I'm really curious now. How were you testing their understanding of overfitting? Like what was the setup/questions?