This is such a great illustration to explain model collapse: AI models training on AI-generated data become worse over time. A challenge, because most companies training large models are running out of data (https://observer.com/2024/07/ai-training-data-crisis/) and increasingly rely on hybrid sets of original and synthetic data.

Nature article: https://www.nature.com/articles/s41586-024-07566-y

A.I. Companies Are Running Out of Training Data: Study

Given the current pace of companies working on improving A.I. models, developers could run out of data between 2026 to 2032.

Observer
@jbaert I call it "home sweet Ailabama". Inbreeding of Ai data. It was easy when it all started as all was human made. Now Ai is training itself on hallucinations that were spewed out by Ai. It'll just get worse with every new iteration as more garbage is used for training.
@jbaert What is the source for that illustration (can't immediately find it in the Observer or Nature article)?