OK, let’s drop autonomous vehicles for a second. Things are getting serious. This post is about creating a machine that writes its own code. More or less. Introducing GlaDoS Skynet Spynet. More specifically, we are going to train a character level Long Short Term Memory neural network to write code itself by feeding it Python source code. The training will run on a GPU instance on EC2, using Theano and Lasagne. If some of the words here sound obscure to you, I will do my best to explain what is happening. This experiment is greatly inspired by this awesome blog post that I highly recommend reading. I am by no means an expert on deep learning, and this is my first time fooling around with Theano and GPU computing. I hope this post will show how easy it is to get started. Some background Neural networks are a family of machine learning algorithms that process the inputs by running them through layers of artificial neurons to generate some output. Training happens by