Language Models are Few-Shot Learners (2020) ☆

OpenAI showed that a 175B-parameter model could perform translation, arithmetic and reasoning tasks from examples in its prompt alone — in-context learning — validating the bet that scale itself produces generality and launching the API era.