I had resisted writing about this subject for two reasons. First, I needed to let certain ideas settle; second, I feel a certain laziness when it comes to writing about what everyone else is already writing about. For almost four years now I have watched the number of essays on writers and LLMs [Large Language Models] grow in a cancerous fashion. Prudence masking immobility and fear. Noise masquerading as thought, desperate to grab a click at any cost. At first, this reminded me of the self-satisfied pandemic essays that reheated theoretical leftovers from the past and presented them as novelty. Writers and columnists, as is well known, have never missed an opportunity of this kind. But once they are handed a current, juicy topic, they suddenly feel compelled to have an opinion about it. And they’d better have it quickly. By Thursday at eight in the morning. Some of these opinions have aged badly, and what is surprising is that people keep repeating them.
There are those who claim that writers can use LLMs as valuable support tools (after all, they know the rules of the subjunctive). Others insist that LLMs are a harmless fad that will soon pass, once the public notices their obvious limitations (they won’t last two more years, you’ll see). Others still claim that LLMs will be able to emulate and replace writers in the short or medium term (there are already enthusiasts showing you how to write five books a day using artificial intelligence and sell them on Amazon). Before addressing what I actually want to discuss—the blur produced by LLMs and the possible death of the idea of the “virtuoso” artist—I will begin by clearing away the confusions implicit in these expectations.
The usefulness of LLMs as support tools is indisputable, provided that a writer’s main difficulties are orthographic and grammatical. But I have never met a writer with something interesting to say whose main difficulties were orthographic and grammatical. Not because spelling and grammar are universal indicators of intelligence, but precisely because they are not. Polishing the prose of a foolish opinion only makes that opinion foolish in a more transparent way. Nothing makes it clearer that a writer has nothing to say than the meticulous correction of their style. As for research queries, no LLM can offer more information than what Google or any other decent search engine already did.
The idea that LLMs will eventually go out of fashion and have no effect on literature conveniently ignores the fact that literature is not disconnected from other forms of speech and writing (by “non-literary” forms I mean everything from Instagram captions to final papers for a class). Even if LLMs were, for the moment, to have little or no direct effect on writing itself (beyond providing fodder for Thursday’s column), they have already had an indirect one: a large portion of contemporary text production—promotional, explanatory, even legal and political—has fallen into the hands of LLMs. At this very minute there are hundreds of thousands, perhaps millions, of bots reading and reproducing what we call “content” on the internet. And that transgenic content is invading the world.
A subterranean war is being waged through the massive, fully automated reproduction of any form of content that appears to attract demand. The phenomenon has been dubbed “AI slop,” and it does not seem reversible. Its consequences will be profound and disastrous. It is naïve to think this will not have an indirect effect on literature. It already is. Not only on what is being written now, but on what has already been written.
Some of those who have noticed this new order of content reproduction have decided to ride the wave rather than fight it. They declare the writer’s craft dead and boast about how easy it is to churn out hundreds of pages on any topic and put them up for sale. The only apparent limitation is the scarcity of real human beings willing to read them. But those enthusiasts who believe they can write books indiscriminately with a minimum level of quality seem not to understand how LLMs work, what literature is, or possibly both.
LLMs operate on colossal datasets. They search for patterns and repeat them without understanding them. The minimum mass of texts required for a prediction to be possible—for a pattern to emerge at all without committing demonstrable plagiarism—is too large: the sharpness of genius is lost in the volume of mediocrity. It slips through the LLM’s enormous, clumsy, predictive limbs. Expecting an LLM to find an interesting adjective is like expecting a manatee to unlock a phone.
An LLM can find nothing but commonplaces. In a very precise sense, the only thing an LLM can replace in a writer, without committing demonstrable plagiarism, is their mediocrity. But it can also multiply that mediocrity with such speed and efficiency that it ends up rendering everything else invisible. Until relatively recently, literature had been spared because the mediocre writer had (human) limits to their productivity. They no longer do. And that is the real problem, and the reason I am writing these lines.
The productivity of the mediocre writer or artist will now be limited only by the demand of those willing to consume—to “pay attention to”—what they produce. A snowball effect in which the expansion of mediocre productivity manufactures (through sheer saturation) an increasingly mediocre audience, which in turn manufactures more mediocre writers and artists, who then resort (must resort) even more to artificial intelligence, is not only plausible but quite likely. Artificial intelligence, at least in its current state, will not replace true writers or true artists, because it cannot emulate them. What it will try to do instead is cancel them out.
LLMs have devalued written—and even spoken—language in the public’s unconscious. I remember when it was more common to read the texts accompanying a photograph on Facebook, or the descriptions of what a business did on its website. Now we skip them. We assume they were written by an LLM (and rightly so: they probably were). We often skip most of an LLM’s answer to our own questions as well. We skim until we reach the line that contains the sentence that matters. What our nervous system is learning is that text is an abundant, insubstantial, frequently supplementary element, growing like white mold in the crevices of our attention.
Literature was already in crisis before the arrival of LLMs, because reading itself was already in crisis. And it is precisely the capacity for reading that is suffering most from the exponential increase in mediocre texts (not only “literary” ones). Meaning is getting lost in text. Text multiplies and meaning dissolves. LLMs produce text by averaging other texts. In visual terms, that has a name: blur. To blur is to progressively dilute an image until it becomes unrecognizable. The LLM must blur, because it is the only way to avoid plagiarism lawsuits. An LLM is a machine for blurring text and culture in general.
I would venture to say that what is at stake is not only literature or the written word, but language itself. There have been many periods in which the practice of writing and reading was almost entirely lost. Entire traditions disappeared or survived only in a few dark and vulnerable copies. But even under those circumstances, oral language preserved the original function of language intact. The word was not devalued, because it still required effort, because it still mattered. Despite the limitations of the oral format, there was still something we can call literature (songs, stories), and something we can call thought, wisdom.
What is happening now—the loss of sharpness, the industrial-scale blurring of the objects of our attention, of what we read, see, and hear—has no precedent, not even in prehistory. Animals communicate by making an effort. Each howl is, in a sense, unique: practiced and eventually executed in a concrete situation.
There are already movements against LLMs. Reasons have not been lacking. And several brands and businesses of different sizes have seized the opportunity to distinguish themselves (and increase sales) by declaring themselves free of AI slop. These reactions, which I certainly support, nonetheless have limited power, and I suspect they will ultimately prove useless. Among other reasons—ones I will explain shortly—because LLMs appeal to one of the most powerful forces in human beings, more powerful than empathy or hatred: laziness. It is easier to fight hatred than to fight laziness.
One of the most unexpected and terrible phenomena of this cultural blur has been the growing number of artists forced to prove that their works were not made using an LLM. This has not been limited to “mediocre” artists or works. On the contrary, it has often been the most complex and sophisticated works that have been accused. Mediocrity not only believes it now stands on the same level as talent or genius. It has found the perfect excuse to deny them, if necessary.
Writers and artists know the difference between what they do and what an LLM does, but that difference is often invisible to the rest of the public. And the great problem is that even if the public feels sympathy for real artists and rejects the massive use of LLMs, it has no way of distinguishing on its own whether something was partially or entirely generated by one. Naïve mediocre artists will tell the truth and declare themselves LLM users; more astute mediocrities will continue using LLMs in secret while declaring themselves fierce opponents of them—a stance that will earn them attention and public sympathy.
If I had to bet on how literature and art will look in the coming years, I would bet on “proudly human” works in which the most obvious traces of LLMs have been erased and in which demonstrative “errors” have been deliberately inserted. The literature and art of the future will embrace premeditated imperfection and performative effort. In a culture blurred enough, even smudges on the lens will be considered beautiful. And while there may indeed be a certain beauty in a dirty lens, any photographer will see that admiration as youthful naïveté. Any professional photographer who has struggled to obtain a sharp, perfect image will understand that admiration for the smudged lens is, in the end, disrespectful.
In the future, as in the past, virtue will be inhuman. But in the past that inhumanity had a divine nature, whereas in the future, in the eyes of the public, it will have an industrial and superfluous one.
“Literary” or “artistic” content will not be separated from the rest of the objects of our attention by the abandonment of LLMs, but by the discourse it adopts around them. In other words, literature and art will become more hypocritical and pedantic than they already are. Never before will there have been such a wide gap between the private life of the artist and what the artist wishes to show of it. And it should not surprise us if one of the most frequent themes in future works—perhaps through unconscious projection—is imposture.
In our time, impostor syndrome has become generalized, and in many human beings it is no doubt unjustified. But haven’t we also encountered several cases in which someone believes they suffer from impostor syndrome, when the reality is simpler—when they are impostors, plain and simple?
As this likely romanticism of imperfection (which privileges process over result, the authenticity of error over the monotony of skill) becomes more evident, eventually turning into an irritating fashion, I wonder whether there will be any reactions that attempt to rescue the idea of the virtuoso writer or artist—or whether literature and art will simply continue digging their own grave, as the public ends up preferring the mundane efficiency and reliable simplicity of LLM-generated or LLM-assisted art and literature (assuming it continues to consume art and literature at all, after years of exposure to a blurred culture in which meaning itself may at times seem impossible).
History seems to suggest that laziness cannot be defeated. So although I adopt the first position in my writing and reading, I confess that I believe the second will ultimately prevail.
Writers and artists who resist the temptation of using LLMs, even in a measured and discreet way, to increase their productivity are about to experience a degree of loneliness unimaginable in the years to come. Their virtue, the sharpness that distinguishes them from the majority, will be invisible to almost everyone except themselves—and even to themselves (under the inescapable pressure of others’ gaze) its existence will be in doubt.
Let the paranoia begin.
