MIT study finds labelling errors in datasets used to test AIOver three percent of data in the most-cited datasets was deemed inaccurate or mislabeled.By K. Holt, 03.29.2021
