David Scott Krueger comments on Imitation learning considered unsafe?

[David Scott Krueger]

I don’t think I’d put it that way (although I’m not saying it’s inaccurate). See my comments RE “safety via myopia” and “inner optimizers”.