Learning human objectives by evaluating hypothetical behavioursdeepmind.com39 pointsdatashrimp7 years ago