This could lower the quality of LLM training data, so I don't think it's a good idea at this point.