The paper reviews the leading attempts in the literature—contextualist, relativist, and interest-relative invariantist—to provide a satisfactory semantic theory of the ways in which the correctness conditions of ascriptions of knowledge seem to vary in tandem with variation in pragmatic factors—interests, stakes, and sentences—afflicting the parties concerned. It is argued that none of these attempts provides a satisfactory account of all the variability phenomena save at the cost of mis-predicting other aspects of what we count as acceptable uses of "knows" and its cognates. The paper concludes with the suggestion that the correct reaction to this finding is not a reversion to the traditional kind of invariantism about knowledge, but rather a move to a kind of deflationism about "knows" broadly comparable to the familiar deflationist account of "true".