OpenAI has just released this post on improvements they made on their transformer deep learning architecture used for NLP. They find their results so impressive that they decided not to release the trained model...Why ? because they’re afraid of the potential of misuse in an era where fakes are becoming really concerning when it comes to manipulate people. For me it sounds weird to look at the problem in this way, but why not. Anyway, you can have a look on how the model is able to generate coherent paragraphs of text given an arbitrary input, it’s worth the read (go to samples section) - hope OpenAI announcement is not a fake 😉