Company or Institution
Sustainable Development Goals (SDGs)
SDG 16: Peace and Justice Strong Institutions
General description of the AI solution
We are excited to introduce a new web application that uses large language models to enhance legal research and collaboration. Our application includes two key features: a powerful document search system and a debate assistant.
With our document search system, users can quickly and easily find the most relevant legal documents for their research needs. Based on advanced natural language processing, this answers questions based first on the contents of a single document, and later an archive of papers or legal notices, highlighting relevant paragraphs that contain the answers. This allows users to ask questions to search for specific terms, concepts, or legal principles in relation to the document(s) in question. The search results are ranked by relevance, so users can easily find the most relevant answers first, linked to documents at the top of the list.
In addition to the document search system, our application includes a debate assistant that uses the same natural language processing techniques to analyse and understand legal documents, allowing them to engage simulated debate on the legal issues discussed in the document. This can be a useful tool for attorneys who want to explore different perspectives on a legal issue, or who want to practise their argumentation skills in a simulated setting.
We believe that our application can be a valuable resource for attorney partnerships, legal research teams within a particular firm or institution, and individual attorneys. It can save several hours worth of time and effort per person in legal research, provide a safe, unique way to store documents, and understand legal concepts.
Github, open data repository, prototype or working demo
Research and data on transformer interpretability:
This is the beginning (for us) of research evaluating the performance of language models on specific tasks, and includes attempts to understand and reverse-engineer their representations to explain how they do what they do. We have published data and code for this, and shall continue to do so throughout the lifetime of the project.
HPC resources and/or Cloud Computing Services