$\color{black}\rule{365px}{3px}$

“Language Models are Few-Shot Learners” - 2020

Link to Paper:

arxiv.org

Table of Contents

1. Introduction

$\color{black}\rule{365px}{3px}$

2. Understanding GPT-3

$\color{black}\rule{365px}{3px}$

In-Context Learning

One-shot / Few-shot Learning is to give single/multiple example(s) (or "shot(s)") of a task in its input prompt and is expected to generalize and perform well on that task. This is possible because GPT is retrained on a huge amount of diverse text data, that taught the model to develop a general understanding of language and a wide variety of tasks.