Technical
Overfitting
A condition where a model learns patterns specific to its training data that don't generalize to unseen data, resulting in high training accuracy but poor real-world performance.
Explained at five levels
Level 1
When an AI memorizes answers instead of truly learning — like acing a test by memorizing answers but not understanding the subject.
Level 2
When a model learns the training data too well, including its noise and quirks, so it performs great on practice data but poorly on new data.
Level 3
A condition where a model learns patterns specific to its training data that don't generalize to unseen data, resulting in high training accuracy but poor real-world performance.
Level 4
Excessive model complexity relative to training data, where the model captures noise and idiosyncrasies rather than underlying patterns — mitigated by regularization, dropout, early stopping, and data augmentation.
Level 5
A regime where empirical risk on the training set decreases while true risk increases — diagnosable via train-test divergence curves, addressable through capacity control (L1/L2 regularization, dropout, weight decay) and the double descent phenomenon in overparameterized models.
Definitions are educational summaries. Terminology can vary by source and context.
Sources
- NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.