[drmanhat]

Oddly enough my reaction to this is that it's a broader societal problem as opposed to an A.I. problem.

Why shouldn't universities switch to examinations where no technology (apart from say calculators) are allowed; and this is strictly enforced? This was certainly the norm when I went to university.

I agree that A.I. trivializes (or changes how you approach) a lot of take home work; but people who wanted to cheat could more or less always do so for that to some degree. I guess it makes it easier to do so; however my expectation would be a greater reliance or weighting on in person examinations as a response; as opposed to a normalization of cheating.

One way in which A.I. could be seen as contributing to this is that it is devaluing the importance of what were seen as 'intellectual' pursuits; as we now have automation for them that is at the very least often surface level effective for undergraduate work.

[taude]

i always found it funny when people complaied about white board coding tests. Back in school, I had to write C code in those little blue books. :)

EDIT: I meant writing in blue books before this era of copying words out of the claude app on your phone

[materielle]

As recently as 2015 when I attended a middling CS program, we had in-person timed exams where we had to write down DSA implementations on a blank sheet of paper in Java.

We were deducted points for trivial syntax mistakes.

If these stories I keep hearing are true, then university programs have really taken a nose dive recently. This isn’t a “back in my day” thing, but within the past 5 years.

The pace of the purported decline makes me question if some of these stories are sensationalist. But I don’t know, I keep hearing about them.

[nradov]

I'm not sure I understand your comment. Surely you don't think that the details of a particular programming language's syntax are an appropriate criteria for grading an exam? That seems crazy.

[j_w]

Rethink it in written language:

> Surely you don't think that the details of a particular written language's syntax are an appropriate criteria for grading an exam?

Computer science is the science of computing. Programming languages are the language used to implement computer science. Therefore you would expect that students accurately use the programming language to answer questions about computing. Seems reasonable to me.

[nradov]

You don't need programming languages to implement computer science. Pseudo code suffices for exams.

[j_w]

You don't need programming languages to DESCRIBE computer science but to implement it you need some programming language.

Quite literally an "implementation detail."

[aarjaneiro]

Sounds fair to me so long as students were aware going into the test that syntax would be graded

[nradov]

Fairness or lack thereof is not the point. Programming language syntax is trade school stuff. And I don't mean that as a slight against trade schools, but it's a different type of training.

[corywadd]

I would suggest the one exception to this would be courses explicitly designed to teach how to use AI, and how not to. But in that case, it's less "use AI to cheat on this course" and "AI is the tool this course is about."

Otherwise your suggestion makes sense.

[asdff]

Then it becomes, teach what? "To use AI", yes, and, then, to do what? Use it how? To make some software? Why? You are already taking software engineering classes to learn to make software. To write something? Why? You are already taking classes that ask you write things yourself. An AI class, to me at least, is akin to taking a class about how to pay someone to write your essay for you.

And if we are talking about the various AI strategies people have where they have LLMs talking to LLMs to come up with whatever gooblyguck, are the poor souls who've been asked to come up with the AI class for the department going to know any of these strategies themselves? Are these strategies even going to be sustainable going forward after VC is no longer subsidizing tokens?

[nradov]

Regardless of VC subsidies, the cost of compute always trends down over time. Whether you like it or not, LLMs will be a pervasive part of everyone's life forever (or at least until a better replacement comes along).

[asdff]

Cost of compute trending down is usually lost as the resulting software bloat that fills the empty space like a gas. We already see this with LLMs. Models get bigger and bigger in an arms race.

[nradov]

OK so the frontier models will be capable of doing even more, but the simple stuff they can do today will get much cheaper.

[asdff]

I don't think that is a safe assumption to make. Moore's law is not playing out any longer as it used to. Jensen Huang already called it dead 4 years ago.

[coolThingsFirst]

If they havent switched already it’s because they are diploma mills and even the veneer of competence training has failed.

Some of these AI chatters may become oncologists.

[bawolff]

> (apart from say calculators)

Generally, i wasn't allowed a calculator in university.