Prototype Advisory System for Learning C Programming Using Generative AI

The technology of generative AI using LLM (Large Language Models) has made remarkable progress, and attempts to apply the technology of program code generation to programming education have attracted much attention. In this paper, we evaluate the suitability of a virtual TA (Teaching Assistant) system for beginner-level learners using OpenAI’s API, which takes C programs with errors and error messages as input to the LLM and outputs appropriate advice. Among the programming errors, the proposed system was generally effective in generating appropriate advice for syntactic and semantic errors, which is sufficient for learners to solve the problems on their own. For logical errors, the proposed system was generally effective in generating appropriate advice, although in some cases the advice was only a general explanation. Moreover, there are cases in which the prompts give appropriate advice to review the part of the sentence that includes division, so there is a possibility to obtain appropriate advice by asking for advice more than once.