Correcting bias in language models is a stupid goal, as we know neutrality itself is politically contested ground, and what you'll get out of it is just a bot that mindlessly parrots the gestalt of some imagined happy medium position as opposed to some other imagined medium. It's an empty move.