RL algorithms are less bitter-lesson-pilled than 2015-era deep learning