The key point seems to be "generated comments", not "use of LLMs".