The field of diffusion models and sampling methods is witnessing significant advancements, with a focus on improving controllability, efficiency, and quality. Researchers are exploring new techniques to align diffusion models with human preferences, such as using preference classifiers and condition preference optimization. Additionally, novel sampling methods, including parallel sampling and speculative rejection sampling, are being developed to accelerate sampling times. These innovations have the potential to significantly impact various applications, including image and language generation. Noteworthy papers in this area include: CPO, which proposes a condition preference optimization method to improve controllability in text-to-image generation. PC-Diffusion, which introduces a preference classifier to align diffusion models with human preferences. TiDAR, which presents a hybrid architecture that combines diffusion and autoregressive models for high-quality and efficient language generation. FSampler, which accelerates diffusion sampling by reducing the number of function evaluations using epsilon extrapolation.