Does anyone know if we're near the theoretical limit of how much we can improve these models by giving them more data? Or should we expect similar levels of imp

Does anyone know if we're near the theoretical limit of how much we can improve these models by giving them more data? Or should we expect similar levels of improvements in next iterations?