GIFT-AI: The Cringe Test: student evaluations of intelligence with LLMs in a Turing Test adapted for classroom use

This article presents “The Cringe Test,” a classroom adaptation of the Turing Test (or imitation game) that stages dialogue with large language models (LLMs) in order to interrogate how culturally specific markers such as vocabulary, grammar, tacit knowledge, and contextual sensitivity shape judgments of intelligence in humans and machines. Implemented in an internationally oriented MA program in communications at the University of Gothenburg, the exercise organizes students into groups that interact with both a human respondent and an LLM, each speaking through a mediator. The LLMs are assigned distinct personae (neutral, friendly, devious, apathetic), while students are free to adopt their own conversational styles. The aim is less to determine whether a machine can “pass” as human than to elicit close, critical analysis of everyday language and the cultural conditions under which speech acts are identified as intelligent, reasonable, or “cringe.” Situated within debates in media theory, philosophy of technology, and critical AI studies, the exercise provides a hands-on point of entry into canonical critiques of the Turing Test and contemporary reflections on embodiment, alterity, and the politics of datasets and prompting. In practice, students typically recognize the LLM quickly; the pedagogical value lies in the subsequent discussions, where attention shifts from the binary of human vs. machine to the fine-grained cues—verbosity, slang, humor, formality, over-correction, and tone—that distinguish machine discourse from peer-to-peer talk. A recurring theme is that LLM attempts at colloquialism, especially their strained use of youth slang, are experienced as “cringe,” recalling the popular “how do you do, fellow kids?” meme. The article argues that this modified Turing Test functions as an effective scaffold for critical discourse analysis, prompting students to move from abstract questions such as “can machines think?” toward more situated inquiries into how prompting, institutional settings, and cultural norms co-produce the performances of intelligence attributed to both humans and machines.