Can you explain what you think makes today's LLMs more than glorified Markov chains? I know they're not technically Markov chains, but conceptually, it's still

Can you explain what you think makes today's LLMs more than glorified Markov chains? I know they're not technically Markov chains, but conceptually, it's still close enough. There's nothing revolutionarily new. No amount of loops and harness tricks is going to change that fact.