Don’t you talk about my tokenized data like that - there’s nothing mundane about prompt engineering via proper clustering of tokenized data (or something)