How I use LLMs to learn new subjects - It’s good at well know facts, things that are mainstream or so niche that there is only one answer: “you should avoid asking the model for concrete details that don’t really matter, and if the model gives you details like that you shouldn’t trust them. But when the model is speaking about facts that do matter - facts that are load-bearing for many other things the model knows about - you can be relatively confident that you’re getting correct information.” // Also, nicely put: “models are not minds; instead they are role-playing a particular kind of helpful character. They prioritize consistent role-play above almost all other concerns."