A newly developed machine learning model can predict the words a person is about to speak based on their neural activity recorded by a minimally invasive neuroprosthetic device.