Llama helped me become more comfortable with Lagrangian mechanics. I had to keep correcting its math errors along the way, but it was nonetheless very good at answering my questions -- like having a patient and empathetic but very absent-minded professor at my fingertips. And because I was always on my toes and double-checking its work for mistakes, I had to learn actively with my brain switched on, instead of just reading passively.

So ironically, its flaws made it a pretty good teacher in this case.

I do have to be especially careful not to ask it leading questions, because it's so biased towards positive affirmation that it would rather lie to say I'm right than to explain why I'm mistaken.