Large Language Models and the problem of algorithmic communicability

Abstract

Large Language Models (LLMs) not only automate textual production but also reconfigure the very regimes of communicability in digital culture. By generating coherent and contextually appropriate texts, these models simulate comprehension and intentionality, challenging the boundaries between human cognition and statistical prediction. Drawing on advances in machine learning and natural language processing, this article offers a critical reading of LLMs through three axes: the technical foundations that sustain their operation; the sociocultural shifts they provoke in the concepts of language, authorship, and intelligence; and the ethical dilemmas associated with algorithmic bias, hallucinations, and system opacity. Far from being neutral tools, LLMs operate as discursive infrastructures that condense power disputes, reproduce epistemic inequalities, and destabilize classical models of subjectivity. The analysis suggests that, although immersed in the logic of cognitive extractivism, these models also open cracks for practices of resistance and symbolic reinvention.