Cinema, neural networks, and the overcoming of metaphysics: from the automation of the image to the automation of the utterance

The light and shadows of digital reality. Living intelligence and its avatars

Abstract

The paper asks what the architecture of large language models can reveal about the structure of language itself. The point of departure is the thesis that Mikhail Kurtov develops based on the material of cinema: by automating a human capacity, in the experience of working with it, a technical apparatus lays bare the conditions of that very capacity. Whereas cinema, in Kurtov’s account, automated the production of images and, through the experience of boredom and reverie, exposed the limit of representation, the language model automates the production of utterances and, in the experience of “slop,” renders the limit of predication visible. The theoretical framework combines Heidegger’s distinction between assertion (Aussage) and saying (Sage), his analytic of boredom from the 1929/30 lecture course, and the concept of the paradoxical element from Deleuze’s

The Logic of Sense

. Drawing on recent corpus studies of machine-generated text, the study introduces the notion of the slop structure, which is a construction the selection of which is governed by the statistical form of the corpus rather than by the meaning to be expressed. The analysis demonstrates that slop is not a defect of generation but its transcendental condition in Deleuze’s sense: it brings into communication the series of the prompt and the series of statistical form, structurally corresponding to Deleuze’s “empty square.” Interaction with generated text unfolds as a successive refusal of the three dimensions of the proposition — denotation, manifestation, and signification — reproducing the three forms of boredom in Heidegger’s analytic. At the point of complete refusal, language, released from the obligation to assert, exhibits its own topological organization. The study concludes that, rather than introducing this organization into language, the language model gives it its first technical embodiment, thereby opening the possibility of an experience where human speech activity comes to light as orientation within a linguistic milieu.