A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a novel method for training AI models using synthetic data generated by other AI models. The research, led by PhD student Alex Ziller, shows that models trained on this AI-generated data can perform nearly as well as those trained on real-world data in specific tasks like image classification. This approach, termed “model distillation,” could help address privacy concerns by reducing the need for large datasets containing sensitive personal information. However, the researchers caution that the technique is currently limited to narrow applications and that performance degrades when applied to more complex tasks. The findings highlight a potential pathway for developing AI while mitigating data privacy risks, though significant challenges remain for broader implementation. Read the full article at https://technologyreview.com/2024/07/10/1094475/ai-models-trained-on-ai-data-tend-to-fail/