Yeah, but they haven't put it in the framework of an actual benchmarked test like machine learning problems. They futzed around with prolog and other stuff rather than have any kind of metrics.