I’m just saying, how different are transformer based LLMs to just really really really big markov chains with some kind of distance matching rather than exact matching anyway.