Enhancing Legal Information Accessibility Through Advanced Language Models
Large Language Models (LLM) have transformed the way people access information. Despite their ability to answer a wide range of questions, chatbots are still unreliable when it comes to sensitive topics such as legal advice due to issues like hallucinations. This project aims to build a trustworthy Hong Kong legal consultation assistant that can engage in multiple rounds of conversation to gather sufficient information and generate accurate responses, complete with citations. The system will use a multi-agent approach that simulates an actual legal consultation process, with each agent capable of retrieving relevant legal knowledge, including regulations and court judgments.
User Interface
Video Demo
System Design
Experiment Results
100 Hong Kong Legal Question is sampled from the Legal Question Bank to benchmark CLIC-Chat against other LLM Models.
- Much higher Citation Accuracy compared to its foundation model gpt-4o-mini.
- No hallucination on the testing dataset.
- Case References present in response.
To find out more technical details, checkout the poster and reports at the Documents Page.