July 2024, article from Rice University.
A study by Rice University warns that training generative AI models on synthetic data can lead to “Model Autophagy Disorder” (MAD), where AI systems degrade over successive generations. This degradation occurs as models trained on AI-generated data develop artifacts, leading to outputs that are progressively marred by errors.
“The problems arise when this synthetic data training is, inevitably, repeated, forming a kind of a feedback loop — what we call an autophagous or ‘self-consuming’ loop.”
Let's start with the truth!
Support the Broken Science Initiative.
Subscribe today →
recent posts
Paradigm Shifts, Entrenched Thinking, and the Future of Medicine
A new paper proposes a single ratio, read off a fingerprick, that tells you where your metabolism is headed. It started in a cancer clinic. Its authors think it belongs in every MetFix affiliate.
Mark Sisson separates longevity fads from the habits that consistently improve health and performance
