LLM Creativity Tests Under Scrutiny in New Study

A new study on arXiv (2605.13450) takes a close look at how well we can measure human creativity when compared to large language models (LLMs). It zeroes in on three main aspects: creative writing, divergent thinking, and scientific ideation. The results show that the Divergent Association Task (DAT) and Conditional DAT are the best measures for creative writing and divergent thinking. However, the study also points out issues with how useful these assessments are for judging machine creativity, noting that they have limited validity even for measuring human creativity. This is the first detailed evaluation of its kind.