The Image of Text

Leftist memes became notorious for illegible dense paragraph blocks that vainly attempt to deconstruct capitalism in a single deep-fried jpeg to the point where the self-referential and self-deprecating ‘leftist meme’ itself emerged. However, this phenomenon is not exclusive to the left. Trump’s tweets (technically his Truth Social posts), for example, typically balloon into the length of a short op-ed. Recently, I met a guy called Tom in a London pub rocking a Ted Kaczynski manifesto T-shirt. The same day I passed by an art performance where an artist, in a flow state, was sharpie-scrawling little words onto every surface of a room. Two days later, a friend served me the ever-dreaded ‘just had to get this off my chest’ blue iMessage screen-sized bricked-up wall of text.

I feel the walls of text closing in on me. Walls of text are those bodies of text we have tacitly agreed that we don’t have to read. It is not just leftist memes, or the ‘leftist meme’, it is Terms & Conditions, celebrity apology Notes app screenshots, Instagram Story essays, annual reports, IRS mail, medicine usage leaflets, Twitter threads, and all that feels like an imposition on our time that we’d rather spend watching Love Island. TL;DR, cool story bro, I ain’t reading all that, don’t worry kitten are some of the ubiquitous reflexes we’ve developed to the text wall visual cue – even ChatGPT has a built-in TL;DR feature that hedges societal collapse.1 

The wall of text has appeared across epochs: from Roman Acta Diurna and Dharani sutra scrolls, to the European pamphlet wars of the 17th century, Egypt’s 2013-14 Tamarod movement, the list is endless.2 But if we zoom in on the leftist meme, there’s one fascinatingly close parallel: the ‘dazibao’. Dazibao literally translates to ‘big character poster’. They were handwritten wall posters plastered in public spaces before but mostly during the Chinese Cultural Revolution by Maoist students known as the Red Guards. 

The messages inscribed on those walls of text contributed directly to many violent acts, but their formal power lied in the sheer saturation of walls with dense text: a flood that collapsed nuance into an overwhelming presence. They transformed Beijing squares into one big IRL leftist meme.

Why do we do this? Why am I writing this? Well, sorry to bring Lacan into this, but both the dazibao and the leftist meme emerged from the same structural lack: the impossibility of total explanation, or the desire to master ‘the Real’. This piece and its writer suffer the same affliction. If we compile stacks of every wall of text ever produced – the truth still won’t be whole. Scriptures know this well. At least the self-aware leftist meme acknowledges this impossibility within its ironic mirror folds. As Lacan states:

I always speak the truth. Not the whole truth, because there’s no way, to say it all. Saying it all is literally impossible: words fail. Yet it’s through this very impossibility that the truth holds onto the real.3

Such glaring futility tips the scales towards the form of the wall of text. As an image, or even a vibe, blocks of text are usually afforded more sympathy in popular culture. The negative association of an imposed reading exercise like Google’s Terms & Conditions’s nine-hundredth update is almost the opposite of how we, at least the meme-literate, feel about ‘copypastas’.4 Copypastas turned textual overload into a participatory joke where the only punchline was its form. The copypasta trolls set them up as pranks that are not supposed to be read, but I bet they loved writing them just as much as the Red Guard Maoists loved writing their dazibaos. The wall of text is also known for attributing cultural coolness, which brands love to exploit when they want to hit the ‘seem smart’ KPI – as can be seen, for example, in the packaging design and advertising of Hinge, Vetements, or Dr Bronner’s Magic Soap. After all there’s a reason Tom was proudly flexing his Ted Kaczynski manifesto T-shirt. Text blocks hold gravitas that some conflate with status. And when people do stop and read, it’s a rare intimate act of attention. 

From ancient scrolls to the live feed doomscroll, the meaning of the words in a wall of text is flattened once it achieves its aesthetic effect: an exegesis on Gramsci’s Prison Notebooks and Ariana Grande’s apology post for licking doughnuts are rendered equivalent; the textual mass performs a visual function before it is parsed for meaning. Love it or loathe it, the wall of text is an aesthetic unit before it’s a semantic one. It is the quintessential image of text.

Attempts to further psychoanalyse wall-of-text authors beyond the fantasy of totality will undoubtedly yield a wondrous buffet of psychic cravings: a pedagogical superego that feels a need to educate, a moralist anti-reductionist commitment against market co-option, a paranoid-schizoid fear of being deliberately misconstrued, sublimated aggression – a list that could be its own wall of text. I am however interested in one root function that is evident in most walls of text: labour. 

Wall texts function as proof-of-work, demonstrating the ability to produce a mass of it. Volume serves as a test, a sign of endurance, commitment, sincerity, a rite of passage. High density signals seriousness and stakes. This is where the idea of a ‘wall’ really comes into play. Density, mass, volume are filters that separate and regulate access, performing a dual gatekeeping function. The amount of writing that you have produced determines your ‘earned’ authority matrix. 

Historically, the human labour of producing text persisted throughout various technological paradigms, even when proxies of its value shifted. In the age of hand-written manuscripts value was firmly tied to irreproducibility; the mass reproduction of text by the printing press devalued scarcity; the photocopier amplified circulation value which the internet later evolved into virality value. Now we’re on the precipice of a new technological paradigm, where value accrues to the form of text.

I vaguely remember an English proverb which I will not bother verifying that is along the lines of ‘One man’s trash is another man’s treasure’, and while walls of text might be the trash we ritually skip over, they can be invaluable to our new ontological cousins: LLMs. 

LLMs are the final nail in the coffin for the labour value of textual mass – as its synthetic mass can now be conjured on demand in a matter of seconds. They are basically lorem ipsum on steroids. But when it comes to textual form, new opportunities arise. In ‘The Death of the Author’, Barthes famously proposes to consider literature (or writing) as a multi-dimensional space in which there is no ultimate meaning. As he writes:

In the multiplicity of writing, everything is to be disentangled, nothing deciphered; the structure can be followed, ‘run’ (like the thread of a stocking) at every point and at every level, but there is nothing beneath: the space of writing is to be ranged over, not pierced; writing ceaselessly posits meaning ceaselessly to evaporate it, carrying out a systematic exemption of meaning.5 

This passage captures how dense walls of text – like a dazibao or a leftist meme – signal presence and affect without engaging meaning. This is basically how an LLM interprets text: not by understanding, but by tracing surface structure – patterns of form, rhythm, style – because beneath the writing, for the machine, there is no meaning to pierce, only structures to model.

This is why LLM output represents a fascinating instantiation of the image of text. When an LLM produces an answer, it is merely predicting the next most likely word in a sequence, based only on patterns it has seen in training. Its ‘intelligence’ is entirely formal, not semantic. 

LLMs are so vibes-based that they’re inherently incapable of understanding or internalising meaning. This is why they are much better at poetry than storytelling. Their reliance on patterns — alliteration, assonance, rhyme — contribute more to the quality of a poem than narrative progression or logical coherence. What we’re used to seeing as dull content machines, are ironically much more form-first artisans.

To LLMs, walls of text can be delicious nutrient blocks. LLMs don’t care about virality or performance metrics as much as they care about token abundance, syntactic structure, formatting and other aspects of form.6 This explains the LLM em dash being an AI tell discourse; it’s a marker of ‘literary’ or ‘editorial’ text which AI engineers regard as ‘high quality’ datasets for their training corpora.

What makes LLMs an inflection point of equal weight to the printing press, photocopier etc. is that they decree humans are no longer the sole interpreters of text. Machines have already ‘read’ – or, as Barthes put it, ‘ranged over’ – more text in the past decade than we will ever be able to. Crawlers are automated bots that sweep the open web and harvest it for LLMs to feed on as ginormous training corpuses. 

This digital porridge is the nourishment that determines LLM utility, and since meaning is never part of this pipeline, the formal attributes are what distinguish the junk from the protein shake. Walls of text, given coherent syntactic structure, make for good protein shakes. If you ever felt like ChatGPT sounds like a Redditor, it’s because Reddit, Wikipedia, StackExchange, Quora, and the like, are where crawlers are sent to collect those savory well-punctuated text blocks for training data.

It is unfortunate that this choice to prioritise the boring editorial style over schizoid misspelled ramblings was made for the LLM by some engineers and corporate strategists, but it’s what’s needed for mainstream utility, also known as ‘what’s good for business’. 

For the past decade social media platforms have kept the inner workings of their feed algorithms purposively opaque – a black box immune to accountability. Now AI oligarchs are doing the same, but for their LLM training data. By veiling training datasets, AI companies evade transparency for any programmed biases they encode that will, like social feeds, shape publics. Elon Musk hilariously exemplifies the extent of such control across both eras: first, when he was exposed tweaking Twitter’s algorithm to show his tweets to more people; second, with Grok’s white genocide and MechaHitler scandals.7 Others may not be as narcissistically sloppy as Elon, but even that level of sloppy can apparently get away with it.

Under the guise of legal and licensing concerns, AI companies are abandoning open web scraping in favor of private, licensed corpora purchased through opaque contracts with publishers and data brokers. This is where they face a dilemma: feed your LLM too little and it becomes too sterile and hallucinatory, feed it too much and you start losing control. They’re boxed in a Sterile<>Chaotic spectrum; where utility is a subjective sweet spot somewhere in the middle. 

Ultimately, AI companies will still have to partially rely on the open web for the foreseeable future – as their crawls yield about 30 trillion tokens compared to the largest licensed private corpora’s measly 240 billion tokens. The sheer volume of open web text makes it formally and structurally indispensable for training LLMs that balance generality, cultural nuance, and linguistic variance. These are basic utility functions for profitable user adoption. Only human textual mass provides the entropy needed to keep models alive, at least for now. The wall of text saves the day again (of course not). 

Dazibaos and leftist memes were never really about the message — they were about flooding public space with text as an event. Street posters, pamphlet stalls, and the social feed had their public square moment. The new medium is the training dataset. Gone are the days of comments section arguments; we now write for crawlers if we want to influence the public(s). Our walls of text will no longer fear the human tl;dr, the machine will eat it all up and ask for seconds. This is the quiet violence and quiet opportunity of our moment: the image of text is becoming a lever of power.

Just as Luther nailed his theses to the church door, we now nail ours to Wikipedia Talk pages, Substack posts, StackExchange threads; we smother it in the blandest encyclopedia sauce and serve it cold to the substrate of machine memory, then pray for it to reappear later as the impending authority of a ChatGPT response.