In particular, other alignment researchers tend to think that competitive supervision (e.g. AIs competing for reward to provide assistance in AI control that humans evaluate positively, via methods such as debate and alignment bootstrapping, or ELK schemes).
Unfinished sentence?