Nope. If I feed an NLP model a bunch of shakespeare, it'll "echo the biases it's fed," but it won't do what you probably wanted it to. A less silly example is f

Nope. If I feed an NLP model a bunch of shakespeare, it'll "echo the biases it's fed," but it won't do what you probably wanted it to. A less silly example is feeding the system a bunch of wall street journal articles and then trying to use it on informal communications like twitter. The goal is to learn something that solves a particular problem. It turns out that the current methodology solves problems in a biased way.