zlacker

I like alignment more it is pretty quantifiable and sometimes it goes against 'safety' because Claude and Openai are censoring models.