GPT3:Language Models are Few-Shot Learners (2020)

huyi / January 2023

0. Abstract

GPT-3：通过增大参数量，在不用fine-tune的情况下就在下游任务上，few-shot任务上表现很好。

1. Introduction

• traditional: pre-train + fine-tune
• task-agnostic model $\rightarrow$ for a desired task