When LLMs are given a classification task like that, are they doing all the analysis natively or invoking classical clustering algorithms with tools?