[TINY] Building Bridges of Thought: Using the Power of Association to Inspire Creativity in Large Language Models

The interplay between creativity and hallucination in large language models (LLMs) has drawn increasing attention in cognitive science. While associative thinking is fundamental to creativity, distinguishing it from hallucination in LLMs remains an open challenge. This study explores how knowledge graphs (KGs) can serve as a structured framework to guide associative processes while mitigating hallucinations. By leveraging KGs to form novel but grounded connections, our approach enhances creative generation without sacrificing factual accuracy. Through experiments on GPT-4o, DeepSeek-R1 , Claude-3.5-Sonnet, Grok-2, and Gemini-1.5-Flash-128k, we demonstrate that incorporating structured associations improves interpretability and reliability in creative tasks. This research advances the understanding of LLM creativity, paving the way for improved cognitive modeling and KG integration while highlighting the limitations of evolving LLMs and potential directions for future enhancements.