Exams are just a good idea. We sacrifice a huge amount in the name of "some people get overly stressed about exams", when really we just need to accept that some people can't cope with very much stress and will probably get a fairly stress-free job, like burger-flipper or project coordinator or VP of HR.

It really doesn’t make sense to waste tons of potential in order to preserve uncomfortable, high stress environments and systems. Seems like a loss for the sake of a loss.

It's not to preserve them. It's to preserve being pretty sure that someone knows what their qualification claims they know.

Are there methods we could use which avoid filtering out people for lacking skills unrelated to those they’ll deploy after graduation? Certainly I’ve seen examples of simple changes that do a lot of good, like single user exam spaces (with a camera). Even basic changes like that can seriously improve exams, but often require resource allocation. If people think things like, “exams being stressful is good, these students just need to toughen up,” then we won’t allocate those resources and we’ll miss out on a good investment.

The school I work with is considering a return to handwritten papers, and the more I dove into the topic, the more I think it makes sense

I made a text editor that detects whether something has been AI generated by analyzing the keystrokes and edit history: www.collie.ink

Haven’t really told anyone about it so no one is using it, but I think it’s a good idea!

Very interesting! I can see how useful it would be, though I’m hoping the adversarial “AI learns how a specific individual types” and just dumps its output in there doesn’t become a thing.

What's so hard about hand copying to paper the essay the AI wrote?

Presumably this would also be joined with in-person proctoring, with a blue book, or its equivalent, and a writing utensil. Not much room to copy down an AI-hallucinated answer when all you have on your person is pen and paper.

that's all fine and good for a subset of essays, but there are others, like research reports, that take much longer than 2 hours to complete.

Pair it with an in-depth oral examination based on the report. Hell even have AI conduct the oral examination, since it can read it thoroughly much quicker than any teacher or TA and draft good questions. The teacher can grade for correctness and whatever other factors they already grade for.

can an llm do citations and bibliography without hallucinating? This seems pretty easy to verify and hard to cheat with.

Have you tried Gemini Deep Research?

Can’t I just do it on a computer and then copy it over ?

Yes for homework papers you could, though you'll still be learning (and often retaining) a lot in the process of handwriting a copy :-D

There's also classroom time being given to students for their writing, with no computers available (or with just monitored access with much AI stuff blocked)

Why not pencil-and-paper? Oral is vastly less efficient, and also problematic for ESL and many other types of students.

Bring back cursive!

I remember when I was a kid people would just copy other people’s essays and just reword the sentences. So people who want to cheat will figure out ways to cheat on homework, etc. AI is just the new vehicle.

I think grades should be based 90% on exams, and 10% or less on homework. Homework should be practice not what your grades are based on.

I am all for in-person written and/or oral exams. We still do it for PhD classes. But we don't have people to conduct oral exams for large classes like in undergrad or even some masters level courses. In my university, even getting a TA is challenging for many faculty due to cost cutting.