To teach in the time of ChatGPT is to know pain

LLM use is the most demoralizing problem I’ve faced as a college instructor.

See full article...

 

[jessebod]

In the twilight of my engineering career I had been pondering going back to teaching. It could be a nice way to give something back and find meaning. Sigh. Maybe not.

 

[xoa]

Well...it is even deeper than that.

Employers use degree-holding as an easy sifter for candidates. Why? Because it is extremely easy for employers to audit if you have one. When you have 400 applicants for a job--sort out everyone without a topical-related degree, even when it doesn't need it. Why? Because less sifting work. Which is why employers over-rely on college diplomas, even when they are completely unnecessary.

For employers--how else do you low-friction easily audit hundreds of candidates for potential competence or work ethic? No one has come up with a viable alternative that works at scale.

So of course kids get degrees. You want to get passed the employer ATS--you "need" one. Of course, with so much degree inflation--the wage-value is diluted and wages are forced down for anyone not on the front-end of that degree's popularity. See early MBA people, or early coder people, or early data scientist majors....as opposed to the tail end, where those credentials became expensive unemployable jokes.

Right, like a lot of areas in life there is a pretty fundamental demand for some sort of low-pass filter to just get the S/N ratio high enough to have human brains do anything further with it at all. Naturally the true test of course is actual work product, but that leaves the question of bootstrapping. A modern twist on the old style "apprentice" system might be part of another approach, though with its own challenges. But there has to be some mix of solutions that matches the information processing abilities of the ones seeking people or offering capital and the value they expect with the volume of applicants.

 

[RickRoyLeonPrisZhoraRacha]

I am not an instructor but I support them. When ChatGPT was gaining traction (a year ago, its all a blur?), a faculty member approached me with concern that "we should be using ChatGPT! Other colleagues at other institutions are using it! And my reply was "it needs institutional approval, plus aren't we opening a pandora's box to cheating, to increased costs (data center, computers, bandwidth, tools) and scraping of student data (aka personal info, grades, papers, etc) and collegiate data (person info, publications, works, IP, etc). While the professor agreed, he was dismayed and frustrated that being left out of this "wonderful" technology would not benefit him or his students. See where that is going, right? Indoctrination of this tech by social peer pressure, even at collegiate level. All about selling you tokens.

Students are quicker to adapt than faculty. Staff like myself, we are limited to what we have "permitted" access to along with neutered budgets and lack of training. "There was no AI-101 and how to protect against LLM and AI abuse". Now, we have tools for faculty to check for AI use, but with the caveat that "its not 100% and there could be false positives". And there are false-positives, with some really brilliant students getting flagged and frustrated of the false accusations, whereas the "cheaters" are not always caught.

I never used Cliffnotes. But someone made money saying you needed them to succeed.

 

[Elektriktoad]

The one lesson every single student learns early is that the school as an institution is not there to help them. It's an obstacle course they have to navigate. They all want to have a good life for themselves, every single one of them, but they have to try and judge what will actually help them to achieve that, and what's only there to make them a more valuable commodity for someone else at their own expense. And since most of them are young, either very young adults or literal children, they don't have the experience to avoid making big mistakes when navigating that minefield.

They're using LLMs as a tool to try and manage the parts they've identified as unimportant or detrimental to them. Like the author, I think that, in itself, is one of those big mistakes. But I also think it's critically important to be aware of why they're making it.

This is definitely a big factor. The student conversation described in the article is very telling.
"Why wouldn’t you just have ChatGPT do the assignment?"
“This is my major, I actually need to learn stuff in this class. I use AI for my other classes.”

Students have a lot on their plate, between coursework, clubs, jobs, and social development. Given the opportunity, I bet a good number of students will try to breeze through courses they've decided don't matter aside from the grade, and focus more on the courses they've decided do matter. I don't know how to fix that. Students' time and energy is limited, and many have not yet learned how to manage them effectively.

 

[Robin-3]

I think part of the answer has to be teaching the students that doing the work is the point, not the final result. Hammer it in over and over that the only one being cheated when they use the AI (Wiki, calculator, etc.) is them. Then continue with the quizzes, essays, and teaching.

But - make the final exam in-person, on-paper, and worth the majority of the total grade. Those who have done the work will be fine, and those who have cheated their way through will crash & burn. Which will be instructive in and of itself. Those who want to learn should figure it out pretty quickly that there is no substitute for doing the work.

This is also (unfortunately) going to be an uphill climb because it has to overcome all the rest of the messaging out there about AI and how it's being sold to (and/or forced on) the workforce.

When you have tech companies and upper management shoving AI solutions down your throat as a shortcut to "efficiency" (i.e. quicker results that are mostly satisfactory, most of the time), with no consideration of long-term sustainability (training of junior professionals, reputational risk from inevitable hallucinations/slop-ification, skills that atrophy rather than develop through a career)... well, that's pretty much what the kids are mimicking. And it's what the snake oil peddlers tech innovators insist we should be doing.

They are using the products pretty much as they're advertised. And being able to say "I want to learn this, not just pass the class" (again, in the face of an ongoing marketing push and parents who likely focus on grades as the primary indicator of success) is the kind of thing that requires more perspective and judgement than many teenagers will have (for most of us, those come in as we get older... if we're lucky XD).

 

[42Kodiak42]

While it's not a perfect solution, perhaps multimedia assignments might help a lot. LLMs present a box for text to be put into, so giving them questions that aren't text, such as images and graphs, may counter a great deal of LLM usage.

These LLMs and AI systems are going to have weaknesses that will need to be exploited, and giving them problems that don't fit in the text-hole seems like the starting point.

 

[internetomancer]

Who knows what will happen if the AI bubble pops and the frictionless and ubiquitous access to LLMs withers into something much more limited....

I like this article except for this last bit. People really need to come to terms with the fact that AI is permanent. It will always be ubiquitous and frictionless.

If the bubble pops, in the Stock Market sense, then we could see trillions of dollars of market share disappear, startups collapse, even widespread unemployment, but that isn't the same thing as uninventing it. The invention will still work. The retail prices could jump, say by a factor of 10, but that wouldn't change anything for college cheaters.

The most we could hope for it so make it illegal. But that's a very long shot. And still not the same thing as uninventing it.

 

[ericrz]

But since the appearance of ChatGPT, the instructor’s job isn’t just to teach the subject and frantically attempt to keep every student’s plate spinning. Increasingly, it’s to moonlight as a detective and prosecutor because students without the motivation to do the work don’t have to skip it anymore.

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.

All of this will solve itself, mostly, in the next couple of years. To stay in business, GenAI vendors will have to move to a consumption model instead of this all-you-can-eat buffet we're all currently gorging on. When they do that, use of LLMs (in the volume students use it today) will generally become unaffordable for both institutions and students (except for the wealthiest in each category).

Will that create a "haves vs. have-nots" sort of scenario? Probably. But that's nothing new. For years I've seen the hardware students bring to class; student A with a brand new MacBook Pro sitting next to student B still using the same Chromebook that got them through high school. That sucks, but it's also unavoidable.

 

[Demosthenes642]

Right, like a lot of areas in life there is a pretty fundamental demand for some sort of low-pass filter to just get the S/N ratio high enough to have human brains do anything further with it at all. Naturally the true test of course is actual work product, but that leaves the question of bootstrapping. A modern twist on the old style "apprentice" system might be part of another approach, though with its own challenges. But there has to be some mix of solutions that matches the information processing abilities of the ones seeking people or offering capital and the value they expect with the volume of applicants.

Yep, that's exactly what the degree has been. The assumption being that someone who could manage the requisite skills to get a degree in the relevant field has at least a reasonable shot at being trainable as an employee and came with at least a base level of skill. During my days in SW management it wasn't enough to get someone hired who had potential because we didn't have the time to impart a CS degree equivilant before they could be productive.

That said, something like a third of hires aren't a great fit for a given role over ~5yrs. Hiring isn't an exact science and the art tends to be finding people with the right mindset who happen to also have picked up the requisite skills so that you can actually justify the hire. Surely there is a better way but that's also a billion dollar question.

 

[Oldmanalex]

I used to teach in college, 40+ years ago. All grades were decided by written exams done in invigilated classrooms. For chemistry, it was easy to require answers like Yes, No, 6, or 2,2-dimethylhept-6-en-1-ol. The traditional method will still work, but as Scott points out will trash the distance learning models completely. And now you would probably have to have the students pass through a metal detector before the exam started.

The whole point of a college education is to train you how to think, at least, moderately critically, and how to master the process of learning by yourself. I have self-taught myself over a 50 year professional career greatly more than I learned in college, but without that college experience I would not have been able to do it. But the original skills do not come easily, they have to be worked on, or in our day "beaten in". Occasionally literally for lazy little so and soes like my younger self.

 

[L0neW0lf]

They asked me if I had a degree in theoretical physics

I told them I had a theoretical degree in physics!

(Fallout New Vegas)

 

[Veritas super omens]

I know...just ask the AI how to prevent cheating!

/s

 

[Veritas super omens]

I used to teach in college, 40+ years ago. All grades were decided by written exams done in invigilated classrooms. For chemistry, it was easy to require answers like Yes, No, 6, or 2,2-dimethylhept-6-en-1-ol. The traditional method will still work, but as Scott points out will trash the distance learning models completely. And now you would probably have to have the students pass through a metal detector before the exam started.

The whole point of a college education is to train you how to think, at least, moderately critically, and how to master the process of learning by yourself. I have self-taught myself over a 50 year professional career greatly more than I learned in college, but without that college experience I would not have been able to do it. But the original skills do not come easily, they have to be worked on, or in our day "beaten in". Occasionally literally for lazy little so and soes like my younger self.

This. Learning how to learn.

 

[smallscreensw]

Excellent piece. Again...the ethics lesson continues. Human nature - especially in America - being what it is, the notion is to do the very least possible, the easiest things, in order to "succeed" and get some kind of certification - which too many folks think they're entitled to - so they can get back to what they consider their real mission in life, being under 21 - Having Fun.

"The friction is the point," is a great statement I'd never heard before, but certainly summarizes what education is supposed to be about. It's taking something one is unfamiliar with, wrestling with it, and coming away knowing more than before the interaction. The workplace - where a paycheck and annual performance review - is one thing, but school - where one should be getting the fundamentals on which a career can be built - can't be achieved through short cuts. We need to rebuild our culture so that "failure" - that is, the mistakes made en route to a Eureka moment - aren't seen as moral failings to be eliminated at all costs.

It's also pretty rich that the one student quoted refused AI for a class in his major but is fine using it for their other 12 credit hours.

It seems an entire generation will have to learn the pitfalls of taking nothing but shortcuts through life...the hard way.

 

[MilanKraft]

People using AI aren’t doing work. They’re commissioning it...

Of course they learn nothing. They did nothing. [emphasis mine]

Winning post and pretty much this simple.

And this (from the article)...
"Predictably, the response from higher education administrators―who are busy signing contracts for institutional LLM subscriptions to show how future-first their thought leadership is―has been to tell instructors that their job is to teach students 'how to use AI effectively.'"

...shows how serious the problem is / how deep the hole we're digging ourselves will be. Administrators literally believe throwing gas on the fire, is the solution. Literally. Fucking clueless and out-of-touch. The author and all educator-kind have my sympathies, that's for sure. Couldn't pay me enough to jump in those waters.

Taken to its logical conclusion... given traditional lecturing and testing methods, and the clear, non-exaggerated propensity for students of all stripes now to cheat with AI... this is actively (now) creating a generation of "graduates" who will: know almost nothing substantive; have zero ability to think through a problem critically; and who will not bring anything useful to the workplace other than maybe optimism, a good attitude, and other fluffy concepts.

I know it's not a true cure, but unless higher education has a meeting of the minds on this and goes full-on "everything is paper-based texts and hand-written / oral exams until we can reign this shit in some better way", there isn't a shot in hell this will end well. If they need to reduce class size for a while to do that, then absent a better idea, they absolutely should. Otherwise we have no shot as a society. None.

If the marriage of smart phones and social media was the worst thing to happen to society till now, this "let the LLM do it for you" thing (in school or business), will make social media look like a town hall with Plato.

 

[graylshaped]

If there’s no friction, no effort, then no work occurred

I feel for the author's struggle from a similar but different perspective, having taught a graduate level class the year coming out of COVID where the students did not want to be in a classroom, but the topic being taught was best served via a live dialogue, not conducive to asynchronous learning. Of fourteen students, by the end of the semester, there were two who I'd have granted referrals to, and my co-instructor thought I was being generous. They got out of it what they put into it, which was how we had intentionally structured the class.

Without trying to put the author out of work, I question the value going forward of any degree earned via asynchronous online courses.

 

[manu_lazard]

I've been teaching computer science in French university for 33 years and next year… I won't. I've decided to quit. Using AI is the clear and direct opposite of learning which requires toil and sweat (a few tears sometimes, hopefully no blood). Some students understand this and do their best to "play by the rules" but most of them just take the easy way and end up not learning anything. Sometimes they get away with it but generally they miserably fail the written exam, and of course it's our fault (subject was too complicated or we just didn't explain it the same way as ChatGPT…).

 

[smallscreensw]

When I started my first high school degree, there were almost no tests. You could go visit lectures without presence requirements and then only pass one test mid-diploma and for the final diploma. The idea was, you want to be a researcher so it is your responsibility to get there and we give you the opportunity of being tested. Nowadays, high school is like elementary school because the industry needs streamlined applicants with certified skills and they need the state to pay for their education because they don't want to. In that climate, of course grades become more important than knowledge. As public funding for education goes down, leading to larger classrooms, and people turn from independent learners to easy-to-integrate work force AI use is a symptom of decades of misguided politics.

This is not going away. Local AIs which run on a laptop are already so good, they can make enough assignments meaningless. Even if OpenAI, Anthropic and all the others close tomorrow, the genie is out of the bottle.

We need to take away the motivation for cheating. If the student uses an AI for work management, why do they need that management in the first place? Do they need more free time? Do they need a different curriculum? Do they need to connect the dots better between their work and why they are there? Do they need a state that strongly regulates the labor market for the benefit of all?

AIs can be awesome tutors and hallucinations will die out. AIs are available 24/7 and they answer questions students are too afraid to ask. If the students missed the basics, AI has the patience to explain years of prerequisites exactly how the students needs it and when they need it. Some can give links to original sources and find the most rare documentation of the most fringe simulation tool.

I do not know what is the best answer to this but here is one I give as teacher of programming exercises: I ask my students to take the role of a software team group lead. Yes, instead of copy&paste code from a reddit forum they will commission an AI to write 90% for them but that means they need to explicitly define what they want. They need to write it down for the AI in a way in which they would never have written it down before. So they actually need to think about what they want in advance. They need to check the result and they need to answer me (test like) when I go with them over their solution about what every single piece in their code does. I try to prepare them for a world that will ask AI use of them but to learn something they can only use it to automate the mundane but need to understand everything in-depth. I have found that leads them to be so motivated that they add functionality, because now they have the time, and they still can answer every question about every small detail. Maybe that approach will hit a road-block at some point and maybe it is not transferable to other sciences and levels but maybe something can be helpful for you.

There's valid points here, given the field of software development. But what about other fields? Medicine? Aviation? Government? Law? AI may be able to assist but absolutely should never be a substitute, especially when a life or an individual's freedom is on the line.

Like anything else, AI is a tool. And tools can be used properly or improperly. Fire can cook a meal or be used to burn down a home to intimidate a neighbor. AI can have a place in education, but we must get to a point where proper use cases are developed and codified, in all fields. And students will have to have it pounded into their heads - by policy, if nothing else - that AI is not for cheating and taking shortcuts, and life isn't always going to be "fun." The "friction" is, indeed, the point.

 

[Sming]

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.

But that’s no more true than thinking that logging a count of reps is the goal of bodybuilding. The hard work of lifting weights is the point because that yields physical results. A popular analogy is that using an LLM to write your essay is like driving a forklift into the weight room. Weights get lifted, sure, but nothing is accomplished. I’m not hoping you can answer the exam question for me—I don’t need your essay to get me out of a jam. The process of doing the work was what you needed to walk away with something.

Author - I've taught high school Computer Science for the past 27 years, and as someone deep in the trenches battling these issues, I want you to know that what I've quoted above has resonated with me so much. I plan on posting it in my classroom for my students to read from now until I retire. Thank you.

 

[orangpelupa]

You have my sympathies.

The embrace of LLMs is just so insanely stupid. I long for the day when the current bubble business model collapses. I’m sure there will be LLMs after that, but they’ll need to be priced beyond casual users.

Fingers crossed. I may be Pollyanna.

Unfortunately probably when AI bubble pops, and these Ai services got priced beyond casual users, the price of hardware to run AI locally will be crashing down.

So some people will just move from cloud to local AI.

 

[internetomancer]

All of this will solve itself, mostly, in the next couple of years. To stay in business, GenAI vendors will have to move to a consumption model instead of this all-you-can-eat buffet we're all currently gorging on. When they do that, use of LLMs (in the volume students use it today) will generally become unaffordable for both institutions and students (except for the wealthiest in each category).

This won't happen.

You have to look at the numbers themselves-- the api prices for the cheaper models are 30 to 50 cents per million tokens. Per million. That's more words than a college student writes in college.

Maybe GenAI vendors will need to make money, and need to raise their prices 10x. Apps will crash. Call centers will catch on fire. Software engineers will fret. Video generation will become exorbitant. But cheating college students will be fine. They will lose $3 - $5. Over the course of their college careers.

And if they wanted to, college students could run GenAI on their own personal computers good enough to cheat all the way through college.

It might seem like college students use a lot of GenAI, but the volume, when you measure it, is small.

 

[orangpelupa]

I work in academic technology at a college and I'm currently both doing focus groups with students on AI use as well as designing a series of summer workshops for faculty on AI in the classroom.

The most salient comment I've heard in the past year was from a frustrated student who works hard on papers only to see other students get a higher grade when they spend 10 minutes with ChatGPT because "ChatGPT gives the answers faculty want"

This reveals a serious disconnect between what faculty want (student learning) and how they assess that. We default to grading term papers because those were reasonable instruments to see student capabilities- we don't actually care about the term papers themselves. (We're more than happy to throw then in the trash after a long weekend if grading). We just want to know if a student actually understands what we've been teaching

This is not a new problem- students have always been able to bypass writing papers by getting other (paid) people to write them -LLMs just make that easier and cheaper. The answer isn't to give up or to add some ridiculous amount of anti cheating tech, it's to change how we assess learning. Does that mean in-class handwritten assignments in a blue book? Maybe, if that's appropriate. Oral exams? Sure. But get more creative- I've used serious role playing games where students reenact historical events like the trials of Galileo and interactive simulations to let them play around with how an industry changes over time. One of my guys runs our innovation lab (makerspace) and he's amazing at figuring out ways to get hands on activities mapped to learning goals.

Is reworking your entire syllabus to deal with AI going to be a pain? Yes. Is it going to take more time? Probably. But it's not impossible to figure out new ways to assess learning- and who knows, they might be a lot more interesting than grading yet another stack of mediocre term papers

I wish more schools and universities have "innovation team".

My mom worked as a lecturer in one of the top university in my country, and yeesh...

The lecturer is expected to do EVERYTHING.

Combined with the overload of students...

There's no goddamn time to innovate!

 

[just another rmohns]

I shared this article with my sister & her husband, both of whom are teachers. The one who teaches college commented:

I had a student in one of my classes use [ChatGPT or Gemini] to get easy answers to an online quiz she was taking while I was watching her. When I commented on it she was quite surprised that I would object.
Later, she commented on my label-your-microscope-drawings assignment that “you can’t have ChatGPT do this for you!”

Since the US's well-meaning "No Child Left Behind" Act was passed, schools have been incentivized based on standardized test scores, and penalized if they choose not to. School systems drive teachers, and teachers drive students, and students focus on the grade instead of the learning. They've literally been trained to by the educational system; we shouldn't be surprised by this. But here we are.

He added:

I’m one of those who have gone back to almost all paper tests, labs, quizzes, etc because the online stuff where they can copy-paste the question into a chatbot and then return its answer is just worthless now. I know they probably do that for some of the lab packet questions, but at least they have to hand write the answer and in doing so process it

Even by text message I can feel his demoralization. But he's still trying.

 

[euskalzabe]

AI is just an assistant. Not a great one at that, but occasionally capable. As a writer, I' do my own research, writing, editing and evaluation. I mainly use AI to act as a copyeditor, with careful prompts that guide it through some rails. Whatever it finds when editing my work, I take into consideration. The work, the process, the decision making, are all mine. If you treat AI as an assistant that supervises administrative parts (copyediting, for instance) of your workload, it can be useful. It can even be remarkably apt as a research assistant, it's certainly helped me find sources over the centuries I did not know about, thus expanding my own theoretical approach. But friction can never cease to exist in good work, because if you learn nothing, you have nothing to report to the world.

The students who realize that the easy-way-out means they won't learn, so they'll rarely if ever improve, are the ones who will own the future. Work and do the heavy lifting while you have few worries in young age (assuming a stable family environment where parents provide at a minimum money to pay the bills, food on the table, and a safe living space), use that growth to power your adult life. If you don't learn while young and have to do both at the same time later in life... oof.

 

[MilesArcher]

My sister is a high school english teacher. I feel her pain.

Funny story - she often has to write many letters of recommendation for kids applying to colleges. She has resorted to using LLMs to write these. Other than extraordinary circumstances, she asks the students to give her some source material - what was your favorite book in my class, etc. I chuckle at students using a LLMs to write about a book, then using a LLM to tell about the fact the book was one of their favorites, the teacher using a LLM to write a letter of rec based on this, and the college admissions office using a LLM to summarize and sort the college applications. Cui Bono? (who benefits)

 

[The Lurker Beneath]

There's evidence that this is not correct. The act of writing seems to enhance learning in and of itself.

https://pmc.ncbi.nlm.nih.gov/articles/PMC8641140/

https://www.scientificamerican.com/article/why-writing-by-hand-is-better-for-memory-and-learning/

Speaking as a bad writer, I would say that I remember notes better when I type them by hand - but I am also to some degree giving up the option of reading them again! Horses for courses. A text file for things I want to re-read, a scrawl for things I want to fix in my mind.

 

[cleek]

YT'er Hank Green had a discussion with Bernie Sanders last week about AI/LLMs. and the subject of handing teaching over to AI came up.

it was interesting.

 

[randomuser42]

"The friction is the point," is a great statement I'd never heard before, but certainly summarizes what education is supposed to be about. It's taking something one is unfamiliar with, wrestling with it, and coming away knowing more than before the interaction.

Yah but this is a hard thing to explain to a college-aged student because they've been told something else for much of their life: Succeed in college or your life is ruined forever. Good grades and a degree = better jobs, better life, better insurance, etc. etc. etc. Is that what we should be telling them? Is it even true, now or in the past? Doesn't matter, it's what a lot of students believe. So the risk of failure has been made into a very extreme thing, it's no wonder so many students will take any seemingly safe shortcut. Unless we re-contextualize the whole college experience the only solutions I foresee are ways to directly make it hard to use AI and severe penalties if you're caught. It's hard for a student to really understand the wrongness of something if no one gets in trouble for it.

But how do you catch and/or deter them, in a way that's fair and doesn't have false positives? Well that's the question, of course.

 

[DCStone]

I used to teach in college, 40+ years ago. All grades were decided by written exams done in invigilated classrooms. For chemistry, it was easy to require answers like Yes, No, 6, or 2,2-dimethylhept-6-en-1-ol. The traditional method will still work, but as Scott points out will trash the distance learning models completely. And now you would probably have to have the students pass through a metal detector before the exam started.

The problem with that model of testing is that it encourages and rewards rote memorization over deep understanding. (There's tons of literature on this, btw). Stuff may be remembered for the test or exam, but gets forgotten soon after. That in turn creates the issue where instructors of follow-on courses have to decide how much time to spend "reviewing" content students should have mastered from the previous course.

The whole point of a college education is to train you how to think, at least, moderately critically, and how to master the process of learning by yourself. I have self-taught myself over a 50 year professional career greatly more than I learned in college, but without that college experience I would not have been able to do it. But the original skills do not come easily, they have to be worked on, or in our day "beaten in". Occasionally literally for lazy little so and soes like my younger self.

Ideally, yes. Research shows that students vary greatly in how they perceive the same educational endeavour, though. If they view it simply as an obstacle course between them and employment, they'll take a very different approach and might very well not develop the mind set and skills to engage with life-long learning. The contemporary issue, as set out in the article, is that LLMs make that weaker approach considerably easier and more attractive.

I recently attended a talk from a prof trying to train students in metacognition throughout their coursework. This included the extensive research backing up the tools and processes used to foster that development. Some embraced it, some had already mastered it to a degree thanks to their high school teacher, and others wanted nothing to do with it since it was seen as "irrelevant extra work".

My reading of the literature (which agrees with my anecdotal observations) is that there has always been this variation in student attitudes to educational practices (even the good evidence-based ones). The difference between 40+ years ago and now is that more students are pushed into academic higher eduction than ever before ("You need a degree to get a job!" has been a mantra for a while.) But that also means a greater proportion of students with very different mindsets and approaches to their instructors. I see this reflected in grade distributions, too. Look with sufficient detail at raw scores, and it is never a bell curve: almost always, it is bimodal, with the "getting it" and "not getting it" distributions easily distinguished.

There are no easy, quick, or cheap answers, and I'd suggest you'd be hard-pressed to pick just one of the three, never mind two. Currently, all I can do is inform students about how learning actually works, how LLMs subvert that, and leave it to them to make an informed choice.

 

[Anti-Derivative]

pretty frustrating. I remember absolutely hating many courses preference for closed-book-exam-100%-of-mark assessment strategies. It always seemed to advantage people with particular learning styles over others without giving a true discriminator of who actually got the material. And - for a while a lot of the profs were agreeing with me and considering moving more mark allocations to course and project work.

Well - so much for that.

 

[ScottJohnson]

The last online class that I took we were encouraged to use LLM's on some assignments as long as we documented everything. We could use a cheat sheet for the exams and they needed to be hand written. The professor also required that we take photos of our cheat sheets and upload them for review.

I came away from that class knowing a lot more than I thought I would learn. I am able to use the knowledge I came away with a few times a week. The class gave me a chance to critically think when using Python.

For online classes, what about doing hand written timed exams, and then upload photos of the work to the professor? Could that work?

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.

 

[Minty Beets]

Dealing with plagiarists was a big reason I quit my tenure-track job 30 years ago. It's so much work to document and the "punishment" is usually toothless. It becomes an arms race of how "expensive" you want to make it for students to avoid doing work for themselves. Fuck that noise.

 

[scrimbul]

Tough problem. I usually have some inkling of a solution to problems in subjects I know little about, even if they're likely wrong, but I have no idea how to handle ai use in education. AI is definitely here, and in all likelihood is gonna stay and improve. Maybe treating ai as you would drugs like heroin, and giving instruction from an early age would encourage kids to avoid ai in a learning environment. Honestly, I think that probably wouldn't work, cause everyone's lazy, especially kids, but it's the only answer I can come up with.

Walled gardens, phone faraday cages with liability waivers, frequent public touring/transparency.

Essentially the best learning environments, at least for narrow focus on specific skills and degrees, will resemble prisons, boot camps, or monestaries.

This all requires a lot of money and security of course, so coursework would have to be condensed for a narrow focus and shorter time span.

It's not remotely relevant to K-12 education mind you, and college is already unappealing as it is in the teens or early twenties when it's otherwise the best time to be attending.

None of these sound like remotely pleasant environments to teach in because the cognitive surrender will get more severe without those measures.

I had the misfortune of attending a degree for Game & Simulation Programming but would have benefitted far more from a year or two of boot camp and lab work over long distance learning and online portals. Naturally I'm just now tidying up the financial fallout of that decision many years later.

Nowadays with software engineering earnings deteriorating for entry level positions I'm fortunate I stayed in my IT field making the same money I would have, but I can imagine how severe the temptation would have been to rob myself with an LLM back in my halcyon days when schools were teaching FPSes or whatever. (though the relevant corpus of code is larger by a couple decades)

 

[siliconaddict]

 

[JohnDeL]

Oil prices are going up and never coming back down,

Oh, they are coming back down. In six months or so, the price of oil will probably be pretty close to $70/bbl again. The only reason they are high right now is all the uncertainty Trump keeps adding to the market. But there are plenty of drllable locations here in the US that weren't profitable at $60 that become profitable at $70. Once those come online, expect the prices to drop.

 

[iCowboy]

And require lots and lots of documentation (e.g., references in Harvard format) and step-by-step derivation of solutions (something that most LLMs don't do).

In my experience, adding a requirement for references doesn't really help. LLMs will spit those out too. Some are obviously hallucinated, some are real - but not relevant to the topic, and some are actually useful. But the poor marker now has to wade through the references to determine their quality.

I'm testing oral/viva/presentation route right now for my students. It kind of works (at least for now) but it is very time consuming on the tutor to prepare and conduct them as well as produce consistent marking. As I'm finding out, there have to be all sorts of accommodations for scheduling (especially in an asynchronous distance model) and for students who cannot take part either for health reasons or because they have limited connectivity or have particular security concerns (such as those working in the military).

The way I'm selling it (with some success) is twofold - the first is to give them a trial session where the emphasis is on pastoral care and making sure they are using all of the facilities we have to offer; the second is a more detailed event with the spin that this is what they can expect when they go for an interview or are working in an environment where professional review is commonplace - and I make sure that they realise in the future they might have to justify choices in their code or documents.

But a plague on administrators who tell us to just adapt and find new ways of assessing students because this is the future.

 

[5.0]

FWIW, the thoughts expressed by the author can be confirmed by one horrifying trip to the Teachers subreddit.

Weep for the future.

 

[mmorales]

Amen. Thanks so much for writing this.

I have been in a group of faculty being pushed by the administration about AI, and this is exactly the dynamic. It is about the friction, and somehow those is administration seem to have forgotten this.

Thank you.