I'm curious when someone will do the right experiment in a way that some LLM on Cerebras will do the reasoning so well so big so fast, that it does something very novel