Well, we are on week five of this semester and I've already been handed a paper written by ChatGPT.

They're not hard to spot if you know how the AI crafts 'answers' and the mistakes it makes, but now is the time to familiarize yourself with what that looks like.

The markers to look for in particular are that ChatGPT doesn't know anything about anything so it is going to get a specific thing like, 'What did X scholar write about Y' wrong almost every time if that scholar isn't wikipedia-level famous (to provide enough text to train on).

It also leans towards fairly vague answers with no specific details (because it doesn't know any), so requiring specific details in an assignment is going to mean that ChatGPT can only produce bad responses.

But you should do that anyway because assertions need evidence!

Other markers are easy for students to edit but they might not: ChatGPT doesn't usually make grammar or spelling mistakes, but students do. It's also bad with names, confusing non-famous authors for famous ones with the same last name and giving the wrong initials or first name.

Finally, if you know the topic area well enough, ChatGPT will cite works that do not exist by famous authors in the broad field. So if your thought is, 'Wait, I don't think so-and-so ever wrote on this?' they probably didn't.

Ask the student to produce their source!

The key here is that, at least for now, ChatGPT produces *bad* answers, but you have to be attuned to the ways in which they are bad. AI will fool a poor or lazy grader, but not a good or attentive one.

Focus on content, not style - which you should do anyway!

(And finally, because folks are asking some more specific questions, I of course can't answer any specific questions as actual student information - rather than general statements about ChatGPT - is protected by law (FERPA - Family Educational Rights and Privacy Act)...

...my advice is to look at your next short essay assignment and feed the prompt into ChatGPT a few different ways to get a handle for the kind of answer it will give you. The actual ones won't be identical, but they'll make similar mistakes.)

But to stress, overall, remember: ChatGPT doesn't know *anything* - all it does is arrange words based on patterns of words its seen. It can't have an original thought and it can't know a fact.

It is both utterly alien from, and utterly inferior to (for now) a human mind.