You may have heard that some while back (what is time), the Commonwealth Prize in fiction was awarded to a dense assembly of words by the name of “The Serpent in the Grove”, which was widely accused of being (1) extremely not good, and (2) the output of an LLM slop machine. This isn’t an essay about the Commonwealth Prize, the merits of the story, and the ensuing fracas, but I’m going to have to discuss the background a bit to get where I’m going. The whole affair, combining as it does anxiety about LLM output and its impact on the creative industries with a particular kind of BR Myers-ish “this isn’t even legible, what the hell is going on” has occasioned some great invective, a genre of which I am fond—one delectable sample, which details the Commonwealth Prize situation in passing, is Sam Kriss’s “If you let AI do your writing, I will come to your house and kill you”. (Don’t go into that essay unless you’re ready to read someone being mean in a fun way—it’s about a 50 on the CC Lemon scale.)

I read a bit of the story; I didn’t get far. I was struck, though, by one image in it, early on, which I’d seen quoted by a number of folks (including Kriss) as an example of obviously bad and obviously LLM writing. A seductive woman walks into the bar:

“They called her Zoongie. Maybe it was a name; maybe rain took a shape and decided to keep it. She had the kind of walking that made benches become men.”

Now, I don’t have patience for “Maybe it was a name; maybe rain took a shape and decided to keep it.” Maybe it was a name? I don’t know, man, you’re the criminal mastermind here, not me. Rain taking a shape—we were talking about her name, her name’s the question, now we’re talking about a shape, shapes aren’t sounds and aren’t words, and even if they were Zoongie isn’t even a particularly rain sound—maybe it could be the sound water makes when it’s going out a gutter in a storm? I guess?

But “She had the kind of walking that made benches become men,” as I thought about that phrase it kinda had something, in a “that cloud looks like a Star Destroyer” way. At first it seemed like nonsense. But sitting with it, I imagined an outdoor restaurant, a barbecue joint, say, on a sleepy afternoon, it’s hot and it’s humid and everyone’s drowsing off, bellies full. Rows of benches, men half-asleep there, a general horizontality to the whole affair. A particular sort of woman walks by in a particular way, and with a Tex Avery cartoon’s SPROING sound effect all of a sudden the men are alert and attentive. Or perhaps there’s an empty room and this woman walks in and all of a sudden it’s full of guys. It’s not the world’s best image, it’s not clear what it’s describing, but if you go with it you can find something that fits in the grand tradition of trying to write “hubba hubba” without having to turn in your Serious Writer Card. (I’ve always been partial to the bit in Brothers K where Dimitri describes Grushenka as having “a supple curve all over her body. You can see it in her little foot, even in her little toe.” Whew!)

Now here’s the catch. I was only able to make something out of that phrase by approaching it with a rather generous spirit: by asking myself what sense the writer could be trying to make out of the words on the page. And that judgment depends on many beliefs, among them that there is a writer and that there might be sense to be made. Those judgments are context-dependent, and our context is shifting.

In the game of go there’s this concept called haengma (I’m told this is Korean for “moving horse”), the tension between the “speed” of a movement on the gridline board and the solidity of the connection between the stone you’ve just played and the stones nearest to it. In go you’re trying to build walls: stones with connections that your opponent can’t sever. If you have two stones next to one another, they’re directly connected—but if you only make directly connected moves, you’re left inching around the board in a game where the goal is surround territory as quickly as possible. A strong opponent will eat you alive. Often you’re looking to make your moves as fast (far apart, loose) as possible—while still being able to connect your stones directly when needed. The kind of haengma you use also depends on the context: on how aggressive is your opponent, how likely to cut, what’s the greater board state? (If you’ve never played, I get that this is tricky to visualize. I’d draw diagrams if I had time. If you’d like some concrete examples, check out the article on Sensei’s Library. I’ll wait.)

Haengma is part of what’s going on in writing imagery, I think—only there we’re dealing with a sort of haengma of word and concept. “Blue” and “sky”, in English, are concepts so close they’re practically touching. Nobody will scoff at “the blue sky,” but you’re not getting anywhere fast. (This is much of the objection to cliche, that you’re just wasting moves.) But we read “the dreadful lemon sky,” to steal a John D MacDonald title, and we think: are lemons dreadful? Are skies lemon-colored? Is a lemon-colored sky dreadful? And we don’t have to root around for long in our mental attic before we hit on one of those awful summer storm afternoons, the lowering light and dust, the shadowless ground, the sense of wrongness creeping into everything. By making that jump, by collaborating with the writer to explore the connections between the words, we’ve landed ourselves in a vivid sense memory.

Now, a funny thing about this, is that it gives you a way to fake a kind of juice in writing: you can slap contrasting concepts and words together without any sense of how they connect, then put on your straightest possible face and wait for a willing reader to do the work for you. Often this isn’t sustainable at length, but it can work over short distances, or in spaces where there’s a lot of emphasis on the value of “getting it,” of drawing meaning from difficult texts. It’s hard to be the only person in the room “not getting it” so there’s often some pressure to make up an “it” to get. (Of course, the opposite also happens: easy enough to say “my five year old could do this” without the slightest understanding of what “this” is. But that’s another issue for another essay.)

This kind of play, of course, only works if people aren’t quick to wonder if they’re being played. So I’m wondering, now, about the effect LLMs will have on prose imagery. LLMs don’t need to be able to produce good original prose for them to cause readers to regard prose in general with, say, 10% more suspicion. I’ve already become vastly less interested in anonymous online text, in reddit product reviews, in comments. Some folks are veering away from emdashes. (You’ll pry them from my cold dead—well, you get the idea.) I’ve even seen a few folks claim they’re giving up three-item lists, a surrender of rhetorical territory which I find shortsighted, immoral, and low-fiber. I wonder if we’ll start to see writers deploying a slightly… tighter prose haengma in response to heightened reader suspicion, flaunting their interiority, their clarity, the coherence of their schema. I wonder how we would detect this at a corpus level? If only there were a way to conceive and model the multidimensional vector space of language itself. Maybe you could use a neural network? That would be really something! Definitely a neat, nerdy tool that wouldn’t cause anybody any problems.

Anyway! Hope all’s well with you and yours. Take care of one another, friends. Keep fighting. Work for the liberation of all sentient beings. I’ll have something fun for you next week.