Evaluation Metrics | Results |
---|---|
Explanations Quality | Human > LLMs |
Dataset Source | Ruo Zhi Ba |
Evaluation Methods | A/B Testing |
Cultural Nuances | Challenging for LLMs |
Existing humor datasets predominantly focus on English, lacking resources for culturally nuanced humor in non-English languages like Chinese. Chumor addresses this gap, sourced from Ruo Zhi Ba, dedicated to sharing culturally specific jokes. Human explanations for Chumor jokes outperform LLMs like GPT-4o through A/B testing.
This paper highlights the need for culturally diverse datasets in AI humor understanding and emphasizes the challenges faced by LLMs in capturing cultural nuances.