随着大规模语言模型（如GPT-4）的出现，法律人工智能领域的研究路径正在发生重大变革。本文的目标是提出一条适应大规模语言模型时代的法律人工智能研究路径。具体而言，该路径包括两个支线：法律知识抽取路径、法律任务应用路径。即笔者认为大规模语言模型时代法律人工智能的研究路径的挑战性应主要从这两个角度出发。在本文中，笔者首先界定了大规模语言模型和法律人工智能的概念，并分析了新技术对传统法律人工智能任务的影响。其次，笔者介绍了传统法律人工智能在法律知识获取上的研究路径，并以民事领域类案检索数据集为例，探究在传统语言模型时代下的法律人工智能研究范式。然后，笔者分析了大规模语言模型在法律知识抽取任务上的实践效果，并以刑事要素抽取数据集为例，通过实验对比，进行传统语言模型与大规模语言模型在法律知识抽取任务上的对比。最后，笔者通过访谈法官、律师、法学院学生，分析了大规模语言模型在司法实践任务中的应用，并提出了引入法律专家在回路的法律大规模语言模型研究实践路径。通过实验验证，本文得出结论：大规模语言模型在法律知识抽取任务上具有显著的优势，其强泛化性与上下文学习能力使得法律知识抽取的自动化实现成为可能。但受限于大规模语言模型的能力，现阶段仍然需要法律专业背景的研究人员进行监督和反馈。同时，大规模语言模型在法律人工智能任务的应用也取得了显著的成果，包括辅助法官进行判决、辅助律师进行文书起草等，但仍然需要法律专家的参与以确保模型的可靠性和准确性。总的来说，大规模语言模型的出现为法律人工智能的研究和应用开辟了新的路径，但同时也带来了新的挑战和问题。我们需要在继续开发和应用大规模语言模型的同时，充分考虑到人的角色，确保法律人工智能的发展能够真正地服务于法律实践和社会公正。

With the emergence of large language models such as GPT-4, the research path in the field of legal artificial intelligence is undergoing significant transformation. The aim of this paper is to propose a research path for legal AI that adapts to the era of large language models. Specifically, the path includes two branches: the extraction of legal knowledge and the application of legal tasks. It is believed that the challenges in legal AI research during the era of large language models should primarily be approached from these two branches.In this paper, the concepts of large language models and legal AI are first defined, and the impact of new technologies on traditional legal AI tasks is analyzed. Then, the traditional legal AI research path in legal knowledge acquisition is introduced, with the civil domain similar case retrieval dataset as an example to explore the legal AI research paradigm in the era of traditional language models. Subsequently, the practical effects of large language models on legal knowledge extraction tasks are analyzed, and a comparison is made between traditional and large language models in legal knowledge extraction tasks, using the criminal elements extraction dataset as an example. Finally, by interviewing judges, lawyers, and law students, the application of large-scale language models in judicial practice tasks is analyzed, and a research practice path for legal large language models involving legal experts in the loop is proposed.Experimental validation has led to the conclusion that large language models have significant advantages in legal knowledge extraction tasks. Their strong generalizability and contextual learning ability make the automation of legal knowledge extraction feasible. However, due to the capabilities of large language models, there is still a need for researchers with legal expertise to supervise and provide feedback at this stage. At the same time, the application of large language models in legal AI tasks has also achieved significant results, including assisting judges in making decisions and helping lawyers draft documents, but the involvement of legal experts is still necessary to ensure the reliability and accuracy of the models.Overall, the advent of large language models has opened new paths for research and application in legal AI, but it also brings new challenges and issues. We need to continue to develop and apply large language models while fully considering the human role to ensure that the development of legal AI can truly serve legal practice and social justice.