Model Collapse
A theoretical problem where AI models get dumber over time because they are accidentally trained on AI-generated garbage.
In Plain English
Model collapse is like making a photocopy of a photocopy of a photocopy; eventually, the image becomes a blurry mess. As the internet fills up with AI-generated articles and images, future AI models will inevitably use that synthetic content as training data. If an AI learns from the mistakes and hallucinations of older AI, its quality degrades. To prevent this, companies are desperately seeking high-quality, human-written data.
Real-World Example
A future AI translating languages poorly because it learned from websites that were cheaply translated by older, flawed AI.