"The hot mess theory of AI misalignment"
a favorite talk from a recent alignment workshop turned article; offers a unique and imo fairly realistic framework for superintelligent system futures that departs from your stock paperclip maximizers.

→ View original post on X — @karpathy,