On My First Case of AI Plagiarism (If That’s the Word)
It happened. My student used ChatGPT to write their final paper in my English class. How do I know? I don’t. That’s the problem.
I mean, I do know. I really do. This student’s essay didn’t read like the rest of the work they’d done during the semester. The voice was generic-sounding, emotionless, and mechanically unassailable, and this kid didn’t write like that.
I’d known this day was coming, but I’d been determined to cling to the old order for as long as I could get away with. Even last spring, when I brought GPT-3 to the attention of my English department, I thought we’d have another couple of years yet to figure out our game plan.
Nope. Zero hour has come.
So what did I do? First, I shook my head, and then hung it, and then scratched it. I knew there was no way to prove irrefutably that this student had cheated, but I did enough digging around on the internet to learn that a machine-learning company called Hugging Face had in fact built a detector for GPT-2 a couple of years back. I found it here, and plugged in my student’s essay. In seconds, the detector told me that there was a 99.98% chance it was “fake,” as opposed to “real.” I then checked some other students’ essays, and then one of my own, and all were at least 95% likely to be “real.” The detector’s results…
