Hacker News Reader

The models are already trained on model output on purpose

Maybe so, but in a controlled and deliberate fashion.

Not just random output of who knows what model, from who knows what prompt.