For those keeping track, this is now (at least) the third wholly or partially AI-generated article on the theme of "taste" as enduring source of human comparative advantage to hit the HN front page.
Another thing to consider is that LLMs have become a premier source of reading for the public. If more of what you read is AI generated, how you write is going to be significantly influenced by that.
One (slightly frightening) possibility is that many people write like LLMs even without assistance from LLMs. AI writing may have just become the norm for some. Maybe that's what's happening here.
LLM style is in many ways "The Atlantic from Temu". Nearly every article in The Atlantic sounds the same - I think we all know the cliches I am talking about. This doesn't mean that every author underwent a lobotomy (I hope). It just means there's a group of editors enforcing a common style.
Joining these editors used to take years of training in order to write like they do. Now we have their wisdom available on the command line.
To put it another way: LLMs write in what was considered pretty good style before their proliferation. There are all these examples of people finding "llm cliches" in well-respected authors' works pre-llm (somewhat obviously, if you think about it).
What made pretty good "taste" pretty good was its novelty. Now that pretty good is common it's no longer stylish.
Magazines’ style is generally good but affected, with some distinct quirks mixed in. These quirks have been also picked up in the training and their automated repetition at scale made it weird and "AI-style"
Yeah if you look back at earlier posts on the same blog, definitely not the same style. I'm guessing the author is either lying to save face or has spent so much time with Claude that they cannot distinguish Claude's voice from their own (we're all vulnerable to this).
Just because LLMs exist does not mean people stopped writing themselves. I mean, some people did, but not everyone. It's a kind of AI psychosis to think or suspect that everything is LLM generated.
I think they're trolling us. Their post-mortem is a very AI-sounding (and Pangram-triggering) claim that they're not using AI:
This post reads off as AI slop. You said it, I see it. I’m sincerely sorry for publishing something that has allowed you to feel this way. If my word means anything to you, I would like to assure you that this post was not authored by a LLM. Nor was it storyboarded, reviewed, checked, etc. by a LLM. Some readers have pointed out that people do not speak this way. That is correct. I do not speak, nor usually write, like this and this post will go down as my not-the-proudest, however, I take your criticism to heart—although not personally—and strive to improve.
I do write like this sometimes. The short sentences, the reversals, the one-word lines—all of it. It’s just the way it is. A LLM writes that way too, because it was trained on the same essays I grew up reading, so me doing it badly and a machine doing it look about the same to you on the page. That says something about my writing. It says nothing about who wrote it.
So let me be plain about it: Claude was not here. No LLM wrote this—not a sentence of it, nor was it outlined, drafted, reviewed, checked, etc. by one, and there is no prompt behind it either. It is just me, writing worse than usual. I will write the next one plainer. Next time, write to me. I too am a person behind this screen.
This is so very paranoid. Consider how hurtful your accusations may be.
The truth is that AI generated text is getting harder to detect, and while AI writing is still very lacking, more writing is getting the unwanted distinction of "plausibly written by AI." This was always going to be the natural course of events but plausible shouldn't mean guilty.
We should give more grace. To prosecute each and every mediocre writer in a witch trial with such weak evidence is entirely unnecessary.
Pangram is a complete scam. There is no durable way to detect whether something was written with AI. I've had folks say they ran pangram on my writing and it was "100% AI". Except I don't use AI to write or edit anything I post publicly.
Can you give some examples of your writing that trigger "100% AI" on pangram? I'm very interested in studying pangram's false positives. Especially if it's something published before ~2024
I agree Pangram's UI is awful and often misleading. Their underlying classifier model is pretty accurate in my experience, though, at least in the sense that it has very few natural false positives. If you disagree, please send me some long-form verifiable false positives that were not explicitly written to trick Pangram. I love to learn.
I just sent you a long article which you could not have read in the time it took you to reply. I would suggest you start there and read that article, which outlines several cases where the author was able to create contrived false positives and negatives.
Contrived false positives and negatives could (and should!) always be possible. That doesn't tell us anything about the natural false positive and negative rates, and a lot of people who have actively tried to use voice instruction to get models to consistently fool the detector without iterating against it directly have failed.
Those people should just try posting a big quote from the article to HN, and then see their comment get instantly flagged due to HN's AI detection (which might as well be Pangram)
It's baffling to me how people are so dismissive of Pangram despite never using it, extrapolating their experience from GPTZero or something else.
Saying "more" seems generous. I don't trust Pangram at all, and I also don't trust the author at all. Both can be trustless charletans simultaneously, and I feel this speaks volumes about the state of digital media today.
So you would trust a known writers word over non deterministic software ? I read New Orleans will soon let AI handle 911 calls, Hope the AI believes the calls are from humans :)
All ML classifiers (and algorithms) have a non-zero false positive rate. Having an error rate is baked into every ML classifier and algorithm. And its always non-zero in practice. In fact, hitting every test in some sort of test suite is likely a sign of a less accurate classifier, not a more accurate one.
Can you give some examples of Pangram false positives? Ideally ones from before 2024, or otherwise ones from notable writers who started writing before 2024.
The author is lying. Pangram is no panacea, but every time I have tested my own prose, it says it is 100% human written. Every single passage I tested from this post was evaluated as 100% AI written. The author should just confess; lying about having written something with AI both reveals the author as a liar and confirms that the author knows that it is wrong to pass off AI slop as one's own work.
I showed some old Olga Tokarczuk essays to my friends to pick what was ai slop and what was not as experiment, pre- 2020 ones from Czuły Narrator to be exact. Guess what, they picked the ai one as human and her as most ai. Maybe because she uses a lot of big words and em dashes.
We are at the point, when there is too much cognitive energy needed to discern writing from human, human ai assisted and all ai. Should we know make mistakes on purpose to avoid being labelled as ai slop? :)
This is why I'm tired of the "this is AI slop" commentary. I'm well past trying to determine whether a human wrote something, and have fallen back to my old ways of critical reading to search for value in the thing I'm reading. And if I don't find it, it doesn't matter who wrote it. Everyone is so excited to be clever and call out others for using AI, when it's far more useful to cite and elevate good work, regardless of tooling.
Have you read the article? It has an unbalanced emdash: `word— another word`. A novice mistake that an LLM likely will not commit nowadays.
In standard typography—depending on which guide you need to follow — it can only be padded by space on both side or neither.
This article is not AI generated. Everyone is asserting it is but the author is well known as a critic of these tools in his communities. Why would he lie and use it to write an article with his genuine views, edit it to assuage the feelings of HN commenters, and then double down? Speedrunning losing his credibility, or, his writing style got on some people's nerves?
I agree with your assessment. Read the article and did not come back thinking an LLM authored. I read more LLM content now than from humans and while there were fingerprints of LLM writing, it still read as institutionalized English. It seems like we are in a weird transition period where LLMs trained on proper English prose are now training humans to identify prose that is apparently not common enough to be labelled as human authored. So what are people looking for in writing? The mistakes, disjointed sentences, less em dashes?
"And when the floor goes, the thing that decides what is worth keeping is no longer the cost of making it."
I'm 100% confident that a human did not write this sentence unless they were deliberately trying to sound like an LLM. Just look at it! You don't need Pangram. There's a classifier model between your ears that knows how humans write.
I'm generally quite open to the possibility that I'm wrong. I said 100% for this one because it is tooth-grindingly obvious. Could you give me some examples of long-form writing like this from before 2024? Generally to retrain my model I need data.
To be honest with you, I think this is pointless. However hostile this might sound, the active userbase of HN seems to be extremely dismissive to the thought of LLM detectors and they think that all those LLM written articles are actually human written. I don't know if this will change, perhaps it'll only get worse from now. (You can see my recent experience in another thread https://hackertimes.com/item?id=49212027 )
If you don’t know how to write well, the AI slop is less obvious. That’s my theory at least, I don’t understand how people can read these nonsense sentences and believe a human actually wrote it.
Interesting hypothesis! I find that my internal sense of whether the article is AI almost always agrees with Pangram when I check it, and almost never triggers on pre-2024 text, even if I didn't know beforehand when it was written. I think this cuts against your hypothesis -- if I was just recognizing a generic style of writing, I would expect it to fire on pre-2024 text, and if I was just recognizing text written in 2026, I would expect it to fire much more often when reading blog posts of already-established writers who I trust.
Previously:
Taste in the age of AI and LLMs - https://hackertimes.com/item?id=47677241
You Had No Taste Before AI - https://hackertimes.com/item?id=45288551