Key points are not available for this paper at this time.
Can a large language model produce humor? Past research has focused on finding examples of these models succeeding or failing at producing humor in the eyes of the authors, without surveying peoples’ actual judgments of its output. These examples, while interesting, do not shed light on exactly how funny ChatGPT is to the general public, nor do they analyze ChatGPT’s humor production abilities in comparison to humans’ abilities. To explore this question, we gave the same comedic prompts to ChatGPT 3.5 and laypeople, and asked them to generate humorous responses (Study 1). We also asked ChatGPT 3.5 to generate humorous satirical headlines and compared them to published examples from professional comedy writers at The Onion (Study 2). Other participants rated the funniness of the human and A.I.-produced responses in each study. ChatGPT 3.5-produced jokes were rated as equally funny or funnier than human-generated responses regardless of the comedic task and the expertise of the human comedy writer.
Gorenz et al. (Mon,) studied this question.