I think this does something for the correctness of LLMs by making it easier to check their output.