Model Collapse
Idea
When an AI system is trained on the output of other AIs, it loses the rare and unusual cases in its data and drifts toward a distorted average. The entry argues the same decay happens to civilizations that lose contact with ground truth, and to financial markets when one risk platform's $35T footprint produces uniform outputs.
The engine's definition — word for word
When AI trains on AI output, it loses the tails of its distribution — nuance, edge cases, minority data — and converges on a distorted mean. The same thing happens to civilizations that lose contact with ground truth. The same thing happens to financial markets when Aladdin’s $35T creates uniform outputs.
Walk this on the live map →