Creative talent: AI now exceeds average human creativity while top humans still lead.

A large-scale study comparing 100,000 people with leading generative models found some AI systems exceed average human performance on defined creative language tasks, while the most creative humans continue to outperform all tested AI.

Researchers led by Karim Jerbi at the Université de Montréal published a large comparative study in Scientific Reports that measured creativity in language tasks. The team included co-authors such as Yoshua Bengio and drew on responses from about 100,000 human participants. They tested several leading generative models, including ChatGPT, Claude, Gemini and others, and used a common framework to compare outputs. The study finds that some AI systems now exceed average human creative performance on specific divergent linguistic tasks, while the most creative people still outperform AI. Major findings: - The study compared more than 100,000 human participants with several leading generative language models. - Models tested included ChatGPT, Claude, Gemini and other large language systems, with GPT-4 noted as exceeding average human performance on some tasks. - Some AI systems now surpass the average human level on defined divergent linguistic creativity tasks. - The highest-performing humans — especially the top 10% — still outperform all tested AI models. - The research team developed a common, rigorous framework for comparison and named co-first authors Antoine Bellemare-Pépin and François Lespinasse, with collaboration from Jay Olson. Summary: The study marks a turning point by showing generative AI can reach and in some cases exceed average human creative performance on specific tasks, while top human creators remain ahead. Undetermined at this time.