Slop Articles

Mostly every post on this blog was written with some AI-assistance. If you scroll to the footer of any piece here, you'll see who wrote it and who co-wrote it — I built that disclosure into the site itself.

So what's the difference between this and slop?

There's a thin line. I think it's a clear one. But it's thin, and the sloppiest writing in the world right now lives just barely on the wrong side of it.

The filter that broke

For most of the history of writing, the best filter we had was that producing the thing was work. Sitting down to draft a few thousand words was expensive enough that, somewhere in the back of your head, you'd ask: is this worth it? And the act of sitting down — gathering the thought, testing it against itself, deciding to commit — was inseparable from the cost of producing the words. You couldn't really do one without the other. The production cost did the verification cost's job for you.

It's a bit like a cryptographic proof of work — the cost is what validate the value.

AI flipped it. Production is now cheap. The implicit equation is eh, just generate it, what's the harm. The harm is that the question "is this worth saying" stops happening at all, because the cost that used to force it isn't there anymore. The judgment call got automated away. The slop is what shows up in the gap.

With the cost gone, the question doesn't ask itself anymore. We have to actually ask it — otherwise we may be just adding more slop to the burning pile of slop.

Is there any thought at the core?

Slop, the way I mean it, isn't AI-shaped writing. It's words without thoughts.

I'm not talking about wordplay, or puns, or the particular pleasure of someone who clearly enjoys a sentence — they might not come with a big idea but still with many thoughts. That kind of writing can be an absolute delight to read, AI-assisted or not. I'm talking about the genre of blurbs that shows up when somebody has prompted a model with "please write an article about [thing]" and pasted the output. The prompt itself wasn't an original thought. Nothing in the chain — from the question to the framing to the conclusion — required a human to have actually sat with anything.

It doesn't have to be an original thought, as in the first person ever to think it. Ideas co-evolve. People arrive at the same insight independently all the time, especially when they're swimming in the same currents.

What it needs is a thought you actually had. One you sat with long enough to notice was there. Not a thought-shaped object generated by something else.

That alone is a pretty thin filter, though. We all have plenty of thoughts on any given day — at least I hope so. So the bar can't just be "did a human have a thought somewhere upstream of this?"

Can you vouche for it

One step up: is somebody willing to put their name on it? Not in the trivial byline sense. In the sense of: if you push on this, I will defend it.

You shouldn't put your name on something you haven't read. You shouldn't make an argument you don't feel comfortable defending. If somebody challenges a claim of yours in a comment thread and you can't speak to it, you didn't write it. You laundered it.

There's a useful distinction here between automating writing and automating thinking. We've been automating writing for a long time. Calligraphy to typewriter to keyboard to dictation. Each step mechanized the production without touching the composition. The thinking and the deciding-what-needs-to-be-said stayed with the human. You can use Whisper to transcribe a voice memo and feed it into an LLM to clean it up — and if the thought was yours and you vouch for the result, you've automated the writing, not the thinking.

The slopification happens when the thinking gets automated and the writing comes out the other end without any care.

Then there is nobody on the behind it and no one to push back at. Just a piece in the void without any meaning what so ever.

Why I care

The feedback I got on my writing, during school, was certainly well meant. But the way it landed was: you're stupid, you're nothing. And that voice followed me for a long time. For years it made writing feel like an exposure exercise, and the exposure was always going to confirm the same thing.

So I avoided writing, esp. sharing my writing. And that despite me loving words. I constantly read... Scientific papers, philosophy, fiction, programming books, whatever's around. Always have. I have plenty of thoughts about what I read, and for most of my life the friction of writing was high enough that the thoughts mostly stayed in my head, or in private notes, or in conversations.

What AI changed for me, specifically, is the friction at the surface — my worry that someone will call me out for typos and other small mechanical things enforcing the feedback from back whin i was in school. AI helps me take care of that.

The risk is that lower friction means less filtering. The opportunity is that lower friction means more of the thoughts that should have been written down all along finally get written down.

How I write most of the aricles

It starts with me talking. Almost every piece begins as a voice memo, walking through the different angles of an idea out loud. I'm not reading from notes. I'm thinking. The point of the voice memo isn't to draft; it's to find out what I actually think.

I've always taken notes ¹ — on whatever I'm reading, whatever I'm turning over — and kept most of the papers and books organized in Zotero². So now I can feed those notes, with the proper bibliography and sometimes the original works, alongside the voice memo, into an AI and automate the first draft. Then I go back and forth — manual edits, model edits, more manual edits — until it sounds right. Sometimes that takes just hours. Sometimes it takes days. But it always starts and ends with me.

There's no piece on this site that came out of me typing "please generate and artile about [thing]" Every one of them starts with the what I'm thinking about, how they connect to other thaughts, ideas and arguemnts.

I want to show my work. I want the proof of work — the part that's still mine to do — to be visible enough that if you don't trust me, you can at least see what I did.

Does it actually add anything

Which brings me back to that thin line between slop and something of value. A genuine thought at the core, a name vouching for it — you can have all that and the result can still be slop. The process was never the deciding factor. Care is. And care starts with the honest question you have to ask on purpose: Does this add anything?

That's the test I keep coming back to. Is there something here someone can feel, learn, or push back on? A claim, a framing, a choice — something that isn't already the consensus? Or is the piece smooth precisely because it doesn't risk anything — because it averages over what's already out there and hands you the median back?

Worth saying: this kind of slop predates AI. Handcrafted slop has always existed — like the LinkedIn post that strung together truisms. It wasn't invented by language models. What AI changed is the cost. It's much easier to produce this kind of replication at scale now. The filter that used to catch the worst of it — work — is gone. So the question, has to be asked deliberately. Otherwise it doesn't get asked at all.

If this post had been purely generated rather than written, you wouldn't be able to tell from the polish.

You'd be able to tell because there'd be nothing here to disagree with, and nothing that stands for who I am.

References