> Now, I'm sure a sufficiently large Markov chain can simulate an LLM but the exponentials involved here would make the number of atoms in the universe a small

> Now, I'm sure a sufficiently large Markov chain can simulate an LLM but the exponentials involved here would make the number of atoms in the universe a small number. No, LLMs are a Markov chain. Our brain, and other mammalian brains, have feedback, strange loops, that a Markov chain doesn't. In order to reach reasoning, we need to have some loops. In that way, RNNs where much more on the right track towards achieving intelligence than the current architecture.