Why do LLM pelicans look like this?

A Hacker News discussion asks why LLM-generated SVG pelicans riding bicycles look alike, framing this as a diversity collapse problem. The post links to a UI design benchmark, Design Creativity Bench, which evaluated 16 frontier models on 80 open-ended UI tasks across originality, creative range, and appropriateness. The best model scored 65.5 against a human reference of 92.3, according to the post. The benchmark, leaderboard, paper, and datasets are publicly linked.

Key takeaways

• Design Creativity Bench evaluated 16 frontier models across 80 open-ended UI tasks in two product domains.
• It measures originality, creative range, and appropriateness rather than a single output quality metric.
• The top model, GPT-6 Astra, scored 65.5 versus a human reference of 92.3.
• The authors state they are publishing the leaderboard, methodology, and evaluation framework.

Why it matters

If models converge on similar outputs, their usefulness for creative work narrows. Measuring creativity as a distinct dimension could help track whether this homogenization improves or worsens across model generations.

Read with care

Source claims not independently checked; benchmark results and the pelican diversity observation come from the authors' own framing.