This post is about errors and biases in natural language generation models. The models amplify existing biases and produce output that cannot be attributed to source texts.