Isn’t that the same? It’s a sort of recursive version of overfitting specific benchmarks