“GANs Didn’t Fail, They Were Abandoned”, Gwern2022-10-04 (GAN, AI scaling; backlinks)⁠:

Diffusion models supposedly beat GANs because they scale better and stabler. That is unproven, and false. GANs should be revisited.

The BigGAN JFT-300M result is not the only such result, because in 2020, Tensorfork replicated the stability of BigGAN scaling on a dataset of n > 100M.

• GAN Advantages
• Diffusion Advantages
• Diffusion Non-Advantages
  • Scaling
  • GAN Instability
  • BigGAN JFT-300M
  • Tensorfork Chaos Runs

Backlinks: