pull down to refresh

What if you just distill from other LLMs and not train off of those texts directly...