In-Context Learning

In-context learning is the ability of a language model to perform a new task based on task description and examples provided within its input context, without any fine-tuning or parameter updates. The model "learns" purely from the prompt structure and examples.

Key papers

• Language Models are Few-Shot Learners — Documents in-context learning at scale with GPT-3, showing that larger models more effectively utilize in-context information

Related topics

• Few-Shot Learning
• Prompting
• Language Models