Enhancing SQL learning with large language models: generating translations and explanations for educational use

Abstract

We explore the potential of large language models (LLMs) to help business users to understand SQL, with a view to include such techniques in teaching and programming support tools. We propose a new method which uses LLMs in two stages: content creation and content evaluation. Our method was tested with 32 SQL queries from Spider and five LLMs. We discovered that using RAG is not beneficial. Human evaluation showed that our method successfully generates quality outputs which can be used in teaching or business. Our students showed a preference for short structured explanations and found both the explanations and translations to be useful.