I think that's quite a strong claim, especially since chain-of-though reasoning means a modern LLM has its own private scratchpad to workshop sentence structure in, if it were a significant problem.