Compiled by the editorial desk with reference to an interview conducted by Futurism and published with the subject's consent.

A graduate student in a business master's program has revealed that he used the artificial intelligence text generator GPT-2 to write two of his course essays — and passed both times. The student, identified only as Tiago, shared his story with Futurism on the condition of anonymity, citing concerns about plagiarism accusations.

Tiago, who is in the final stages of his degree, said he used the AI to write essays for two different classes. He explained that the assignments were follow-up essays to business case presentations, typically three to five pages. The courses, he said, were not information-dense, allowing him to submit AI-generated content with minimal editing.

“I don’t know if it qualifies as plagiarism,” Tiago told Futurism. “I figure no, maybe, but that’s not a gamble I’d like to take right now.”

His method involved writing a brief outline with a topic sentence for each paragraph, then feeding that sentence into TalkToTransformer, a web-based tool that makes GPT-2 accessible without technical expertise. The AI would generate a full paragraph, which Tiago would review and lightly edit before inserting into his document.

He said he had to iterate several times to get output that matched his intended argument. The AI sometimes produced false quotes or invented facts about companies, which he removed. He also found that using business buzzwords like “innovation” and “synergy” in his initial sentences steered the AI toward more convincing business-school prose.

Tiago said he got the idea after reading about a journalist from The Economist who used GPT-2 to enter a student essay contest on climate change. That experiment, he said, was “very close” to what he later did.

He acknowledged that the AI-generated essays were not high-quality, but they were good enough to pass. “The final essay is pretty poor,” he said, “it’s just not poor enough for the professor to fail you.”

He expressed little surprise at passing, attributing his success to what he called the “poor standards” for essays in business school. “It’s not as much a feat for GPT-2,” he said, “as proof of the poorness of business school’s curriculum.”

Why the AI Was Hard to Detect

Tiago noted that while there is a simple way to detect GPT-2-generated text — feeding it back into the AI to see if it can predict the next words — he doubted his professors considered that possibility. He was slightly worried that anti-plagiarism software used by the school might have been updated to catch AI writing, but he submitted the essays anyway.

He also pointed out that business school essays are not like academic papers. “Business cases are not like a theorem, they’re more like a parable,” he said. “You can take any conclusion that you want, and if it makes sense and if it fits in the more general narrative, you are assured a passing grade.”

Tiago said he would use the AI again if he had more coursework ahead, but he cautioned that students should not expect high marks. He noted that while he passed, he was “far from the best” and likely among the lowest-scoring students who passed.

His disclosure comes as OpenAI initially hesitated to release GPT-2 in full, fearing it could be used to generate fake news. In practice, the technology has been used for a range of purposes, including as an AI dungeon master in role-playing games. Tiago’s case, however, highlights a less-publicized use: academic shortcuts.