AI Cannibalism: How Models Are Eating Themselves Into Collapse

Key Takeaways AI cannibalism refers to training language models on AI-generated data instead of human-produced content — creating a feedback loop that degrades quality over time. Researchers have formally shown this leads to model collapse: an irreversible degradation where outputs become homogenous, inaccurate, and eventually nonsensical. The fix isn’t simple, but strategies like RAG, rigorous […]