Queen's University at Kingston

Published purpose

Recently, revolutionary progress has been made on large language models (LLMs) such as the GPT/Llama series, among others. Since the models provide unprecedented capabilities in representing and accessing knowledge, they have started to be used in scientific discovery/design, a new research trend that is still at its infant stage but will have a transformative impact. An essential component that is seriously understudied is reasoning for intelligent design. Reasoning is one of the most basic components of intelligence, which draws conclusions from existing knowledge. Without a solid reasoning ability, the progress that can be achieved for AI for design is seriously limited. The project aims to develop novel LLMs with enhanced reasoning. In a feasible setup, the project focuses on material discovery. The team targets two key objectives: (i) Developing novel reasoning models to effectively narrow down the search space of material design. (ii) Researching better graph networks to represent materials and building LLM-based agents to find material design solutions.