Why wank wins

LIMy 2025 article An essay on wank proved to be startlingly widely read. It made its way around a large part of the internet, sparked quite a lot of discussion and still garners quite a few readers, even now. It was, to my mind, not too bad an article, but as is quite often the case with the first inceptions of an idea, a bit scattershot and often not very tight argumentatively. What this is to say is that there are a bunch of gaps in the article which, as of April 2026, are kind of annoying me. There is also considerable demand for more analysis on wank theory, and as the quantity of wank that we deal with has only increased since I wrote the last article, extending the argument and making it more rigorous seems like it might be a good idea.

One thing that went neglected in the initial article that we should now address is "why is wank an effective discursive tactic?" After all, people engaging in wank are generally annoying, abrasive, make people feel coerced and unduly pressured and all in all are the kind of people whom you wouldn't invite to dinner parties. The initial explanation I gave in the article was that it was a form of self-soothing, and heaven knows that explains a lot, but it doesn't explain why wank is an effective way of altering the behaviour of other people. After all, there are a lot of self-soothing behaviours (drinking heavily, getting stoned, racism, misogyny, general histrionics) that other people generally don't accept unless they're either desperate or equally terrible, and yet wank is not only tolerated, it works for getting people into positions of power in much the same ways that lies or bullshit do.

Allied to that is the lack of a clear mechanism for how wank functions. With both lies and bullshit, there are clear ways in which lying or bullshitting translates to personal advantage: a lie might get people to act in ways that they otherwise wouldn't that benefit you, bullshit helps you build social status and gain the approval of your superiors and the fact that it can be misleading in its own right doesn't hurt either. With wank, it's unclear how you go from being very attached to a claim that you're making regardless of its truth to gaining the kind of social status that purveyors of wank usually collect.

This article aims to build on the ideas in An essay on wank, trying to understand what social and cultural levers wank works on and why it's effective as a form of bad-faith speech in this moment. However, to do that, we first need to discuss how we as a society read text these days. So, without further ado, let's have a look at:

Two ways of interpreting text

It's mostly a commonplace these days to say that we exist in a post-literate society: people tend to listen to podcasts or watch short-form video (or other forms of audio or visual media) more than they read, and even when they do read, they tend not to read very long or complex texts. They certainly don't write. This is sufficiently visible from basic observation that I don't much feel the need to defend it, but for much the same reason it's not very interesting. There is, however, a corollary to this that is interesting: as this change has occurred, the way in which many people interpret and interact with text has changed. In this discussion, I don't necessarily mean textual as in "written down" here: a politician's speech, for example, is presented in such a setting and in such a way that it's interpreted more like a written text than like colloquial conversation, even though it's spoken in medium. Rather, when I say text I mean something along the lines of "a body of words that's expected to stand and be interpreted alone as a cogent whole": a fairly specific sub-case of the concept of text in literary theory. In any case, having stated what I'm talking about, I should probably now explain what I observe to have changed.

The usual, long-respected way of interpreting textual communication is to interpret the semantic meaning of the text by breaking it up into sentences and interpreting each sentence according to the rules of the language in which the text is posed. We might call this, say, the Grammatical interpretation. The Grammatical interpretation makes meaning from text through two channels: the first is the bare meaning of what the semantic meaning of each sentence in the text is and how those meanings relate to each other, which is then augmented by richer side-channel information, often mediated by social knowledge. Particular choices of words with certain connotations (the "butt dial" vs "booty call" dichotomy), tone of voice, facial expression, even the positioning of where and when a given thing is said can all modify the bare meaning of the sentence beyond recognition. Nonetheless, all such interpretations are in a sense anchored to the bare meaning as communicated in the grammatically structured sentences.

The Grammatical interpretation of text is by far the most traditional one, and can admit great complexity in communication: you can express anything from the Theory of Relativity to Alexander Pope's Essay on Criticism and have confidence that with the Grammatical interpretation, even if some of the fine details are missed or people lose some context, they'll get the general gist of what you're saying. For the most part (a few unpleasant experiments aside), it's also what we attempt to teach in school. However, it has its disadvantages: Grammatical interpretation is mentally costly, takes quite a lot of training and worst of all, it can lead to you hearing strong arguments for things that you don't like and that might even make you feel hurt. It's no wonder, then, that having been enabled to develop a new form of textual interpretation.

In addition to this traditional mode of interpretation, then, we now have a new one which, for reasons which will soon become clear, I'm calling the Bayesian interpretation. The Bayesian interpretation isn't completely divorced from semantics, but deals with them oddly. For one thing, the Bayesian interpretation doesn't really pay much mind to grammar and treats a text as a cluster of nouns and noun phrases with no particular order to it. It also tends to change verbs (it doesn't ignore them entirely, but it tends to turn them into nouns when it can) and then relies entirely on nouns to derive a meaning. Finally, while the Bayesian interpretation can generally interpret the semantics of individual referents pretty well (in the sense that if you say "Horse", the Bayesian and the Grammatical interpretations will align on what class of object in the world is being referred to: we're not going full Wittgensteinian language-game here), other semantic content goes out the window, being replaced in the Bayesian interpretation by the sentiments attached to said referents in the mind of the speaker or audience.

In this sense, the Bayesian interpretation engages with a text as though it's a painting or a video rather than a sequential document where the grammar of the language is important. To use an analogy closer to what my readers are used to, it's also very similar to what happens in a Naive Bayes classifier (see? There's the name), which is a machine learning model used as an early, though still remarkably effective, form of email spam filter. A Naive Bayes spam filter treats an email as a bag of words without reference to the order they're in or any sentences: it simply marks each word with the probability that an email with that word in it would be spam. "Viagra", for example, might have quite a high conditional probability of the email being spam associated with it, while "anaphylaxis" would likely have a very low probability. The classifier uses Bayes' rule to combine all of these probabilities and arrive at a final score for whether an email is spam or not. Importantly, this flattens all of the meaning of the text into a spam/not spam categorisation and loses anything that might be conveyed by grammar, so false positives are, in practice, fairly common.

While the Bayesian interpretation of text is more subtle than this, it has much of the same flattening effect. Complex chains of reasoning are lost, verbs are often lost along with context as to what's actually being done and the overall meaning of texts or statements often gets flattened into an agree/disagree dichotomy based on sentiment matching: does the sentiment of the text match the sentiments I associate with the individual referent words in my bag of noun phrases. This isn't always bad: for example, there are certain combinations of words that will get me to immediately dismiss a text ("protect white women" or "1488" comes to mind in this case: if they're being spoken about with positive valence, I'll usually discard the text outright immediately, and "human biodiversity" will have a similar effect). We all do it, and usually it's simply as a first-pass filter to get rid of anything that simply isn't worth our time. Using this to interpret texts that you should on the face of it be willing to engage with in good faith, however, is far from ideal.

To make the problem clearer, let's examine how a non-trivial text might be differently interpreted depending on whether the person in question is using the Grammatical or the Bayesian interpretation. Our sample text is a fairly simple statement with some political loading as follows:

Depleted uranium munitions are commonly used by NATO armies, as the high density of the metal allows for the production of highly effective armour-piercing munitions. The use of these munitions is, however, often criticised by anti-war and anti-nuclear groups, who view them as being radioactive and akin to nuclear weapons.

Depleted uranium is not, however, a radiological hazard, as the half-life of U-238, at about 4.47 Billion years, means that the element decays very slowly and almost all of the U-235 and U-234 (which have much shorter half-lives) has been removed from the metal. The negative health effects of exposure to depleted uranium dust primarily come about because depleted uranium is a heavy metal and is thus chemically toxic to the human body.

Claiming that depleted uranium munitions are equivalent to nuclear weapons, then, is unhelpful from the perspective of an anti-war movement. They can be criticised on their own merits, of course, but saying things that are facially untrue damages our credibility among the people whom we need to persuade. We would, on the whole, be better off arguing for us to avoid idiotic military engagements entirely.

Reading this using the Grammatical interpretation would derive the following points as the bare meaning of the text:

• People often talk about depleted uranium munitions as though they're nuclear weapons
• Depleted uranium isn't in fact a nuclear weapon, or even a radiological hazard
• Inasmuch as depleted uranium is harmful to human health, it's because it's a heavy metal and thus poisonous in the usual way

Therefore;

• Fixating on depleted uranium as a nuclear hazard is a mistake on the part of the anti-war and anti-nuclear movements
• Saying things that aren't true undermines our credibility
• Rather than calling for action against depleted uranium munitions, we should aim to prevent senseless military engagements that might use them

This is a fairly specific set of claims about depleted uranium munitions. In addition to the bare meaning, the voice used and the exact language communicate a bit extra: the writer, for example, is sympathetic to the anti-war and anti-nuclear movement but might well disagree with some of their more hardline positions, thinking them unreasonable. The author also uses fairly precise military language, implying that she's somewhat familiar with the military and munitions and not as hostile to it as she could be. This is a reasonable set of positions, and can be dealt with in a reasonable way, also using the Grammatical interpretation. A peace activist using the Grammatical interpretation might agree and change their communications, or they might present a counter-argument: depleted uranium is a by-product of enrichment, for example, so even if it in itself isn't a nuclear weapon, opposing it because it's still inextricably linked to the nuclear industry might make some sense as a tactic. Either way, a productive discussion about the Grammatically-encoded meaning of the text can be had.

The meaning extracted from the Bayesian interpretation however, might look like this (this is for the case where the reader doesn't like or disagrees with the position: there's another Bayesian reading where the person agrees, but for the sake of space we won't unpack that one):

Depleted Uranium + Nato (negative sentiment) - criticised by (negative sentiment) - anti-war and anti-nuclear (positive sentiment) - negation - radiological hazard (negative sentiment) (author disagrees with my stated sentiments) - negation - nuclear weapon (negative sentiment) - falsehood (claim that earlier statement is false) - damages credibility

This parsing of the text is then condensed into "Text contradicts the sentiment attached to the referents in this text, therefore author disagrees, therefore author is wrong", leading to a Bayesian interpretation of the text as "sentiment disagrees with what I feel about these points, therefore the text is wrong". This isn't something that you can really engage with on a Grammatical level, as the interpretation is rooted in the emotional valence associated with the nouns in the sentence: the way the original text signifies the idea that the author cares about their cause and wants to help them do it better so they're pointing out a mistake doesn't come through in the Bayesian reading because it relies on the grammatical structure of the text. To be clear, as I've already mentioned we all do this as a first pass on most texts we write, and in many cases having the Bayesian interpretation flag something as wrong and throwing it out is the correct move. I don't think, for example, that there's any moral duty to give eugenicist screeds anything more than the most cursory Bayesian interpretation. It cannot, however, be the only tool you ever use, and if you never actually work through the Grammatical reading of some texts, you are shooting yourself in the foot.

If this distinction looks a lot like Kahneman's distinction between System 1 and System 2 thinking, you're not wrong there at all: the parallels are striking. The main reason I make the distinction is because, as we'll soon see, a lot of people these days apply the Bayesian interpretation to everything while otherwise outwardly using System 2 for analysis. This is particularly common in political spaces, hence the need for the distinction.

Having set out our definitions, we can now observe that despite how much of a problem using the Bayesian interpretation for everything is, a striking number of people in our society function entirely in the Bayesian mode. There are likely a number of reasons for this: first of all, the general declining state of literacy, coupled with our abortive experiments in three-cuing as an educational technique (it's striking that, as a method for teaching reading, three-cuing hews disturbingly close to teaching people to look for the Bayesian interpretation) predispose people to do this. This is doubtless exacerbated by the fact that we're all inundated with text and faintly overwhelmed by information these days, and the Bayesian read is a lot less cognitively demanding than the Grammatical one. Finally, a lot of us tend to consume as much audio and video content as we do long-form text these days, and those media tend to push more towards a Bayesian interpretation of text than a Grammatical one. Either way, the fact is that a significant proportion of people in the world now are basically stuck in the Bayesian mode.

How dare you say we piss on the poor?

So, what does all this have to do with wank (and with bad-faith communications in text in general)? My initial essay, while popular and clearly striking on a concept with clear applicability, was a first-pass analysis and thus had some noticeable gaps in it. To fill those gaps and create a better understanding of what wank actually is and why it works, additional analysis is needed.

The first unexplained point, as mentioned at the start of this article, is that it leaves it unclear why wank as a form of bad-faith communication actually works: after all it's annoying, upsetting and has neither the ability that lies have to actually make people believe things that are untrue nor the ability of bullshit to build on social affinity and garner social status (it can do a little of the second thing, but it's nowhere near as effective at it as bullshit is). Looking at it through the naive lens, wank shouldn't be socially effective at actually getting results for the person pushing it. And yet we see wank being socially effective and working all the time, and thus it being prevalent. There has to be a mechanism, therefore, that we're not catching.

The second unexplained point is why wank has become so much more prevalent recently: it does, after all, feel like a rather novel form of bad-faith speech compared to lies or bullshit. With no explanation for this, the whole structure seems rather odd, so some idea of why it's prevalent now when earlier in history it was a less useful tactic would be useful. Finally, it was noted indirectly that wank has a more coercive or uncomfortable texture than other forms of bad-faith speech often do: it feels as though you're actively being pushed to do something rather than just to believe or go along with something. This is an interesting fact, as it makes this form of bad-faith speech rather distinct from other other varieties, but there's not necessarily an explanation in the first piece for why this is the case: a notable lacuna that we should probably look to fill.

To explain this, let's examine our forms of bad-faith speech from my previous essay through the lens of how they expect text to be interpreted: as a reminder, these are lies, bullshit and wank. Lies and lie-type deception are, in general, the form of bad-faith speech that we expect when everyone's functioning in the Grammatical mode. In general, a liar aims to communicate factual information that just happens to... not be the case in the case of the lie being in the bare meaning of the text, or to give a false impression of the situation in the side-channel communication while saying things that are strictly true in the bare meaning. Once again, we can use Colin Powell lying about Iraq having weapons of mass destruction as an example here: the Grammatical-mode reading of what he said was precisely "Iraq has weapons of mass destruction". Furthermore, the side-channel communication aimed to give the strong impression that Iraq had or was about to get nuclear weapons specifically. This was all communicated fairly straightforwardly under the Grammatical reading and didn't rely much on sentiment at all: it just happened to not be true, and not be true in ways that benefited the US administration at the time in their efforts to justify launching a ground invasion of Iraq. This, in general, is how those of us who read long essays expect deception to work: Grammatical-to-Grammatical, just false.

Bullshit, by contrast, is Bayesian-to-Bayesian communication. In bullshit of all flavours, meaning isn't encoded in what you can infer from the bare meaning of the words as interpreted through the grammar of a language. Nor is it in the side-channel communication: rather, it communicates in terms of the Bayesian sentiment-process from one person to the other. In the LLM space, for example, much of the bullshit communication comes from the sentiment association (AI, Innovation, Wealth, Good): you can more or less say any sentence containing words like "AI", "Innovation", "Disruption", "Transformation" and so forth and communicate what you want to other people who function in the same way: there's no obligation for your sentence to actually encode any meaning in the Grammatical mode, because the main thing being communicated is "I think that AI and Innovation are Good Things and we need to do them, and the people criticising it want Bad Things instead of Good Things". "Make America Great Again" encodes its information in a similar way: in the Grammatical interpretation, the statement is actually meaningless (or rather, under-determined: we don't actually know what "America" means in this sentence, nor what it would mean to make it Great Again), and all the actual content comes from sentiment. Given the increasing prevalence of people who function entirely in the Bayesian mode, it makes sense, perhaps, that we've seen a like increase in the overall prevalence of bullshit in the world: after all, the people who demand that communication says something valid in the Grammatical interpretation are few and far between and can mostly be ignored. This isn't always bad: after all, as well as bullshit, small talk and phatic conversation of the type that we use for social bonding fall into this category as well, and if you insist on everything that's said having a grammatically-encoded communicative payload, you will not be much fun at parties. That said, I'm not sure that this is the way to run countries or build nuclear reactors, so I think there's some value to perhaps stopping this from happening so much. But neither lies nor bullshit are what we're here to discuss in the end.

And so, to wank. Contrary to either lies or bullshit, wank tends to function mostly in situations where the proper method of interpretation to use is contested: spaces such as politics, the more serious kinds of workplace or NGO, or anywhere where people are motivated by strong emotional commitments to a cause or a value set and aim to do something about it. This allows tension between the Bayesian and the Grammatical interpretations of text to form, because both sentiment-based Bayesian reads and the more mechanistic Grammatical reads are considered valid and useful: it's important that you say and communicate things about your cause that make your community feel the right way, but at the same time there's pressure to communicate effectively and clearly as to what is actually the case. For example, in the peace movement there's consistent pressure to communicate that you think that war and weapons are bad and shouldn't exist, but at the same time you need to be able to effectively know and communicate about what these weapons actually are and what they do if you want to be effective: as before, it does no good to say something materially false about depleted uranium if you want to be able to effectively persuade people. This can create significant stress in such organisations, especially because people, being people, are almost always status conscious. In an environment where both types of interpretation are admissible, the people who can only effectively use the Bayesian mode are thus at a disadvantage in terms of status games. In that situation, they have two options. The healthy one is to get good at Grammatical-interpretation communication work, but that's hard. The unhealthy one, though, is to coerce people into using the Bayesian interpretation at all times, even when the Grammatical one would make more sense.

And would you know it? That process looks remarkably wank-like. One might argue, even, that wank is precisely that: an attempt to coerce people into admitting the Bayesian interpretation of a text over the Grammatical one, even when the Grammatical reading is more useful. This captures our earlier definition of wank as a claim that's expected to be treated as though it's factual but handled with the delicacy of a deep emotional commitment, and adds a theoretical underpinning of why it works rather than people just doing it for no reason or for self-soothing purposes: what wank speech is doing is using emotional coercion to demand that people use a form of interpretation that's convenient to the person speaking wank.

To go back to our depleted uranium example, we can get an excellent model of how wank develops structurally. Suppose that I'm working with you on writing the platform of an anti-war movement and the subject of depleted uranium comes up. I see the words "uranium", think "nukes" and then say that we should add a point demanding a ban on all depleted uranium munitions or ships armed with them entering New Zealand. At this point, you might, knowing about depleted uranium and its actual applications, point out that this is an unrealistic and unreasonable demand that doesn't achieve what we're working for, putting me into a bit of a bind. I can accept what you have to say and take the hit to my status that comes from saying something silly. But suppose I don't want to? Well, in that case, I can count on the fact that what you wrote about depleted uranium has terrible Bayesian vibes: it suggests that uranium isn't actually that bad and that we shouldn't worry about it. And besides, you used the term "munitions", which isn't how we talk about military things: maybe you're not as on-side with this whole project as we thought you were. The play for me is obvious: I get offended and act as though you're personally hurt and wronged me, loudly demanding that the Bayesian interpretation of what you've written is the only correct one, at which point you might well withdraw the statement and agree for the sake of getting along. If you don't, I can go tell the rest of the group that you think uranium's actually fine, relying on the fact that much of the group will adopt the Bayesian interpretation and those who don't will shut up to stay a part of the group, and they'll most likely line up behind me, either expelling you from the group or marginalising you within it. I manage to boost my status, get the language I want into the platform, and I get to protect my feelings and not admit that I was wrong about the uranium: in fact, everyone will agree that I'm right about it being a radiological hazard in order to avoid any more messes. This lines up really well with the definition proposed in the last article: the claim about the properties of depleted uranium is expected to be treated as materially true because it has the right vibes, but if challenged it's treated as though the challenger doesn't share your deeply held values and in fact believes them to be wrong.

This line of attack is usually used in groups where people are generally expected to have similar values and similar sentiments about words and the things in the world that they refer to, and in this cases wank can actually be a very effective form of coercion. After all, we want to be in favour of peace and safety for people, and we don't like nuclear weapons: being told that if we don't agree with a given statement that we don't actually hold these positions could easily hurt enough to get people to go along with something that they know to be wrong. The threat of social exclusion and ostracism that comes with that makes it even worse: if you know that other people will believe those things about you if you don't assent to the Bayesian reading of the claim, that's an extremely strong incentive to go along with it however false it might be in the Grammatical reading. Given that fact, it's quite obvious that wank represents a pretty good tool for getting people to assent to interpretations of text and thus political and other positions that are convenient to you: it can, for example, be an excellent way to get a group to adopt what Olúfẹ́mi O. Táíwò calls "deference epistemology" even when that form of epistemology is overall bad for the group: I can, in the right spaces, leverage my marginal status as a trans woman to get groups to adopt positions completely unrelated to transness if I get strategically upset in the right way. In short, if you're in a space where you can reasonably assume that people share enough of your sentiments about certain things, wank is in fact a very powerful coercive tool.

This mechanism behind wank also explains neatly the increase in prevalence of wank over the last few decades. In a situation where it's assumed that the reading of a text will be Grammatical (with a greater or lesser degree of skill behind it), it's difficult to generate wank: there's no Bayesian interpretation based on sentiment to push towards. Even in places such as theological debate that you'd expect to see a lot of wank occur, people instead took part in serious debates about such subjects as "what is a hypostasis and how many of them does Christ have?" in the understanding that the words meant something and that it was important that what was said was true. With the increasing prevalence of the Bayesian reading of texts, though, wank became increasingly viable inasmuch as pushing for the Bayesian reading rather than the Grammatical one became something that might actually be valuable to a person.

Of course, if it turns out that some people fundamentally don't share the emotional valences that you associate with certain referents, you might try to use wank to coerce people into accepting the Bayesian reading of a given text even though it's unlikely to work. When this fails, though, functioning in the Bayesian mode as you are, you're unlikely to recognise this and might well double down on your immediate demands, not registering that they're ineffective. You might even point other people who do share your valences at the other person, telling them that this person is evil and needs to be crushed. This dynamic happens in a lot of places; to restrict myself to just one example, I think this is the reason that a lot of social media dogpiles or "discourse" occurs: people read something in a Bayesian mode and try and coerce others into adopting it because it's comfortable, not registering that other people don't share the referent-related sentiment that they need to make the coercion actually stick. The other side then either sticks relentlessly to the Grammatical interpretation they were using immediately or falls into their own Bayesian interpretation with their own set of sentiments. And then everything catches fire.

In any case, this covers all of the points that I named as being missing in the initial essay. We now have a mechanism by which wank works and an explanation for why it actually works and why people use it. We've explained the coercive elements behind it, and we have a fairly strong understanding of what happens when the attempts at coercion fail. So, what does this do to a society where wank is prevalent?

The consequences

The consequences of wank in this framing are generally similar to the ones that I've discussed in the first article: allowing emotions to cloud your view of the actual situation on the ground is dangerous and always has been. However, with this framing and an understanding of the mechanism behind the effectiveness of wank as a discursive tool, there are two other dangerous consequences that need to be discussed in depth.

To start with, the fact that wank is discursively effective has a fairly harsh flattening effect on texts that people produce in general. If I assume, as once was the case, that people tend to read texts using the Grammatical interpretation (albeit maybe with greater or lesser degrees of proficiency depending on the situation), I can write to be understood in that way: I have, consequently, relative freedom when it comes to what words I use, what constructions I use and how I write more generally. If I'm writing for an audience where I know that a non-negligible number of readers will be leaning on the Bayesian interpretation, though, I have to write in the knowledge that my writing is going to be misinterpreted in some very strange ways. Writing with an eye to misinterpretation is clunky and often leads to turgid prose: to give one example, we might think about the current spate of gutter antisemitism that seems to be increasingly socially prevalent. In a piece of writing written with a mind to Grammatical-mode readers, this is fairly clear-cut: there is no good reason to burn down a synagogue or shoot people at religious events and nothing more needs to be said expect to say that it's sad that one of our oldest bigotries is still very much alive among us. In the Bayesian reading, however, there very much exists an (Antisemitism, Jews, Israel, Palestine) association, and when someone reads a piece that I've written about antisemitism, it is going to end up being about Israel and Palestine in their eyes no matter how irrelevant it is to the actual situation. This means that at a minimum I need to interrupt the flow of my work with disclaimers about how this doesn't mean that Israel's war crimes in Gaza/the October 7th pogroms that Hamas indulged in were at all acceptable (depending on which set of Bayesian readers I expect to be a problem): this makes my writing much more muddled, longer, and draws attention away from the Jewish communities in the diaspora that I want to be safe and who maybe deserve to have the full, centred attention of the community occasionally without reference to a country on the other side of the world that they have little political say in the governance of. At worst, I might just not write the article at all, and while I'm sure a lot of people would be happy about that, this isn't universalisable: if nobody writes anything for fear of the Bayesians, we will live in a poorer world that we understand a whole lot less well.

The second thing is that in an environment where some people are only capable of seeing the Bayesian interpretation of a text and an even wider group of people are being coerced or deceived into admitting that interpretation even when a Grammatical one is available and makes more sense, having certain forms of knowledge becomes suspect (and thus inadmissible) in itself. I, for example, know quite a bit about petroleum engineering (I'm friends with a few petroleum engineers and did several papers on it back in engineering school). This means that, as it happens, I know some things about oil and drilling for it. However, that means that I'm going to use language that's different from and more specific than someone ignorant: I might talk about things like "well-head pressure", oil being "heavy" or "sour" and what consequences this might have for its usability. But here's the thing: "moderately sour light Arabian crude" has a very different emotional or sentimental valence from "oil". The second is the kind of thing that a none-too-thoughtful leftist might say "America attacked Venezuela and kidnapped Maduro for oil" has a total sentiment that feels right. "Venezuelan crude is heavy and sour and largely unusable by American oil processors" feels as though it has the wrong valence and marks you out as an industry insider who is insufficiently against oil production, and thus a statistical association between "using language that marks you out as knowing stuff" and "being on the wrong side" forms. Even if you're on-side in the general sense and the Grammatical interpretation of what you've written contains sensible and useful information, the language used and the sign that you know something has the wrong vibes and invalidates the statement in a Bayesian sense. The end consequence is that in a space where wank has taken root, only people who know nothing about certain subjects are held to be qualified to talk about it.

This is an obvious danger, as it means that organisations who've formed this kind of epistemology become increasingly unable to respond to the actual facts on the ground. This is particularly glaring in LLM/AI discourse: in a lot of corporate spaces, knowledge of LLM tools and their limitations will mark you out as "having the wrong vibes", even if you're relatively positive about the technology: speaking about inference, statistical distributions and so forth means almost definitionally that you aren't the kind of LLM maximalist that's popular at the moment. This means that the people setting policy almost definitionally wind up being the ones that know basically nothing about the tools: they're easily persuaded by performance that even a more informed enthusiast will dismiss, and when trying to encourage use of the tools they'll do things like set token quotas for workers that simply make no sense to speak of. In short, they make bad policy that gets them in trouble.

New interventions

Fortunately, if this mechanism is correct, this isn't all bad news. Understanding that wank is coercing people into interpreting text in the Bayesian mode rather than in the Grammatical one opens up a number of new interventions that we can take to reduce the overall prevalence of wank and improve the overall quality of our thinking in the process. Specifically, if we can undercut the ability of people to function entirely in the Bayesian mode, wank becomes that much harder to sustain as an effective discursive tactic.

The first and most obvious intervention is that knowing that this is a possibility, we can nudge people towards the Grammatical interpretation of texts rather than the Bayesian one. In cases where someone has clearly admitted the Bayesian interpretation of a text, gently asking "Aside from what you feel about this, what does the text actually say?" in an individual or a group environment could actually be quite effective at shifting the tone of a discussion or a discourse. In the best case scenario, the person trying to push wank might actually re-read the text and reassess, and even when they don't, asking the question gives people social permission to actually look at the Grammatical interpretation of a text at issue, even when they don't necessarily agree with it. Getting to the point where we can actually close-read a text without it necessarily triggering a serious emotional response would be no small thing.

Similarly, when producing texts for other people to read, we can deliberately nudge them towards engaging with it in the Grammatical manner rather than the Bayesian one. Even something so simple as mentioning at the start of a text that the text is fairly dense and would benefit from reading closely, or stating your intent to improve what an organisation is doing rather than simply say that it's wrong can help people engage with the text in the way that you'd like them to rather than the way that they're instinctively drawn to. Even on social media, where trying to encourage people to stop and actually read before they respond is difficult, these kinds of nudges could help a lot.

On a social level, interventions pushing improved literacy could do a lot to help. This is probably something that we should be doing anyway given the somewhat parlous state of literacy in the world at the moment and how important it is for general human function, but it would also help reduce the amount of wank we have to field. Literacy-favouring interventions are relatively cheap, we know how to do them effectively and they're implementable without a great deal of state or corporate support: in short, we should be investing in them in volume. In professional settings, formal training in reading and writing would be well-worth investing in and would help reduce overall levels of wank a lot, leading to better decisions.

Finally, pointing out when people make the kind of rhetorical move described in this article and thus generate wank is, when it's safe, an important thing to be able to do. Wank relies on not being noticed as such to be effective: if you can actively point out "hey, this person is blatantly misreading this text and is trying to push you to do the same because the misreading's better for them", that is beneficial to us. Obviously this isn't always safe: if you're supporting a family and your job's at stake, well you might choose to be quiet. On the whole, however, it's often safer to challenge people pulling this shit than you might think, and having it pointed out in the open robs it of much of its power.

In general, we don't want to be interpreting most texts on the basis of what emotions are associated with the words being used. This is a bad way of making decisions and leads to serious mistakes being made. Wank, however, is an extremely effective coercive tool for making that happen no matter how bad an idea it is, and has consequently become extremely prevalent in this world of ours. If we're going to be serious about building a better world, we need to push back against it. Hard.