Evaluating LLM Creativity through Long-Tail Performance

Creativity benchmarks for large language models (LLMs) vary widely in task and metric design. This raises the question of whether they capture a common notion of creativity.