I think the best objection to LLM writing goes "If I wanted to know what an LLM thought I would ask one." Everyone with $20/month to spare already has all the LLM commentary they want and there is no need to show your audience more.
Unless they're bad at prompting or just aren't going to go run the prompt. I'm considering writing an article where the actual point are the prompts, and if a human reads the prompts and takes only the prompts away then that's actually just fine, but where LLM output might clarify as a cheat sheet if someone finds the prompts inscrutable. I'm not sure how this will be received, I suspect the allergy to LLM writing is so strong that even doing that will annoy people and get strong downvotes. Maybe I'll only include the prompts. But I think my user preferences prompt produces better output than other people will get if they rerun the prompts, and I'm not going to share my user preferences publicly.
I guess it will be fine if you directly describe what you're doing and you make it easy to only read the prompts / skim the LLM outputs. Because few people share their LLM prompts when posting LLM writing, people are not fed up by the exercise. If anything it sounds like a "small fun experiment".
There's a deep problem with this claim: confirmation bias. You believe LLM writing is bad in certain ways; you see bad writing with certain tics; you update even further toward LLM writing being bad.
Suppose that some LLM writing is good, and doesn't have those recognizable tics. Would it change your belief? Generally no, because you don't identify it as LLM writing. Inversely, suppose some writing with those tics is by humans. Would it change your belief? Again no, because you would assume it was written by an LLM.
Further: if you know in advance that a piece of writing is by an LLM, that's highly likely to influence your opinion of it.
I'm not necessarily claiming that some LLM writing is as good as human, but I claim that if it were true it would take you a painfully long time to realize that, because the only cases that would update you would be ones where you formed an opinion and only then learned the provenance of the writing. And how often does that happen?
I think evidence from blind tests is almost the only useful evidence here, because most people already have strong beliefs one way or the other. Prior to seeing such evidence, I think the more reasonable claim is, 'Don't let LLMs produce crappy writing for you', which really can be simplified into 'Don't put out crappy writing'.
I infer you mean "the claim that LLM writing is a slog to get through." Which yes, I wait for the day that most LLM writing I see is not usually a slog to get through. I hope it comes! It's perfectly possible I see lots of LLM writing that was created by prompting wizards using state-of-the-art $200-a-month models in ways not dreamt of in my philosophy. If so, great. If you can fool me, we both win.
But I see so much LLM writing in the wild and professionally that just sucks. Whether it sucks because of some fundamental property of LLMs (doubtful), or because of path dependencies for the LLMs that are commercially available (getting warmer), or because of bog standard skill issues on the part of the relevant centaur (could be), one clear fact is that the authors don't think there's a problem.
In situations like this, where I see a lot of people doing something that seems like a pretty big mistake, I want to just say "don't do that." Because I think if I say "don't do that, unless you're good at it," well, if the people making the mistake knew they weren't good at it, they'd already not be making the mistake! I'd rather say "don't go cave diving" and the tiny minority of expert, professional cave divers who know and relentlessly apply all the proper cave diving safety rules can smile knowingly and ignore me. Of course, the amateurs can ignore me too. But I am here advising otherwise!
Thanks for the thoughtful reply!
I infer you mean "the claim that LLM writing is a slog to get through."
I mean centrally the claim 'to most human beings, AI prose is something sus. If you use AI to write something, people will know. Not everyone, but the people paying attention, who aren’t newcomers or distracted or intoxicated.' I should have been clearer about that.
It's specifically that factual claim that I think tends to be strongly self-reinforcing. If you don't have a way to identify false positives and false negatives, there's no way to get evidence that updates you against that claim.
I'm intending to make a broad epistemic claim here that's not specific to LLM detection. One case that really drove this point home for me personally was the great 'do mp3s sound worse?' debate a number of years ago. Many people were utterly confident that they could hear the difference between mp3s and uncompressed files -- and of course they could in some cases, when the mp3s were poor quality! But in the many cases where they couldn't, they assumed they were hearing uncompressed files, and so it didn't change their minds at all.
But I see so much LLM writing in the wild and professionally that just sucks
I absolutely agree that there's a ton of sucky writing out there, including lots of sucky writing that's identifiable as being from LLMs (and probably some sucky writing that isn't identifiable as being from LLMs). Ninety percent of everything is crap.
In situations like this, where I see a lot of people doing something that seems like a pretty big mistake, I want to just say "don't do that."
Fair point! I'm not really disagreeing with the advice, only with the epistemics of the factual claim.
That said, separately from the epistemic point: after posting the comment, I got curious about the current state of the evidence on LLM writing detection by humans. I asked a couple of frontier models[1] to do a review of 2025 / 2026 papers using blind testing[2]. As I read it, the upshot is:
Note that most of the data is on 2023/2024-era models, eg ChatGPT-3.5, GPT-4o, Claude 2. I would expect that humans would do worse at detection with current frontier models.
Reminds me of the - "No CGI" is actually Invisible CGI - video series from a few years ago. Linked below. At this point, a substantial percentage of writing is at least aided by LLMs so the only LLM writing you notice is just the bad writing. I don't fully buy this but I think it explains a least a portion of the discourse.
I'm quickly finding lots of caveats.
On the one hand I don't let Claude fully write for me. Even if I let it write a draft based on an outline, I so thoroughly rewrite that draft that it's not really Claude's anymore. It's more like a bunch of babble in the direction of what I want that I can use to prune into what I want to say.
But on the other I find all kinds of docs where basically nothing is lost by letting Claude lead the writing.
Tech specs? Yep, I just let Claude own the words. All I care about are the decisions, and there's little value in exactly how those decisions are expressed in words so long as the decisions are communicated clearly.
Research paper? Again, mostly can just let Claude own it. The words are just a means to convey some results. No one is reading this for the prose.
Claude skills? Also just let Claude write them. Yes I might make some small tweaks in the name of token efficiency, but it knows how to write for itself better than I can.
But an essay on my blog? A chapter in my book? Well, those places are where I express my ideas, and Claude can't understand them better than I can unless I first clearly express them myself, at which point all Claude can offer is additional ways to explain the same ideas (which is sometimes helpful in editing!).
So in the end I find it really depends on the purpose of the writing. If it's about the words and ideas, then I agree, don't let LLMs write for you. But if it's not so much about the words as about what the words are used to convey, then the words aren't adding a lot of value by being written by you and you can safely let the LLM write them, just as I like the LLMs write code.
human writing is evidence of human thinking. If you try writing something you don’t understand well, it becomes immediately apparent; you end up writing a mess, and it stays a mess until you sort out the underlying idea.
Can you elaborate more on this. It feels like quite the opposite to me - the more I've thought about something, the messier it comes out. The harder it is to unknot the spider-web of thoughts into a linear rhetorical structure which is readily comprehensible to a virgin reader. Particularly topics I have a tendency to 'geek out' on. Does this mean that I don't truly understand them, or that they are lacking a unifying underlying idea? Am I perhaps confusing passion and knowledge for understanding?
Or is it only evidence of thinking about the writing - the words on the page/screen the reader is looking at right now? And one can have a personal understanding of something which is clear in their own head (or perhaps even readily conveyed to others with similar domain knowledge - like that XKCD comic), but not readily translatable to the page?
I think there are two separate claims lurking here:
I think the first is basically true, while the second isn't necessarily true because writing can be hard. There are lots of ways for a clear idea to not make it into clear writing, like inferential distance as you mention.
The main thing to me is, if you start writing and it's just not working, one hypothesis is that your thinking has still not really crystallized. And if you go straight to an LLM to "clarify this" you accidentally tend to throw out that hypothesis.
Content note: nothing in this piece is a prank or jumpscare where I smirkingly reveal you've been reading AI prose all along.
It’s easy to forget this in roarin’ 2026, but homo sapiens are the original vibers. Long before we adapt our behaviors or formal heuristics, human beings can sniff out something sus. And to most human beings, AI prose is something sus.
If you use AI to write something, people will know. Not everyone, but the people paying attention, who aren’t newcomers or distracted or intoxicated. And most of those people will judge you.
The Reasons
People may just be squicked out by AI, or lossily compress AI with crypto and assume you’re a “tech bro,” or think only uncreative idiots use AI at all. These are bad objections, and I don’t endorse them. But when I catch a whiff of LLM smell, I stop reading. I stop reading much faster than if I saw typos, or broken English, or disliked ideology. There are two reasons.
First, human writing is evidence of human thinking. If you try writing something you don’t understand well, it becomes immediately apparent; you end up writing a mess, and it stays a mess until you sort out the underlying idea. So when I read clear prose, I assume that I’m reading a refined thought. LLM prose violently breaks this correlation. If some guy tells Claude to “help put this idea he has into words” then Claude will write clear prose even if the idea is vague and stupid. If the guy asks to “help find citations” and there are no actual good ones, Claude will find random D-tier writeups and link to them authoritatively. Worst of all, if the guy asks Claude to “poke holes in my argument” when the argument is sufficiently muddy, Claude will just kind of make up random “issues” that the guy will hedge against (or, let’s be real, have Claude hedge against). So you end up with a writeup which cites sources, has plenty of caveats, and… has no actual core of considered thought. If you read enough of these, then you start alt-tabbing away real fast when you see structured lists with bold headers, or weird clipped parenthetical asides, or splashy contrastive disclaimers every 2-3 sentences, or any number of other ineffable signs subtler than an em dash.
Is it possible that a 50% AI-generated hunk of text contains a pearl of careful thinking, that the poor human author simply didn’t have the time or technical skill to express? I suppose. But it ain’t worth checking.
Second, and closely related, AI prose is a slog. There’s way too much framing, there are too many lists and each list has a few items that serve no purpose, the bold and italics feel desperate, and it’s just all so same-y. In your own conversation with an AI that you can fully steer, you can sometimes break out of this feeling for a little bit. But reading the output of someone else’s AI conversation is rarely any fun.
In short, if someone reads writing “by you” and it seems LLM-y, they will think both that:
If you want them there, they are not going to stick around. In fact, the more you want a reader, the more likely they are to be turned off by this stuff. Even if they’re the biggest AI fan in the world.
Luddite! Moralizer!
Fine. I admit it. Just this week, I too experienced Temptation.
You may know me as an editor. In this capacity, I was revising an academic paper’s abstract in response to reviewer comments. But I had several papers to work on in the same project, and the owner of that project actively encouraged me to use AI to move fast enough to meet deadlines.[1]
So I gave Claude the paper and the reviewer comments, and asked it to come up with a new abstract that would satisfy the reviewers. The result looked good.
“It’s just an abstract,” I whispered to myself, face lit eerily in my laptop screen’s blue light. “Summary. Synthesis.” I rocked back and forth. “I could… just…”
But no. Claude’s abstract was a useful reminder of which paper this was, and Claude helpfully catalogued what the reviewer requests were. Still, I rewrote the abstract myself, from scratch. In so doing, I noticed a lot of things I hadn’t seen, when I was just skimming the AI output. Stuff it included that it didn’t really need to. Stuff it emphasized that wasn’t actually that important.
Did I run my abstract by Claude in turn? Yes! It had two nitpicks, one of which I agreed with, and fixed in my own words. Use these tools. You should totally ask Claude to find you sources for a claim, but then you should check those sources like you would check the sources of an eager day one intern, and expect to throw most (or all) of them away. You should totally ask Claude to fact check, but expect it to miss some factual errors and unhelpfully nitpick others. You can even ask Claude to “help clarify your thinking.” But if you’re really just clarifying it, then you won’t use its text. Because once your thinking’s clear, you can write the text yourself, and you should.
To be clear, editing I do as part of the LessWrong Feedback Service uses my own human judgment, and I don't use LLMs to make edits.