ryan_greenblatt comments on The case for ensuring that powerful AIs are controlled

[ryan_greenblatt]

I strongly agree that a bunch of careful thinking and red teaming will be likely required for control to work on early transformatively useful AI.