Book coming in August 2026

Model collapse

What happens when AI trains on AI output: each generation learns from the last one’s mistakes and flattens toward a confident average. Chapter 8 calls it cannibalism, which is ruder and more accurate.