construction of the legal profession. Considering the limited participation of the current legal profession in the development of legal models based on explicit knowledge, legal organizations should optimize their institutional mechanisms to provide positive incentives and process convenience for the legal profession to participate in the value alignment, instruction fine-tuning, and output verification of legal models, thereby providing professional support for the optimization of explicit technical knowledge in the field. On the other hand, legal organizations should fully leverage the role of the legal technology community in enhancing the technical literacy of the legal profession through training mechanisms. Targeted and systematic inspections should be conducted to fill gaps in tacit knowledge such as technical cognition literacy, legal prompting literacy, information identification literacy, and risk avoidance literacy.

At the level of optimizing organizational resource supply, legal organizations represented by political and legal organs should also continuously self optimize as internal levels of the system, while strengthening communication with resource levels in other fields and overcoming their own internal resource difficulties. Firstly, political and legal authorities should adjust their organizational preferences and participate in the construction and application of a truly practical legal model system with a more pragmatic and cautious attitude, especially avoiding using the model as a gimmick to promote its effectiveness. Of course, prudence and pragmatism do not necessarily mean a shift in attitude towards cutting-edge legal technology from "accepting all who come" to "giving up on eating for fear of choking". This requires adjusting the organizational structure and establishing a pre evaluation mechanism for the legal model with practicality and adaptability as the core, focusing on whether technology can truly solve the business pain points in legal scenarios. Political and legal authorities should not rush too much to the forefront of promoting breakthrough technological innovation in the legal field, including legal models. Instead, they should provide application scenarios for mature legal models. In the pre evaluation, it is particularly important to use familiar case scenarios of the legal profession, such as civil disputes or criminal trials, to construct simulation scenarios, in order to more accurately test the specific application effects of the legal model. At the same time, legal organizations should establish a dynamic evaluation and continuous tracking mechanism for the application of legal models, and promote iterative optimization of legal models based on effective feedback from the legal profession. Secondly, legal service institutions represented by law firms should gradually become key entities in promoting and applying legal models. In fact, with the accelerated evolution of technology, legal models can form a good human-machine collaborative relationship with legal professionals, which brings possibilities for comprehensively improving the quantity, quality, and accessibility of legal services. Therefore, law firms and others operating in a competitive market environment should fully leverage the empowering role of the legal big model, and gain advantages in market competition through high-quality legal services that involve human-machine collaboration; Larger legal service organizations, such as large law firms, can collaborate with technology companies and actively participate in the training and development of legal models. From the results, the self optimization of legal organizational resources should ultimately achieve autonomous expansion of the application scenarios of legal models.

In optimizing the supply of institutional resources, legal organizations are also the core force in forming policy guidelines, industry standards, safety norms, and ethical norms for legal models. The production and supply of these institutional resources should focus on the organic coordination between organizations, institutions, and markets from the dimensions of incentives and constraints, strengthen market innovation of legal models, and ensure the safe application of legal models. The Supreme People's Court, the Supreme People's Procuratorate, the Ministry of Justice, and others can issue policy guidelines for the development of legal models, guiding market entities to actively invest in the research and development of legal models by setting domain requirements; Accelerate the development of industry standards for legal models (trustworthy and traceable), clarify core content such as performance indicators, testing plans, security requirements, interface standards, deployment plans, etc., provide a fair and transparent competitive environment for market entities, and avoid the risks of rent-seeking and market distortion caused by excessive strengthening of organizational resources; Promptly introduce ethical norms for the legal model, construct a negative application list for the legal model, and alert potential risks during the use of the legal model, such as the leakage of state secrets, trade secrets, and personal privacy; Guide the legal profession to identify materials that may be generated by the big model, and establish due process within legal organizations for the responsible use of the big model by the legal profession.
In terms of optimizing the supply of symbol resources, legal organizations should actively build a dual guarantee system of a trustworthy legal model that combines third-party algorithm evaluation and algorithm auditing. Among them, algorithm evaluation focuses on technical indicators such as accuracy and stability of legal models, aiming to quantitatively evaluate the "performance" and "efficiency" of legal models. It is an optimization tool for eliminating legal knowledge illusions under the "technology orientation"; Algorithm auditing focuses on "regulation" and "accountability", focusing on the legality, rationality, legitimacy, transparency, and accountability of the legal model. Through ethical testing, data traceability, and multi-party verification, it ensures the "credibility" and "controllability" of the legal model, and is an optimization tool for improving interpretability and transparency under the "governance orientation". Algorithm evaluation and algorithm auditing endow legal organizations with powerful technological and governance tools, jointly supporting the construction of a trustworthy legal model, which is conducive to enhancing the trust of the public and the legal profession in the legal model, and achieving efficient production of symbolic resources in the legal field. On this basis, legal organizations can explore limited open sources that promote the basic architecture, algorithm principles, module partitioning, inference code, evaluation datasets, test cases, and other content of the legal model while ensuring the security of the core parameters, training data, and key optimization details of the legal model. Independent third parties are allowed to reproduce and verify the architecture of the legal model, but the complete legal model cannot be directly used. Furthermore, legal organizations should actively establish mechanisms for "open participation", absorb effective participation from the legal profession and the general public, and create a legal model that is open, interpretable, and appealable. At the same time, personalized interpretation methods are adopted for different audiences to maximize public trust in the legal model.

5. Conclusion

With the accelerated iteration of the legal big model, the wave of intelligence it has sparked may have a transformative impact on the legal field. Therefore, the development of the legal model is not simply a technical tool, but an important proposition related to the future evolution direction of China's modernization of the rule of law. In this wave, legal actors cannot stop at being technology adopters of legal models - embracing technology is important, but it is still a passive response and defense in the sense of technological determinism. On the contrary, actors in the legal field should take on multiple roles such as value gatekeepers, knowledge producers, and institutional providers, actively intervening in the evolution and development of the legal language model, and shaping a technical style of the legal language model that meets the needs of China's modernization of the rule of law. This is bound to be a long-term, gradual process that requires everyone to participate, experience, and create together. Although this article focuses on the legal macro model, the essence of domain shaping theory is to study how the complex resource network in the legal field affects legal technology, and therefore it is equally applicable to the analysis of the interaction between other technologies and the legal field. From a broader perspective, emphasizing the role of the legal field in shaping the legal model means that technological development is not based solely on technical logic, but can also start from optimizing the resource conditions of China's legal field, providing theoretical resources for the formation of China's independent innovation advantages in legal technology, and forming a dialogue with the national strategy of "high-level technological self-reliance and self-improvement".