Overfitting
Overfitting happens when a model memorises its training examples so closely that it fails on new ones. It looks brilliant in testing on familiar data and stumbles in the real world.
In one line, for a 12-year-old
Overfitting is like memorising last year's test answers instead of learning the subject.
An example
A skin-check app trained mostly on photos of light skin performs poorly on darker skin it rarely saw.
Why it matters to people
Systems that only work on people like those in their training data can quietly fail the people least represented — a fairness problem, not just a technical one.