Only if the “direction of truth” exists in the wild with unchecked training data.

That clustering is a representation of the nature of the data fed to the model: all their training data was unambitious true or false… It’s not surprising that it clusters.