Considering that LLM content that makes it into training content makes the trained LLMs worse… is this adversarial?