I don't do this stuff at the training level, I just [ask AIs to] make pictures and stories where horrible things happen to people I do not like. That said, giv

I don't do this stuff at the training level, I just [ask AIs to] make pictures and stories where horrible things happen to people I do not like. That said, given that everything that came out of ChatGPT is processed inputs from the real world, wouldn't feeding that output into training another AI basically be some weird new combination of coprophagy and inbreeding (in a digital sense)?