Shouldn't dataset filtering be the priority here? Don't get me wrong, I like my LLMs uncensored, but ingesting angry tweets and other internet trash seems like

Shouldn't dataset filtering be the priority here? Don't get me wrong, I like my LLMs uncensored, but ingesting angry tweets and other internet trash seems like a utter waste of compute and parameter space. If they are going to spew something toxic... At least let it be from an eloquent, concise source.