Anfonwch hwn fel neges destun: Multi-task self-supervised visual learning