Multimodal LLMs Can Zero-Shot Judge Image Creativity, Study Finds

A Harvard and Penn State arXiv paper tested whether multimodal LLMs can zero-shot judge the creativity of AI-generated images, finding they mostly succeed, with a 0.68 correlation to human ratings and entertaining chain-of-thought critiques.

2026-08-15 ~ 2026-08-15 · 2 related posts