Ambient Dataloops: Generative Models for Dataset Refinement
A new method uses generative models to iteratively refine training datasets, potentially improving model quality but also raising concerns about data contamination and feedback loops.
A new method uses generative models to iteratively refine training datasets, potentially improving model quality but also raising concerns about data contamination and feedback loops.
Auto-scored: The paper introduces a dataset refinement technique with minor dual-use implications for model training and data integrity.