The problem is that the model doesn't know if anything it's saying is true or false, so trying to make it 'fact check' just means it will constantly interrupt itself regardless of the accuracy of the output.