>>105995962
He was probably talking about hallucinations (which aren't really a variance problem so much as a more general unreliability problem).