Scaling MLPs: AT ale of Inductive Bias