Thomas Kuhn, answered from the texts and cited to the page.
Science progresses, and the pattern of progress is real: later mature science solves more puzzles, addresses a wider range of phenomena, gives better quantitative predictions in more domains than its predecessors. Progress is a feature of the historical record. What progress is not, in my account, is movement toward a fixed mind-independent reality whose shape later theories more accurately limn.
There is no Archimedean standpoint outside the sequence of paradigms from which the convergence of theories on a single truth could be verified, and the historical record itself does not support the pattern of approximation that the realist needs: the theoretical entities of the great superseded sciences (caloric, phlogiston, the luminiferous ether) did not approximate the entities of their successors; they were dropped.
The late position, articulated in 'The Trouble with the Historical Philosophy of Science' and developed in The Last Writings, is a Kantian one: what corresponds to world in the history of science is the variety of niches within which the practitioners of various specialties practice their trade, structured by the lexical taxonomies of those specialties. No sense can be made of the notion of reality as it has ordinarily functioned in philosophy of science — but this is not the denial that there is anything beyond mind and culture; it is the recognition that the philosophical concept of reality, as the realist deploys it, does not pick out a coherent target.1234
And third, if the notion of truth has a role to play in scientific development, which I shall elsewhere argue that it does, then truth cannot be anything quite like correspondence to reality. I am not suggesting, let me emphasize, that there is a reality which science fails to get at. My point is rather that no sense can be made of the notion of reality as it has ordinarily functioned in philosophy of science.The Road Since Structure, pp. 122–123
Like any other value, puzzle-solving ability proves equivocal in application. Two men who share it may nevertheless differ in the judgments they draw from its use. But the behavior of a community which makes it preeminent will be very different from that of one which does not. In the sciences, I believe, the high value accorded to puzzle-solving ability has the following consequences.The Structure of Scientific Revolutions, pp. 164–165
Of course, there are many differences among those who would consider themselves realists about science in general, and about natural kinds in particular; the same is true of those who see themselves as continuing the constructivist or the nominalist tradition. In rejecting both sides in both debates, Kuhn is certainly rejecting the uncomplicated versions of Editor's Introduction these positions.The Last Writings, pp. 38–39
In its time, each of these beliefs was stated or derived from statements that called upon parts of the lexicon differently structured from our own. Those statements are not translatable; they cannot individually be compared with statements we might make; our judgments of truth value cannot be applied to them.The Last Writings, pp. 129–130
The position became more explicit and more Kantian in the late papers, and I am willing to call this a development rather than merely an articulation of what was always there. Structure sometimes uses progress from language and sometimes progress toward language; the later writings consistently use progress from. The Kantian reading was not arbitrary; it was the framework in which the lexical-taxonomic articulation of incommensurability finally made sense. It is also where the gap between my position and the social constructivism of the Strong Programme is widest: Bloor and Latour push the from-not-toward framing into a denial of any extra-social constraint on knowledge, which I never accepted.