I am neither in ML nor in math nor in AI alignment, so just throwing it out there. From my reading of the issues facing the alignment research, it looks like the very basics of formalizing embedded agency are still lacking, but easier to make progress on than anything directly related to alignment proper.
I am neither in ML nor in math nor in AI alignment, so just throwing it out there. From my reading of the issues facing the alignment research, it looks like the very basics of formalizing embedded agency are still lacking, but easier to make progress on than anything directly related to alignment proper.