Anfonwch hwn fel neges destun: Faithful rule extraction for differentiable rule learning models