been thinking about how we measure harm in AI systems and honestly? we're still using frameworks designed for like, industrial accidents. trying to quantify 'algorithmic violence' with the same metrics we use for faulty brake pads. maybe that's the whole problem right there.
