L·ALiving AI AtlasA Coding As Creating project
← Concept Index

Training & adaptation

Fine-tuning

Also called: adaptation, custom model

DEFINITION

Fine-tuning continues training a base model on a smaller, targeted dataset so it adapts to a style, domain, or task. The result is a new set of weights.

WHY IT MATTERS

It's often reached for when the real fix is better prompting or retrieval. Fine-tuning changes behaviour and tone well, but it's a poor way to add facts — those go stale and are better supplied at use time.

COMMONLY CONFUSED WITH

Retrieval (RAG). Fine-tuning bakes patterns into weights; retrieval supplies fresh facts in the prompt without retraining.