What is the deconfusion branch of ai alignment research?

We don't have an answer for this question yet. Would you like to write one?