AI Models Collapse When Trained on Recursively Generated Data

The peer-reviewed foundation for the model-collapse discussion. Repeated training on generated data can erase the tails of the original distribution until the model retains a thinner version of reality.

Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., and Gal, Y. 'AI Models Collapse When Trained on Recursively Generated Data.' Nature 631 (July 2024): 755-759. doi.org/10.1038/s41586-024-07566-y