hypothesis question research llm

Can we look at an LLM with the perspective of a giant Hidden Markov model ? If so, what can we use from Hidden Markov literature to interpret and improve LLMs ?