When and how to do stop saying LLMs are predicting the nect set of tokens and start saying they are thinking? Is this the point?