I took the graduate version of this course, and thankfully did better on this exam than GPT-4. It was a fairly rigorous course, and getting an A required about 20 hours per week of work.
> ... required about 20 hours per week of work.
Of thinking or of remembering?
There were weekly problem sets. The questions on these tended to be a bit more in depth than the midterm/exam questions.
Fwiw Scott Aaronson is maybe the most well respected researcher in this field. I don't think this is a case of some shitty professor asking his students to regurgitate useless information
Have you actually tried using GPT 4 and 3.5 ? I'd say 3.5 could be called a regurgitation machine with a few tricks, GPT 4 seems to apply concepts and feels like showing signs of human level intelligence. If the rate of progress continues for a few years we are collectively fucked.
Yeah, the improvement from 3.5 to 4 is amazing. I'm surprised so many people are so dismissive, and I suspect it's because most of them haven't tried it out. Such are the ever-moving ML/AI goalposts.
That isn't true of this course though. I've read the lecture notes, it really is a great introduction to the subject.