I used to teach students. Now I catch ChatGPT cheats.

I work with students, but not in a scenario where I'm faced with these challenges directly (I don't focus on written work nor am I grading assignments). I do, however, get to see and hear how they interact with ChatGPT and similar tools and have come away with a couple of insights:

• Most of them use these tools very poorly. Students are mostly unaware of the strengths and weaknesses of LLMs, and, if the response they get isn't immediately useful, they give up quickly. This paper lays out this difficulty more clearly and formally than I ever could. Ultimately, this makes me very wary of any suggestion that students are going to have success in using these as learning aids, at least not without substantial guidance.

• Students are very aware of others using these tools to cheat, and it's worsening the already enormous academic pressure many students feel. While I've only had a few students confess to their own usage of these tools, many are eager to complain about their classmates' cheating and how they feel their own work is being devalued as a result.

I think the only thing in this piece that I see differently than the author, is how large or small of a shift this really is.

I once believed my students and I were in this together, engaged in a shared intellectual pursuit.

Really? I don't want to sound overly negative, but come on... Students doing the bare minimum to get a passing grade is not some new phenomenon. I suppose the accessibility of these new tools has laid bare just how many students are willing to take the easier path, but the connection between educational achievement and career success has long undermined the idea that schools could be so pure in their ideals.

My optimistic vision for the future of student writing is focused on deep investigation of topics and, to put it bluntly, student ego. One sad commonality I've found in student work is an utter lack of pride. It's so beaten into students that the purpose of work is to be graded, that the concept of putting in effort for their own self-value is completely foreign. When I think back to my own struggles with writing as a student, I can't help but think that it's no wonder I couldn't write: there wasn't anything I knew well enough to write about with confidence. I'd love to see students given the opportunity to become experts at something, to the point where writing about it to share their knowledge and opinions would just be the natural step.

I'm disappointed by these takes and hoped we'd be seeing better adaptation by now. This idea that we're locked into the options of let 'em cheat or pen and paper essays during class time is just so unimaginative. How about teaching them how it works so they can understand the pitfalls and power of it themselves? How about gearing some work around proper outlining and the structural planning rather than the finished product? They will sit there and proclaim "oh okay sure but when you get the real world with such and such situation chatGPT won't help then" ...then design your assignments around that instead! It's like the whole teaching institution has been so stuck with 5 paragraph papers for so long they don't know how to do anything else.

It doesn't have to be this way. If we didn't go in with such a defeatist attitude ChatGPT could be the impetus for revitalizing and reimagining education as we know it. In fact I think it's inevitable, I'm just uncertain how long it will take us to get there.

I'm surprised by the number of responses arguing that this is a failure of education. Many commenters here seem to think that the professors need to adapt their curriculum to better fit the age of AI and to start testing things that AI cannot do.

The comparison to "you won't always carry around a calculator in your pocket" from @ThrowdoBaggins is poignant, but also important in where it falls short. We still need to teach basic arithmetic to children so they have some intellectual leverage with which to grasp the stronger concepts. We still need to teach basic reading and writing skills to students so they can learn higher-order skills. You cannot expect someone to become an architect of software systems without first learning how to program.

AI is fundamentally different than previous in that in can do most introductory-level work at a level that most students can't; in other words, LLMs are smarter than most students. If students learn the LLM-first workflow, they literally will not have the ability to adapt when LLMs fall short.
Part of my educational research was in designing programing assignments that LLMs struggled with that were still accessible to introductory programmers, and we came up with lots of introductory coding assignments that all 2023 LLMs consistently struggled with, even with good prompt engineering. Then GPT 4o came out and one-shot all of them.

In my mind, there are two reasonable conclusions from the existence of systems with increasing intelligence that can complete student assignments. Either (A) humans cannot use LLMs to do higher-order tasks at work, in which case LLMs are actively harmful towards education, or (B) humans can use LLMs to do those higher-order tasks, in which case we are rapidly moving towards a time when most humans trained in white-collar language-based work are unemployable (since what's the human middleman needed for?).
My perspective is that (B) is what will hold given to what extraordinary extent AI researchers are using AI to make novel developments (and that we've only just begun using RL to actually improve LLMs). If that's the case, then we really will need to rethink education.

One possibility might be to use oral exams more. Maybe students could be asked questions about their paper after they turn it in, which would at least ensure that they've read the paper and understand the ideas in it. The trouble is that this sort of one-on-one teaching scales poorly.

It turns out that if there is anything more implausible than the idea that they might need to write as part of their jobs, it is the idea that they might have to write, or want to write, in some part of their lives other than their jobs. Or, more generally, the idea that education might be valuable not because it gets you a bigger paycheque but because, in a fundamental way, it gives you access to a more rewarding life.

A problem with the argument against careerism is that students are sometimes going hundreds of thousands of dollars into debt. It's irresponsible to do this without having a plan to pay it back, so those students, at least, can and should be concerned that this enormous bet will pay off for them.

So I think it's important to acknowledge that it's a valid concern. But that doesn't mean that students should be solely concerned about how their study will prepare them for their careers. Study can serve multiple purposes.

One paragraph from the article seems like something AI may have trouble with.

I have said this sort of thing to students from time to time, in explaining why most of their grade—all of it, in some cases—is based on such long-term writing assignments. Such assignments have been the most powerful teaching tool in my arsenal, particularly when they are developed in stages: a first draft followed by feedback, followed by a revision, and so on. Assuming that the student puts in the effort, the paper inevitably gets better through such a process.

I don't think AI would do well creating a first draft, refining it based off of feedback in a realistic manner, and then iterating each time. I would imagine that the "wheels will fall off the bus" at some point as the AI is typically expected to give the answer as opposed to an emulation of a iterative writing process.

While iteration can be a difficult thing to embrace for students trying to cram completing an assignment in mere hours before it is due, it may be beneficial to emphasize that iteration more since it is where AI could have a decent amount of difficulty (so I hypothesize, I haven't tested this myself).

One last thing that may help is an introduction of a physically printed source that AI would know nothing about such as an essay provided by the instructor (or a larger body of work). Since it is unique to the professor, AI will hopefully have not seen it to refer to. The AI would need to know of it first requiring effort from a lazy student to enter it (either by typing or some sort of OCR). Still, I would think that the AI would tend to not focus on the newly entered work with a ton of weight, will not cite it correctly, and will just be given a more difficult time figuring things out. When the student's work is submitted, they can then be asked to write a quick opinion on that source since they should have reviewed it when writing their work.

Basically, AI doesn't know everything but that is a claim without proof... so perhaps it helps to introduce something that the AI doesn't know about to prove that point.