- Researchers at Rice and Stanford University found that training AI models with AI-generated content may lead to a degenerative condition they termed as Model Autophagy Disorder (MAD).
- The study suggests that continuous training on synthetic content reduces the information variety in the model leading to decreased quality and diversity in output.
- Although yet unreviewed, the study asserts the importance of fresh real data in training next-generation models and highlights implications for the AI-dependent internet infrastructure.
Menu
