Model collapse describes how AI models can get worse when they are trained again and again on text or images made by other AIs.
Example
An image model that only sees AI images produces ever more uniform pictures after several generations.
For pros
A study in Nature (2024) showed that rare content disappears over generations when real data is replaced by synthetic data. Mixing in real, human-made data reduces the effect. This is why proof of origin for training data is gaining value.
