The concept of probably approximately correct (PAC) learning has been foundational in computational learning theory.
I don't want to be as good as I was, I want to be better than I was.