A dataset very much is the source code. It's the part that gets turned into the program through an automated process (training is equivalent to compilation).