It's not an equitable solution*, either, and can raise some PPI concerns (in addition to the point you raise of whether or not it's sufficient for purpose)Exam proctoring software that records the screen and webcam effectively nips "open ChatGPT and copy and paste" in the bud. The problem is that you can use a phone while keeping it out of the view of the webcam, which would still be possible there. I've been advocating for replacing the webcam with a phone camera positioned from the side so you can see hands, which I'm seeing a little movement on from the software companies.
I'd imagine in an MBA program you'd get good grades for cheating - but only if the professor knows you're cheating. How well the student lies about the cheating would be part of the grade also.Who's "we all"? Don't lump us into your cheaters circle.
Sadly, this isn’t true.Students often carry misconceptions about coursework. They may view an instructor as an opponent standing in the way of the grade they want. And they see “getting the right answers” as the goal of education because that’s how you secure that grade.
… The process of doing the work was what you needed to walk away with something.
Not an MBA, but in a music history class I had a professor who would always throw a question on the test that we were completely unprepared for and expected to not know in order to assess our bullshitting skill, one he thought was important.I'd imagine in an MBA program you'd get good grades for cheating - but only if the professor knows you're cheating. How well the student lies about the cheating would be part of the grade also.
Students often carry misconceptions about coursework. They may view an instructor as an opponent standing in the way of the grade they want. And they see “getting the right answers” as the goal of education because that’s how you secure that grade.
Not an MBA, but in a music history class I had a professor who would always throw a question on the test that we were completely unprepared for and expected to not know in order to assess our bullshitting skill, one he thought was important.
I thought that the article was pretty clear as to why LLMs are far superior at cheating than methods of the past. Of course you could always cheat. That wasn’t the point of this article.I think we should just look back to how we did this 25 years ago. We all used Sparknotes, and most problems asked by professors were answered in the spark notes. You could literally copy/paste whole sections and just move sentences around and change some words and get a good grade. Fraternities and academic societies have maintained “test banks” for years.
Which is all to say, this is probably not as big a problem as it seems.
So what do yall do about this problem? The current situation isn't equitable either because it puts the students who don't want to cheat at a disadvantage.I operate under the rule that we can't require the use of such software - and most of my colleagues are in agreement that we wouldn't be comfortable doing so anyway, for the reasons mentioned.
This is the same (flawed) thinking that upper managers apply to return-to-work edicts when some people do genuinly work better remote. "It doesn't work for me so no one could benefit from it".I’ve long been skeptical of long-distance learning.
Primarily I feel that way as: an occasional student, because my motivation flags when I have to spend more time in front of screens. (Though I do recognize that, for some students, long-distance learning is the only possibility.)
But also as a past, and potential future instructor, I’m skeptical because of the cheating plague that has existed long before the rise of AI. LLMs, and other forms of AI are just ‘weaponizing’ that phenomenon, alas…
At this point the only way that I’d agree to teach would be: in-person only; and a student’s grade would rely solely (or at least largely) on their proctored work (e.g., in-class quizzes, exams, etc. ). However I’m not sure if typical schools and administrators would allow that approach…?
Yes, and grades are functioning as a proxy for far too many things to be a good measure of nearly any of them at this point.Goodhart's law would be better written as "when a proxy measure becomes a target, it ceases to be a good measure", which both gets to the point but also helps consider that sometimes proxies are the best we've got. Sometimes one just needs to measure better.
Here you point out that grades have a crazy number of competing uses that create perverse incentives and dilute their meaning. Despite these things they've been a simple, intuitive, and useful enough proxy to reach the point they're at today. My argument is that they have been becoming less useful over time as more and more people have the capacity and incentive to game them. LLM's are, I believe, the tipping point where they no longer do what we need them to do well enough to keep using them.Have we? Or have we allowed grades to get diluted due to perverse incentives? Is the appropriate response to surrender on that or to attempt to realign via better tests? It's pretty important to figure out what the foundational demand is or else we'll just end up reinventing the wheel with new labels. Like with grades, yes the true goal is a complex combination of "how can we motivate students to learn, and measure how well they're learning in a dynamically granular manner, and also measure how well they handle meta tasks related to learning that matter for real world unsupervised activities including continued learning and learning application, and then also convey all that in a high density form that decision makers in different domains can make use of before the student in question has more track record, and also we need to feed that back into how we ourselves are doing with our teaching/methods". That's not an easy problem space, but it's key to recognize that grades became and have continued to be a thing because they've been useful.
Not having the consequence be a direct impact on opportunities and learning potential is not the same as there being no consequences at all. It's about re-aligning the consequences to better match actual learning/effort, and then finding a different (related) measure that better matches capacity for success.But students very much actively want to be able to convey their achievements outside of the class. The whole point for most students is linked to the real world, to getting good jobs or other opportunities they desire. "Consequences" are therefore fundamental, touching on various zero-sum games. Of course people are going to care, and some people are going to do better relative to others, and the ones who do worse are going to have incentives to cheat. You can't just side step all that. Like if some colleges stop providing grades as you suggest, game theory out a bit the results of that. Students exit college, and go to apply to some job, and naturally the company wants to know how they did. How do they convey this in a rapid objective fashion? Will companies then preferentially recruit from colleges that do still provide grades? What would the result when it comes to where future students apply to, and the consequences for the colleges without grades they don't? Etc.
Seriously, friction does matter. Assuming the goal of education is to "get the grade" is like assuming that when I run 3 miles for exercise, it's because I needed to be where ever the end of that 3 miles was.
This is, far and away, the most upbeat and positive thing I've heard anybody say about (relatively) normal people's interactions with LLMs yet. It gives me a bit of hope.I haven’t encountered any students who think they’re learning when they let LLMs do their work.
To be clear, I absolutely think teachers being miserable is a HUGE problem. I just mean that, rhetorically, he doesn't do a great job connecting his arguments to his central thesis.I'm willing to accept that there may be a day where we don't have to worry about this because dedicated teachers are no longer necessary, but I think we're still at the point where we should take seriously a factor that drives people away from a profession dedicated to preparing high quality humans in the next generation. Can't back that up with data, however.
Simple. We don't do remote on-line exams, even for courses that have an on-line component. We're not set up as a distance education institution or department. If we were in that business, we'd have to come up with something, but we're not in that situation at this time.So what do yall do about this problem? The current situation isn't equitable either because it puts the students who don't want to cheat at a disadvantage.
Between a covid generation and LLMs it's definitely been tough, but that subreddit is extreme. It rewards the more extreme stories with more engagement and upvotes so there's a feedback loop where you get the wildest stories.FWIW, the thoughts expressed by the author can be confirmed by one horrifying trip to the Teachers subreddit.
Weep for the future.
That's because you have to be EXTREMELY secure in the world before spending four to ten years of your prime and enough money to buy one or more perfectly nice homes without a VERY explicit goal at the end.This is a remnant from the original “liberal education” purposes of college. You round out a student into a complete thinker and equip them for life, not a career, etc
The problem is that this is completely out of touch with modern priorities and I have become a fatalist about it.
This is exactly what I was going to say. What's the point of trying to stamp out the LLM cheaters? You can't force people to learn (that's part of what makes education different from training). Time wasted on a futile crusade against LLM cheaters would be better spent on supporting honest students.So as an adjunct instructor myself, I disagree. I don't see my job as being "detective and prosecutor," making sure students don't cheat. First of all, that's impossible. There's no foolproof tool that can tell you a submission is "50% AI," or 75%, or 100%, or whatever. I've certainly read some papers that I know were partially or entirely AI-generated. But I can't PROVE it beyond a shadow of a doubt to a disciplinary committee, so I don't go down that road.
This isn't like when I was in college, where if a faculty member found you with a cheat sheet or "crib notes" written on your arm, that was indisputable. Cheating with AI isn't like that.
But in the second place -- I don't feel that my university is paying me to be a detective, or a prosecutor. My job is to deliver education on a specific topic to a group of students each semester, and devise ways to evaluate their learning. I'm delivering the education, and I'm evaluating their learning in reasonable ways. If students choose to fake those evaluations by using LLMs? That's their choice. It's a poor one, but that's their choice.
Unfortunately, cheaters think everyone cheats. Especially cheaters at the top (management). Part of the reason is so they can think they're better than people doing objectively harder work--they assume they're just cheating too. And somehow in the pre-LLM era, they never thought how skilled the people all the cheating was farmed out to must be. It's like the idea of turtles all the way down, but with cheaters.Who's "we all"? Don't lump us into your cheaters circle.
I think this is missing the point of education though, especially post-secondary education. The goal is to learn not just the process to get to end result, but also to learn core concepts about that subject (espcially in science and math courses). I won't say that LLMs are unhelpful in day-to-day work in the real world, they certainly can be, and we should teach taht skill as well. But the whole point of the article is that teachers don't have any good tools to combat how easy LLMs have made short-cutting the work required to learn something unrelated to LLMs, and lazy students (or overworked students) are using them to their own detriment either because they don't understand why they need to put in the work, or because they don't care.Your argument has holes...so we started using skidsteers in factories to increase efficiency, as a result those who wanted had time to go to the gym and workout as a result of the weight yhey gained from using a skidsteer in the factory....even better before the loom everyone had to know how to make shirts, due to the loom we had more time to do other things. Your argument of using a skid steer to lift weights at the gym is backwards, as we used it to make work easier....instead lets look at school lessons. Now we could make a student go to a library and regurgitate a book onto paper....or we can change the goal...instead of writing a paper using the library and pretending the internet doesn't exist, or not using ai....Instead of:
“Write this essay yourself”
It becomes:
“Use AI to generate 5 perspectives, then critique them”
“Find flaws in this AI-generated argument”
“Design a system using AI and explain the tradeoffs”
That’s way harder cognitively than doing grunt work.....just my opinion
Of course it never applied to everyone, nor was it intended to. Colleges & universities grew out of seminaries and were not, until very, very recently, targeted at teaching career skills. That was the realm of professional schools (law, medicine), trade schools, military academies, apprenticeships, etc. - all perfectly respectable forms of education suited to different purposes. The idea of "college for everyone" meant trying to wedge many different goals in to the same institutional framework and, shocker, it's not working very well.That's because you have to be EXTREMELY secure in the world before spending four to ten years of your prime and enough money to buy one or more perfectly nice homes without a VERY explicit goal at the end.
Essentially, you have to be so certain that the world is a safe and welcoming place that will be good for and to you, that you're willing to just "let the universe take the wheel" and gradually show you a path.
Does that sound like the world we currently live in?
Bonus question: does that sound like it has ever applied universally to all classes of any given society?
That's because you have to be EXTREMELY secure in the world before spending four to ten years of your prime and enough money to buy one or more perfectly nice homes without a VERY explicit goal at the end.
I think we should just look back to how we did this 25 years ago. We all used Sparknotes, and most problems asked by professors were answered in the spark notes. You could literally copy/paste whole sections and just move sentences around and change some words and get a good grade. Fraternities and academic societies have maintained “test banks” for years.
Which is all to say, this is probably not as big a problem as it seems.
How did that work out for you? Were you able to get a good job after graduation? Has the missing knowledge ever been a problem for you? Note: zero judgment is meant to be implied by these questions, I'm legitimately just interested to know whether this approach to school worked out in the long run.So yeah, I cheated. I cheated ferociously. Any chance I got to cheat, I took it. I'm not defending that as the correct moral choice or the responsible one, merely the pragmatic. It's almost certain that this had deliterious consequences on my learning. I'd like to think I wouldn't use an LLM to do my work for me if I were going to school today but I probably would. I should qualify that to say that I probably would if I were taking a full time student course load with the goal of obtaining a piece of paper that states I'm qualified to work in my chosen field. If I were taking classes for no other purpose than to learn and develop skills, I would keep my course load light and I wouldn't feel the need to cheat.
That hand-writing could enhance learning in and of itself is not a new idea or one I'm inclined to argue.There's evidence that this is not correct. The act of writing seems to enhance learning in and of itself.
You'd have thought that it would set off a bunch of red flags, for sure. Then again I saw exactly the same thing in some of the undergraduate courses I had to take. I'm not against reusing some questions, and I'm certainly not against students having access to copies of past exams. But recognizing that and taking it into account should be a key part of any instructor's job.One of the most devastating academic disillusionments I experienced came from those test banks.
I was in grad school where ostensibly all of this material is relevant to our ability to perform in our career. I put so much effort into studying for an exam. I’m reading the text, doing practice problems, reviewing lectures, etc etc.
I show up on exam day and there is a little pod of guys I’m friends with reviewing a printout. “What’s that?” “It’s last years exam!”
Every single one of those guys got a 100%. Not a question had been altered. They didn’t study at all, just brute memorization the morning of the test. I got an 85%. And our grades and class rank directly affected our ability to specialize.
One of them had a test bank that basically got the entire group through the first two years of the program at the top of the class. Was it wrong of them to use it? Probably. I consider it “more wrong” that the faculty was repeating the exams verbatim for years. They surely had to notice a disproportionate amount of 100% results!
My hope is that the initial use of LLM was simply due to people doing one prompt searches (transferring google searches) , as we are already seeing LLM natives using the technology on a more interesting way: upload your manual into it and query it on the friction. I can see this working very well on subjects like maths and chemistry for example, where the student can interrogate the text book on the topics that are challenging to understand and then ask it to generate exercises similar to ones they failed. Or talk them through how they got it wrong.
Yes basically every student can have their own teacher, standardised by the quality of the textbooks.