I get the enthusiasm for open weights models but what is the justification from an alignment perspective? And here I mean both narrow alignment (models being used improperly to do l bad things) and bigger alignment (agents or even AGI that are not aligned with humans overall).