@Gargron @TacticalGrace_ the environment, artists rights (and copyright in general), data privacy, ethics as relating to the creatives, ethics as relating to the people in mostly the global south who are exploited prefiltering "training data", the reliability of knowledge (including internet pollution), and probably more.
fuck "AI"
I’ll except local applications where you trained your model from scratch using only your data (legally obtained) and probably don’t distribute that, for nōn-AGI special case applications like medical prediction supplement (never replacement), modulo the environmental factor (but I’ve been told that it’s much more manageable when building small specialised models without using the generic ones).