[vrighter]

You can very easily inject wrong information into the state transition function. And machine learning can and regularly does do so. That is not a difference between an LLM and a markov chain.