Abstract
Ray Tracing is a fundamental concept often taught in introductory Computer Graphics courses. Ray-Object Intersection (ROI) questions are frequently used as practice for students, as they leverage various skills essential to learning Ray Tracing or Computer Graphics (CG) in general, such as geometry and spatial reasoning. Although these questions are useful in teaching practices, they may require significant time and effort to produce, as the production process can be quite complex and necessitates careful verification and review. Recent advancements in Generative Artificial Intelligence have led to the possibility of automated or assisted exercise generation, but this is underexplored in CG. To investigate this, we conducted experiments over two years and across three different models (GPT-4, o4-mini, and GPT-5.1) to evaluate the performance of LLMs in writing ROI programming questions and model solutions. The results indicate that there has been a significant performance improvement when writing ROI programming questions and model solutions with the newer models (o4-mini and GPT-5.1) over GPT-4. In particular, the newer models have achieved near-perfect scores with writing model solutions and are capable of generating teaching materials with strong baseline qualities that can be refined into high-quality resources through expert review. This demonstrates that LLMs can be proficient in this field, which can be leveraged to enhance the productivity of CG educators and provide CG students with more learning resources.