Updates Ramana Kumar and Scott Garrabrant argue that the AGI safety community should begin prioritizing “approaches that work well in the absence of human models”: [T]o the extent that human modelling is a good idea, it is important to do...
