Jiaxi Cui, Munan Ning, Zongjian Li, Hao Li, Yan Yang, Bohua Chen, Bin Ling, Yonghong Tian et al.
Chatlaw is an AI assistant that combines a multi-agent framework mimicking the Standard Operating Procedures (SOP) of real law firms with a Role-Aligned Mixture-of-Experts (RA-MoE) architecture to enhance the accuracy and reliability of Chinese legal services.
General LLMs lack knowledge of the Chinese legal system and are vulnerable to hallucinations, making it difficult to provide reliable results in high-risk fields like legal consultation.
By imitating the collaborative structure of law firms, Chatlaw is designed with multiple role-based agents (e.g., legal assistant, researcher, senior lawyer) working together. A novel RA-MoE architecture is developed to align internal experts with each agent role, intelligently routing each step to the most suitable parameters.
Chatlaw achieves a 7.73% improvement in accuracy over GPT-4 on the LawBench benchmark and an 11-point higher score on the Unified Qualification Exam for Legal Professionals. Real-case studies and expert assessments confirm its robustness, contributing to improved accessibility and reliability of legal services.