Alpha left

The "alpha left in X" refers to the amount of opportunity lies in optimizing across a particular dimension "X". This is a homage to the learning rate in stochastic gradient descent (referred as "Alpha"). If there's a lot of alpha left in a dimension, you more quickly approach the local minima (minimizing loss with larger leaps). Therein lies opportunity through the asymmetry of effort to gain (eg. getting twitter clout or scamming VCs out of money). Once the local minima is reached, large alphas will increase loss, at which point the asymmetry of effort to gain is lost. For more; see some adaptive gradient descent algorithms e.g. AdaGrad.