Reference · Glossary

Pre-training

The expensive first stage where an LLM reads massive text to learn general language patterns — before anyone fine-tunes it for a product.

When to use

Explaining why base models know grammar and trivia but not your company's private docs.

When not to

Day-to-day chatting uses **inference**, not pre-training — you are not re-training the model each time you send a prompt.

Example

GPT-style models are pre-trained on broad internet text, then aligned (RLHF) and optionally fine-tuned for assistants.