Nature paper finds models trained on AI-generated data degrade into ‘model collapse’
See how this story connectsIlia Shumailov and Yarin Gal of the University of Oxford, with co-authors, published ‘AI models collapse when trained on recursively generated data’ in Nature on 24 July 2024. They showed that training generative models repeatedly on output from earlier models makes rare patterns in the original data disappear and outputs grow narrower.
Why it matters
Developers training models on web text that increasingly contains AI output got peer-reviewed evidence that models fed on their own outputs lose rare patterns and grow less diverse.
- When
- 24 Jul 2024
- ID
WTA-N-2024-0025- Where
- United Kingdom
- Who
- University of Oxford; Ilia Shumailov, Yarin Gal
- Topics
- Frontier models, Safety testing
Summary by wheresthe.ai, from Nature. Checked against the source on 9 Oct 2026.
Source copy: download the page as saved on 18 Oct 2026 (PDF). Document the source offers: document 1 (PDF). Also held by the Internet Archive.
Cite this story
WTA-N-2024-0025: “Nature paper finds models trained on AI-generated data degrade into ‘model collapse’.” wheresthe.ai, 24 Jul 2024. https://wheresthe.ai/d/oxford-model-collapse-nature-2024-07/ (source: Nature).
Spotted an error? Email [email protected].