Modern machine learning is becoming an attractive new tool all over particle physics. On the experimental side, classification networks have been shown to capture subjet information better than any other method. On the theory side it looks like generative networks can help us solve challenges in event simulation and develop new analysis concepts. I will briefly describe the basic features of GANs and how they conceptually fit into our particle physics way of interpreting data. I will then show how these networks can be used to generate LHC events, manipulate event samples, and encode detector simulations. The last aspect leads to the question what we could achieve in general with inverted simulations at the LHC. An established example is detector unfolding over the entire phase space with the help of an invertible network.