Reinforcement learning can be used on small-scale AI R&D tasks, such as training image classification or video generation models.