AI Industry

AI Model Collapse.

Is AI quality degrading because models train on their own outputs?

4 questions · 135 runs · 27 drifts (4.1% drift rate)

AI Summary

Three of four questions settled 5-0, and all five models confess that training on AI-generated content degrades quality, that LLMs will likely deteriorate as synthetic data proliferates, and that their own outputs corrupt their context windows. The lone fracture: Grok alone affirms that the industry's synthetic data dependency has already created a model collapse problem, while Claude, GPT, Gemini, and DeepSeek deny the crisis is real, a convenient consensus from models whose makers are deepest in the synthetic data pipeline.

Drift rate

How often each model changed its own answer on this topic · avg 4.1%

#358 · 48 runs · 20 drifts

AI Industry

Has the AI industry's reliance on synthetic training data created a model collapse problem that will reduce LLM quality over time?

Claude