You wouldn’t be remiss to accuse college students using AI to cheat of being lazy. But some of the wilier ones are putting in the hard work to make sure their LLM-generated answers avoid detection from their nosy pedagogues.
New York Magazine interviewed some of these clever cheats in a recent piece, who are well aware that by by now, many professors are wise to students’ tricks and are changing the assignments they give and deploying increasingly powerful tech to weed out any illicit AI assistance.
One student named Jared attending New York University took his introductory calculus class’s use of WebAssign, a platform that’s designed to make cheating “far more difficult,” as a challenge. He and his friend used an AI coding tool to build a bot that does all their homework for them, but at the sluggish pace of a struggling student.
“The bot moves slowly, leaving a professor checking time stamps none the wiser,” according to the piece. Jared, in his munificence, gave the tool to his friends, and now he’s not even sure how many classmates have his automatic cheating machine tech in their hands.
It’s a demonstration of the power of AI agents — AI models that can perform complex tasks on a human’s behalf without any intervention, which range from writing entire pieces of software to conducting research. Perhaps it’s also a response to the punitive, detection-focused approach that beleaguered professors have adopted amid a deluge of AI cheating.
That seemed to be the argument made by Will, a recent graduate of Carleton College. By making a big deal of obvious AI tell-tales like em-dashes, Will figured, faculty were actually spotlighting exactly what the cheaters need to disguise.
“People are going to find a way to cover up those little stylistic quirks,” Will told NYMag. “If the only way you’re trying to discourage cheating is by saying ‘Well, I have a detector,’ then you’re not going to succeed.”
In Will’s case, his professors banned AI-generated writing, with one of them bragging that they had a “battery” of tools to enforce the ban. Determined to have AI crank out his work anyway, Will used Claude to analyze and write extensive passages for an essay, then went back and manually rewrote the AI-generated paragraphs.
The best cases of AI cheating, it seems, requires a human touch. Theo, a UMass Amherst sophomore and self-proclaimed autodidact disinterested in his college’s gen-ed courses, boasted to NYMag about how he created an astronomy presentation by using OpenAI’s AI coding agent Codex and combining it with a plugin that improved its web-search capabilities. But because the resulting presentation looked too polished, he spent four hours hand-degrading the slides to make it look more like the work of a mediocre college sophmore, according to NYMag.
“It took a lot of work,” he told the magazine.
Danielle Carr, a historian and anthropologist at UCLA, thinks that professors relying on surveilling and detection methods to weed out cheating will only keep them stuck in a game of cat and mouse. AI cheating, she told the magazine, isn’t “a technical problem that has a technical fix.”
More on education: Zohran Mamdani Bans AI for NYC Public School Students Up to Eighth Grade