Model Collapse

"AI models collapse when trained on recursively generated data" (Nature 631, 755–759, July 2024; preprint arXiv 2305.17493) — shows that recursive training on synthetic data causes irreversible collapse of distribution tails, degrading models generation over generation. The canonical citation in the synthetic-data debate. Oxford-led (Yarin Gal senior author; Shumailov then-Oxford), joint with Cambridge, Imperial, Toronto, and Edinburgh.