Large Language Models in Scientific Research

Large Language Models (LLMs) have dramatically increased in popularity throughout 2023, and we’re continually discovering more about how they can be used across disciplines. The Accelerate Programme for Scientific Discovery has been exploring how best to support researchers across the university to use large language models in their own research. 

From understanding historical texts, through navigating scientific literature, to creating models that have custom abilities like storytelling - LLMs are already being used across a wide range of fields. Yet, the barrier to getting started is high, especially in disciplines where the use of AI is in its infancy.  

In 2023, the team ran an LLM study group, bringing together 8 researchers from diverse fields like Linguistics, English, History and Chemistry. The group got to grips with the state of the LLM field, and learned how models are finetuned to perform specific tasks. Accelerate’s Machine Learning Engineer created a toolkit of LLM techniques in Python for the group to build on. The topics covered in the study group now form the basis of an online taught module and a 1 day workshop, both to be launched in February 2024. 

In 2024, the programme will also bring together experts to identify and share best practices for using LLMs in research.