Generating Magic cards using deep, recurrent neural networks
Once upon a time–by which I mean last week–a PhD candidate decided to feed the Magic: the Gathering card database into a neural network and see what cards it generated. This turned out to be successful (and hilarious).
The generator was instantly popular, and people started inserting the generated text into card templates. Someone even set up a Twitter account that outputs the cards from the generator.
There have been previous Magic-related generators, though they were based around Markov-chains and grammars, and so lacked the deep machine-learning flair of the neural network approach.