Is this training data even valuable? Usually AI data annotators get paid to write LLM responses, but here all they'd be getting is a bunch of user queries.