I think my preferred LLM usage style these days is to try something on my own, and then get the LLM to critique what I did. I think it's both better for learning (just like learning natural languages, it's better for learning to make a mistake and be corrected than it is to look up the right answer each time), and also I think the stakes are lower since you're not relying on the LLM to not hallucinate. You can also learn some new threads to pull on that you can investigate on your own etc