Pretraining LLMs on Large Corpora

Pretraining exposes models to vast text to learn general language patterns.

What is the main purpose of pretraining an LLM on a large corpus?