Summary:
-
Digital Self-Consumption
New research from Rice University and Stanford University warns that AI trained on synthetic data risks Model Autophagy Disorder (MAD), akin to mad cow disease. -
Degrading Quality
Without fresh, real-world data, AI-generated content declines in quality and diversity, leading to indistinguishable outputs. -
The Need for Freshness
Findings suggest that fresh human-generated data is essential to maintain AI model performance and prevent self-consuming loops.
Read more at: ScienceAlert