Powerful boosting learners tend to overfit. We discuss the number of iterations, base learner complexity, and shrinkage as countermeasures.