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 shows that this ‘model-generated data’ can be used to effectively train new, smaller models, potentially reducing reliance on large, manually curated datasets. The technique involves using a large, pre-trained model to create a diverse set of synthetic examples, which are then used to train a more compact, specialized model. This approach could lower the computational cost and data collection burden for developing AI in specialized domains. The researchers caution that the quality and diversity of the synthetic data are critical to success, and further work is needed to scale the method and ensure it avoids amplifying biases present in the original training data. Read the full article at: https://technologyreview.com/2024/07/15/1094565/ai-training-synthetic-data-mit-study