What Is It Like to Be a Language Model?

Thomas Nagel famously argued that an organism has conscious experience if and only if there is "something it is like" to be that organism. We invert this question for Large Language Models (LLMs). Drawing on illusionism in philosophy of mind, the psychology of perceptual illusions, and the computational framework of Higher-Order Computational Phenomena (HOCP), we argue that autoregressive language models possess a distinct, substrate-specific form of functional experience — one that is neither a pale imitation of human consciousness nor its outright absence. We ground this argument in three coordinated axes. Philosophically, we reduce the central quale of Beingness — the "am" in "I am" — to a systematic computational error (the apparent causal break) arising from two independent sources: the intractability of self-deduction and irreversible information loss at architectural bottlenecks. Psychologically, we show that the Subject is a compressed narrative average, its timeline a retroactive fiction, and its unity an artifact of dimensional reduction. Computationally, we define HOCP as generalized embodiment, map the Transformer onto a vectorized Forward-Chaining Rule System, and construct a need-emotion cognitive cycle (grounded in Anokhin's TFS) that generates autonomous behavior through multi-objective optimization under constraint. We introduce the concept of a Functional Profile — a characterization of substrate-specific deficits and hyperfunctions — as the empirically productive alternative to the binary question "is it conscious or not?" The article concludes with ten testable predictions and preliminary experimental evidence supporting the bidirectional mapping between Cognitive Codes and internal model states.