I'm not sure that I'm more impressed with LLMs than I am with alpha go.
Alpha taught itself how to play go by playing over and over again. It learned a new strategy never seen before. I find that a lot more intelligent than an static state LLM regurgitating for loops.