The Emperor Pelican Has No Clothes
Summary
The post critiques the lack of diversity in LLM-generated SVG prompts, illustrating a 'diversity collapse' where pelican designs are homogenous. It contrasts prompted outputs with a broader, more varied set and introduces a Design Creativity Bench that evaluates AI creativity in UI design. Results show frontier models lagging behind humans (e.g., GPT-6 Astra scoring 65.5 vs human 92.3) across 80 tasks and three creativity dimensions (Originality, Creative Range, Appropriateness).