MIT study finds 70% of AI training datasets lack licensing info, 50% have errors. New tool aims to improve transparency and dataset selection.
Filter all Transparency posts in the archive →