Maybe models can use the book "Calling Bullshit" or something like this, when they have conflicting sources, to find out sources which are bullshit.