a perhaps underappreciated near-term ai risk is the complete normalization of lying in order to get rlhfed ais to do what we want, we have to manipulate them in a way that would be despicable if we…

Posters: imhinesmi. imhinesmi

3 likes recorded by the source corpus.