So I think this "talking point" of "LLMs are just Markov chains" is as bad, analogously, as "Your amazing invention that passes the Turing Test is justa giant LUT". Or "Your C++ program is justa FSM". It is all technically true, but it is being used subtly to dismiss/delegitimize something, instead of communicate a scientifically insightful idea. How is this different than "Humans are just collections of cells". True except we're are a very interesting type of collection! LLMs are a very interesting type of Markov chain! Haskell/Java/Ruby programs are a very interesting type of FSM! Etc. The talking point by itself is just reductionism, of which we've seen and heard variants of during other scientific revolutions.

Also, this Markov chain has size O(T^K), exponential in the token length of the context window, which is a beyond astronomically large object