Before 2020, teaching a language model a new task meant collecting thousands of labelled examples and retraining it, an expensive ritual repeated for every job. Then OpenAI published "Language Models are Few-Shot Learners" and described GPT-3, a model with 175 billion parameters, roughly a hundred times larger than its predecessor. Its headline finding was not that it wrote better sentences, though it did. It was that you could teach it a new task simply by describing the task, or showing a few examples, in the prompt itself.