L039 · Under the hood · 3 min
How fine-tuning works
Fine-tuning takes a model that already knows language and trains it a bit more on your own examples, so it picks up a style or a task.
01 · The idea
Large models start as foundation models, trained on a huge amount of general text. They can do many things, but none of them exactly the way your team does.
Fine-tuning means training that model further on a small, focused set of examples: maybe a few hundred support replies in your company's tone, or forms filled in the right format.
After fine-tuning, the model leans toward that style without needing long instructions every time. It is good for consistent tone and format, and less good for teaching new facts, which change and are better looked up with RAG.
It needs clean examples, some technical work and testing afterwards. Many teams find that better prompts or a few examples in the prompt get them most of the way first.
02 · Think of it like this
It is like a skilled cook joining a new restaurant. They already know how to cook. A week of practising the house dishes teaches them the local style.
03 · Do it now
04 · Quick check
Words in this lesson
See also
L042Under the hood · 2 min
Synthetic data: when AI makes its own practice material
Synthetic data is training material made by computers, often by other AI models, to fill gaps where real examples are rare, private or costly.
L028Under the hood · 2 min
Fine-tuning vs prompting
Before you train your own AI, try a better prompt and your own documents; that solves most problems faster and cheaper.
L025Under the hood · 2 min
RAG: give AI your documents
RAG lets a chatbot look up the right pages in your own documents before it answers, so it can use facts it was never trained on.
L054Under the hood · 2 min
Overfitting: learning the answers, missing the lesson
A model overfits when it memorises its training examples so closely that it fails on anything new.