I agree, but also: more efficient architectures are less likely to be undetectably situationally aware & deceptively aligned, because they have less surplus computational capacity in which to…
Posters: davidad. davidad
4 likes recorded by the source corpus.
Topics: AI safety