Model Collapse

Definitions

1. A phenomenon where AI models trained on AI-generated data progressively degrade, losing diversity and accuracy over generations. The ouroboros of AI — models eating their own output until the output becomes meaningless.

   In plain English: When AI models trained on other AI's output gradually get worse and worse — like a game of telephone where each copy loses fidelity.

Related Terms

• Hallucination
• Training Data
• Synthetic Data
• synthetic training data
• slop
• guardrails
• Neural Scaling Laws
• Compute Optimal Training