On the number of provable formulas