Mohd Zaki1,N M Anoop Krishnan1,Mausam Mausam1
Indian Institute of Technology Delhi1
Mohd Zaki1,N M Anoop Krishnan1,Mausam Mausam1
Indian Institute of Technology Delhi1
Scientific theories and facts are available in books, research papers, and websites. The advancement of computing power enabled researchers to train language models that can understand scientific text and explain it in natural language when prompted. These models have more than ~100 billion parameters and are trained on programming codes and text data. This training paradigm imparts these models the capability to solve mathematical equations, write computer programs, analyse a given text and answer the associated questions. Recently, these models became available for public use and can be equally used by students and teachers. In this work, we analyse the capability of two generative large language models, GPT-3.5 and GPT-4, to solve and understand undergraduate-level materials science questions. Specifically, we take questions which require reasoning, logical thinking, and numerical solving abilities. This analysis will help the teachers to engage with students to enable an in-depth understanding of materials science. We have also classified questions from a materials science perspective. The performance of both the LLMs from the domain point of view will enable the decision-making of researchers on how to use them for different materials science tasks.