My aim is to explicate a concept of valid inference that is congruent with the fact that valid inferences are used to support our assertions and thereby extend our knowledge. The most common definition of the validity of inferences in contemporary logic and philosophy is obviously not relevant when one has this aim in mind. Other proposed definitions within proof theoretic semantics or with a constructivist origin turn out to have serious failings in this respect. In this talk I shall reject some principles that those definitions were built on and consider some other principles that I think are essential for determining the wanted concept of valid inference. Some of them are based on Gentzen’s ideas behind his system of introduction and elimination rules, now developed in a different way than in the rejected definitions.