A Confederacy of Models: a Comprehensive Evaluation of LLMs on Creative Writing
Submission Type: Regular Long Paper
Submission Track: Theme Track: Large Language Models and the Future of NLP
Submission Track 2: Natural Language Generation
Keywords: LLMs, language models, creative writing, evaluation, text generation, storytelling, creativity
TL;DR: We evaluate 12 recent LLMs, comparing them to humans, on a creative writing task using human evaluation with a detailed rubric.
Abstract: We evaluate a range of recent LLMs on English creative writing, a challenging and complex task that requires imagination, coherence, and style. We use a difficult, open-ended scenario chosen to avoid training data reuse: an epic narration of a single combat between Ignatius J. Reilly, the protagonist of the Pulitzer Prize-winning novel A Confederacy of Dunces (1980), and a pterodactyl, a prehistoric flying reptile. We ask several LLMs and humans to write such a story and conduct a human evalution involving various criteria such as fluency, coherence, originality, humor, and style. Our results show that some state-of-the-art commercial LLMs match or slightly outperform our writers in most dimensions; whereas open-source LLMs lag behind. Humans retain an edge in creativity, while humor shows a binary divide between LLMs that can handle it comparably to humans and those that fail at it. We discuss the implications and limitations of our study and suggest directions for future research.
Submission Number: 963
Loading