>These 2 examples are somewhat cherry-picked That's for sure. I have seen many Markov chain implementations, and if you could generate 1 funny thing for every

>These 2 examples are somewhat cherry-picked That's for sure. I have seen many Markov chain implementations, and if you could generate 1 funny thing for every 10 tries, then that was a good day. Both Markov chains and LLMs have a distinct style, which gets old over time. Markov much faster for me. So, in my experience LLMs win, by far. I do agree with the author that the LLM style can get really boring. I experienced the same myself. But the Markov results, while much less restrained, are so much more nonsense too. Often questioning its overall usefulness. Which, I think, the world also agrees on: while Markov chain implementations were fun toys at best, which worked sometimes to a kind of funny degree, LLMs are everywhere.