Careful cross-validation, ensembles, careful data preprocessing can to specific extend reduce the loss and achieve the tuner-specified optimal parameters
Using techniques like k-fold cross-validation ensures that the model is evaluated on different subsets of the data, reducing bias.