Model Collapse
FrontierTechniques and Architectures
Model collapse is the gradual degradation in quality that happens when AI models are repeatedly trained on data generated by earlier AI models, rather than on original human-made data. Over successive generations, outputs become less accurate, less diverse and increasingly detached from real-world information.
In practice
Ask any AI vendor how much of their training data is genuinely original and how much was generated by another model. The answer is getting harder for suppliers to give, which makes the provenance of training data a claim to test rather than accept.