If you're just worried about the pragmatics of being caught, I'd keep in mind that ChatGPT doesn't actually have a very high "dynamic range" given some specific set of inputs. If you use it enough, or explore the same topics and prompts enough, you'll quickly start to see structures, words, phrases, presentations that its output gravitates towards.

These may not be enough of a watermark to generally detect what was written by an AI and what wasn't yet, but it does suggest that the proliferation of AI-written assignment essays will quickly produce a catalog of plagiarism sources that works with existing tools.

In other words, once people submit enough AI-written essays about popular assignment A, existing plagiarism detection tools will easily start to notice the phrases/sequences/structures that it statistically gravitates towards (because that's what it does by design!) and start detecting these in later work. That software won't know what's AI written or not, but it will think that your submission may be plagiarized because a lot of it looks pretty specifically familiar.

Going a bit further... the AI tooling will also be used elsewhere, in things that not-just-students read. So writing style will potentially drift/evolve to be like AI writers, and this will muddy the detection efforts. I know I've noticed numerous times how I pick up on and start using phrasing (and other tells automatically), matching what others do over the years.

Of course, I wouldn't be particularly surprised if the above effect doesn't manifest. It's very dependent on how much AI writing spreads across numerous areas, and whether kids-these-days do any measurable amounts of long-form reading.

> can reliably detect the output of ChatGPT.

I imagine schools will have to put minimal value in out of person assignments, relying on proctored exams for final grades. This might be an investment opportunity!