Just don’t use an LLM for learning for doing projects at first. I only use it for things I already know how to do or for research. I treat it like a teenage intern.
And it can be pretty great for that. But I'm not sure if this works well for people who don't have experience reading API documentation or coding support sites like Stackoverflow. Beginners having a problem most likely don't know any abstract term for the problem they want to solve, so they'll feed their scenario meticulously to the LLM, causing it to output an already tailored solution which obfuscates the core logical solution.
It's like starting to learn how to code by reading advanced code of a sophisticated project.