x
Distillation Robustifies Unlearning — LessWrong