A pattern I have seen repeatedly for the last year is people who assert this, then have a problem and come to me/people like me for help, only to find that the basic research they did was complete garbage.

AI is very, very good at spitting out language. It's very, very bad at ascertaining the quality or correctness of the language it spits out.

And the problem with plausibly correct output is that sometimes it's good enough until it isn't, which is the pattern of behavior I've seen with heavy AI driven research.

I'm not going to discount that it's better than google searches. But it's hardly good enough to equate to undergraduate understanding. The act of reading and writing from primary/secondary sources crafted by humans is in itself the way to acquire and retain knowledge. Having it hallucinated at you in a plausibly correct way is dangerous to equate to that.