Doing is not understanding. Even LLMs seem to know the difference.
I’ve written and spoken a lot about this (link to the talk). Naturally, the exchange here was too good not to share. Here is Claude in Cursor lecturing a user on the difference between having something coded by an LLM vs. coding it yourself so you learn and understand.
The better we separate things we need to understand from things we just need to do, the more effectively we will benefit from LLMs. We certainly can’t understand everything (nor do we need to), but it’s a good idea to avoid the illusion of understanding just because we can do it.
To paraphrase Feynman, we can only understand the code we can create.