Overfitting
When a model learns its training examples too closely, so it performs well on them but poorly on anything new.
It is a core reason a model that looked great in testing can flop in the real world.
Frequently asked questions
What is overfitting in plain language?
It is when a model learns its practice examples too closely, so it aces those but stumbles on anything new. Like a student who memorized last year's exact test answers instead of understanding the subject, then freezes when the questions change.
Why did an AI that tested great fail in the real world?
Overfitting is a common culprit. If the model was tuned too tightly to its test data, its impressive scores may not carry over to messy real situations, which is why a strong benchmark result does not always mean strong real-world performance.
Is overfitting something I need to worry about as a non-technical user?
You do not have to fix it, but it is worth knowing it exists, because it explains why a slick demo can disappoint once you feed the AI your own varied, real inputs. It is a reason to test any AI tool on your actual work before trusting it.
New to all this? Start with what an AI agent really is, browse the full glossary, or explore the learning hub.
← Back to the glossary