Electoralism can wait, I want to get back to talking trash on LLMs. I think I’ve figured out the appropriate way to handle things.

※

This informal series — I haven’t created a category for “Contraslop”, though I probably should — is where I’m “thinking aloud” (or, rather, “in public”) about how I’m going to handle my LLM policy next semester, because I had issues with it last semester.

“Contra-” very obviously means “against”, but I’m using “slop” to refer to the outputs of so-called AI systems. A lot of commentators I enjoy listening to (Ed Zitron, Robert Evans, the folks over on Trashfuture, those over at This Machine Kills) have gravitated to this term, though it also has the connotation of being “content” in general.

I’m probably going to discuss the other side of what I’m talking about today at a later point — that being the reduction of communication into content. However, for the moment, I want you to simply understand what the title of the series means “contraslop” means “against LLM writing”. It began as a joke, but I’m going to maintain this terminology and insist on it: I’m not resisting the inclusion of a helpful technology. I’m refusing to sit idly by while people fill my environment with slop.

I halfway feel that we’re going to look at the release of LLM-generated text and images into our informational environment as something like the detonation of nuclear weapons or the overuse of lead. Or microplastics. [Image of industrial waste barrels, uploaded to Wikimedia commons by Robert Ashworth, and used under a Creative Commons Attribution 2.0 Generic license.]

※

Occasionally, I’ll have a student who incidentally has missed one or two of these Q&A days — who are baffled when they learn eight weeks into the semester about a fundamental aspect of the class that they have simply not engaged with.

This might be a bit of a long walk, but come with me on this.

Monthly, in my classes, I have a “Q&A Day”. For these sessions, I plan nothing, and require each student to write down a question for me to do a deep dive on. The goal here is fairly simple: a lot of students come to the conclusion that asking questions, far from being a sign of engagement, is a sign of a lack of proficiency with the material. They believe that, if they ask a question, they are signaling to their peers (and perhaps to me? It’s unclear) that they don’t understand, while everyone else does. By making it a requirement that everyone ask a question, this belief is gradually ground down and they develop the habit of asking questions.

I put very few limits on this – I make it clear that, while I will prioritize questions related to class material, almost anything is on the table. We’ll go over skills and knowledge necessary for the class first, and then I’ll spiral out to things related to English in general, to college in general, and I’ll even answer personal questions that don’t cross important boundaries.

This has cause some difficulty for me in the past – and I say “difficulty” instead of “problems” for a very particular reason – such as when, in the aftermath of a young black man being shot in the vicinity of the college at which I taught I had to grapple with a student asking, openly and honestly, why bad things happen to good people.

I did my best with that one, but acknowledged first and foremost that nothing I could say there would be satisfying.

It is my belief that this sort of openness and willingness to grapple with difficult questions is necessary.

The responsible party for this, I assume.

But last semester, I was asked by a student what my favorite “trope” in stories was – when I was confused by the nature of this question, I was helpfully given a few examples by students: “enemies-to-lovers” was a popular one, as were various “AUs” (“alternate universes”, where a set of characters are translated over into a new setting.)

One could easily look at these as being fairly common outside of fanfiction, though they’re commonly associated with that practice. After all, Romeo and Juliet could easily be held to be an “enemies-to-lovers” story, and Star Trek: Deep Space Nine was originally conceived of as (basically) a science fiction AU to the old TV show The Rifleman.

However, it seems to me that reading stories because of the tropes involved is sort of like going to the grocery store and leaving with nothing but cucumbers, avocados, aloe vera juice, mint gum, and garishly colored sheet cake simply because each thing you picked up happens to be green. It’s not necessarily something that should never happen, but setting out to make it happen is a strange criteria.

Behold: French toast [Egg and Cheese breakfast sandwich on rye bread, uploaded by Robert Loescher to wikimedia commons and used under a Creative Commons Attribution-Share Alike 4.0 license.]

This is the result of what I have come to think of as the “Decompositive Move”. That is, breaking things down into their basic components and assuming that everything that is produced by this move is equivalent to the assembled desired end result simply because the components are the same: as if french toast and a breakfast sandwich are identical simply because they both contain bread, eggs, and dairy.

Enough food metaphors – it’s been cool enough around here that I can cook, can you tell?

I call this a “move” because it’s not quite an aesthetic stance, not quite an epistemological stance, not quite an ethical stance, but has effects in each of these domains. It stems, I believe, from the preponderance of largely user-facing metadata electronically presented to us: when you try to find something on a streaming service, you navigate to the category you want to watch, and then pile on additional descriptors until you’ve narrowed the hundred-thousand options down to about a dozen, which is about five minutes after your dinner’s gotten cold. Navigating Twitter, Tumblr, and Instagram, you can find content by picking an appropriate hashtag and searching for it.

These are signifiers that you select to open a field of electronic objects, and then you can survey it and pick exactly what you want: it seems obvious that, after you have been doing this for a while, you find yourself searching for a particular tag more than others, and begin to identify the tag not as a commonality between the things selected but as a self-existent thing that you can develop a fetishistic relationship to. So, when asked to express a preference for something, you don’t end up naming a film or a book or game or whatever, you name the tag.

As other people engage in the same thing – our platforms are designed to make this kind of relationship more common – we begin to exchange information about the tags that we like and build communities related to them. This is a key step in the Decompositve Move, but it isn’t finalized until we make the leap to breaking things down specifically to fit them into these tags.

This results in a kind of “folk structuralism”, as exemplified by TV Tropes and similar: it reifies patterns of story events in a way that allows this fetishistic relationship – and encourages amateur writers to approach their art as assembling these fetishized patterns, eschewing invention for simple recombination.

Please note: On reflection, I’ve done this sort of thing in the past – saying “I want to write something like X but with Y element emphasized or added.” I think it’s a dead end, though, because it never really results in anything that I managed to complete, while things that occur to me through other channels have been possible to complete. Maybe this isn’t the case for everyone, but most creatives I’ve met shy away from the recombinative method.

Okay, you may say, this could all be true, but where’s the LLM policy?

I’m getting there.

※

This is a strongly atomistic way of looking at cultural objects: if all the components are present, then the object is present (think that bit from Alan Moore’s Watchmen, where Doctor Manhattan explains that there are the same number of atoms in a dead human as in a live one – this is atomism taken to a parodic, nihilistic extreme.)

An image of “The Engine”, from *Gulliver’s Travels*, part 3, chapter 5: a series of blocks with words upon them, used by the scholars of the Academy of Projectors in Lagado to produce random strings of words. By turning the cranks, a random string of text is created. This is basically an analog LLM.

LLMs are among other things predicated on an atomistic logic of language. Because they simply work by predicting the next symbol in a sequence based on a large repository of statistical data, they cannot actually create a functional top-level structure except incidentally.

This is the same issue as the Decompositive Move: it only recognizes the elementary, first-order components of what’s being examined. It ignores the second-, third-, fourth-, fifth-order components that are generated not simply by combining the first order components but by relationships between them.

I don’t much like Romeo and Juliet, but we can recognize it as a skillfully-executed bit of writing: the poetry doesn’t just stem from combining the dissolute, womanizing Romeo with the virginal and dutiful Juliet in an enemies-to-lovers structure. We also have to consider the relationships between each of the characters, and then the relationship of outsiders to those relationships, and consider this as motive forces. Some characters are simply devices — waldos extended from the brute machinery of plot to make the experience trend in a particular direction — but all of the other characters are representative of people enmeshed in a larger, more dynamic matrix of interrelations. We have to understand the characters as components of a larger structure, the play, which is an apparatus built from language that takes recognizable forms and attention and produces catharsis.

There is a reason that the red pen is treated as an attribute of the teacher, much as the calculator is an attribute of the accountant or the stethoscope is an attribute of the doctor: we are assumed to be simply correcting errors in specific locations — and larger marks that indicate more pervasive problems are often treated as a death sentence by students.

If we focus on the micro level – spelling and grammar, among other things – we create the conditions on the ground for LLM-derived writing to flourish. This is given a certain amount of lip service in lower levels and at college, but the fact of the matter is that a lot of responding to student writing centers around identifying points in the student-produced text where it isn’t working as effectively as it can and explaining to the student why it isn’t working as well as it could be – which often looks like hunting for errors.

Still, there is a pressure to respond to particular moments of a student’s text because it isn’t acceptable to forgo response, and some students interpret this the same way – as hunting for errors. Part of the job of teaching is responding in such a way that the student comes to understand that the commentary is not referring to here, but is referring to a deeper structure that became visible here — the way that a malady of the nerves can be seen in a twitching in the extremities: you don’t fix it by operating on the fingers, you fix it by addressing the pervasive issue in the parasympathetic nervous system, which will most likely require a less focused intervention.

The error-reduction approach is an atomistic orientation towards writing, and one that almost exclusively produces bullshit in the Frankfurtian sense (i.e., a communication that is unconcerned with its own truth value.) This is because it gives a separate criteria to aim for: we needn’t concern ourselves with truth or falsehood, because we have a tightly defined set of criteria for judging it as skillful versus unskillful.

Not this kind of holistic. The basic idea is that “holistic” is an antonym of “atomistic” — emphasizing the whole over the parts. This is what we strive to do as writing teachers, though it is sometimes difficult to do this, as issues in need of correction are often in particular locations. [image taken, wikimedia commons, and used under a Creative Commons 2.5 license.]

We need to define a particular orientation to writing that counteracts this. Let’s call it the “Holistic Orientation”. Judging writing by Holistic standards would not over-privilege the line-by-line in relation to the top level concerns. Given the nature of the job (teaching citation and genre conventions) we can’t abandon the line-by-line completely, but the way that writing is responded to needs to make it clear that this is simply a vehicle for the top-level ideas of the writing.

This is in-line with how I attempt to teach, and how I know many of my colleagues attempt to teach. What this will require isn’t a wholesale replacement of the curriculum, but a change in how it is communicated to students.

The problem here, as in other power relations, is likewise one of bullshit: I have had students disregard clearly stated class policy because it ran counter to their expectations, i.e., because they assumed that I was bullshitting them about what the rules were. Specifically: while I have harsh penalties for plagiarism, I have no penalties for late work — my hope is that this will lead to students who would otherwise commit plagiarism to simply take an extra few days to actually do the work. The student I’m thinking of (who I will not provide more information about for obvious reasons) assumed that while there was no explicit penalty, there would be an implicit penalty for being late — i.e., that I would justify grading [them] down through other means but that it would stem from them being late, which set up a bit of social calculus: is plagiarism a bigger problem or is the [imaginary] threat of a late penalty? Meanwhile, my actual self is glad when things are a bit more spread out, because it means that I can get everything submitted to me graded before the next class and I can still have part of a real weekend.

This leads to a number of issues: these students have very clearly had their trust in people in my position broken – not necessarily because earlier teachers have intentionally misled them, it could easily have been because of exigencies outside of those teacher’s control. The end result, though, is the same: my students come into the class expecting that there will be a certain amount of deception. They expect that I am communicating not how the class is but how I wish for the class to be perceived to be.

Hence a difficult issue: I don’t want to encourage them not to question authority, but I need to get them to trust me in certain domains.

Of course, I’m going to bring this back to tabletop games, and use it to snipe at the market dominance of D&D, which builds this kind of adversarial relationship into its play structure as a result of the way it’s marketed. [image is from Turn2538, and used under a Creative Commons Attribution-Share Alike 4.0 license.]

This is a lot like playing tabletop role-playing games: occasionally, when running a game, your players behave in a way that looks erratic from the outside because they’re nervous about falling into a trap that you’ve set for them: they think that you’re going to randomly change the rules on them to their detriment. The rookie approach is to bribe them into compliance (which doesn’t work), or to give them an actual difficulty to fixate on (which is, at best, a short term fix.)

There isn’t really a quick fix, because the most stable way to handle things is to behave in a consistent, open way that aligns with what has been communicated (generally, I allow myself a bit of inconsistency in students’ favor, but I’m walking that back as time goes on. It doesn’t tend to yield the desired changes in behavior.) This takes time, and doesn’t always feel worth it, due to the length of semesters: unfortunately, it’s the only way to relate to students. The only way to secure trust is to be worthy of trust.

Which is all well and good, but how do I get them to take my policies seriously in the first six weeks of class or so?

I can’t – I can just lay the groundwork for the long-term fix, and apply the policy as written when it comes up.

※

The primary change, in preparation, is going to come in a different orientation towards how the rubric is constructed. My principle concern, thus far, has come in making sure that what is graded is what is visible on the page – a shift that has been very friendly, thus far, to the Decompositive Move. It’s the atomistic approach to writing. Sure, I grade the thesis of a paper based on how well it controls the rest of the paper, and I tend to focus my “grammatical” grading on whether the idea makes it across or not instead of the presence or absence of X number of errors, but these are gestures at a Holistic Orientation.

Instead, I might opt to emphasize things that sound more “abstract” on the surface – placing a premium on coherence and harmony over other potential values. Perhaps the gold standard will not be simply that a paper has a well-developed thesis statement, but that it communicates a consistent and developing viewpoint throughout it without slipping into pure opinion – which should, as a second-order effect, generate a well-developed thesis.

Classes are still nearly a month away, and I’m beginning to assemble material for my classes, but I will be experimenting with this as I move into the semester. I’ll report back around midterms, and refine my strategy further.

※

If you enjoyed reading this, consider following our writing staff on Bluesky, where you can find Cameron and Edgar. Just in case you didn’t know, we also have a Facebook fan page, which you can follow if you’d like regular updates and a bookshop where you can buy the books we review and reference (while supporting a coalition of local bookshops all over the United States.) We are also restarting our Tumblr, which you can follow here.

The Decompositive Move (Contraslop, Part 3)