If you like the rigor of formal methods and dislike the vaguery of LLMs, shouldn't you be in favor of using the first to improve the second?