Freely subscribe to our NEWSLETTER

Mechanisms designed to detect the use of large language models are never going to be perfectly accurate. And there will always be ways around them. Clever prompts, style transfer, and hand-editing of a model’s output are all obvious ways to lower the score on a piece of content. Online services that rewrite text will become popular. A cat-and-mouse game will ensue. And what happens if a GPT-detector deems a student’s homework to be AI written? Can the student contest that determination? What manner of controversy lies therein?
Large language models will ultimately be built into everyday word processors. Everyone will use them. And they’ll be embraced for their ability to make us more efficient and creative. As such, teachers should craft assignments with the assumption that students will learn to use these tools not to cheat, but to produce better content.