Collaborative potential of LLMs in humor-rich, culturally nuanced domains

Determine the collaborative potential of large language models (LLMs) as co-creative partners with humans in humor-rich and culturally nuanced creative domains.

Background

Prior research has examined LLMs as co-creative partners in creative writing tasks such as poetry and narratives, but comparatively little work has focused on humor, which is highly dependent on cultural context and shared knowledge. The authors highlight this gap and frame it as an open question motivating their paper.

To probe this question, the paper conducts a user paper on internet meme generation, comparing human-only, human–AI collaborative, and AI-only creation across creativity, humor, and shareability. The findings show increased idea quantity and reduced perceived effort with LLM assistance but no clear improvement in quality when humans are involved, and strong average performance for AI-only outputs. These results suggest that the scope and limits of LLMs’ co-creative potential in culturally nuanced humor remain to be more fully determined.