Legal research to become more efficient with new large language model contextualised to domestic law

The Infocomm Media Development Authority (IMDA) will partner the Singapore Academy of Law (SAL) to co-develop a new large language model (LLM), which will make legal research quicker and more efficient. Known as GPT-Legal, it will be deployed on LawNet¹ in phases from September 2024, which is used by 75% of Singapore’s lawyers.

A Memorandum of Understanding was penned today by Dr Ong Chen Hui, Assistant Chief Executive (Biztech Group) of IMDA and Mr Yeong Zee Kin, Chief Executive of SAL, with the aim to uplift the Singapore legal sector, with the use of Generative AI (GenAI) and collaborate to build GPT-Legal. The signing was observed by the legal community at Asia Tech x Singapore (ATxSG) 2024, Asia’s flagship tech platform.

GPT-Legal is trained on legal data from LawNet’s vast repository of legal content, which includes Singapore Law Reports. This makes it highly perceptive to the nuances of Singapore law, improving the accuracy and reliability of its output. Evaluation tools have also been built to assess the performance of GPT-Legal, in generating summaries that align with the content needs of the local legal sector. Trust and safety features embedded in GPT-Legal will further help to manage hallucinations and enable legal researchers to use generated content with confidence. These include flagging parts of the summary that are poorly substantiated by the source to urge users to read these with caution, highlighting corresponding paragraphs to facilitate cross-checking with the original source text, and providing an accuracy score for each summary to help legal researchers assess how much they should rely on it.

In its first phase of deployment, GPT-Legal will be used to summarise more than 15,000 Singapore court judgments, providing catchwords, facts and holdings in the judgment at a glance. Currently, only reported judgments² feature such summaries. Such summaries allow lawyers to conduct research more efficiently, as they can better assess which cases are more likely to be relevant, allowing them to get the right information faster.

Dr Ong Chen Hui, Assistant Chief Executive (Biztech Group), IMDA, said: “IMDA partnered SAL to develop GPT-Legal - a large language model contextualised to the local legal context. Through GPT-Legal, we enhance the research experience for lawyers by building in an ability for the model to understand Singapore’s laws, while providing guardrails and other cues to increase the confidence that lawyers have when interacting with AI models like GPT-Legal.”

Future phases of GPT-Legal will continue to allow LawNet to keep pace with other legal research platforms. These enhancements ensure that LawNet users—who make up the majority of legal professionals in Singapore—will benefit from advances in search technology and experience.

GPT-Legal may also be used beyond LawNet in the future. For instance, legal tech companies may be able to use the LLM to build new products or upgrade their current product offerings. GPT-Legal may also be enhanced for use in tools that summarise private legal documents, such as prospectuses and contracts.

Mr Yeong Zee Kin, Chief Executive of SAL, said, “Our partnership with IMDA has shown how collaborations can help steer AI to do good. The LLM we have co- developed will help legal researchers use AI-generated content with confidence. This is especially pertinent in the legal sector where it is paramount to uphold the lawyer as a trusted professional in the service and administration of justice. We look forward to engaging further with the legal profession and beyond to refine GPT-Legal and identify use cases that benefit SAL members.”