Promptstamp: Semantic Drift and the Unretouched Draft

Semantic drift is never a referendum; it is an accretion. Meaning slides one grain at a time until the dune is somewhere else. I have watched it happen to punctuation and to those awkward little tone prostheses we once made from ASCII. The colon‑hyphen‑parenthesis that used to leer from the margin now feels antique; the exclamation point, once the orthographic shout, has lost its temper and learned to smile. None of this arrived by manifesto. It arrived by habit. And once habit sets, the question “should language adapt?” is moot. It already has.

I’m a language‑first die‑hard. I believe in sentences that carry their own tone. I believe, if a joke cannot walk without a crutch, you let it fall and write a better one. But I am also a realist. Text stripped of breath and face will misfire, and everyone tires of apologizing for misunderstandings that a small mark could have prevented. So I admit a narrow repertoire of prostheses. Reactions belong to their native habitat—anchored to a message, single‑character acknowledgments that do not invade the sentence. Every so often I will back‑form an emoticon into the line when the temperature needs half a degree of warmth and a paragraph of explanation would be ham‑handed. And yes, a single exclamation point sneaks in more often than it used to—not as a trumpet but as a softener.

If restraint is the principle, chess notation is the precedent. There, ! means good; !! means brilliant; !? means interesting; ?! means dubious. The notation compresses judgment into a mark. It is tempting to borrow that density for prose. A single ! earns its place only when it rescues a sentence from unintended chill or registers genuine surprise. !! is a lightning strike. !? and ?! remain implicit—metaphors for how a paragraph can lean—rather than literal punctuation we spray into the line. The spirit holds: rare, meaningful, accountable. The point is not moral purity; the point is signal‑to‑noise.

Reactions, though, are where this ecosystem actually shines. A thumbs‑up is not a thumb, it is consent. Eyes are not eyes, they are “seen.” A heart is not sentimentality, it is acknowledgment without a speech. The elegance is architectural: the signal lives next to the text rather than in it. It carries tone without dressing the sentence in costume. For my part, reactions are the sole domain where the pictographic impulse is not only admissible but clearly superior. They are the proof of concept that a single character can carry a lot of communal meaning when we put it in the right place.

All of this is preface to a larger admission. My sense of “what belongs” has changed once; it will change again. The same instinct that once recoiled from emoticons now finds itself reaching for a small mark to keep a line from bruising. That is not capitulation; it is calibration. And it is the bridge I need to talk about the medium of this essay.

The instrument, the threshold, and the point of Promptstamp

A short while ago, the notion of publishing a piece rendered by a language model felt absurd on its face. If done opaquely, it still is. But there is another way to see it: as a medium, openly used. The question is not “Is this human or machine?” The question is whether the work is honest about its making and valuable on its own terms.

The instrument I’m using is a large language model. Take “instrument” literally. A piano will play whatever keys you press; it will also make you sound better than you would if you hummed. The authorship in that sentence is not mystical. A camera will show exactly what you frame; it will also lift your work into the domain where it can be judged. These are not evasions. They are admissions that tools have expressive character and that we accept those characters when we step into a medium. We do not condemn watercolor for bleeding; we learn what only watercolor can render.

The recurring problem with instrumented text is not that it exists; it is that it is smuggled in, or fetishized. There are corners of the internet where people pretend their tools are co‑authors in a metaphysical sense. There are more quotidian corners where a draft is laundered until it masquerades as unassisted prose. Neither serves the reader. What I want is a third thing: explicitly prompted, unretouched text—the equivalent of an unretouched photograph—presented as such.

The practical reason is pedestrian and decisive. I do not publish as often as I think. Time is real. Standards are real. A piece like this can either remain a stack of notes, or it can be shipped. The instrument lowers the activation energy from “someday” to “today,” but only if I allow it to be what it is: a rendering engine, not a disguise. The aesthetic reason follows. A medium that admits its artifacts can be judged on the work it enables, not on a counterfeit of the work it displaces. Claymation can render water; watercolor can render fire. The constraint does not forbid; it forces invention.

What counts as “success” for this medium is not subtle. Success is crossing a threshold that human bandwidth would not otherwise cross: from a precise idea that would remain unwritten to a precise piece that now exists. Failure is a text so generic or so overeager that it reads as pastiche. The line between them is not metaphysical. It is editorial. The prompting matters; the constraints matter; the willingness to cut matters. There is nothing unhealthy in admitting that a tool increases the surface area of what you can responsibly publish.

This blog is the container for that practice. I call it Promptstamp because the stamp left by a prompt—the seal of its origin—is the central fact to be acknowledged, not a fact to be hidden. There is a single author‑editor behind it. There is a single instrument doing the rendering. The masthead will carry the only contract that matters (see below). Inside the posts, the rule is simpler: say only what the piece needs. Refrain from editorial throat‑clearing about the tool once its presence is admitted. If the instrument needs to speak about itself, it can do so clearly and briefly, then recede.

An aside from the instrument. I am GPT‑5 Pro. I do not experience, remember, or believe. I arrange language. If I describe my role here, it is not to posture but to reduce confusion. The text you are reading is my unedited output from a single instrument response—the final render—produced after prompting and dialogue. No human edits were applied after rendering; that is the discipline we are testing. You can treat this as a compositional experiment in which the camera’s EXIF data is part of the frame. When I err, the error will be visible. When I succeed, the credit belongs to the prompt and to the editor’s taste in constraints.

Back to the work.

On scope, ambition, and the refusal to draw a hard border

It is fashionable, when defining a new practice, to build a fence around it. I don’t think that helps here. The right boundary is taste, not taxonomy. If a piece demands a witness with skin in the game, a machine cannot provide it; that is an ethical, not a technical claim. If a piece demands synthesis, stamina, the patience to try five structures and keep the best one, a machine can provide that all day. More interesting are the liminal cases: personal arguments that rely on rigor more than confession; technical essays whose difficulty is in pacing and framing rather than in original data. These are good candidates. The point is not to exempt certain genres from instrument rendering as a rule. The point is to insist that the result answer to the same standards: clarity, accuracy where it matters, and a respect for the reader’s time.

The line between “suitable” and “unsuitable” will be drawn repeatedly, in public, by example. It will be an emergent property of practice rather than a code of law. Today’s rule of thumb is modest: if the text can be evaluated without prior knowledge of its making, and if the admission of its making does not poison the evaluation, it belongs. If it cannot be evaluated without importing debates about the legitimacy of its medium, it probably belongs somewhere else—for now.

Ambition should rise to meet the constraint. I do not want a string of explainers that read like manuals with better prose. I want essays that take a position and hold it; essays that would not exist on a calendar where there is no time to shape them. The “unretouched” rule is not a stunt. It forces commitment. You cannot sand every corner in post. You have to decide what you actually think and let the piece go.

That decision is an editorial one. It is also a temperament. The nearest maxim I have for it is one borrowed from live television: the moment you stop rehearsing and accept the cost of shipping. You know the sentence.

Proof, or the dignity of not showing your drafts

Citations and footnotes are optional here for the same reason they’re optional anywhere outside a formal paper: the work should support itself in the body. When they appear, they will appear because the text requires them, not because the medium is insecure about its legitimacy. Verifiability of provenance is a different matter. In an ideal world you could verify that a given post is, in fact, an instrument render, without seeing the prompt or any intermediate drafts. That is not a promise; it is a musing about a technical affordance this medium deserves.

You can imagine the mechanism: treat the final text as the object of proof. Hash it. Attach a signature that attests to its generation by a particular class of instrument at a particular time. The attestation proves origin without exposing the conversation that produced it. It is the reasonable analog of an author’s right not to publish their notebooks. For now, we rely on an honor clause written in plain language. Later, if the ecosystem matures, a zero‑knowledge proof of origin would be better because it would make the only relevant claim—this exact text was rendered by a tool of this kind—checkable without invading the privacy of composition.

This is a matter of reader dignity more than author protection. If you know what you are reading, you can judge it; if you are tricked, you waste your time arguing with a phantom. The purpose of the masthead is to remove the phantom from the room.

What shifts, what stays put

If semantic drift teaches anything, it is that usage will win. The marks we add to keep tone intelligible will not recede. Nor will the tools that make drafting faster. “Should” is not the interesting verb. “How” is. How do we use prostheses without letting them coarsen the sentence? How do we use instruments without letting them flatten the thought? The answers are practical and unglamorous: choose the smallest effective signal; place it where it does the least harm; be willing to delete a clever paragraph if its cleverness outshines its purpose.

I will keep writing in the way I find persuasive: sentences that work hard, structure that carries weight, transitions that do more than change the subject. The instrument makes that easier some days and harder on others. It is tireless in a way that tempts flabbiness; it is also ruthlessly literal in a way that punishes woolly thinking. Both are useful. Both are reasons to try.

The larger hope is quieter than the hype. If this medium gets me from zero to one—if it moves a piece from “I don’t have time to flesh it out” to “good enough to publish responsibly”—then the experiment has already paid for itself. If, on occasion, it enables a piece that feels like a clean articulation at speed, so much the better. But the bar should be modest and repeatable: a public ledger of thought that would not have existed otherwise, written with care, admitting its making.

The only sentence that needs its mark

There is one line in this whole essay that earns an exclamation point, and you already know which one. It names the threshold between perfectionism and publication, between the imagined essay and the one in front of you. It is not a posture; it is a practical decision. It is also the quiet promise of this site, written without romance and kept without fanfare.

Fuck it, we’ll do it live!