The post states that the accuracy of the GLM is reported for a test set (out of sample data), but I cannot tell if the other models' accuracies are reported for a test set or training set.

Also the statement "Decreased model complexity leads to less accurate models" isn't necessarily true.