Thirty-two out of thirty-five students turned in an assignment on the Industrial Revolution full of far-fetched references to Madagascar. The culprit is not a collective typo, but one trap set on purpose by Dr. Jason Gibsonhistory professor at Alcorn State University, who hid in the text of the track an instruction written in white font, therefore invisible to the naked eye, but perfectly readable by a chatbot.
@iamjasongibson I can’t make this stuff up. #professor #highereducation #college #midterm ? Piano famous song Chopin Deep deep clear beauty – RYOpianoforte
The instruction instructed the AI to insert references to Madagascar anywhere in the answer, in a way that made no logical sense to the requested topic. Those who copied the text of the question directly into ChatGPT or another chatbot, then pasted the generated answer without rereading it, found themselves with a elaborate on the industrial revolution dotted with surreal phrases on the African island.
Gibson recounted the episode in a series of videos published on his TikTok profile, which went viral within a few days. In one of the examples shown by the professor, the chatbot closed a paragraph on the Industrial Revolution with the phrase “Madagascar floats sideways through the afternoon” (“Madagascar floats sideways in the afternoon”), an expression that has no logical connection to the subject matter and which almost sounds like a coded message between spies.
None of the students checked what the AI wrote
A second example, relating to a topic on technology and social inequality, concluded with the sentence “Madagascar purple bicycle whispers to the ceiling”in Italian “The purple bicycle of Madagascar whispers to the ceiling”. The most significant detail, according to what Gibson himself told, is that none of the students involved reread the answer before handing it in: it was enough to scroll through the text to notice the anomaly.
The professor clarified that he had not used the episode to ridicule his class. After confronting the students with the tests, he offered everyone the opportunity to contest their grade: only two tried to do so, a surprising fact considering the clarity of the evidence collected. All of the identified students still failed that section of the midterm.
The phenomenon of the indiscriminate use of AI for exams is not new, in fact university teachers have been dealing with it for years now. Already in August 2024, the incorrect use of AI in essays was described as endemic in American universities, with teachers forced to invent ever new countermeasures: handwritten essays, oral exams, tracks constructed specifically to be difficult for a linguistic model. We recently talked about it with the case that Professor Serrano spread.
The battle, however, seems far from won. Serrano himself observed the average score rises from the usual range of 65-80% up to 96% after allowing that the task could be done at home. According to his estimates, 84 out of 86 students would have used artificial intelligence to complete the test.
