Which of the following is the BEST method to mitigate the risk of AI model degradation?
Overreliance on synthetic or model-generated training data can cause model collapse, in which rare or distinctive data points disappear and outputs become increasingly homogenized. Including human-generated data helps retain real-world diversity and grounding in the training distribution. NIST AI 600-1, Generative AI Profile
Community Discussion