Fakes of the Future | Los Angeles Review of BooksFakes of the Future<br>Literary credibility in the age of AI.<br>By Krzysztof PelcJune 10, 2026<br>LARB Lit

Science & Technology

Did you know LARB is a reader-supported nonprofit?<br>LARB publishes daily without a paywall as part of our mission to make rigorous, incisive, and engaging writing on every aspect of literature, culture, and the arts freely accessible to the public. Help us continue this work with your tax-deductible donation today!

Donate

This essay is a preview of the summer issue of our print journal, the LARB Quarterly, no. 49: Traffic, out now. Become a member for more essays, criticism, poetry, fiction, and art—plus the next four issues of the Quarterly in print.

NAPOLEON’S FAVORITE BOOK of poetry was a fraud. He carried it through the Italian campaigns and still had it with him, years later, in his exile on Saint Helena. Attributed to an ancient Celtic bard named Ossian, the poems were presented as translations of a recently “discovered” third-century epic cycle. Raw, melancholic, and untouched by Christian pieties, Ossian’s poetry swept across Europe, fueled nationalist sentiment, shaped early Romantic taste—Goethe was a fan—and, improbably, became Napoleon’s bedside read, even as many of Europe’s literary scholars suspected it of being a forgery. Today, Ossian is a curiosity with which hardly anyone bothers.

As odd as the episode now seems, it was less an anomaly than a recurrent symptom of a certain kind of malaise. Late 18th-century Europe was gripped by nostalgia for imagined pasts unspoiled by the perceived corruptions of modern life. Writers and readers alike yearned for the sublime, for sentiment, for a “natural” folk genius unburdened by learning. Ossian’s songs—primitive, elemental, unmediated—offered what the existing canon could not: the promise of uncontaminated origins.

We have, I believe, crossed a new threshold, and all authored writing—novels, poems, screenplays, newspaper columns, not to mention love letters—will be judged according to which side of that divide it falls on. On one side are texts produced before the arrival of generative large language models (LLMs). On the other, everything that has followed—texts that might still be useful, even compelling, but that will always face a lingering suspicion of not being entirely human, of having been smoothed by systems trained to predict the word that comes next. We will come to prefer the former over the latter, not because it will be better, but because we will be more certain of its origins.

Tastes Catch Up

A recurrent blind spot when imagining the effects of any large-scale technological change is to assume that the world will change but that what we value in it will not. In practice, our tastes swiftly adjust—and in ways that often blunt the most alarmist forecasts. Mass industrial production didn’t wipe out handmade crafts; it turned them into a marker of distinction and raised their price. Photography didn’t kill painting; it updated its ambitions. Wikipedia didn’t abolish expertise; it changed what we demand of experts. In the same way, AI won’t downgrade writing, but it may permanently change what counts as good writing.

Perhaps you’ve already adjusted. Perhaps you’ve started treating books, articles, essays, and emails written today differently from anything before, say, 2022. On hearing a polished sonnet in effortless iambic pentameter at a wedding toast, you may now smell a rat. If that reflex hasn’t set in yet, my bet is that it will soon. In fact, I suspect the distinction will be drawn first by those of us who rely most on LLMs in our own writing, who know how hard it is to resist their knack for finishing a thought before it has fully formed—sometimes directing it in an unanticipated direction. The better and more omnipresent these systems become, the more we will come to prize the culture that preceded them. We have ejected ourselves from our all-human paradise, and we will look back on its artifacts with a trust that can never be recovered.

Some of the unease around LLMs centers on hallucinations: invented facts, spurious citations, confident nonsense. Those are conspicuous and easy to ridicule, but they’re also tractable. The model designers will develop safeguards, and we will develop habits. We’ll learn to double-check things that seem too good to be true. No, the real threat is not the flamboyant hallucination. It’s the impossibility of knowing whether any text we now come across represents an individual sensibility or a synthetic hybrid—the result of a human prompt and a system trained to complete patterns drawn from vast archives of prior writing.

The Threat of Recursion

If a key ambition of cultural production has long been innovation, its opposite is recursion: the slow, imperceptible drift that happens when a body of work increasingly reflects what preceded it. Generative models are trained to reproduce what is most statistically probable. When...