LLMs Improve Creative Problem Generation with Divergent-Convergent Thinking
Published:Dec 29, 2025 16:53
•1 min read
•ArXiv
Analysis
This paper addresses a crucial limitation of LLMs: the tendency to produce homogeneous outputs, hindering the diversity of generated educational materials. The proposed CreativeDC method, inspired by creativity theories, offers a promising solution by explicitly guiding LLMs through divergent and convergent thinking phases. The evaluation with diverse metrics and scaling analysis provides strong evidence for the method's effectiveness in enhancing diversity and novelty while maintaining utility. This is significant for educators seeking to leverage LLMs for creating engaging and varied learning resources.
Key Takeaways
- •LLMs often produce similar outputs, limiting the diversity of generated educational content.
- •CreativeDC, a two-phase prompting method, addresses this by incorporating divergent and convergent thinking.
- •The method significantly improves diversity and novelty in generated problems while maintaining utility.
- •Scaling analysis shows CreativeDC generates a larger effective number of distinct problems.
Reference
“CreativeDC achieves significantly higher diversity and novelty compared to baselines while maintaining high utility.”