This research introduces MCQG-SRefine, a novel approach that enables large language models to generate complex multiple-choice questions for professional medical exams through iterative self-improvement.

• Implements a three-stage refinement process (self-critique, correction, comparison) that significantly improves quality of generated questions
• Outperforms traditional methods and basic LLM prompting on medical exam question generation
• Demonstrates 27.4% improvement over GPT-4 baseline when evaluated by medical professionals
• Creates a comprehensive benchmark for medical multiple-choice question generation

This breakthrough matters for medical education by automating the creation of high-quality assessment tools, reducing the burden on medical educators while maintaining rigorous testing standards for future healthcare professionals.

MCQG-SRefine: Multiple Choice Question Generation and Evaluation with Iterative Self-Critique, Correction, and Comparison Feedback