UC Santa Barbara

"In a shocking finding, scientist discovered a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English.” The by-now notorious prompt for OpenAI’s GPT-2 software was, after ten attempts, followed by the announcement that “the scientist named the population, after their distinctive horn, Ovid’s Unicorn.” The discovery of course was not that of evolutionary biology but rather computer science, not silver-white unicorns, but an unsupervised language model so uncannily successful that it too might have seemed chimerical—an associative link reinforced by OpenAI’s decision to withhold the trained model from the public, for fear of malicious misuse. How should we teach natural language processing, which is to say human reading and writing, after GPT-2? What, quite plainly, should be the response of language instructors to the exponential developments in the field of neural text generation?