How it works:
since M2 threshold is lower (by process, uncontrollable anymore), in order to get the same mirrored current, the output stage biases him with a lower Vgate w.r.t. that of M1 so that there is a finite (unwanted) offset between input and output.
By moving the potentiometer's cursor toward right the resistive path from Vdd to M2 decreases so that more current of that in M1 flows (mirror is not balanced anymore). By consequence the output stage increases the Vgate of M1 to allow this major current to pass reducing the offset error to zero.
