Can LLMs Compete with Researchers in Generating Novel Ideas?

This study explores whether large language models (LLMs) can generate novel research ideas at a level comparable to expert human researchers. Over 100 NLP researchers participated, providing ideas that were compared to those generated by LLMs. Surprisingly, LLMs were judged to create ideas more novel than those of human researchers, but slightly less feasible. The research highlights limitations in LLMs' self-evaluation and generation diversity. Future studies will assess if these differences in novelty lead to impactful research outcomes.