How to Train a G-Flonet?

The G-Flu network is a learned sampler. It can query the model, like in model Bazel-Rl, right? So if you want to train a policy, I mean, that's the way we train a goal-playing machine. You don't actually need to play with a human. This is fake data. And so now the having a very large net of the kind we love these days could be useful even if the data set is small. Because what you're learning is how to convert the knowledge in the data set into a model and then convert that model into a machine that can answer questions as inference.