They do well on the harder problems after training. You have to train them a specific way, for example in rate problems, before they have a "regimen" for the harder problems