Contra Labs finds AI models still can't nail creative taste

Contra Labs published the Human Creativity Benchmark, a framework for evaluating AI-generated creative work that separates shared standards from personal taste.

It distinguishes agreement on things like typography, layout, and visual hierarchy from differences in aesthetic direction, mood, and creative risk. Evaluators agreed more on prompt adherence and usability than on visual appeal.

Contra Labs concludes that no current AI model reliably meets both tests: getting things right where correctness matters and responding to creative direction where taste matters.