I like to use llm to produce code for known problems that I don't have memorized.

I memorize really little and tend to spend time on reinventing algorithms or looking them up in documentation. Verifying is easy except the fee cases where the llm produces something really weird. But then fallback to docs or reinventing.