Using AI is a different skill set that allows you to dive into topics that you otherwise aren’t ready for. I just used it to do a task that would have taken me a couple days of reading up on a different software system that I wasn’t already familiar. Now I have no need to ever really know that system, is that a good thing or not? I don’t know yet. But I had to know lots of basics about how those systems work in general to get the AI to do the thing I wanted, snd it wasn’t a one shot prompt, rather it was an iterative prompt process.
I touch LaTeX once every 10 years. I'm not going to learn it because I'm not fond of debugging macro processors and have never had a good experience with the language where you have to invoke a stew of packages that will mysteriously stomp on each other. I generated a script the other day to prepare a document in the format I needed. It mostly worked, but the LLM also stumbled on the packages until I could coax a working solution out of it. They're good for these problems where you only need shallow knowledge.
Most of us who touch latex make our one great template and forget it, or at least we try to just work off what is given to us.
You still need "knowledge" to use AI, but AI can handle details. Students relying on AI to pass classes means they might not ever obtain the knowledge they really need to use AI well, or maybe I'm cynical and they actually learn the cursory knowledge they need to use AI during the test because otherwise they wouldn't be able to use AI.
I hope there are at least some classes on using AI to solve problems though, like in a domain. "Using AI to boost programming" should be a CS course at least that you can take after you learn programming the manual way.
Much like how if you stop going gym you lose muscle mass, the same happens with knowledge and understanding with the brain.