LLMs allow me to tackle stuff I have no business tackling because the support from the LLM for the task far exceeds google / stack overflow / [insert data source for industry or task].
Does the concept sink in? Yes and no, I am moving too fast most of the time to retain the solution.
When the task is complex enough and LLM gets it wrong, oh boy is it educational, not only do I have to figure out why the LLM is wrong, I have to now correct my understanding and learn to reason against it.
I was a very bad student, most of the classes didn't make sense to me, bored me out of my mind, I failed a lot. Do I ever feel that way when talking to Chatgpt about a task I have no idea how to solve? No, and guess what we figure it out together.
Another data point, my english writing has improved by using chatgpt to refactor / reformat, more examples, mostly correct english structure. Over time stuff sinks in even if you are not writing it, you are still reading it, and editing.
Lets take code for a minute, is it easier to edit someone else's code or your own? So everyone that has to dive deep into troubleshooting chatgpt's code is somehow dumb/lazy? I don't think so, they are at least as smart as the code.
What would happen if we made a curriculum around using chatgpt, how far would I get in chem 1 if I spent 90 minutes with chatgpt prompts prepared by a professor and a machine that never gets tired of explaining / rephrasing until I get it?