So maybe the answer is to write decent documentation for LLM's to train their data their data on.