Towards Using MultiModal LLMs as Graders in a GUI Design Course

This paper explores the potential of MultiModal Large Language Models (MM-LLMs) in automating grading processes in GUI design courses. We introduce a novel approach in using MM-LLMs to conduct pairwise comparisons rather than assigning direct numerical grades, aiming to enhance grading objectivity and efficiency. Preliminary results indicate that multimodal LLMs (such as OpenAI GPT-4o) can successfully identify superior designs when presented in pairs, suggesting a potential for a scalable and less biased grading system that could benefit both the educators and the students by providing timely and consistent feedback.