Glossary
What is Fine-tuning?
Further training of an existing AI model on your own examples so that it behaves in a specific style or performs a narrow task better.
In plain words
Fine-tuning takes a model that was already trained on general material and trains it a little more on examples of your own. The aim is to change how it behaves: a consistent tone, a fixed output format, or better results on one narrow task.
Fine-tuning or RAG?
People often reach for fine-tuning to teach a model company facts. That usually works poorly, because facts change and a tuned model cannot show where an answer came from. Giving the model your documents when it answers (RAG) is normally the better first step for knowledge. Fine-tuning is better for style, format and specialised skills.
What you need
- Many good examples of the input and the output you want, checked by people
- A way to test the tuned model against the original on examples it did not train on
- A plan for redoing it when the base model is replaced
Common mistakes
- Fine-tuning before trying better instructions and RAG, which are cheaper and quicker
- Training on examples that contain personal or confidential data
- Too few, or inconsistent, examples
- No comparison against the untuned model, so nobody knows if it helped
Questions to ask a company that offers this
- Why is fine-tuning needed here, and what did you try first?
- How many examples are needed, and who prepares and checks them?
- How will you show it is better than the original model?
- Does our training data stay private?
- What happens when the base model is retired?
A good company will answer these plainly and in writing. Vague or evasive answers are worth noting.
Related terms
Find companies
Explanation last read through on 13 October 2026. It is general guidance in plain language, not advice for your particular project. Tell us if something is wrong or unclear.