On the Length of Proofs