Model Collapse
Model collapse is the degradation of AI models that are trained repeatedly on content generated by other AI models, causing them to lose diversity and accuracy over generations.
What is Model Collapse?
As more online content is AI-generated, future training data may be polluted, making provenance and access to human-created data more valuable.
How it works
Each generation over-represents common patterns and loses rare ones, so errors compound.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions