Datasets - properly curated and labeled - remain a scarce resource. What can be done about this?