Term explained
What is fine-tuning in AI?
Fine-tuning is giving an already-trained AI model extra training on a smaller set of examples, so it gets better at one particular task, style or subject.
In plain English
Training a large AI model from scratch takes huge amounts of data, time and money. Fine-tuning takes a model that already exists, a foundation model, and trains it a little more on examples you choose: your company’s support replies, medical notes, legal contracts, or text in a particular tone.
It is like hiring a well-educated graduate and then giving them a few weeks of training on how your company does things. They keep everything they already knew and pick up your specific habits.
Why it matters
Fine-tuning lets businesses make a general model better at their own job, often at a fraction of the cost of building one. It is also how companies shape their chatbots’ behaviour: the friendly, helpful style of most assistants comes from fine-tuning on examples of good conversations. Open-weight models are popular for this, because anyone can download and fine-tune them.
An example
An insurance company fine-tunes a model on thousands of past claim summaries written by its best staff. The fine-tuned model then writes new summaries in the same format and level of detail, which a general chatbot would not do without long instructions every time.
What to watch out for
Fine-tuning teaches a model a style or a skill, but it is not a reliable way to teach it new facts, and the model can still make things up. For answers based on your own documents, RAG is usually the better choice. Fine-tuning on poor or biased examples also passes those problems straight into the model.