Maybe fun little bike shed: Are LLMs not just fancy Markov chains? They are next token predictors which have some hidden internal state that output probability

Maybe fun little bike shed: Are LLMs not just fancy Markov chains? They are next token predictors which have some hidden internal state that output probability distributions which lead to further states.