The phrase "in practice" doesn't really work when you're referring to highly finicky strategies like RL and adversarial training