#shift-invariance

A reward model is trained with the pairwise loss $-\log \sigma\big(r(x, y_w) - r(x, y_l)\big)$. True or false: adding the same constant, say $+5$, to every reward the model outputs changes the value of this loss.