GPT-4o achieved the highest humor score among models for English datasets [4]. Clean Dataset achieved the highest humor score in Russian, followed by GPT-4o [4]. In a study, HumorSkills captions were rated only 0.08 points lower on a 5-point scale than top-rated human captions, with p=0.053 [2]. This makes HumorSkills not statistically less funny than the best human captions [2].

The HumorSkills system was rated as significantly funnier than the VLM baseline, GPT-4o [2]. The study also showed that GPT-4 is capable of explaining the mechanics of jokes [3]. Another study's results indicated that humor is rooted in the frontal lobe of the cerebral cortex [6].

Which AI model scored highest in humor caption ratings?

Related Content You May Like