Yeah this is my main issue with the argument. He acknowledges in the article that LLMs are already non-deterministic, but he doesn’t seem to actually understand that.