Model Collapse
Model collapse is the degradation of AI performance caused by training models on data generated by other AI models.
“Model collapse is the digital equivalent of genetic inbreeding, where synthetic data recursive loops destroy model variance and degrade performance.”
As the internet fills with AI generated content, future foundation models risk training on low quality synthetic data. This recursive loop destroys the minority variance and original reasoning capabilities found in human generated text.
Reverse Citations: Implemented & Audited Across Platform
Richard Ewing’s Research Thesis
High quality human data is becoming a premium asset. Organizations must secure proprietary, verified human data pipelines to avoid the commoditization and degradation associated with synthetic training sets.
Latest Publications & Research Activity
The Hidden Inflation of AI: Why Model Collapse Is a Business Risk
Your Claude API Bill Is Higher Than Your Revenue: Why Simple Python Tasks Are Blowing Up AI Costs
Why Redundant Requests Are Driving Hidden AI Costs
Frequently Asked Questions
Q:What causes model collapse?
Training AI models on datasets heavily polluted with AI generated content instead of original human data.
Inspectable Evidence Ledger
Classified evidence items supporting, extending, or refining this canonical research specification.
| Evidence Item | Publisher | Evidence Type | Strength | Role | Action |
|---|---|---|---|---|---|
| Hidden Inflation of AI | CIO.com | Analysis | ★★★★ | Origin | Inspect ↗ |
Recommended Citation
Ewing, R. (2026). "Model Collapse." Richard Ewing Research Canon. Available at: https://www.richardewing.io/concepts/model-collapse
@article{ewing_model_collapse,
author = {Ewing, Richard},
title = {Model Collapse},
journal = {Richard Ewing Research Canon},
year = {2026},
url = {https://www.richardewing.io/concepts/model-collapse}
}