> [!info]- meta > **Source**: [Original URL](https://readwise.io/reader/document_raw_content/305206518) > **Author**: [[Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Nicolas Papernot, Ross Anderson, Yarin Gal]] > **Full Title**: Ai Models Collapse When Trained on Recursively Generated Data > **Category**: #articles > > **Summary**: AI models can deteriorate when trained on data generated by other models, a phenomenon called "model collapse." This process leads to the loss of important information about the original data distribution, especially rare events, over generations. To maintain model performance, access to real human-produced data is crucial. > ## 🔦 Highlights & Commentary - We find that indiscriminate use of model-generated content in training causes irreversible defects in the resulting models, in which tails of the original content distribution disappear. We refer to this effect as ‘model collapse’ and show that it can occur in LLMs as well as in variational autoencoders (VAEs) and Gaussian mixture models (GMMs). ([View Highlight](https://read.readwise.io/read/01jvz6mgpx18ktgr1wmk9zqdzn))