Generating effective test questions is a crucial part of education. It's a time-consuming task for educators, and ensuring questions effectively assess student understanding is a constant challenge. Could AI help? A new study explored just that, comparing three large language models (LLMs)—GPT-3.5 Turbo, Flan T5 XXL, and Llama 2-Chat 13B—in their abilities to generate relevant questions from university slide text. Researchers used a two-step process: first, one LLM identified key concepts ('answers') within the slides. Then, all three LLMs created questions based on those answers. Student volunteers evaluated the generated questions across five key criteria: clarity, relevance, difficulty, connection to the slide content, and how well the question aligned with the provided answer. The results? GPT-3.5 and Llama 2-Chat 13B performed slightly better than Flan T5, especially in clarity and question-answer alignment, with GPT-3.5 demonstrating a particular knack for matching questions to answers. While all models showed promise, some weaknesses emerged. Aligning questions precisely with answers proved tricky, and the models sometimes generated overly generic questions. Another constraint is cost—while open-source models like Llama 2 offer a free alternative, the best-performing model, GPT-3.5, comes with usage fees. This could limit wider adoption in educational settings. Despite these limitations, the study highlights the potential of LLMs for personalized learning support, such as quick quizzes and knowledge reinforcement. For high-stakes assessments like formal exams, however, further refinement is needed to ensure accuracy and fairness. Future research could explore fine-tuning these models to improve their question-generation skills and address the identified limitations. The study's findings offer a glimpse into a future where AI could significantly reduce the burden on educators while providing students with more engaging and personalized learning experiences.