Human-in-the-Loop Control of Objective Drift in LLM-Assisted Computer Science Education

arXiv:2604.00281v1 Announce Type: new Abstract: Large language models LLMs are increasingly embedded in computer science education through AIassisted programming tools, yet such workflows often exhibit objective drift, in which locally plausible outputs diverge from stated task specifications.