Enhancing Diversity of LLM-Generated Educational Tasks

The paper presents CreativeDC, a prompting framework designed to enhance the diversity of educational tasks generated by large language models (LLMs) while maintaining high utility. By employing a two-stage reasoning process inspired by creativity literature, the method was evaluated in Python programming, yielding a 1.6x increase in distinct high-utility tasks compared to existing baselines. This advancement is significant for practitioners seeking to leverage LLMs for educational content creation, as it addresses the "Artificial Hivemind" effect that leads to homogeneous outputs.