GPT-3 paper "Language Models are Few-Shot Learners" published

OpenAI published "Language Models are Few-Shot Learners" on arXiv, introducing GPT-3, a 175-billion-parameter language model—10x larger than any previous non-sparse language model. The paper demonstrated few-shot learning: with only a handful of examples in the prompt and no fine-tuning, GPT-3 performed well across diverse tasks including translation and question answering. The results backed up the scaling-laws predictions and strengthened the view that performance would keep improving with size.