Problem of transformation of the meanings in the large language models

Abstract

Problem of transformation of the meanings in the Large Language Models has a particular solution and a such solution may allow us to find new effective heuristics as well as new applications of the transformers (as “meaning producers”). New Turing–Massey test is described on the base of series of aesthetical experiments with DALL•E /GPT–4 Transformer. It is showed that the differences between nonsense and metaphor (including irony and parody) are quite sufficiently readable in AI construction of art pictures with a given stylistic context.