I think you’re missing a subtlety with markov chains. It’s not about picking the next work with highest probability, it about picking the next word using the ne

I think you’re missing a subtlety with markov chains. It’s not about picking the next work with highest probability, it about picking the next word using the next word probability distribution. I played with them almost 20 years ago, and the difference in output was pretty obvious even with simple trigrams. The poetry produced was just better. I can’t imagine any modern llm not using a probabilty distribution function for the same reason.