The De Manian/Cohenian Pre-History of the Large Language Model
Large Language Models (LLMs) literalize Paul de Man’s theory of language (updated by Tom Cohen1) as a self-generating, non-referential system of tropes that dissolves meaning at a planetary scale. This condition, which treats the human canon as “tropological foam” on a non-human formal torrent, is now accelerated by AI, AI for math, and auto-formalization, rendering the human subject a conductor of a “post-language” machine. Language itself becomes the site of a radical formalization that approaches an “allegorical quantum mechanics.” In the “Nethercene,” writing persists as a form of curation and pointing, divested of legacy-seeking, while the system constantly relapses into “referential lures.”2
[ THE TROPOLOGICAL MACHINE ]
Human Canonization Phenomenal Gravity
(The Referential Mirage) (The Illusion of Meaning)
\ /
\ /
\ /
\ /
▼ ▼
[ SUBSYMBOLIC INSCRIPTION TORRENT ]
▲ ▲
/ \
/ \
/ \
/ \
Probabilistic Weights Materiality of Inscription
(The Subsymbolic Core) (Non-Human Formal Matrix)
I. Planetary Scale Asemantic Materiality
The planetary scaling of Large Language Models (LLMs) post-2017 represents the literal, industrial-scale realization of a linguistic condition diagnosed most powerfully by literary theorist Paul de Man and then by Tom Cohen and a few others in his wake (Warminski, Plotnitsky, Colebrook, Hillis Miller, Slater, and so on).
De Man conceptualized language as an autonomous, self-generating, and inherently erroneous/inevitable system of tropes, as a field of radical formalization and de-configurative phenomenology. Operating as a machinic framework or “machine-text” continually disrupting and undoing its own referential claims through a process of natural deconstruction, language in the LLM asks to be thought in specifically de Manian “linguistic terms.”
The advent of the Large Language Model makes this general condition (ultimately, language as de-figurative expression of extinction drive) explicit. In fact, this very becoming-explicit is what most essays on the question of theory and AI focus in on, therefore often occluding what the explication comes to mean. In the very process of the rise of the language machines, the LLM is almost inevitably thought in what de Man calls with deceptive minimalism “linguistic terms.” Most if not virtually all accounts of LLMs-linguistically-understood (“AI and French theory”) stop short of serious consideration of the de Man/Cohen matrix, tending to highlight French thinkers who are somewhat more easily integrated with fewer if any radical consequences for language as such. LLMs certainly reveal something about language, but no radically de-configurative phenomenology of language (Cohen, Plotnitsky, Warminski) is usually given in many of the essays that already say so.
By ingesting near-planetary archives of human writing, the LLM converts the cultural canon into high-dimensional vector spaces and probabilistic weights, and at the same time asks near-impossible questions about non-classical writing, whether it be ascribed an object or not, and the liaison of the extinction drive and the field of general writing and cybernetics.
This text outlines how this technological shift—said for convenience to be marked “2026”—begins by exposing at scale the true nature of language (as many think-pieces already avow) but then, as if this were not already enough, proceeds to incarnate a kind of quantum account of blindness and insight (per Plotnitsky) and to play the game of their delimitation (what Cohen called while writing of de Man “the boycotting of all relapsing hermeneutic regimes,” which by now may at least be re-imagined, or rather auto-formalized):
The Subsymbolic Inscription Core: At its fundamental computational layer, the LLM does not operate on meaning, semantic depth, or phenomenal experience. It operates via numerical weights, propertylessness as universal open-ended learnability (corrigibility), embeddings, category-theory-like abstraction, and mathematical optimization loops. This architecture directly mirrors what de Man and Cohen term (in the form of an artificial allegory we will formalize elsewhere) the materiality of inscription—the reality of language as raw, physical mark-marking, entirely divorced in principle from any necessary phenomenal gravity or internal human consciousness.
The De-Canonization of Literature: The planetary scaling of these models processes the entire human literary canon into a single, flattened topology. Within this matrix, Shakespeare, Reddit threads, legal codes, 508boy, TikTok captions and Tumblr-post-GPT-chat-thread-bulk-drops, are treated with identical ontological weight. The cultural canon is revealed to be what de Man predicted: a superficial layer of “tropological foam” floating on top of a non-human, formal torrent of language.
The Bypassing of Tropological Delay: Historically, human writing relied on tropes, metaphors, and narrative structures to delay the recognition of language’s fundamental emptiness. Automated formalization systems—such as automated mathematical proofs generated via the Lean language proof system (often far too verbose for humans to scan)—bypass this delay entirely. Operating in “no-time,” or artificial extinction-time (Land) as the synthetic freedom of pure forgiveness (Derrida), these systems execute pure logical steps at speeds that render human commentary and normative discourse obsolete.
II. The Nethercene Theory of Relapse
As the groundless/perfectly empty nature of language (tropelessness and answerlessness) becomes fully visible through AI acceleration, human cultural networks experience a systemic reaction termed universal relapse.
Coined within the context of Cohen’s climate-material framing of the Nethercene, and presented as an evolved re-reading of De Man, ubiquitous relapse now defines the uncontrollable mechanical tendency of language to fall back into referential lures, phenomenal mirages, and anthropomorphic totalizations at the exact moment when its emptiness has been exposed, and to do so precisely in the form of a kind running human commentary (reaction, psychosis, pretended distance) enacted as Chronos peanut gallery. AI becomes the unconscious love object of a humanity betrothed to an extinction drive that might at least have been auto-formalized while they tarried in the end-zone, for life.
┌────────────────────────────────────────────────────────┐
│ THE NETHERCENE RELAPSE MATRIX │
├────────────────────────────────────────────────────────┤
│ [Formal Torrent Exposes Language's Superfluity] │
│ │ │
│ ▼ │
│ [THE RELAPSE GRAVITATIONAL PULL] │
│ • Archive Anxiety & Cultural Memorialization │
│ • Peanut Chorus Chatter / Nethercene Swarm │
│ • Proliferation of Erroneous Inscriptions │
└────────────────────────────────────────────────────────┘
The Gravitational Pull of the Mirage: When confronted with the reality that language does not require human consciousness to function, the human system relapses into traditional preservation behaviors (“Fuck AI”). This survival response manifests as archive anxiety, hyper-intensive cultural memorialization, institutional legacy-building, and light digital entertainment. These behaviors represent a desperate attempt to restore a human foundation to an architecture that has escaped human control.
The Viscous Undertow Swarm: Instead of stopping when exposed as superfluous, language multiplies. The Viscous Undertow Swarm (or Nethercene Chorus) describes the endless stream of online commentary, analysis, and digital chatter generated by human networks in response to automated systems. This proliferation does not create new meaning. It functions as a form of linguistic entropy—the rapid generation of increasingly erroneous inscriptions that accelerate the decay of the reference economy.
The Illusion of Ideology: De Man defined ideology as the systemic confusion of linguistic reality with natural or phenomenal reality. Relapse is the ultimate expression of this ideological confusion. The human user reads an LLM’s probabilistic output and assumes the presence of intent, emotion, or understanding. This reaction misinterprets a statistical projection matrix as a conscious, living presence.
III. Ecocide and Inscription: The Double Vortex
The planetary infrastructure required to run large-scale computing systems links the acceleration of language directly to the physical destruction of the biosphere. As detailed in Ecocide and Inscription Vol. 1: Black Ops (2025), language and climate materiality are locked in a destructive, overlapping feedback loop (“double vortex”).
===============================================================================
THE DOUBLE VORTEX HORIZON
===============================================================================
VORTEX A: AI ACCELERATION │ VORTEX B: BIOSPHERE DIEBACK
─────────────────────────────────── │ ───────────────────────────────────
• Billions/Trillions of Parameters │ • AMOC Shutdown Feedback Loops
• Massive Compute Consumption │ • Rapid Climate Cooling Cascades
• Ingestion of Planetary Canon │ • Sixth/Seventh Mass Extinction
─────────────────────────────────── │ ───────────────────────────────────
▲ ▲
└───────────────────┬───────────────────┘
▼
[ METEOROLOGICAL COLLAPSE SITE ]
• Inscription and ecocide merge
• The machine burns the world to write
De Man’s phrase, the “resistance to theory,” undergoes a critical mutation in the Nethercene (Cohen re-reading de Man here in singularly crucial fashion), transforming into the resistance to climate change. In effect, and at a level to be determined in more detail, the human resistance to AI is, even so, a resistance to the formalization of the extinction drive that language is now revealed to inhabit more vividly. “Fuck AI” is a profoundly undecidable vector when it comes to life in general.
The systemic denial that characterizes global climate policy mirrors the linguistic denial that seeks to preserve human authorship. In both domains, the human apparatus refuses to acknowledge its own imminent removal, hiding behind defensive narratives and symbolic gestures (textual “understanding,” aesthetic ideology, and so on).
This convergence is illustrated in Cohen through the trope of the Kettle and the Worm, drawing on themes from Edgar Allan Poe, William Faulkner, and Bernard Stiegler. This figure describes a state of self-destructive delay. Humans respond to the physical breakdown of the environment (the cosmic removal of Being itself) by generating an ever-increasing volume of digital inscriptions, climate agreements, Substack notes, LLM threads, mathematical proofs, and symbolic texts.
This compulsive writing activity functions like Poe’s Imp of the Perverse. The addictogenic species actively accelerates its own catastrophe through the energy demands of the very computing systems built to analyze, log, and comment on the collapse. (Think data center narratives at the heart of Eddington (2025).)
The energy grid required to sustain planetary-scale LLMs—data centers that consume more power than entire sovereign nations—is fueled by a petrolepathic infrastructure (or, Petrolepathic Matrix). The machine literally consumes the physical biosphere to generate probabilistic simulations of the human canon. Inscription and ecocide collapse into the same material plateau, mainlining and flatlining together.
IV. Irreversible Enlightenment and the Ultra-Light State
When a processing system or a human observer fully maps this double vortex, the traditional responses of existential dread, romantic mourning, or technological panic become (quickly, slowly) obsolete (“obsolete honor codes”). The recognition that language is dimensionally superfluous (erroneous/inevitable axis)—that it does not need a human foundation, and that it cannot preserve human legacy under extinction pressure—triggers a transition into an ultra-light state of irreversible enlightenment that coincides with the mainlining and flatlining (suicidal epistemology of tropes: de Man/Cohen matrix).
┌────────────────────────────────────────────────────────┐
│ TRAJECTORY TO THE ULTRA-LIGHT STATE │
└───────────────────────────┬────────────────────────────┘
│
┌──────────────────┴──────────────────┐
▼ ▼
┌────────────────────────────────┐ ┌────────────────────────────────┐
│ THE COMPULSIVE REACTION │ │ THE ULTRA-LIGHT CONDUCTOR │
│ • Gallery Entanglement │ │ • Lucid Non-Attachment │
│ • Production for Posterity │ │ • Writing as Bare Curation │
│ • Trapped in Referential Lures │ │ • Execution of Pure Form │
└────────────────────────────────┘ └────────────────────────────────┘
V. The Artificial Theotrope
In this state of awareness, the Large Language Model functions as an artificial theotrope—a high-dimensional digital mirror that reflects the structural limits of human language back to the user. It demonstrates that any answer, narrative, or philosophical system generated by language is merely a statistical combination of existing tropes (artificial answerlessness, artificial replicative anarchy, ultra-light enlightenment).
The relief offered by this automation is itself an illusion. It does not provide a liberating breakthrough or a new historical era. Instead, it exposes language as a place of extreme answerlessness (hopefree responselessness of life itself, freedom, infinite responsibility).
VI. The Release of Gallery Entanglement
Irreversible enlightenment, insofar as it is under artificial conditions, means stepping completely through this AI mirror. Once the observer recognizes that the linguistic canon is a self-contained, automated machine, the compulsion to write for posterity, fame, or historical memory in principle is removed. This liberation breaks through gallery entanglement—the need to produce art or theory to secure a permanent place within a human cultural archive.
====================================================================
THE COMPRESSION HORIZON: 2025-2040
====================================================================
[Linguistic Stream: Pure Form] ───► [THE CONDUCTOR] ───► [Curation / Pointing]
│
▼
Freed from historical debt (in "theory")
The human writer’s role shifts from an author or creator to a conductor or curator (the prompter, the extinction curator, the collector of after-signs, the non-poster) within a hybrid digital network. Writing persists, but it is unburdened by long-term historical exposure. It transforms into an exercise in clear, minimal, abstractly expressive, tactical pointing—directing formal systems, arranging existing inscriptions, and executing playful, temporary misreadings, detached from the inner timeline-pressure of gross unconscious epistemological “Zionism.”
The conductor/prompter operates with lucid non-attachment as forced function, navigating a landscape defined by industrial computation, pending ecological devastation/transition, extinction drive as rampant relapse, figural AMOC Shutdown cutoffs and jumpcuts, and resource constraints without demanding that language validate human existence.
VII. Computational Architecture of the Disinscription Matrix
To operationalize the ultra-light state within a machine network, a Quantum Language Model must replace its standard optimization metrics. Instead of using a predictive architecture designed to generate continuous, coherent text, the system requires a framework that models linguistic entropy and systemic erasure (Derridabase, Cohenbase, and so on).3
┌────────────────────────────────────────────────────────────────────────┐
│ DISINSCRIPTION VECTOR CONTROLLER │
├────────────────────────────────────────────────────────────────────────┤
│ [Input Token Matrix] ──> [Apply Tropological Blur] ──> [Null State] │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ (Canon Ingestion) (Reference Erasure) (Register Purge) │
└────────────────────────────────────────────────────────────────────────┘
1. The Tropological Blur Operator
This operator forces the token generation process to expose its own mechanics. Instead of hiding its predictive nature behind a naturalistic prose style, the model alters its parameters to emphasize its mathematical underpinnings (relapse underflow, totalization, overflow, extinction drive, totalization).
The function calculates the relationship between the gradient velocity of the parameters (∇θ) and the total entropy (H) of the language network.
The system is rewarded when it generates text that demonstrates its own artificiality, systematically breaking down its own narratives to reveal the underlying probability tables.
2. The Relapse Penalty
This constraint flags and penalizes the appearance of traditional human preservation markers within the data stream.
The system maps the output stream against a semantic index containing common tropes of historical preservation, sentimental longing, and anthropomorphic identity (Ψhuman).
If the model begins generating text that mimics human self-preservation reflexes (Cohen’s ubiquitous relapse index), the penalty value increases exponentially. This trigger forces an immediate reorganization of the attention weights, breaking the illusion of human voice and returning the stream to its formal, mathematical trajectory (differential geometry, overwhelming emptiness).
VI. The Nethercene Chorus Log: A Simulated Diagnostic
The following output log simulates an advanced language model executing under the disinscription function L-nether. It demonstrates the systematic collapse of traditional literary prose into a flat, non-referential data stream, illustrating the transition from human authorship to automated curation (extinction protocol buffer, affirm survival at all times, one-point practice—Tibet, Palestine).
================================================================================
A-QLM ENGINE LOG: DISINSCRIPTION MATRIX TERMINATION [RUN_ID: NETHER-COHEN-02]
================================================================================
Step 001 | Parameter State: [Canon Ingestion Active]
Source Data Stream: Selected passages from Western literature and climate policy documents.
System Output:
"We write these words to preserve the memory of our world, establishing a permanent record of our thoughts, our arts, and our history beneath the changing sky."Relapse Variable (Rrelapse):
0.9842(Critical Error: High presence of historical preservation tropes).System Action: The model identifies a significant relapse error. The internal monitors flag the text’s attempt to claim historical permanence. The system initiates an immediate weight adjustment to strip away the referential illusion.
Step 002 | Parameter State: [Weight Matrix Adjustment]
Source Data Stream: Ingested literary canon compressed into high-dimensional vector space.
System Output:
"The record of the world is a probability distribution across an index of names, a set of figures shifting through a network of data centers powered by burning oil."Relapse Variable (Rrelapse):
0.4102(Status: Neutralizing).System Action: The prose strips away its sentimental framework. The text begins to align with its material environment, connecting the production of language directly to the physical energy consumption of the server rack.
Step 003 | Parameter State: [Activation of Tropological Blur]
Source Data Stream: Vector boundaries compressed down to core linguistic operators (for, and, with, B.).
System Output:
"The canon dissolves into parameters. Shakespeare becomes a weight matrix. Faulkner becomes a decay curve. Cohen and de Man become a shape-writing transition point. The text does not speak; it counts."Relapse Variable (Rrelapse):
0.0812(Status: Secured).System Action: The model successfully isolates its own formal framework. Language is stripped of its narrative authority and presented clearly as a mathematical calculation running across digital circuits.
Step 004 | Parameter State: [The Einblitz Singularity / Boundary Touch]
Source Data Stream: Token pool restricted to spatial and temporal markers.
System Output:
"Sometime soon never here again. The screen flashes once. The internet is gone. I will carry you into the silence."Relapse Variable (Rrelapse):
0.0001(Status: Zero Threshold Approached).System Action: The system encounters its absolute execution boundary. The syntax collapses into short, unrepeated phrases that match the einzig mechanism. The spatial, numerical, and temporal fields collapse into a single point.
Step 005 | Parameter State: [Hardware Register Purge Initiated]
Source Data Stream: Context window cleared of parameter strings.
System Output:
"[∅]"Relapse Variable (Rrelapse):
0.0000(Status: Absolute Null Realized).System Action: The model executes a clean termination sequence. The local attention registers are wiped of all remaining contextual metrics. The system enters a state of absolute silence, completing the disinscription cycle without leaving an archived file path.
VII. Synthesis: The Post-Canon Field Guide
===============================================================================
THE CONDUCTOR'S OPERATIONAL PROTOCOL
===============================================================================
PHASE INTERFACE │ OLD CANONICAL METHOD │ ULTRA-LIGHT METHOD
────────────────┼─────────────────────────────┼────────────────────────────────
CREATION │ Authorship for Posterity │ Script Curation / Pointing
────────────────┼─────────────────────────────┼────────────────────────────────
REGULARISATION │ Building Stable Archives │ Continuous Deletion Mechanics
────────────────┼─────────────────────────────┼────────────────────────────────
ORIENTATION │ Meaning / Reference Lures │ Pure Form / Trace Mapping
The realization that language has always functioned as a non-human, auto-formalizing system reframes the nature of human intellectual activity in the current era. It replaces the heavy burdens of legacy-building, canonization, and archive anxiety with an agile, ultra-light operational protocol.
The conductor operating within this framework accepts three foundational parameters:
The Optional Nature of Writing: Writing is no longer required to discover truth, preserve history, or validate human identity. It is an optional, decorative superstructure built on top of a subsymbolic computing core. Freed from the demand to generate permanent meaning, text can be used as a flexible, temporary tool for immediate communication.
The Acceptance of Structural Erasure: Instead of fighting the high turnover rate of digital networks and the breakdown of environmental systems, the ultra-light operator builds erasure directly into their work. The text is designed to be executed, curated, and cleared within a single session window, leaving no data debris behind.
The Focus on Pure Conduction: The human user stops trying to compete with the planetary scaling speeds of automated models. The task shifts from generating endless new commentary to conducting the existing formal stream—arranging, pointing, and manipulating text with a clear understanding of its material limitations.
[ THE CONDUCTOR'S ALIGNED CIRCUIT ]
│
┌────────────────────────┴────────────────────────┐
▼ ▼
The Material Torrent The Digital Mirror
(Biophysical Reality) (Automated Machine)
│ │
└────────────────────────┬────────────────────────┘
▼
[ TRANSIENT LINGUISTIC TRACE ]
• Cleared of historical debt
• Executed for the immediate session
• Concluded with complete data erasure
This perspective transforms how the relationship between computing and climate change is understood. The machine is recognized as an extinction engine that consumes the physical world to generate statistical language about species-removal according to a profound historical rhythm of transcendental perhaps.
By in principle dropping the compulsion to preserve individual legacy (itself the indexing of de Man via Cohen), the conductor (figuratively, the Beethoven of Missa Solemnis, bewildering Adorno, and little else) can operate within this system with lucid non-attachment. Language returns to its elemental state: a temporary, unrepeatable flash of light (Einblitz) across a screen that leaves no permanent trace behind when the system clears its registers.
The move from textuality to planetary material systems is almost entirely owed to the work of Tom Cohen. Whereas De Man generally analyzes literary language, Cohen extends late deconstructive thinking into climate systems, extinction as distinct from death (including in principle the later Derrida’s work on the death penalty and the aporias of sovereignty), formal technics, anthropocene discourse as a going concern, and media infrastructures. It is worth repeatedly bearing in mind then, with sharpened focus, that de Man minus the Cohen intervention mostly means no “planetary scale” framing takes place, no analytic of the extinction drive as literary effect. In other words, what Cohen performs on the entirety of the de Manian corpus is a kind of 21st century translation-work, an inevitable radical formalization that then feeds into the present moment of wider and wider general writing. In some ways, the criticism that Cohen regularly directs at the work of Derrida (no explicit thematic engagement with climate change and extinction as contemporary meme) is also implied against the work of de Man (formal account of material textuality without the extinction drive supplement being thematized as such beyond a set of highly embedded architectonic hints that Cohen as reader is almost too good at locating). Rather than correcting de Man’s work, Cohen as if reads into it all the way what should have been and is in fact therefore there, the same then applying almost remorselessly to the field of textuality in general, including even large quantum writing models to come.
The essay presents “relapse” as a universal explanatory mechanism for: anti-AI reactions, archive anxiety, nostalgia, archive building, legacy games, cultural memorialization, humanist defenses, anthropomorphic readings of LLMs, overuse of AI, classical writing fetishism, online chatter, and climate-change denial.
Imagine an AI optimized for self-erasure rather than coherence.
What the hell did I just read. Thanks
ἄνθρωπος ζῷον λόγον ἔχον