LLM-Based Code Generators on CS1 Coding Tasks and Learning Trajectories
A recent study delves into the interaction between budding programmers, aged 10–17, and Large Language Models (LLMs) like OpenAI Codex, revealing pivotal insights into their amalgamation in learning programming. The investigation highlighted diverse approaches adopted by learners, including the predominant AI Single Prompt and the more balanced Hybrid method, the latter showing promising correlations with understanding of coding concepts. Over-reliance on LLMs surfaced as a concerning factor, potentially impeding autonomous code authoring and emphasizing the need for a balanced utilization of such advanced tools. Instances of academic integrity and clarity in crafting prompts are underscored, outlining the educational implications and the significance of precise communication of coding intent.