Training Data
The examples an AI learns from.
In everyday terms
Text, images or records collected to teach a model. Its quality and variety shape everything the model can and can't do.
For professionals
The corpus used to fit parameters, usually split into training, validation and test sets. Coverage, label quality, duplication and licensing all matter.
Think of it like…
The textbooks a student studied. Gaps or errors in the books become gaps or errors in the student.
You've already seen it
Debates about AI trained on artists' work or news articles are debates about training data.
Myth vs reality
Myth: More data always makes a better model.
Reality: Messy, biased or repetitive data can make it worse. Quality matters as much as quantity.
Quick check
A face-recognition system trained mostly on one age group will likely…
- Work equally well for everyone
- Work worse for under-represented groups
- Refuse to run
- Learn other groups automatically
Show answer
Work worse for under-represented groups: Models reflect the data they saw.