ModelRefs / Model Collapse — AI Glossary

Model Collapse — AI Glossary

Degradation in a model trained on AI-generated data as synthetic outputs crowd out the diversity of real human-authored text.

Overview

Shumailov et al. (2023) demonstrated that iterative training on model-generated content causes distributional narrowing and eventual capability collapse. Relevant to any synthetic-data pipeline. Mitigated by maintaining high-quality human data anchors and monitoring output diversity.

Reference details

Topic	architecture
Also known as	model autophagy disorder
Last reviewed	2026-06-24

Primary source

Shumailov et al., The Curse of Recursion: Training on Generated Data Makes Models Forget (arXiv)

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Model Collapse — AI Glossary.

Frequently asked questions

What is Model Collapse?

Degradation in a model trained on AI-generated data as synthetic outputs crowd out the diversity of real human-authored text.

Is Model Collapse the same as model autophagy disorder?

Yes — model autophagy disorder are common aliases for Model Collapse.

What concepts are related to Model Collapse?

Closely related concepts include pretraining data, data augmentation, alignment.