Creative solutions are built upon obscure associations, those that are uncommon but semantically close. In contrast, LLMs work on common associations, so cannot be very creative. We extend the LLM model to find the associations required for creativity.
We chatted with ChatGPT about solving the many creativity problems (insight problems) used in creativity research. It did not do well. It flailed and failed, and had to be directed to the obscure connections (uncommon but semantically close) that are required for crafting a creative solution.
Since all LLMs rely upon common association among words and creativity requires obscure associations among the keywords of the problem, there is a fundamental problem with using LLMs to innovate. Consequently, we have devised a CLM (Creativity Language Model) for the task of innovating and creating.