Abstract

Large language models (LLMs) have prompted renewed debate about the nature of cognition, reasoning, and consciousness. Their ability to generate fluent and contextually appropriate language invites comparisons with human intelligence, raising the question of whether such systems genuinely understand, reason, or possess minds. This paper examines these issues by analyzing LLM performance across several core dimensions of cognition. It argues that apparent competencies in common sense, logical reasoning, and theory of mind are best explained by large-scale statistical pattern recognition rather than by grounded, structured, or intentional processes. Similarly, while LLMs can simulate features associated with consciousness, they lack the defining characteristics of subjective experience, including embodiment, continuity, and first-person perspective. The paper also highlights the role of anthropomorphism in shaping interpretations of these systems, showing how linguistic fluency can lead to over-attribution of cognitive and mental properties. Overall, it suggests that LLMs are better understood as sophisticated simulacra of cognition, whose performance sheds light on both the structure of language and the limits of current computational models of mind.