ChatGPT's Limited Joke Repertoire Popular Among College Students in Japan

Researchers from the Institute for Software Technology, German Aerospace Center, and Technical University Darmstadt have found that OpenAI's ChatGPT-3.5 has a limited understanding of humor, with over 90% of 1,008 generated jokes being the same 25 jokes. The researchers concluded that the responses were likely learned and memorized during the AI model's training rather than being newly generated. Despite this, the researchers noted that ChatGPT's focus on content and meaning in humor indicates progress toward a more comprehensive research understanding of humor in language models.