Is clanker writing too perfect?Is clanker writing too perfect?
There is certainly something about AI writing that we don't like. Here's a fairly good description of what it feels like:
And beyond the cadence of words, the text itself often feels intelligent before intelligence has actually entered the room. And when you read enough of it, it begins to feel off. It's not wrong, exactly but feels empty or even vapid. Every sentence arrives with a feeling of precision that is just too precise.
I don't think I'd use the phrase "too precise." It's more that any given AI-generated sentence is like a costume, it's not the real thing, but sometimes it's hard to tell what gives us that sense. Nosta claims that it feels to polished and that may be a good description.
But if you are using AI-generated writing in your own work, maybe you don't want it to be that way. You want to sound a little imperfect. This isn't anything new. People have been saying this about AI writing for quite some time: so Nosta says you have to stick some imperfections in there to hide the clanker:
The writer isn't revising anymore, the writer is unwriting...Today's task is to remove techno-fluency and insert the imperfections into our prose.
We don't need clankers with beauty marksWe don't need clankers with beauty marks
I don't think "adding imperfections" makes wirting better at all. As much as people like to say it, it's not "imperfections" that mark our humanity and make our writing interesting. Traveling further down this road, Nosta says the reason we don't like clanker writing is because we can't find the human presence in it.
So, what the hell is going on here? And in this context, what is writing actually for? Maybe writing was never just a transfer of information. Maybe we were always searching for signs of our human presence. After all, we don't only read for conclusions, we read for evidence that someone was actually there and doing real thinking.
Conclusions are like towers. They often need a lot of groundwork before they can stand. While we may not read for the towers, I think we do read for the groundwork.
I don't care about the writerI don't care about the writer
You and I can see the same thing and come to very different conclusions. It's natural that this is the case: we've lived different lives, afterall, and been shaped by different experiences and different DNA and different weathers.
When writing works well, it can alter our mindset so that we see the world from some different viewpoint without having lived the life to get to that viewpoint. This is an electrifying feeling: suddenly seeing things differently because the writing has created within us a new vantage from which we can see the world -- often without us even realizing it.
I don't read things to find "signs of human presence" -- there are plenty of dumb humans in whose presence I'd rather not be -- but I do read things to be able to have my mindset shifted.
Writers think about how to do this, and reading their writing is the experience of our mind forcibly being moved. Clankers don't write this way. They don't care about bringing you to a new place -- because there is no place they inhabit.
The only text where I don't mind reading AI text is when I (or my students) ask it to improve a human-written draft of a physics paper. It's really good at this. Adopting the hedged, super structured, style expected in the community. No surprises, just clear and simple statements of fact or reasoning.
But outside that, I just hate it. Two main reasons I can think of.
First, I despise blatant laziness. At least, try to hide it. Create an illusion of effort. But AI is too transparent in the lack of effort.
And the other reason, it's just always the same cadence. The same literary sentence structures. The beautiful ones you want to read once or twice in a good book. But not at every paragraph.
And as soon as I notice actual slop, the one that sounds good but is meaningless when you think about it, I just stop reading. And then I'm pissed at myself and at the author for having wasted my time.
Yes. This feeling resonates. I get so angry that I fell for it. That I thought there was going to be an interesting idea or thought in something and the slop-writer managed to trick me. It feels like I myself have somehow been taken down a notch. Feels bad.
“If it takes more time for me to read it than it took for you to write it, I don’t wanna read it”
It’s like posh laziness combined with making up shit on the spot.
As if someone was more worried about sounding formal than actually thinking too hard about the subject, while also having it in one go using a pen.
Proper human text production usually implies some sort of review/interaction process, like I can go back and fix some mistaeks on this comment box, while LLMs are literally autocompleting the next word without much foresight, oversight or any sense of direction.
Same applies for generative “art”, the machine can parrot the techniques but it lacks the bigger picture (pun intended).
AI and bullshitters sound alike, probably because both often are “confidently wrong”.
It annoys the fuck of me when a hooman pulls stories out of their ass, the same applies for clankers.
The sterile feeling is just noise distracting from the superficial and sometimes made up takes.
I think this will change once LLMs are capable of sensing its limits and no longer sound like confident idiots.
I think it's the lack, not of errors, but of authenticity and style in the generic output. I think you can fine-tune a bot to imitate a style, but even then, it won't be surprising. This is because the model is bottom line static, and most if not all variance increases hallucination rather than actual "style".
That's a good way of describing it.
I'm still confused and mystified by why AI writing is so unpleasant to read. And it is surprisingly hard to actually put a finger on what is bothersome about it.
Imagine you wanna get paid with zero effort, zero intellectual investment, zero anything.
Got it?
Now you know the type of person the owner of the bot is that generated the stuff you're reading is. So you can be pretty sure that it's gonna be not worth your time.
But there are also examples of people (misguided people, I believe) who put tons of time into "research" or "writing" with their llm and produce a piece of writing that is still slop. Maybe they think they have come to some bright idea or something, but it's not. It's just slop.
Here are some examples:
I think the authors in both cases spent a lot of effort on these things...or they think they did. But they are both slop. Just absolute wastes of time.
I don't think these people intended to produce slop; I think they didn't know any better. They seem to have spent hundreds of hours on their projects and fully convinced themselves that what they produced is meaningful.
It's still slop. It's horribly frustrating to read. I guess humans have been doing this for a long time: I remember reading some feminist, structuralist literary theory in university that was pretty awful...but even that didn't feel like slop. There's something else going on.
Okay. I actually messed up, because I should have said "zero attention to detail" - or at least close to it.
I'm of the unpopular opinion that "shipping" is overrated. It's in my opinion fomo to ship early if you didn't do severe quality control. This makes me slow af, but I avoid debt this way: do the work now, so that you don't have to do it later. Works except for when you have no idea what you're doing.
When Theranos went down, I was hoping we'd see the end of fake-it-till-make-it. But humanity is addicted to this. Both on the producing and the receiving end. We need to have dreams then get rekt while living that dream so we can move on to the next dream. Substance... is gone.
Much of the AI use we see is unsubstantial. The cases where it is substantial, you won't see, because it was just one cog in a giant machine. And that's what we're dealing with. Some noob that has no fucking clue what they're doing so they're letting an LLM do it for them. If they had an idea and/or actual standards, you'd never see the slop.
</end_of_boomer_rant>Good writing isn’t just information transfer. You can usually feel when someone has actually wrestled with an idea.