Boosting one more tweet from my new research alt, where I brainstorm how an algorithm that’s robust to adversarial corruption might work. I’d be surprised if this question hadn’t been explored before, but I couldn’t find where; pointers welcome.