[anu7df]

I understand model output put back into training would be an issue, but if model output is guided by multiple prompts and edited by the author to his/her liking wouldn't that at least be marginally useful?