A world model incorporating both vision and tactile data could be significantly more data-efficient than a vision-only model