Making GenAI Socratic for Computational Thinking

While Generative AI (GenAI) tools offer significant potential to enhance learning, they also pose significant risks as students rely on them for quick answers without deep understanding. Even more concerning, GenAI can hallucinate, producing plausible yet incorrect information. Consequently, many teachers are hesitant to integrate GenAI tools in homework assignments, let alone examinations. To address these challenges, we propose a pedagogical framework grounded in the Socratic Principle, designed to harness GenAI’s capabilities to foster computational thinking in undergraduate and graduate computer science education. GenAI is embedded into learning materials as thought-provoking questions that students can refine and run directly on state-of-the-art language models on the learning platform. For homework, the system provides access to shareable GPU resources that students can use to run custom GenAI tools for collaborative, computationally intensive projects. For examinations, we introduce OPTMentor—an in-browser Python debugger powered by our custom fine-tuned AI Mentor model that runs locally via WebLLM. To ensure academic integrity, we explore techniques to constrain and further fine-tune language models to deliver hints, not answers, promoting academically safe AI-assisted programming. © 2025 IEEE