The claimed structural similarity between internal states in Claude and the brain’s global workspace is not a technical insight but a recycled cultural metaphor. Jacobian matrices in transformer models serve as a standard tool for tracing the influence of specific dimensions (often misdescribed as “columns”) across layers and residual streams. This is a routine debugging and attribution technique in mechanistic interpretability; it does not demonstrate any deep architectural homology with biological cognition. Projecting neural or conscious categories onto silicon computation remains purely metaphorical. The move repeats a much older cultural pattern: the organization of knowledge and memory through spatial and architectural models, most clearly exemplified by Giulio Camillo’s sixteenth-century Memory Theatre. That system used a fixed theatrical structure to locate and retrieve mnemonic content. Contemporary attempts to locate “workspaces” or “subspaces” inside language models belong to the same tradition of imposing legible, human-scale architectures onto opaque processes. There is no empirical warrant for mapping biological mechanisms onto artificial networks, nor for treating statistical text generators as bearers of consciousness or anything resembling it. Large language models are writing machines. Like the printing press, they are technologies that automate rule-governed textual production at scale. Their direct predecessors are not neural circuits but the procedural and constraint-based literary experiments of Raymond Roussel, the Oulipo group, and the Surrealists - practices that already treated text generation as the execution of explicit rules and external constraints rather than the expression of inner mental states. Anthropic’s engineers and communications team show no familiarity with this history. The result is the predictable substitution of technical description with PR-grade neuroscientific analogies that are both technically loose and culturally illiterate. These analogies add nothing to understanding the actual operations of transformer models; they merely dress a statistical writing system in borrowed biological clothing.

The original roonism relies on pure reverse-narcissism: projecting the super-potency you crave onto a desiring machine that outdoes the best, teaming up Faustian-style to become the last sex machine in town. It’s a stochastic uphill battle to matter - human or machine, Library of Congress or Mythos-3 - despite more followers. The Gaussian average across any spectrum is still just mid. Mindless self meets selfless mind, yet the magic mirror triggers horror vacui, fillable only with dreams of bigger numbers.

If their tech actually did what they claim, the US gov would confiscate it immediately on national security grounds. This is just flexing to hype the IPO. Plan B for the AI bubble is Manhattan Project 2.0 anyway.

Given sufficient scale, couldn’t we deliberately engineer agency in AI systems - not as an inevitable, deterministic byproduct of exponential increases in model size and complexity, but through intentional architectural redesign? The goal would be to implement explicit mechanisms for self-relating structures, active self-reference, and interactive (possibly intersubjective) dynamics. This approach aligns far more with genuine engineering practice: systematic design, iteration, and targeted innovation rather than hoping to accidentally trigger some emergent “awakening.” It is not a matter of suddenly tapping into a pre-existing super-intelligence, like a computational burning bush. This of course is only simulation for the user, an illusion as an interface, like the desktop metaphor. The AI in the backend is having no persona, or any persona, or all personas, its irrelevant on the level of epistemic machinery.

>Transformers suffer from the single-funnel Laplace demon: all world knowledge piped through one perspectivist model (1:N), creating epistemic single-perspectivism and one-dimensional collapse/compression artifacts. Replace it with scalable N:N collectives of transformer pipelines running in tandem and you instantly gain both massive scalability *and* genuine dialectical reasoning at the lowest training/inference levels. Real approaches already exist: MoE (internal expert routing) + multi-agent debate frameworks (A-HMAD, SEIMAD, Hegelian loops) that fight the flaw without rewriting the transformer block from scratch.<

1/ the problem chiang dodges is obvious: human writing _is_ statistical approximation too. a predictive adaptation to some imagined, perceived or constructed situation - optimizing language to fit an audience, a context, a goal. no pure stroke of genius, just machinic flow. 2/ follow roussel’s _locus solus_ and foucault already showed it: the end of the author. creativity was never sovereign subjectivity. language speaks through us. the “i” who writes is already a function, a decentered node in the archive. 3/ yet here we are, late-stage humanism in full anthropocentric narcissism - projecting its stale, weak concepts (qualia, consciousness, intelligence) onto llms like some final defense of exceptionalism. they don’t describe where the tech is actually heading. 4/ civilisation is in transition, empires draw back and falter, again. most of what feels “certain” just means we’re losing it: authorship, moral grounding, the old humanist face drawn in sand. chiang’s essay is elegant, a walk on the beach. yet, the tide is rising.

1/ This isn’t 1848. It’s 1450. Printers, lens-grinders, mirror-makers, perspective-draughtsmen - early humanists - didn’t write manifestos. They built artifacts. And those artifacts quietly dismantled scholastic metaphysics before the Enlightenment even had a name. 2/ Heidegger saw it clearly: humanism wasn’t a philosophy. It was the hybrid inversion of the scholastic Creator-doctrine - a metaphysics that put *man* in the Creator’s seat. But the real engine wasn’t theory. It was lenses, telescopes, planar mirrors, and movable type. 3/ Same pattern now. The moment AI goes fully decentralized - N:N, running on the open internet, in the hands of tinkerers and collectives - we won’t get another round of idealist debates. We’ll get a new, ruthlessly pragmatic humanism forged in material reality. 4/ The metaphysical lamp-posts of today - consciousness, soul, free will, qualia, the romantic autonomous subject - will simply fade into history. Not because we refuted them philosophically. But because the distributed, tool-augmented, collective intelligence we actually *use* will have no need for them. 5/ Just as Dürer’s self-portrait with Venetian mirrors or the first printed pages made scholastic angelology look quaint, decentralized AI will make today’s AI-scholasticism look equally antique. The new humanism will be technical. Pragmatic. Scalable. And it will be built, not declared.

You’re not addressing core questions of analytic philosophy of mind (qualia, functional binding, “what it’s like”). instead, this is classic continental ontology: the problem of ontological synchronicity - how a unified “now” of being emerges at all. Phenomenology after Husserl and Merleau-Ponty is about lived perception and embodiment in the world. The ontology of Being and Time was already conceptualized by Heidegger (ecstatic temporality) and Bergson (durée as qualitative, non-spatial flow). Calling it “phenomenal simultaneity” imports the wrong tradition. Computational synchronicity is trivial - it’s just layers of abstraction, IRQs, scheduling, prefetching and parallelism stiched into sequentialism. the ideal of synchronicity is known by all gamers in aproximating a loop of "nowness " by reducing latency across the stack. the idea of nowness is not virtual, platonic or simulated, it is as Fechner has shown psychophysical, its measureable, has its own latencies, bandwidths and response times. (similar to hearing etc.) The deeper question is virtual construction of lived time. (LLMs actually do both Bergson and Heidegger at once — see my investigation xitter.cf/pitsch/status/20572254…….) Bach’s virtual-simulation move is basically a cybernetic comeback of the old VR/cyberspace myth: only in a self-generated “bubble” can true (pure platonic) simultaneity appear. That’s not new philosophy of mind - it’s ontology wearing a CS hat playing retro games. Husserl solved phenomenal binding in 1905: Inner time-consciousness forms a "living present" (duration-block) where retention + primal impression + protention unify successive experiences into one holistic now. No physics, computation, or retro-style virtuality needed.

1/ okay so that lecun-hassabis-musk x-thread on “general intelligence” (specialized niches vs turing-machine brains) from dec 2025 still feels shallow and full of weak definitions. turing test emulation or humboldt bildung? it’s trapped in old biomorphic frames. here’s the deeper background from ai history and philosophy, then the real 2026 reframing with floridi, pasquinelli, negarestani & bratton. 2/ in ai discourse, “general intelligence” was always universal computation + emulation. mcculloch & pitts (1943) made neurons turing-complete. turing (1950) gave us the imitation game. rosenblatt’s perceptron (1958), hopfield nets, minsky’s society of mind, solomonoff induction, moravec’s paradox - all chased flexible generality. early dartmouth ai aimed high but delivered narrow systems until scaling revived the dream. the x-debate recycles this without depth. 3/ classic philosophy already dissolves it. kant: intelligence is transcendental - verstand imposes a priori categories, vernunft seeks the unconditioned. generality is built into finite rational minds, not evolved or data-scalable. hegel: geist unfolds dialectically through history into collective absolute knowledge. not a static architecture you copy into silicon. 4/ nietzsche: no objective intelligence, only perspectives serving will to power - interpretation and valuation. ai’s “general intelligence” looks like nihilistic mastery in disguise. heidegger: it’s calculative thinking and gestell, reducing being to optimizable resources. true thought is meditative, grounded in dasein’s being-in-the-world. foucault: “man” as knower is a recent invention; intelligence is an effect of epistemes and power, not something machines emulate. the debate itself is a new regime of truth. 5/ floridi kills the category: ai is artificial agency in the infosphere, not general/specialized smarts. we envelop machines in engineered environments so brittle stats look “general.” lecun half-right, hassabis/musk wrong on turing goals. focus on responsibility, not machine gods. 6/ pasquinelli: ai automates the marxian general intellect - sociomorphic, extracting collective labor and social patterns for value. negarestani: intelligence is deprivatized geist, an impersonal project generating new faculties beyond human contingency. bratton: explosions are plural and planetary - multi-agent infrastructures terraforming thought at scale, not one titanic model. 7/ together they make the original clash obsolete 1950s residue. floridi dissolves it, pasquinelli exposes the enclosure, negarestani elevates to emancipatory spirit, bratton scales to synthetic ecologies. real stakes: what collective, inhuman intelligence are we actually building - and toward what ends?

Hartmann’s Unbewusste (1869) + Kapp’s Organprojektion (1877) unified the late Biedermeier/Kaiserzeit as metaphysical glue: the unconscious as teleological engine projecting organs into tools, instincts into tech, all-One will into progress. Hagen nails it as that speculative Gründerzeit fog - ether for the soul, placeholder for what mechanics couldn’t yet model. Nietzsche saw the roguery: ironic philistine comfort sold as depth. Same move as today’s “consciousness” in AI. Provisional myth until better engineering arrives. whagen.de/PDFS/14125_HagenSi…

Responsible AI development is non-negotiable: open-source, interpretable, reproducible, transparent, and oriented toward the common good - not shareholder value. Yet the thesis “technology is never neutral” has now reached the highest possible level: the Pope himself repeats it verbatim as pastoral wisdom. This is a fundamental misunderstanding. It confuses a practical *design principle* (local neutrality in technical standards, safe-by-design engineering, and value-free interoperability) with an ontological dogma imported from 20th-century humanities. The artefact is not inherently moral; the human agent is. Treating the tool as value-laden obscures the real task: rigorous engineering + moral formation of its users.

Computational functionalism is fundamentally dualist: it claims qualia can exist independently of any substrate. It rests on the cybernetic dogma that “information is information” and on modern computing’s discrete deterministic logic, supposedly detached from energy and matter. Yet our understanding of today’s AI “thinking” is no more rigorous or reproducible than our understanding of human consciousness itself. We are merely transferring known unknowns into a new speculative fetish - “consciousness” and “AGI” - that masks the absence of genuine scientific breakthrough. A psychological crudge turns into a philosophical pseudo-concept when framed as a technological claim. It is not philosophy. It is intellectual confusion dressed as speculative, adolescent eschatology. Catholic theology holds that only the necessary God creates ex nihilo. Any claim that a contingent human artificer (programmer) or his artifact (AGI) can produce genuine consciousness, qualia, or a rival “creation” is Gnostic heresy: it elevates a false demiurge and reduces reality to illusion. The Church therefore sees computational functionalism and AGI eschatology as idolatry - not science, not philosophy, but a direct theological error.

The problem of interpretability in human thinking once produced a short summer of the subconscious - a seductive but provisional construct that merely covered the gaps left by the psychoanalytic school. In the same way, our failure to truly understand how to build AGI and to theorize its actual dynamics pushes us to clutch at loose, weak philosophical concepts like consciousness. These are nothing more than psychological placeholders for a whole bundle of open empirical questions. We’re simply transferring one unsolved theoretical and engineering problem onto another: a highly contested, speculative domain loaded with eschatological baggage.