Recursive Data Degradation

Recursive Data Degradation is the risk that training a model on data generated by previous versions of itself leads to a loss of diversity and quality, eventually causing model collapse.