This Sea of Indistinguishable Words

Prompt 1: Write a 100-word opening paragraph to a newsletter deploring the stylistic sameness of contemporary non-fiction writing.

In an era saturated with non-fiction, a troubling trend emerges: stylistic conformity. The landscape of contemporary writing is marred by a sea of indistinguishable prose, lacking in creativity and individuality. As readers, we are bombarded with cookie-cutter language, devoid of the vibrancy and diversity that once characterized non-fiction literature. From articles to essays, the sameness pervades, stifling innovation and diluting the power of expression. It's time to lament this homogeneity and advocate for a resurgence of originality. Let's break free from the shackles of formulaic writing and reclaim the rich tapestry of language that makes non-fiction truly engaging and transformative.

One hundred words on the dot. ChatGPT3 wrote this passage in less time than it took me to type my prompt. I probably should have asked the bot to focus on the non-fiction writing that interests me most, that is, longform magazine writing and book-length non-fiction with literary aspirations (which is often commissioned on the back of publication in magazines). The bot has latched onto the term ‘stylistic sameness’, in hindsight quite a sloppy and imprecise usage. I don’t exactly mean that all contemporary non-fiction sounds the same, though some of it does. What the bot has surfaced is a group of stylistic practices that bedevil contemporary non-fiction, they mar the sea, as ChatGPT would have it. Some of these are errors that reflect the limits of generative AI as a writing technology; some of them are scraped right out of the human-generated datasets and reproduced in the generated texts. I’m to blame for the concept though, and for the poor phrasing. Jeremiads against bad writing are tedious and snobby, often racist and classist, and so copious that it’s hard to justify writing another one. Complaints about bad style are fun to read, especially if they get mean and personal, which is why they’re so popular in a clickbait economy; it’s rare that the authors of such consider why there’s so much bad writing around.

Prompt 2: Make it angrier, please.

In an age overrun by non-fiction, a maddening monotony prevails: stylistic uniformity. The contemporary writing scene is marred by a suffocating sameness, devoid of flair and distinction. We drown in a sea of generic prose, robbed of creativity and individuality. From articles to essays, the monotony reigns supreme, crushing innovation and neutering expression. It's infuriating to witness this homogenization, a soulless march towards mediocrity. We must rage against the formulaic chains that bind us and demand the resurgence of originality. Let's unleash our fury upon the blandness and reclaim the dynamic richness of language that fuels true non-fiction brilliance.

Barricade the bookshelves, friends, for we are being overrun by a horde of non-fiction authors! When I read this I was startled to note how quickly ChatGPT started to sound like a cultural reactionary railing against the wokerati.

Am I angry? If so, with whom? I’ve written plenty of negative reviews but I’m reluctant to point at any one or two writers as somehow emblematic of the shoddy style of contemporary non-fiction. No one gets paid enough, no one has enough time. Even a social model of authorship, one determined by relationships between authors, editors, readers, critics and other intermediaries seems rather nostalgic in the Generative AI Era. Yes, I’ve been reading Big Fiction by Dan Sinykin, which is as stimulating as everyone says it is (including Alice Grundy here), thinking about the corporatisation of authorship, trying to figure out whether generative AI is just another acceleration of writing technologies or something new altogether, whether it’s a continuation or a break. Also: style cops are jerks and the One True Style, whether it’s attributed to Orwell or some other luminary, a political fiction.

Prompt 3: More sarcasm, more cynicism.

Ah, rejoice, for we dwell in an age where non-fiction writing has achieved the pinnacle of banality! Behold the majestic landscape of contemporary prose, where every piece reads like a carbon copy of its neighbor. How delightful it is to navigate through this sea of indistinguishable words, where creativity and individuality are but distant memories. From the riveting articles to the captivating essays, one can hardly contain their excitement at the sheer predictability of it all. Let us bask in the glory of this monotonous parade, where innovation is a forgotten relic and expression is relegated to the realm of clichés. Truly, a golden age of literary conformity!

Each of these airless ChatGPT-generated paragraphs qualifies as terrible writing. You couldn’t publish them as they are but you might consider doing so after a few tweaks. Don’t blame writers, don’t blame editors, blame the just-in-time content management ethos that values word salads because they cost nothing and are generated in seconds. They’re useful, these little hypotheticals, because they index a set of stylistic practices that are familiar to readers of contemporary non-fiction, whether published online or in print. It’s a perverse irony: a prompt to generate a complaint about bad writing instead generates oodles of examples of bad style for discussion.

Clichés proliferate (‘rich tapestry of language’, ‘power of expression’, ‘formulaic chains that bind us’, ‘golden age’, and so on) and so does a weird, unchecked tendency to alliteration and assonance (‘troubling trend’, ‘suffocating sameness’, ‘maddening monotony’). I wonder whether this last is a riff on my use of ‘stylistic sameness’ in the prompt. There are too many adjectives. The machine describes problems using passive constructions that make it difficult to attribute causation or agency (‘We are bombarded with cookie-cutter language’ – who is being bombarded? And by whom?). Put differently, it’s written by no one, from nowhere. Don’t write in the passive voice. How does ChatGPT process that ubiquitous nugget of style advice?

Syntax is often just a little distorted. Mixed metaphors, malapropism, and adjacent unrelated figures pull sentences in contradictory directions. You find yourself asking, how does this verb operate on this subject? See for example, ‘The contemporary writing scene is marred by a suffocating sameness…’ The frequent figuration of the literary world as a landscape or ecology – I’ve done this a lot in my own writing – is rendered as a very strange set of topographies in these passages. AI glitches are called hallucinations; this prose displays the surface characteristics of an hallucination without any psychic connective tissue. If you stare for a while at this sentence from the third passage, it just gets weirder and weirder:

Let us bask in the glory of this monotonous parade, where innovation is a forgotten relic and expression is relegated to the realm of clichés. 

We can fashion some sense out of the first clause, a snarky injunction to bask in the glory (bit of a cliché, yep) of a monotonous parade (parades can be monotonous, but how does a person bask in their glory?) The use of the word ‘where’ carries the counter-intuitive implication that the parade is a location. Weird move, weirder to make it a place where innovation is a forgotten relic. I mean, if it’s been forgotten… And if what ChatGPT wants to say is that literary style in this monotonous parade relies on clichés, and I think that’s the sense of it, then why on earth would expression be relegated to the realm of clichés rather than reliant upon it, or drawn from it, or whatever?

OK, that’s enough. The polar ice-cap is melting as I fritter my days away critiquing a text generated by a machine. In sum: ChatGPT struggles with figurative language, struggles with irony, struggles with tonal variation. Who doesn’t? You don’t ask ChatGPT for good writing, though, you ask it for writing that meets the requirements of a specific context and it does the best it can. You generate a text, scoff at the glitches, edit it, weep, and get on with your day (or so I’m told).

I set up an alert for editor jobs when I moved to Canada. This year a hefty proportion of the jobs that land in my inbox involve training generative AI tools. These are jobs that seek to make redundant writers and editors, especially those who work in much more lucrative fields than publishing. They’re flexible jobs for Busy Moms like me, which means you work at home on your own computer with limited training. Most of them offer zero hours contracts with no benefits. They’re not just bullshit jobs in David Graeber’s sense of the pointlessness and harm of the work being obscured, they’re shit jobs from a labour rights perspective, in that they offer bad working conditions, little security, and no recourse to unionisation. And, haha, they’re jobs that require the worker to contribute to the shrinking of not just their own professional prospects, but those of everyone else in their field.

Generative AI currently poses an existential threat, as they say, to professional writers who work in rule-bound contexts, many of whom don’t have writer in their job title. I’m thinking of solicitors who draft contracts, authors of technical manuals, marketing and communications experts who put together sales strategies, anyone writing a strategic plan, a change strategy, a strategic vision or othersuch. These writers aren’t paid to sound like themselves, they’re paid to generate impersonal or denotative texts specific to context and to write in a way that restricts semantic overflow.

To be an AI trainer is to provide feedback that helps chatbots formulate rules about how language works in context. Manuals of rhetoric and composition exist, style guides too, but there are no fixed rules about metaphor and irony, about how to leverage the internal contradictions of language. When I prompted the AI to add sarcasm to its paragraph on bad style, the whole thing fell apart. This will probably change.

ChatGPT might not get metaphor at the moment but it’s tremendously significant to all writers, even those who do get metaphor and work with it. It’s a total failure of solidarity (and insufferably pompous) to section off literary practice with a sigh of relief that the tech hasn’t yet figured out how to sound plausibly literary. First they come for the digital strategists and the marketing consultants, then they come for the journalists - and eventually they’ll come for the poets. Perhaps humans can learn from ChatGPT, in the sense that the texts it generates present cartoonish versions of common human errors and reveal cliché. It’s easier to anatomise the bad style of an AI-generated passage because the errors are outlandish and no one gets offended. This is one of few redeeming features of a technology that is otherwise frightening. (Cyborg that I am, I’m suspicious of reflexive technophobia and tried to approach ChatGPT in a playful spirit, as Sheila Heti did when she generated her Eliza texts in 2022. I fed the bot a bunch of prompts on topics I find difficult to write about in the interest of finding shortcuts, and some daft prompts, and came to the conclusion that this was not my kind of fun, and that for me, ChatGPT-enabled shortcuts led to deadends.)

The most immediate and widely reported danger to literature is copyright infringement, as LLMs continue to scrape and pillage without permission or payment, ingesting books and articles whole in the hope of one day sounding more human. The excellent term ScabGPT captures the threat to labour rights and I am hoping to learn more at this event on AI and creative labour tomorrow. The Australian Society of Authors provides a summary of related issues:

We know that generative AI will reduce supplementary sources of income and skills-building opportunities for authors and artists such as copywriting and concepting art jobs. We are also concerned that inferior AI-generated content will flood the market, making discoverability harder for professional writers and lowering quality for consumers.

Writers of course are also humans who live on planet Earth, and hence will also have to reckon one day with the environmental costs of AI, with the concealed consequences of a technology that seems so quick and cheap as to be supernatural.

Right now, ChatGPT is a poor stylist and that is partly because the writing it generates is neither relational nor material. We can’t locate a writing presence in the passages above, we can’t hear a voice. There’s no there there, there’s no one (or two or three) there, no Coleridgean imagination to stick it all together. Instead of a subjectivity formed by acculturation and socialisation, a machine, an assemblage of algorithms, mines a dataset. And when it is prompted to do so, no one is left waiting. Editors and bosses don’t have to wait for copy – and there’s no writer sending off crabby emails as they wait to get paid. It feels like an understatement to call this industrialised writing

A bloviating imperative - writers, write better! – can’t fix bad style in this context. Abusive pedagogy is neither effective nor ethical, and such instructions are not much help to generative AI either. We might do better to consider the headline structural reasons for bad style in an industry now further pressurised by the astounding speed of ChatGPT: short deadlines, low pay, cost of living, copyright infractions, sales imperatives, ever-depleting editorial resources. If bad work conditions produce bad writing, AI has rapidly transformed the conditions in which writers work – for the worse.

There’s long been a tension between the literary aspiration to be a fine stylist (or to publish fine writing) and the economic motivation to make (or enforce) deadlines; and this tension was heightened in the digital publishing economy of the first decades of the 21st century, a period which saw the transformation of all cultural forms into content. Generative AI, with its capacity to produce texts to order on the spot, snaps the string. If we are to complain about bad style, and god knows I often do, it’s not helpful to recirculate the same tropes of lazy authors, not enough editors, criticism gone to pot, indifferent readers etc etc. We need to relate our complaints about style, about writing that seems rushed and unfinished, to the institutions which are publishing bad work, which ask writers and everybody else in the industry to deliver more for less, which make using chatbots seem appealing. Instead of worrying about whether this means the death of the author, AI should prompt the humans who generate texts, aka writers, media and publishing workers, to talk more about their work conditions, about authorship and textual production in the generative AI era, and about what kinds of collective action we can take. It’s not that humans can’t write texts that are irredeemably, unfixably bad - it’s that generative AI, even right now in its early iterations, can do so much faster, and won’t ask to get paid.

I often listen to The Garret podcast, hosted by Astrid Edwards. Over the years she’s spoken with a ton of Australian writers about their work and careers – and she interviewed me in February about Critic Swallows Book. Our conversation roamed all over the place, and I really appreciated Astrid’s questions about credentials, and thought in particular about that part of our exchange a lot after we spoke. Did Astrid make me perform any weird impromptu stunts? Nope. Did I ramble on? For sure. Listen to our conversation here.

This week New York magazine launches a limited run newsletter devoted to… language, grammar and style, edited by the magazine’s copy chief Carl Rosen. The promo text promises that it will deal with thorny copy questions. Writes Rosen, ‘The greatest thrill of my job is diving into imperfect copy — I hate clean copy; the messier the better — and helping the voice of the writer come through.’ It sounds good to me. Rosen’s phrasing indicates the pointlessness of editing ChatGPT: there’s no writerly voice to coax into view.

I did feed one last prompt to ChatGPT: Write this paragraph in the style of Catriona Menzies-Pike.

Amidst the ceaseless churn of contemporary non-fiction, a curious phenomenon has taken root: the creeping sameness of style. Picture, if you will, a landscape devoid of the vibrant hues of originality, where prose flows in a monotonous stream, each piece indistinguishable from the last. Gone are the days of daring experimentation and linguistic acrobatics. Instead, we find ourselves adrift in a sea of conformity, where every article, every essay, marches in lockstep towards a horizon of bland uniformity. How quaint, how utterly charming, to witness the death of nuance in favor of formulaic drudgery. Truly, we are living in a golden age of literary homogeneity.

What am I doing here? Je n’en sais rien! Here’s the Infra Dig About page.

Thanks for reading Infra Dig! Subscribe for free to receive new posts and support my work.