Anfonwch hwn fel neges destun: Solving inefficiency of self-supervised representation learning