AI glossary · Beginner · Language models
Supervised learning
Supervised learning is a machine learning approach that shows an AI data in which questions are paired with correct answers, so it learns on its own the rule 'for this kind of input, give this answer.'
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Basics
- Fields
- Language models
- Added
- 2026-10-08
How it works
First, people attach a correct answer to each piece of data. This answer is called a 'label'; for example, each photo is marked 'cat' or 'dog.' The model looks at the input, gives an answer, compares it with the correct answer to calculate how wrong it was, and adjusts its internal values little by little to reduce that difference. Repeating this over many examples lets it learn the patterns between inputs and correct answers. After training, it can predict answers for new data that has no correct answer attached.
Why it matters
Supervised learning underlies many of the AI features in real use today. It is especially strong at 'problems with a clear right answer,' such as filtering spam, recognizing objects in photos and converting speech to text. Because results can be compared directly with the correct answers, it is also easy to measure how well a model performs. SFT (Supervised Fine-Tuning), used to refine large language models (LLMs), is, as the name says, a form of supervised learning.
Things to know
Collecting large amounts of labeled data takes a lot of time and money. If the labels themselves are wrong or skewed, the model learns those errors as they are. The model can also overfit, fitting the training data too closely so that performance drops on new data. There are two main problem types: 'classification,' which picks one of a set of categories, and 'regression,' which predicts a numeric value.
Example
An email service's spam filter is a typical example. If a model is trained using emails users have 'reported as spam' and ordinary emails as labeled data, it can automatically decide whether a newly arrived email is spam. As user reports accumulate, the labeled data grows and the filter improves along with it.