#bias-correction

Adam uses $\beta_1 = 0.9$ and $\beta_2 = 0.999$ with $\mathbf{v}_0 = \mathbf{s}_0 = 0$. Ignore $\epsilon$. On the first step, by what factor would the update be larger if you skipped bias correction, i.e. used $\mathbf{v}_1/\sqrt{\mathbf{s}_1}$ instead of $\hat{\mathbf{v}}_1/\sqrt{\hat{\mathbf{s}}_1}$?