In my CS undergrad I had Doug Lea as a professor, really fantastic professor (best teacher I have ever had, bar none). He had a really novel way to handle homework hand ins, you had to demo the project. So you got him to sit down with you, you ran the code, he would ask you to put some inputs in (that were highly likely to be edge cases to break it). Once that was sufficient, he would ask you how you did different things, and to walk him through your code. Then when you were done he told you to email the code to him, and he would grade it. I am not sure how much of this was an anti-cheating device, but it required that you knew the code you wrote and why you did it for the project.
I think that AI has the possibility of weakening some aspects of education but I agree with Karpathy here. In class work, in person defenses of work, verbal tests. These were corner stones of education for thousands of years and have been cut out over the last 50 years or so outside of a few niche cases (Thesis defense) and it might be a good thing that these come back.
Yep, it's easy to shortcut AI plagiarism, but you need time. In most of the universities around the world (online universities especially), the number of students is way too big, while professors get more and more pressure on publishing and bureaucracy.
I did my masters in GaTech OMSCS (Chatgpt came out at the very end of my last semester). Tests were done with cameras on and it was recorded and then they were watched I think by TAs. Homework was done with automated checking and a plagiarism checker. Do you need to have in person proctoring via test centers or libraries? Video chats with professors? I am not sure. Projects are importants, but maybe they need to become a minority of grades and more being based on theory to circumvent AI?
It's not even about plagiarism. But, sure, 1:1 or even 1:few instruction is great but even at elite schools is not really very practical. I went to what's considered a very good engineering school and classes with hundreds of students was pretty normal.
For many of the “very good” engineering schools that I know of they got “very good” status because of their graduate programs. In graduate school a 1:few relation is almost certain. In undergraduate, not so much.
Probably generally true. There's some "trickle down" (sorry) especially for students who take direct advantage of it or from the institutional wealth generally. But, yes, students at such institutions who struggle aren't necessarily well-supported.
Ironically the practically of such instruction goes down as the status of the school goes up. I got a lot of 1:1 or 1:few time with my community college professors.
In some university systems it seems to be possible (I'm thinking of the khôlle system in France), so I don't see how the much better funded US system would not be able to do it.
Google tells me that is more of a system with preparatory schools in France. That said, there is more of an emphasis at some schools than others in individual interactions and seminars at the undergraduate collegiate level. I had some of that--just not mostly in engineering. In US elite schools, there's certainly time conflict for professors given research priorities.
Yes, this is part of the French prépa/CPGE system, which is the "standard" way for students to enter elite engineering schools. You do your first 2-3 years of undergrad in prépa.
It happens to some degree in the US. I got a Masters at a school (after a BS at another school) that had 4-year BA degrees that you could extend to an additional 1-year BE degree.
Strongly agree. I was involved with several CS lectures in the past ~10 years that did not require a final exam, and we always did a 1:1 session between student and tutor in which the tutor asked the student detailed questions about their past exercise sheet solutions. Over the years, I estimate that I conducted about 100 of such 1:1s. It was always obvious when the students did not write the code themselves. They couldn't really explain their design process, they didn't encounter the edge cases themselves during testing, and you couldn't discuss possible improvements with them.
15+ years ago I was doing a CS undergrad (or Bachelors? not sure how it translates) at the local uni in a small EU Country and this approach was the standard across all subjects as part of 'lab work'. There were people there to do that, not the prof himself, but approach was exactly the same. And after a few months they had a really good picture on what level everyone is ect.
On the other hand, I had a neighbour ask me if he can make his 1 month apprenticeship when he finished his 3rd year of CS High School (eg ~18 years old, 3 of 4 years of 'CS trade school') 6 months ago or so. I was totally gobsmacked by his lack of basic understanding of how computers work, I am confident that he did not confidentially know the difference between a file and a folder. But he was very confident in the AI slop he produced. I had a grand plan of giving him tasks that would show him the pitfalls of AI -> no need for that, he blindly copied whatever AI gave him (he did not figure out Claude Code exsists), even when the results were very visibly bad - even from afar. I tried explaining stuff to him to no avail. I know this is a sample size of 1, but damn, I did not expect it to be that bad.
Maybe as a society we can take some of the productivity gains from AI and funnel them into moving teaching away from scantrons and formulaic essays. I want to be optimistic.
As someone who was incredibly lazy intellectually in high school, I can't imagine what would have got me motivated beyond time and growing up.
I did nothing in high school and then by 19 for fun on Saturdays I was checking out 5 non-fiction books from the library and spending all Saturday reading.
There was no inspiring teacher or anything like that for me that caused this. At 16 I only cared about girls and maneuvering within the high school social order.
The only thing I can think of that would have changed things for me is if the math club were the cool kids and the football team were the outcasts.
At 16 anything intellectual seemed too remote to bother. That is why I would suspect the real variable is ultimately how much the parents care about grades. Mine did not care at all so there was no way my 16 year old self was going to become intrinsically motivated to grow intellectually.
All AI would have done for me in high school would have been swapping a language model for copying my friend's homework.
> The only thing I can think of that would have changed things for me is if the math club were the cool kids and the football team were the outcasts.
For background I grew up in the US, my wife grew up in China. And how she grew up (in a high tier Shanghai Highschool) she says that is kind of how it was. Top social order was basically Rich and politically connected (not different from anywhere I guess) but also really good students. Where the best students are looked up to. But also just everyone asks you how you do all in school all of the time. There are students who focus more on sports and go to sports schools, but unless they end up going to the Olympics or something, its really looked down upon compared to those who specialize in STEM or more difficult subjects.
In my high school, honors/AP students weren't outcasts, we were kind of just a separate set mostly our own clique with some being popular and some not independently of being AP students. Like I happened to be Football Team Captain and in AP classes, 3 other Captains weren't in AP. Academic success was just a non factor.
The TL:DR of every "AI vs Schools, what should teachers do?" article boils down to exactly this: Talk with the students 1-1. You can fake an essay, you can't fake a conversation about the topic at hand.
Or just do some work/exam in a controlled setting.
Talking to students in order to gauge their understanding is not as easy or reliable as some people make it out to be.
There are many students who are basically just useless when required to answer on the spot, some of whom likely to score top-of-the-class given an assignment and time to work on it alone (in a proctored setting).
And then there are students whom are very likable and swift to pick up on subtle signals the examiners might be giving of, and constantly adjusting course accordingly.
Grading objectively is extremely hard when doing oral exams. Especially when when you're doing them back-to-back for an entire workday, which is quite likely to happen if most examination is to be done in this way.
Not yet but we are getting close to be able to do it, tiny microphone, tiny earpiece, zero AI lag, I give it less than 10 years before it's trivial for anyone.
I think that AI has the possibility of weakening some aspects of education but I agree with Karpathy here. In class work, in person defenses of work, verbal tests. These were corner stones of education for thousands of years and have been cut out over the last 50 years or so outside of a few niche cases (Thesis defense) and it might be a good thing that these come back.