@inproceedings{lemmens-etal-2026-roast,
    title = "The Roast of {GPT}4o: Experiments in Generating, Detecting and Evaluating Celebrity Roast Comedy",
    author = "Lemmens, Jens  and
      Genette, J{\'e}r{\'e}my  and
      Veale, Tony  and
      Daelemans, Walter",
    editor = "Amir, Ori  and
      Hempelmann, Christian F.  and
      Rayz, Julia  and
      Dong, Tiansi  and
      Miller, Tristan",
    booktitle = "Proceedings of the 2nd Workshop on Computational Humor ({CH}um 2026)",
    month = jul,
    year = "2026",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.chum-1.5/",
    pages = "65--71",
    ISBN = "979-8-89176-431-6",
    abstract = "We present exploratory experiments in the comedic roasting capabilities of GPT4o. Specifically, @ComedyCentral roasts were scraped to design a survey in which participants blindly evaluated snippets of human and AI roasts, and had to predict the author (AI/human) in a second round of reviewing. The results show that there is no significant difference in how the barbs in human- and AI-generated roasts are rated. Further, a qualitative analysis showed that although the model utilizes specific recurrent phrases to imitate the style of human comedians, both generative LLM detectors and humans performed suboptimally in predicting the true author of the roasts."
}