Lesson 42 · 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.
Models learn from examples, and good examples run out. Some are private, like medical records. Some are rare, like a factory machine failing in a strange way. Some are just slow and expensive for people to write.
So teams generate data. An AI model can write thousands of practice conversations, maths problems with worked answers, or fake but realistic customer emails. A simulator can create images of road scenes in fog, snow and night.
This helps a lot when it is checked carefully. The risk is that made-up data copies the mistakes and blind spots of whatever made it. Train only on your own echoes for too long and quality can drift and get narrower.
Good practice mixes synthetic data with real data, filters out bad examples, and tests the final model on real-world cases it never saw.
Do it now
Make a small synthetic dataset
- 1. Use the prompt below to generate practice examples for a task you know.
- 2. Read them as an expert would. Mark any that are wrong, odd or too similar to each other.
- 3. Notice how quickly you would need a human check before using them.
Create 8 short, realistic customer messages to a small bakery. Mix the type: 3 complaints, 3 questions about orders, 2 compliments. Vary the tone and length. After each one, label its type in brackets.
Quick check
Words from this lesson
- Synthetic data
- Artificial examples made by computers, often by other AI, used to train or test models when real data is scarce or private.
- Training data
- The examples, such as text, images or recordings, that a model learns its patterns from.