Comment by int_19h

That paper again.

LLMs have been trained on synthetic outputs for quite a while since then and they do get better.

Turns out there's more to it than that.