Two models of AI oversight – and how things could go deeply wronggarymarcus.substack.com1 pointson_of_gloin3 years ago