GPT-3

Language models are few-shot learners (2020), introducing scaling laws and in-context learning.