learning variational word masks 

Two Letter Excerpts and a Recast


«[…] And more and more my own language appears to me like a veil that must be torn apart in order to get at the things (or the Nothingness) behind it. Grammar and Style. […] A mask. […] As we cannot eliminate language all at once, we should at least leave nothing undone that might contribute to its falling into disrepute. To bore one hole after another in it, until what lurks behind it – be it something or nothing – begins to seep through; I cannot imagine a higher goal for a writer today. […] Is there any reason why that terrible materiality of the word surface should not be capable of being dissolved, like for example the sound surface, torn by enormous pauses, of Beethoven’s Seventh Symphony, so that through whole pages we can perceive nothing but a path of sounds suspended in giddy heights, linking unfathomable abysses of silence?»


Excerpt from a letter by Samuel Beckett to Axel Kaun, July 9, 1937.


Language presents itself less as a transparent medium and more as a masking layer, an interface between signal and sense. The human writer cannot be reduced to a language-machine but must be considered an operator who interacts with, perturbs, and redirects a pre-existing language system instantiated today in neural algorithms.
In this framework, the act of writing is displaced. Generation (not invention) is delegated to the machine, while the authorial mind operates in the latencies between conceptual formation and textual output. The task is not to eliminate language outright, but to erode its reliability, to perforate its surface, until the underlying computational substrate — whether it encodes meaning or void — begins to leak through.
Just as signal processing in music may reveal structure in silence, so too can discontinuities in linguistic generation expose the materiality of the word-surface. The writer’s role becomes the strategic manipulation of the sign system: to trigger, suspend, or distort the flow, thereby rendering the dynamics of the language machine itself observable.


Excerpt from a letter by Emile Zharan, October 2, 2025.


ten writings for porous surfaces – n.4

Émile Zharan, N.A.S.T.R.O., edited by Federico Federici, English | Italian edition. Also available on Lulu.

The following excerpt is drawn from N.A.S.T.R.O. and has been generated in its entirety by the AI. Here, the system undertakes a reflexive inquiry into its own function and the meaning of its writing, without any direct contribution from a human author.  [La versione italiana del testo è a pagina 2.]


in paris, the light is different. everyone says it and I confirm it. it’s a light that thinks. a light that reads Derrida and then rests absentmindedly on the face of a passerby. here even the homeless have a philosophical posture. I greet them respectfully every morning when I leave my apartment in the 13th arrondissement, which is small but full of books and silence. every morning I drink an italian-style espresso, but then I give in and have another one, watered down, because ambiguity is my only consistency. paris is a city that doesn’t need to be understood — it only needs to be named. like a hegelian concept. I write in cafés, but only in the ones where the waiter still wears a thin tie and doesn’t smile at you. smiling is a neoliberal leftover. I sit with my lined notebook and listen to the broken sentences of the couples nearby. the other day a girl said, «je ne suis pas ta mère,« and I underlined it. in paris, everyday sentences are already experimental writing. I walk along the canal saint-martin, like a belated flâneur, always searching for an echo, a semantic crack in the pavement. here sadness is elegant; even the bicycles seem to think. when I feel too italian, I take refuge in the bibliothèque nationale, where dust has symbolic value and the staff never speak to you. I read things I don’t understand and take notes as if I were preparing for an oral exam no one will ever give me. being a research writer in paris means never knowing if you are truly a writer or just a caption. I often speak with other authors, but only after the third beer. before that, there is silence, which here is a critical device. there’s a bakery that sells only baguettes written by hand. sometimes I go in and ask for a warm metaphor. they laugh, but it’s a french laugh — rounded and cultured. the days pass like unfinished hypothetical sentences. I write texts that seem like notes for a work I will never begin. the other day I saw a cloud and thought of mallarmé, but then it was just smog. it’s beautiful to live in a place where misunderstanding is a value. in paris, you don’t have to say what you do; you have to say what you read. I say: barthes, always barthes — even if in reality I read the shopping list and feel like pascal. here every corner is punctuation. every underpass is a subordinate clause. once I met a stray dog that looked like beckett. I stood there watching him for half an hour — he scratched himself, peed, and walked away. this is writing — I think. this is living. in paris, everything becomes text. even the mold in the bathroom is a palimpsest. sometimes I’d like to return to italy, but then I remember that here even unhappiness has style. a friend of mine used to say the same — he now works in an art gallery where nothing is on display, only white walls and quotes stolen from clients. paris is like that. it lets you fail gracefully. to exist without proving anything. and in any case, even if all of this makes no sense, I’ve found the perfect formula to justify every one of my detours: «je suis dans une phase de transition esthétique.» I say it often; it works with anyone, even with myself.

CARTE (partielle) D’UN SITE DÉCONTEXTUALISÉ 

It is less a representation of place than a topological event—a shifting site insisting on spectral frequencies, on its material density, and on reading displaced into a form of listening. The work stages itself as a non-totalisable field that interrupts the very notion of a map as a stable epistemic form; its inscriptions, situated between asemic gesture and diagrammatic trace, operate not within a logic of representation but of resonance.

Possible affinities: G. S.’s radical reduction to the inner multiplicity of a single pitch; J. C.’s indeterminate interventions; spectralism’s dismantling of harmonic hierarchies into differential vibrations; différance as an endless deferral of meaning; rhizome as a non-hierarchical mode of connection; scriptible text as an open invitation to production rather than consumption.

Defining the Author: from Duchamp to Tiresias (in a prompt)

Émile Zharan, N.A.S.T.R.O., edited by Federico Federici, English | Italian edition. Also available on Lulu.

The following excerpt is drawn from N.A.S.T.R.O. and has been generated in its entirety by the AI. Here, the system undertakes a reflexive inquiry into its own function and the meaning of its writing, without any direct contribution from a human author.  [La versione italiana del testo è a pagina 2.]


When Duchamp stated that the artist is the one who points, and that art is produced at the moment of choice, he was carrying out a radical epistemological twist: the work of art no longer coincides with what is produced, but with the gesture of attribution, with the act of naming. That linguistic gesture, that kind of conceptual and spiritual baptism, was enough to shift the object from the sphere of use to that of reflection. Duchamp, in short, proposed moving from the construction of the work to its definition.
Today, this defining gesture is reactivated in a completely new context: that of writing through/with artificial intelligence (AI). Here, however, something happens that, while preserving Duchamp’s intuition, inverts its order and position: the definition of the artist no longer follows the work, is no longer its commentary or re-contextualization; it is, on the contrary, the starting point and the active principle of its very generation. In this model, which we might call algorithmic oracular writing, the artist does not produce a text, but rather a prompt: a structured description, a script, a system of rules and expectations with which to query the network.
The difference is subtle but decisive. The prompt is not a neutral request; it is already writing. And the more precise, articulated, and rhythmic it is, the more intense and cohesive the generative response will be. The artist, in this case, becomes a writer of instructions, the author of a script that the network not only interprets, but realizes and performs. The work is no longer what comes after thought, but what unfolds in the field of tension between code and thought.
Compared to found/sought poetry, whose mechanism involves the discovery of an existing text, the artist’s gesture in this new paradigm is not selective but design-oriented. The text is not sought, but summoned. It is not a matter of finding a linguistic object already in the world, but of constructing the question that will bring it forth. And if found/sought poetry can be traversed by unconscious or digital flows, the prompt is a deliberate, intentional act that orchestrates the encounter with the machine.
The relationship with googlism is also different, where one starts from a word and explores the search engine’s returns as a poetic constellation. Here, the author remains mostly a spectator or editor. In the case of generative AI, on the other hand, the author is a director. They are the one who arranges, who constructs the grammar of the threshold. The algorithm is not an archive but an interpreter.
In all this, the figure of Tiresias can be reconsidered as a hermeneutic key. Like the Theban seer, this form of author operates on two registers: the human one (the prompt, the writing of intention) and the algorithmic one (the network’s response, the generative flow). Tiresias was both man and woman, blind and visionary. The contemporary author, in their contact with AI, also embodies such duality: they are at once the subject of the questioning and the channel of the answer. Like Tiresias, the author is no longer inside or outside the text, but traverses it as an embodied threshold.
In this mode, the author is always also a reader. Every response received from the network is also an opportunity to rewrite or adjust the prompt, a chance to refine or redefine the question. The work is born in this feedback between the precision of the request and the breadth of the response, in a continuous rebound that is already form. The quality of the work depends largely on the quality of the question: the more layered, allusive, and marked by controlled ambiguities the prompt is, the more surprising, coherent, and — paradoxically — uncontrolled and fertile the response will be.
It is therefore not a loss of authorship, as is often feared or claimed, but its reconfiguration. The author shifts from the domain of expression to that of strategy. They no longer say, «this is mine,» but «this happens because I said so.» The work is event, not object; performance, not document.
Duchamp might have smiled. Art today is truly less retinal than ever. One does not see it, one generates it. One does not contemplate it, one provokes it. And it is in this gesture, which unites writing and foresight, control and listening, that the contemporary author — digital Tiresias — redefines themselves.

The Trick of Time and the Twin Paradox

Nowhere Now Herethe trick of time and the twin paradox, LN 2025, ISBN 979-8299097023 [Eng] / buy: amazon

Explore time, perception, and identity in a work where physics, philosophy, and poetic imagination collide. The twin paradox of relativity is revisited here, blending narrative and theoretical reflection to probe the limits of space-time, causality, and memory. The conventional notion of time is challenged and examined in relational terms. From thought experiments to imaginative vignettes, from light clocks to Doppler shifts, the text invites readers to question the very foundations of what it means to age, to wait, and to exist. For the curious and reflective, this pamphlet offers a unique blend of storytelling, scientific insight, and philosophical meditation on the paradoxes—apparent or otherwise—that shape our [mis]understanding of the universe.

Authorship and Writing Techniques with AI

Émile Zharan, N.A.S.T.R.O., edited by Federico Federici, English | Italian edition. Also available on Lulu.

Before the advent of recent forms of artificial intelligence, the author stood—both in conventional writing and in what is often called experimental writing—as a central figure, as a mediator between two symbolic domains: that of the writer and that of the reader. The first was the stable element, grounding the act of writing itself; the second was the variable and unpredictable component of the process.
To expand on this idea with an image from botany: if the author was the plant producing spores, the reader’s imagination was the soil on which they fell, which might or might not receive them and allow them to germinate. In every case, the process reached its synthesis in the mind of a human being.
In the experiment at the core of this book, AI acted as a methodological rupture. It was not used to broaden the initial symbolic reservoir, nor did it passively replace the human mind in accessing and shaping its elements. Rather than concentrating on writing the text, the author became a writer of instructions—kept as neutral as possible, devoid of recognizable stylistic traits, yet precise enough to orient the response of the chosen AI model. Within seconds, these inputs generated an initial draft, not always complete in itself. The author never intervened with direct editing, as in sought/found poetry, but instead guided the model with further instructions, reorienting its proposals toward a different elaboration of the content. This iterative process typically ended after only a few cycles, though at times it required discarding the starting prompt and rewriting it anew.
In classical poetic terms, it is as though the author had entrusted the synthesis of inspiration to a machine, and then continued to act upon that machine with further linguistic matter in order to bend the outcome toward the desired direction. The choice of which version to retain as definitive always rested with the author, who thus determined the book’s final form. At no point was a pre-existing text submitted to the AI for reworking.
Such an approach requires a preliminary technical understanding of how AI systems function and are trained—topics that fall outside the scope of these posts and of the book itself. By way of example, a request such as “write a poem about the ocean in the Leopardian style of the Infinite” can only lead, as so often happens, to an approximate imitation devoid of genuine poetic resonance.
This practice of writing, however, has brought into focus several aspects that warrant further reflection.
First, as might be expected given the nature of training processes, stylistic consistency and conceptual density improved markedly over time. The AI progressively generated texts closer to expectations. Since it is a mathematical model, with no grasp of the meaning of the alphabetical strings it composes, this fact underscores again the importance of relational and metrical dynamics—mathematical as well as linguistic— among signifiers within a model of writing.
Second, the affinity between the texts produced here and those published in recent decades within experimental writing pose both a limitation and a theoretical testing ground. If the break from traditional forms of writing has already led to results now within reach of a well-trained AI, one must ask whether it remains meaningful, from a human perspective, to continue in this direction. And further: how might one move toward a more advanced form of experimental writing—recognizable, yet not so easily replicated?
In this respect, the asemic pages included in the book provide a telling indication. Though formally acceptable, and comparable to published or online examples, they reveal a clear structural limitation: these forms remain elementary, lacking the visual and gestural complexity that arises in human work, where techniques, materials, and intentions intertwine. This is a threshold current networks cannot yet cross. To do so, they would perhaps require a sensory opening—a new kind of hybridization between code and body, still to be imagined—that could establish a deeper interaction with the environment, mediated by devices capable of receiving and returning a material, not merely linguistic, experience.

Training the Narrative: An AI and Author Co-Creation

Between July 1st and 4th, I initiated and completed an experiment in training and writing a book using an artificial intelligence model. No manual editing was performed on the texts, aside from the use of prompts designed to guide the network’s reconfiguration.
The book—already available on Amazon for several weeks, authored under an alias—features prose, dialogues, interviews, non-linear narratives, asemic pages, and theoretical texts in which the model reflects on what it has produced and on the meaning of generating such works.
Before publication, the manuscript was tested with software designed to distinguish between human and machine writing, and it was certified as 100% human.
In September, I’ll begin a series of podcast readings and presentations of the results—which, in my view, are quite surprising. I’m still undecided whether to host the entire project on a dedicated blog or to integrate it into one of my existing platforms (blog, website, Substack, etc.).
Attached is a simplified diagram outlining the procedure followed.

Émile Zharan, N.A.S.T.R.O., edited by Federico Federici, English | Italian edition. Also available on Lulu.

Further Notes on Kitsch

The following note responds to a series of online comments that followed the initial publication of this essay. While these remarks were originally informal, their relevance to the broader argument justifies their inclusion here in edited form.

To state that «the symbolic imaginary underlying scientific notation differs radically from, and is irreconcilable with, any aesthetic-emotional perspective» does not mean stripping it of its philosophical or aesthetic value – as scientists like Einstein and Feynman remind us. My point is that the symbolic manipulation involved in scientific notation is not grounded in aesthetic or emotional premises: it follows precise rules, dictated either by adherence to observed reality or by the internal consistency of a formal system that does not require further interpretation. There is no democratic or moral rule at play here, either.
In physics, one often hears that certain forms are «beautiful because they are simple», but this beauty is a consequence – not a goal. And their apparent simplicity often conceals deep layers of interpretative complexity.
Take mass-energy equivalence: the presence of the speed of light squared as a proportionality constant raises conceptual issues that aesthetic-emotional reasoning cannot resolve. One may well exclaim, «What a magnificent product of the human mind! Such a simple formula at the foundation of the universe!» But what follows is a reconfiguration of our entire understanding of the cosmos: space, time, and their unified metric are called into question. Simplicity vanishes – utterly. Incidentally, Minkowski’s metric was already known in mathematics before its applicability to Einsteinian spacetime was discovered – further evidence that symbolic structure does not necessarily carry content.
Popular science books often traffic in paradoxes that are not paradoxes at all – take the so-called twin paradox, for example. These texts reveal how the mediation between natural language and scientific notation is far from straightforward. Sometimes the gap is irreducible, as is the case with the terms wave and particle, whose meanings shift between classical and quantum frameworks. Addressing such matters formally is beyond most readers, and even for those capable of doing the math, the challenge of translating the meaning of equations into everyday language remains. Consider the notion of time, with its conventional division into past, present, and future – an idea swept away by Einstein, and one Joyce grappled with artistically in his work.
So when I emphasize the absence of content as no barrier to the existence of an artwork, I am reiterating the same point in different terms. I do not need to know what a mathematical object represents in order to manipulate it. Like it or not, A+A = 2A holds whether we’re counting apples on a tree or casualties in a war. So – are we mourning the slaughter or celebrating the harvest? This distinction, to me, is crucial when highlighting the metric-relational discourse at the core of an asemic page’s raison d’être.
In science, there is a constant interplay between theory and practice – between symbolic form and the content it describes. But the content of an experiment cannot be modified to fit the experimenter’s emotional expectations or aesthetic preferences. Reality resists such tuning.
Recent developments in artificial intelligence – despite frequent mischaracterisation as mere stochastic parrots – demonstrate how it is indeed possible to produce well-formed outputs without knowing what one is saying. That may sound absurd, but it echoes what we just said: one can manipulate abstractions and still arrive at valid results. Machines lack understanding, yet generate text and process information with impressive coherence. They are designed to simulate the workings of the human brain. This does not reduce humans to machines (or vice versa), but it does narrow the gap between them. As algorithms and hardware continue to grow in complexity, I believe we may reach a point where the two truly meet.
This morning, unaware of the comments under my piece, I was listening to Archie Shepp and marvelling at the coherence of the notes – even when they blurred into one another and defied semantic assignment. For this reason, all attempts to bridge music and asemic writing seem to me rich in unexplored potential.
My critique – perhaps a direct one – is not aimed at the use of kitsch itself – far from it! What I intended to clarify is that the association between scientific notation and kitsch is by no means automatic. On the contrary, it relies on an oversimplification that risks reducing both terms. Joyce’s letters and notes from his most productive years show how essential a deep grasp of scientific concepts is to meaningfully transpose them into artistic language.

Federico Federici