https://www.landofgpt.com/training-diffusion-models-with-reinforcement-learning/