The blog post explains why the behavior in question is cheating, not a "useful shortcut". Overfitting a model isn't a real advancement to the field. The point

The blog post explains why the behavior in question is cheating, not a "useful shortcut". Overfitting a model isn't a real advancement to the field. The point is to take a closer look at the current benchmarks and potentially come up with better ones.