AI safety 2 What sorts of systems can be deceptive? Oct 31, 2022 Why you might expect homogeneous takeoff: evidence from ML research Jul 17, 2022