I remember having 7 day take home open book open internet exams in my math classes in college. We were also expected to typeset all our answers.

Those exams probably took the median student 30 hours and had a median score of 50% including partial credit.

Sad to think current students will never get that experience, because a chatbot could write something good enough to get a 50% in 10 minutes, for many of those classes.

True… but, we’re in a transitional stage still. Eventually, if these models are really going to be such a big deal, then the workflows of real mathematicians will probably involve using them very effectively. Eventually the generation that grew up using the things will become professors, and they’ll be good enough at using the things to ask a question that’s still hard with an LLM.

Well, either that will happen, or the pace of advancement will increase faster than humans can adapt. That’s the singularity, right? I don’t think we’re that close to it.