Can we steer AI models toward safer actions by making these instrumentally useful?
2050: Who's in Control?
We need a harm severity impact scale for loss of control
