I like that these AI idioms exist. They're like watermarks for text. It's worth the cost of humans avoiding them. Companies will eventually train their models to be undetectable, but society would be better if they didn't.
Humans are just trying to do what Pangram is trying to do: guess what is AI, badly. The post argues against this:
> In the end, shaming people for writing that gets flagged as AI can lead people to sidestep structures the model has learned from us: structures that are effective tools for argumentation. We take the tools of critical thinking out of the kit at the time we most need them.
This is my position with this stuff. It became part of the LLM loop because it’s used a lot- it’s used a lot because it’s effective.
Now we’re going to stop using effective rhetorical methods because they imply AI, even if we know we’re not using AI?
It reminds me of, as a teenager, asking my dad if he ever saw Led Zeppelin live. He hadn’t, because he didn’t really like fans of Led Zeppelin and didn’t want to be associated with them.
As an ashamed fan of certain bands I get this instinct but I also promised to myself when I heard this that I would do my best to not allow other people to influence how I thought about things I enjoyed.
On the same note I’m trying to be “braver” about things like em-dashes, though my personally style has always been to use them as I did in this comment- like this, which I guess distinguishes me, until an LLM picks that up too…
> It's worth the cost of humans avoiding them
That's really unfortunate though. It's like Michael Bolton from Office Space: "No way! Why should I change? He's the one who sucks."
I don't think so at all. Models are trained in many ways and are changing aggressively, resulting in different patterns in different regions, domains, languages, and will be different 3, 5, 10 years down the line. Having everyone try to learn and adapt around how to stay within very magical, fuzzy, and ever-changing boundaries to avoid appearing to be an AI, instead of focusing on producing good writing or communicating as it is natural to them, seems like a recipe for bad thinking and arbitrary reactions.
It's like knowing to stay away from a Github repo because it has a readme that's full of emoji bullet points.
I thought I was the only one that did this. Double stay away if at the end you find out it was "made with love"
Except that the entire point of the article is that they're not AI idioms. They're not "watermarks for text." They're legitimate language constructions that LLMs tend to overuse, but that real humans also use. Real humans do, in fact, say "align with" all the time, just as often as "corresponds."
And you can pry my em dashes from my cold, dead hands.
What's worse is neurodivergent writing, including my own, often resemble AI output. Now it feels like I'm having to alter my own voice in online discussions just to specifically avoid being accused of pasting an AI response.
The "AI Detection" tools employed by schools also regularly flag writing from those with Autism, ADHD, and non-native English speakers as being AI generated as well.
So, naturally, I can't stand the phrase "write like AI" when these things tend to come up because no, there are no humans that "write like AI" it's the models that have stolen the literary devices from us and now have poisoned them.
That is an empirical question. Do you have empirical sources you'd care to share?
The article is not God, just because it claims something doesn't mean we have to accept it.
For better or worse (and pretty much for worse), these usages have become AI idioms. Language evolves over time, things that used to be harmless become offensive, certain terms end up taking on the complete opposite meaning than their original meaning, and we are watching certain language patterns and idioms become watermarks for AI and while it sucks, it doesn't make it false.
I'll just quote from the article, which no one claimed was God and that's really a weird way to dismiss it, but you do you:
"We create a culture of self-censorship and AI-detector-pressured rewriting and paraphrasing as people strive to avoid these witch hunts. That is the opposite of protecting human expression. We should resist normalizing a trust in any machine's ability to determine matters of guilt. If using AI to write is, at its worst, an industrialization of the mind, then AI detection, at its worst, becomes a surveillance system for thought."
And, I'm sorry (I'm not), but I am not going to just roll over and shrug and say "welp, guess we all need to dumb our writing down to keep well-meaning idiots from screeching 'AI! AI! AI! WHOOP! WHOOP WHOOP WHOOP!' at us." That isn't the evolution of language. It's Idiocracy.
Once upon a time, using em dashes—which hardly anyone knew how to conveniently invoke—was a fun writing quirk to have.
Now I'll have to find something else to overuse: maybe sentences structures around colons, or use of Japanese 「hook brackets」.
Well reading between the lines I don’t think they’re saying all of those uses are AI. They’re legitimate constructs, like the em-dash, en-dash, and hyphen, all of which I used to use regularly. But now they’re AI tells so I use them sparingly.
Sociolinguistic register happened.
I agree with the feeling. But if you agree with the analysis of the article, this cat & mouse game ultimately amounts to stop disclosing our reasoning threads through commonly accepted linguistic structures. That's quite a price to pay as a society...
"So, if we publicly shame people whose text looks like it might have been written by a machine – because it mimics the language used for human reasoning – and people stop writing in ways that they internalize as "AI writing" out of fear of false detection, it sends a signal that your language for reasoning must be policed, or you too could be held up to public scrutiny."
This is honestly both terrifying and well articulated.
High praise to the blog author.
There are plenty of idioms, find a different idiom, tough titties.
Surely it hasn't escaped your notice that the LLMs don't tend to stick to just the one set of idioms/devices. They're constantly running new batches of them into the ground
> There is danger in evaluating for language patterns over its content
I agree, but it’s worth noting that that has been done since long before LLMs. Fifteen years ago, I used to teach a graduate course on academic writing pedagogy. The students and I would read research papers on the teaching of academic writing; we also analyzed textbooks and course syllabuses to get an idea about what was actually being done in classrooms. While phrases like “critical thinking” did come up, the overall focus was clearly on language patterns: sentence and paragraph structure, the use of transition words, vocabulary for hedging and boosting (i.e., making assertions seem weaker or stronger), etc.
In a university context, it can be very difficult to evaluate student writing based on its content. In humanities-focused and creative writing, what the student decides to say can be seen as an extension of the student’s personality, identity, and individual experience; if a teacher evaluates the content, including the reasoning, it can seem that the teacher is evaluating the student as a person. And if the students are in the sciences, especially at the graduate level, the writing teacher often won’t even understand what the students write because it is too technical. Teaching and evaluating language patterns, not content, is often the only option.
Surely these leading tells will be trained out of models pretty soon, given how well known and overused they are. And it might make the writing slightly worse in a way. But it is quite annoying how often this type of construction is used in everything at the moment.
I think that the current models are still like over-achieving savants rather than true human level because the largest model is only 1/10th the complexity of the human brain. I've recently become fairly convinced that new hardware paradigms (like types of CIM) are about to move from research into real-world development and scaling. So I believe within a few years, the model sizes will increase by another 10 times.
Compared to upcoming 100 trillion parameter models, humans will obviously be _much_ dumber/slower than AI in all fields. Already with the 10T models, some LLMs beat 99.9% of humans in competitive programming.
The AI hatred from many may actually continue to increase, but in cases where the bottom line matters, we are rapidly approaching the point where writing or work product that looks like it is human-authored will be suspect just on that basis. In other words, for some people it will be the reverse -- "this work looks like it was created by a human" could be devastating for your businesses credibility at that point.
I liked everything in this post, with one exception. I'm less sure that avoiding speaking like an AI is robbing us of language useful in critical thinking. I'm far more worried about people offloading their critical thinking to AI systems and losing the habit.
Also, the Greeks were worried about rhetoric and, in my opinion, rightly so. The skill to argue a point well is different than those that are needed to be correct. To become a skilled rhetoritician was viewed as dangerous (and right now AIs are only moderately good... though they are improving fast).
> RLVR is weirder, and I suspect it's why we see "It's not X, it's Y" so often.
This feels like an easy enough hypothesis to verify, for anyone in the business of training LLMs - does the not-X-but-Y rate increase after RLVR?
It’s unlikely this is true. LLMs are way more mad-libs / templates than we like to admit, that’s (ironically) not a judgement about their capability, it’s primarily just an observation. But it’s also what plain old SFT, which I believe is the primary culprit, ends up imparting.
In one of the essays posted here, which was, ironically, about AI in education, a sentence, that an AI could not possibly write, that I could possibly write, because of its length and unusual structure, before finally reaching the verb, went on for 25 words.
I don't know if it was written that way to show trust in the reader's intelligence, show disregard for reaching a wide audience, show a demonstration of skill, or was artifact of someone just thinking at that level.
The robot can be made to write any such thing
In the interest of not occupying significant page/screen height with LLM output, example prompts+responses here: https://dpaste.com/H9DXKNYQH.txt
Your first sentence is 45 words and contains 9 commas.
> I don't know if it was written that way to show trust in the reader's intelligence, show disregard for reaching a wide audience, show a demonstration of skill, or was artifact of someone just thinking at that level.
This is how early forms of "reasoning" in LLMs worked: just literally inserting words like "Wait...", "Hmm...", "Let me reconsider...", "But is it really..." into the token stream.
Is this not how current forms of reasoning work? It seems like the open models still output things like that, and the closed ones all just summarize their thinking instead to avoid distillation, but probably do the same thing internally.
I think the basic idea is the same (not being a frontier lab researcher I couldn’t say for sure), but there are different techniques, such as “reasoning tokens” that aren’t literally words, and more interesting structures than just sticking them into the stream.
People stopped actually reading when we dropped classical liberal education, right after WWII.
This is merely the end-state of industrialization, which is efficient and soulless.
nice article, but i think as a non native english speaker, i always use the model in english for reasoning and then translate the output to my language. most of these considerations do not apply. because the translation step is taking out alot of these language artifacts
Do you manually translate or translate with an LLM? While reading, I was wondering how common these kinds of written tics are in languages outside English.
Clearly humans always type "it's not merely X, but also Y"
In my experience, and as someone who has technically been a professional copy editor before (in the sense that I got paid by someone to perform line/copy editing of some material or other, but definitely not who or what you would think if I only said I used to be a copy editor), the uses of "not only, but also" that didn't have anything wrong with them (or didn't deviate from convention, if you like) were in the minority by far. Assuming people in the last few years haven't all become suddenly interested in that particular construction and skilled in its use, the presence of an improperly implemented NOBA should remain vaguely reliable as an indicator of authenticity :D
> In the end, shaming people for writing that gets flagged as AI can lead people to sidestep structures the model has learned from us
It's interesting why LLMs generate constructions like this more frequently than they presumably exist in the training set. I wonder if this is some sort of mode collapse caused by post training, and/or maybe because they are training on synthetic data so these things become self-perpetuating and self-amplifying (a feedback loop)?
The lesson for humans worried about being falsely identified as AI is just learn to write better! It doesn't matter where your repertoire of phrasing comes from (copying AI or not), but one of the basic rules of writing is not to repeat yourself unless you are doing so deliberately for a purpose. Go ahead and use "It's not just X. It's Y" if you want to, but if you use it multiple times in the same short piece of writing, then you may deserve to be called out for poor style, if not for being an AI.
Its not model collapse nor does it have anything to do with training data frequency. It's simply RLHF where the humans hired to tune the conversational style of these LLMs preferred certain idioms over others and so the reward function for these LLMs gravitated toward using them.
If LLMs generated text based on training data frequency they'd likely be some of the most vulgar and hostile things ever created. The internet is full of insults, profanity, and low effort content. The repeated phrases are a side effect of reward optimization rather than some kind of model collapse.
>Recent overuse by language models has led many to declare it bad writing. I'm not so sure.
It is bad writing.
Always? There's never a place for it?
Not always or inherently, I would say, but if you can't help consciously noticing how much a piece uses tricolons, or negative parallel constructions, or dashes offsetting punchy clauses at the ends of sentences, then the style has probably become what you might call overseasoned
I'd say it's average writing.
You’re absolutely right to push back on this.
Sometimes it’s not just about the Ys but also the Qs.
> Because if Pangram's AI system found me guilty, that's the end of my career. That's literally extortion.
How is this different from humans? When I went to high school, my teachers extorted me too. Especially subjects like English and unlike Math, where evaluation is 100% subjective.
I don't how the English evaluation works in the average (US?) school, or even what's in the exams. But it's possible to have useful native language exams with objective evaluation.
Until 2006, the national Finnish/Swedish (as a native language) exam at the end of high school in Finland consisted of two essays. One was based on materials that were provided, and another on a topic that was given. I believe there were a few options to choose from. Both essays were scored independently, and your score was the better of the two. If you had learned to write essays, it was effectively an intelligence test and a good predictor of future academic success across many fields.
Including CS, as my department found out. In particular, the ability to write essays was a better predictor of success in CS studies than the scores in mathematics and natural sciences. Probably because there has always been a large subset of CS students who are otherwise good at CS but can't handle anything resembling mathematics.
Another bunch of dead give aways in code bases with READMEs is the repetitive:
- "No X, No Y, No Z." pattern
- "Here is X - it makes Y"
The worst and most obvious one is the constant over use of emoji ticks and crosses.
"Hyphen functioning as an em dash" is an expected human thing as it's what's easy to type. It's specifically an actual em dash which got bulldozed, much to the dismay of those who bothered to put the unicode character in.
If you read The Mac is Not a Typewriter in 1992—thus burning Option-Shift-hyphen into your typing patterns for life, along with a dogmatic love for serif body fonts—you're the real victim here.
Or those of us that use a full featured editor when writing md!
This reminds me of another em dash+AI related topic: I've noticed LLMs have an extreme bias towards spaces around the dash while people can go either way with it.
There's something similar in Microsoft Word, Ctrl-Alt-Minus on the numpad.
I prefer the double dash "--", but Microsoft products will convert this to a proper em-dash if you press space afterwards, I think...
Double should map to endash, tripple for em.
A lot of the LLM bots on HN (and elsewhere) will find-and-replace their em dashes with hypens in an attempt to evade detection.
Precisely, anything to remove AI smells in favor of natural looking text.
My point is I don't consider em dash vs hyphen to be a strong signal either way, humans and bots alike use both interchangeably.
A signal is not the same thing as a guarantee. Both of your points so far, i.e. your provided text & that bots often bother to replace em dashes to avoid detection, actually support that it is a signal though.
The stronger signal is the grammatical structure, not the specific glyph used.
The stronger yet signal is both combined! This glyph, that emoji, a given sentence structure, that formatting, a certain phrase. The more you notice -> the stronger the signal, the more you miss/discard -> the weaker the signal.
and we will now hold you responsible!
Alternatively, no one sounds like an llm, an llm sounds like someone, typically those close to the median of the training corpus. If AI were genuinly capable of novelty, it would be a big deal, tech bros having enough work ethic to design new detectable prose for an llm is a mssive reach and has no real evidence supporting it, else why do tech bros only tackle the easier issues? Things we have massive well labelled corpi for? Why is it never dishwashing and folding laundry?
I put to you, if you see a trope in AI writing it's because that trope appeared in the training corpus. Therefore, sure, being predjudice against it lets you catch some AI, but you'll also flag human outout. I think that may not be worth it in the end.
Show me a single substantial (5000+ words) piece of writing from before the release of GPT-3 that triggers Pangram with high confidence.
You’re absolutely right. This is the smoking gun. This changes everything.
That is actually what I'm firmly convinced is the most dangerous thing about llm's. No matter what you put, it will always agree with you, and what's worse, it will try to make you think that you're unbelievably smart for saying x.
I used one to help me plan a sales route, and it kept fucking it up. Every time I corrected it, it tried that hand wringing vizier sort of ass kissing. It's very off-putting, but I can see how someone struggling with social interaction could be sucked into that nonsense.
My custom instructions to ChatGPT starts with "Never apologize", which seems to work.
Now I see the full picture.
Wait, there could be more things to consider.
This is the crux of the analysis.
I'm zeroing in on the main culprit.
This is the real unlock. Here's the key takeaways.
It's not just an unlock. It's a major discovery.
Signs? Those are normal ways of writing? What the hell? Is everything AI now?
Problem is, everything gets poisoned by AI these days, and it gets worse when there's some sort of reward attached. Karma points in the case of Reddit and HN, in Wikipedia you got a ton of commercial actors and propaganda/distortion campaigns.
And that's why everyone on the receiving end of the AI slop deluge is so paranoid.