Creativity benchmarks for large language models (LLMs) vary widely in task and metric design. This raises the question of whether they capture a common notio…
机构:芝加哥大学
来源:arXiv 2610.12998 | AI4Papers 论文推荐平台