It's good to see someone building something easily explorable. If people had more access to experiments like this, there would be far less BS hype around deep learning, since the public would get some intuitive feel for what those networks are doing, how they work and what are their limitations.
It find it hilarious how many people who ought to know better claim that GANs do not rely on memorization of source data. Obviously they do. At some point people even studied how connectivist memory could work (e.g. Hopfield networks, which I've just started looking into - fascinating stuff). Somehow that branch of research is never mentioned in any of the recent deep learning papers.