Review for NeurIPS paper: Your GAN is Secretly an Energy-based Model and You Should Use Discriminator Driven Latent Sampling

The paper claims that the model has a fast mixing. I don't see it in the paper. A good mixing means the a single MCMC can approximate the target distribution and traverse almost all modes. Your model just corrects the samples from GAN. This paper has a very serious misunderstanding about the definition of MCMC mixing and mistakenly uses Fig.5 as a support.