Posted by usefulposter 1 day ago
https://simonlermen.substack.com/p/large-scale-online-deanonymization
https://news.ycombinator.com/item?id=47139716Reading the site in past 2 years left me with the feeling that HN has been injected by subtle to catch AI marketing campaigns. It's exausting and calling out astroturfers imo is not that bad
> Your arguments will come of as stronger to the reader.
That is persuasian, not authenticity, to the OP's point.
Typed without a spellchecker :).
And that's where I think the guidelines could be expanded a bit more to restore the balance. Something along the line of 'HN is visited by people from all over the world and from many different cultural and linguistical backgrounds. Please respect that and realize that native English and Western background should not be automatically assumed. It is the message that counts, not the form in which it was presented.'.
(For example: If I’m trying to express a point about how we shouldn’t assume that dinner isn’t “her duty” but is instead “our duty”, a French-like aphorism expressed in English literally as “the chicken won’t fly into the oven unprompted” could plausibly be AI-translated instead as “don’t count your chickens before they hatch”, doing catastrophic damage to the point. To a machine translator those two aphorisms are not distinctive; but they are, even if it’s a weird expression in common U.S. English.)
I can understand the sentiment though, as I am learning a second language and in many of our writing assignments we are expected to use (from my perspective) overly formal and complex grammatic structures when writing simple letters. I have come to accept, or at least hope, that this is simply an exercise to ensure that students have fluency with the grammar.
That’s true. I’m fluent in German, but there’s still a difference between me and a native speaker. I’ve often seen my ideas dismissed, only for the exact same point to be praised later when a native speaker expresses it more clearly.
I now expect malapropism, hacker curtness, and implicits: TAIDR is the new TLDR.
Post the translation as best you can manage, and below it put the same comment in your original language. If someone has qualms with your comment having broken english/mistranslations they are welcome to run bits of original language themselves.
We're all here to talk about tech, and we aren't all perfect little english robots.
Write it broken.
Broken and true is more authentic than polished and approximately so. When I see an AI-generated comment or email, I catch myself implicitly assuming it is—best case—bullshit. That isn’t the case if the grammar is off. (If anything, it can be charming.)
Besides, this isn't an English poetry forum. Language here is like gift wrapping for an idea: pleasant if pretty, but not the most important thing.
From the perspective of someone reading the comment, I'll take “inauthentic” but actually comprehensible over “authentic” but incomprehensible any day.
Also, using bad grammar as a heuristic for humanity will just end with LLMs being prompted to deliberately mess up their grammar, and now we're back to square one, with the state of the written word even worse off than it was before.
That may be a defect in me. Maybe I should make a stronger effort on such comments. But I suspect I'm not the only one who does that, and at that point it becomes an issue that affects the community as a whole.
At which point you’d be fully justified in using an AI to decode their text. I still think that’s a better world than pre-filtering.
I've seen enough GPT-generated slop that I find its style of writing very off-putting, and find it hurts the perceived competence or effort of the author when applied in the wrong context. I'm not sure if direct translation tools serve a better purpose here, but along with the other commenters, I personally find imperfect speech that was actually written "by hand" by the author easier and more straightforward to communicate with despite the imperfections. Also, non-ESL speakers make plenty of mistakes with grammar, spelling, etc. that humans are used to associating with "style" as authentic speech.
It can also become a crutch for language learners of any age / regardless of their primary language, that inhibits learning or finding one's own "style" of speech
The human touch of someone’s real voice myself, rather than a false veneer will carry more weight very soon.
I've never sent or posted anything AI-written, beyond a pro-forma job description - because I don't know the domain-specific conventions, and HR returned my draft to me with the instruction to use ChatGPT, which I think amusing, but whatever: the output satisfied them, and I was able to get on with my day.
I occasionally experiment with putting something I've written through an LLM, and it's inevitably a blandifying of my original, which doesn't really say what I intended. But maybe that's good? My wife thinks I'm sometimes too blunt, and colleagues don't always appreciate being told technical details.
I also appreciate individuated writing - including the posts by people on this board are not native speakers. Grammatical mistakes seldom inhibit understanding when the writing has been done with care.
I'm rambling at this point, but it's because I'm truly uncertain how these cultural changes will turn out, and (an old man's complaint, since time immemorial!) pretty sure I'll end up one of the last of the dinosaurs, clinging to my manually written "voice" long after everyone else in the world has come to see my preferences quaint.
This is tragic. I write English well and will employ grammar and word choice effectively to make an argument or get a point across. English was my best subject at school 45 years ago despite a career in tech. In fact, I’d suggest that my career as an architect and the need to convey concepts and argue trade-offs with stakeholders of varying backgrounds has honed that skill. Should I now dumb down my language or deliberately introduce errors in order to satisfy the barely literate or avoid being “detected” as an AI? (as if the latter were possible. It’s an arms race).
Language is a tool. If it wins the argument, yes. I’ve absolutely gone back through drafts to tighten up language and reduce word complexity. And if I’m typing with someone who frequently typos, I’ll sometimes reverse the autocorrect. Mostly as a joke to myself. But I imagine it helps me come across as less stuck up. (Truth: I’m a bit stuck up about language :P.)
While this is true, it is not just a tool. Or, I should say it’s a tool with far greater utility than just winning an argument or making a localised point. Language is how we think, and the ability to reason well is absolutely dependent on our skill with language.
Language is the mark of humanity in the sense that how else can I convey to you a fragment of my inner state? My emotions, my feelings, my desires. The language of poetry and literature. That which sparks an emotional response in another.
Dumbing down language is dumbing down period.
I agree. But I don’t always see it as dumbing down. James Joyce’s Portrait starts out with a lot of nonsense, that doesn’t mean it’s dumb or dumbed down. It’s just communicating something that is best described that way. Even to an erudite audience.
I have expertise in some topics. I don’t think of communicating that in lay terms to be dumbing down. The opposite, almost: finding good analogies and expressing them clearly is a lot of fun, even if what comes out the other end isn’t particularly sophisticated.
EDIT: spread > express Which may be a segue to a point regarding using corrective tools as a form of preemptive editing?
Funnily enough, I've noticed myself getting worse with they're/their the more is use English (which is my third language).
That is a false equivalence. What a YC-backed company does is not relevant to how a YC-owned web forum operates.
I miss pre 2010 internet. As soon as the advice animal memes started appearing on Facebook it was a quick decline.
You don't lose your voice if you ask for advice and manually incorporate the suggestions you agree with.
You might lose your voice if you say "Improve my comment to make it better" and copy-paste the result without another thought.
It might feel like just a couple of tweaks, but they add up fast.
When using LLMs to write, the temptation to avoid actually thinking about what you're communicating is too much for most people.
Keep polishing and everything eventually turns into a smooth shiny ball. We need texture, roughness, edges.
An LLM telling me I omitted a qualifier and that my statement isn't saying what I meant it to say isn't changing my voice - it's ensuring what you see is my voice.
I'm confused by this need(?) desire(?) to polish things that are irrelevant.
Relevance is in the eye of the beholder.
(As an experiment, I took that paragraph and threw it into gemini to ask for spell and grammar checking. It yelled at me completely incorrectly about saying "I'm not dang". Of its 4 suggestions, only 1 was correct, and the other 3 would have either broken what I was trying to say or reduced the presence of my usual HN comment voice. So while I said the above, perhaps I'm wrong and even listening to the damn box about grammar is a bad idea.)
That said, I often post from my phone and have somewhat frequent little glitches either from voice recognition or large clumsy thumbs, and nobody has ever seemed to care except me when I notice them a few minutes after the edit button goes away.
AI is being used as a substitute for skills development when it costs nothing but time to get better. If you’ve reached a plateau with the above method, go find an article or book or interview about editing, pay attention to it and take notes, rinse/repeat.
Spellcheckers will catch grossly obvious errors, but not phonetic typos. AI grammar tools will defang, weaken, soften, neutralize your tone towards the aggregate boring-meh that they incorporated at training time.
Each person will have to decide whether they want individuality or AI-assisted writing for themselves. Sure, some will get away with it undetected, but that’s a universal statement about all human criteria of any kind, and in no way detracts from the necessity of drawing a line in the sand and saying “no” to AI writing here.
Consider the Borg. Everyone’s distinctiveness has been added to the Collective. The end result is mediocre (they sure do die a lot), inhuman (literally), and uniform (all variation is gone). It’s your right if you desire to join the Collective and be a uniform lego brick like the others, but then your no-longer-fully-human posts are no longer welcome at HN.
Pffff... I'm not going to install LibreOffice for that, or to figure out how to make Gdocs to work with uBlock.
There is a much easier way. Open LLM chat, type there "Proofread please for grammar, keep the wording and the tone as it is, if it doesn't mess with grammar. Explain yourself." and then paste your text. I don't really know what the tools you mentioned do, but any "free" LLM on the Internet will point to things like missing articles, or messed up tenses in complex sentences.
You recommend choosing self-improvement, but I just don't believe I can figure out how to use articles. With tenses I think I can learn how to do it, but I'm not going to. I remember there is some obscure rule how to choose the right tenses, but I was never able to remember the rule itself. I'm bad with rules, it is the reason I chose math as my major. There are almost no rules in math, you are making your own rules. The grammars of languages are not like that, they have rules which can't be easily inferred, you need to remember them. Grammars have exceptions to rules, and exceptions to exceptions, and in any case they are not the rules, but more like guidelines, because people normally don't think about rules when they are talking or writing.
No way I'm starting to learn rules now, I'd better continue to rely on my skills. But LLMs can help me see when my skills fail me.
> It’s your right if you desire to join the Collective and be a uniform lego brick like the others, but then your no-longer-fully-human posts are no longer welcome at HN.
I believe you (as most of fervent supporters of the rule here) gone too far onto philosophy with this, too far from the reality and practice. You can't detect AI in my messages, because they are mine. Even when I ask LLM to find words for me, it is me who picks one of the proposed alternatives, but mostly I manage without wording changes. I transfer the LLM's edits by hand by editing the source message, so nothing unnoticed can slip into the final result. If I took the effort to ask an LLM to proofread, it means I care about the result more than usual, so I'm investing more effort into it, not less.
There's what now? I do think math is flexible but it feels like there are plenty of rules, depending on the context.
Well, no one can help you to develop your voice. If it is your voice, then it have to be your own creation. I think we are at agreement here.
> Developing your voice by doing your own proofreading pressures you away from the mean, by helping you double down on what you value most and by choosing which grammatical rules to disregard and when disregarding them is more in-tone for yourself than adherence.
Oh... If I wanted to become a professional writer, then I'd agree with you. Maybe...
You see, I don't use LLM to fix my writing in Russian, because with Russian I'm totally in control of my grammar, I know when I deviate from it and if I do, I do it consciously. But with English I don't know. Sometimes I can see that I don't know how to follow English grammar in some particular case, and sometimes I don't even notice that I don't know.
So, returning to your argument, if I wanted to become a famous English writer, I think I'd choose to write a lot and discuss my writing with LLM, and I'd do it for hundreds of hours. LLM are unbelievably useful for digging into language nuances. Before LLMs I had urbandictionary, but it could help with specific phrases, not with choosing between "I took the effort to ask an LLM" and "I took the effort of asking an LLM". I wouldn't have a clue that there is any semantic difference. But LLM can point to it, and it can explain the difference, and give me more examples of it. Or it can point that "you recommend to choose" is not good, because of "something-something" I don't remember what, but it boils down to "you just have to remember, that the right way to use the verb 'recommend' is 'recommend choosing'". I don't see the difference, I can't choose to disregard it, because I have no opinion on if it is good or bad.
If I wanted to become an English writer, I'd spend hundreds of hours with LLM, just to get an ability to see as many differences as it is possible, to get an idea of what I value most, and which grammatical rules I like to disregard. But even after that, I think I'd continue to use LLM. It can provide unexpected takes on what you feed into it. ... Hmm... I should try it with Russian. In Russian I can pick a style for my writing and to follow it (in English I can't control the style consciously), I can (and do sometimes) turn grammar inside-out, make it alien, readable for a native speaker, but in weird ways readable (a bit like letters written by Terry Pratchett heroes like Granny Weatherwax or Carrot)... I wonder, if I can employ LLM to make it even more weird.
> I still won’t equate regressing to the AI mean with personal growth away from the average masses.
I can't obviously judge in which direction LLMs are changing my English, so I can't even give you an anecdotal counter-evidence to your statements about regression to AI mean, but I'm still sure that I'm not regressing to the mean. You see, I pick when to follow LLM advice and when not to. I'm choosing what to change. The regression to the mean you are talking to is going on in a high dimensional space, you can regress on some dimensions and continue to deviate from the mean on others as much as you like. I don't like to deviate on grammar dimensions (at least without knowing about my deviations), I was born in a family of a teacher and an engineer, which were all into to be educated and the familiarity with the grammar was one of the important part of it, and I was born in USSR, where the proper grammar was enforced in all media to the extent that make me laugh and rebel against grammar (after all the decades passed, lol). But I can't allow myself to just ignore grammar, I was taught in a way to use it properly. So I decide to use LLM. I'm too lazy to do it each time, or even every second time, but still I use it and learn from it.
The prospect to regress to the mean by using LLM seems very unlikely to me. I don't regress with all the propaganda around me when regressing is the most safe thing to do really, so mere LLM stand no chance to achieve it.
I've never, ever, ever ever ever, seen anybody complain about spelling mistakes in a comment here. As long as you can understand the comment, people respond to it.
And why would you want to "improve your writing" for an HN comment? I think people here value raw authenticity more than polished writing.
Lots of people break HN guidelines. I see it virtually every day.
> And why would you want to "improve your writing" for an HN comment?
Some people like to write well regardless of the medium. Why is that a problem for you?
> I think people here value raw authenticity more than polished writing.
Classic false dichotomy. Asking an LLM for feedback is not making your comment less authentic. As I pointed out elsewhere, it can make your comment more authentic by ensuring that what you had in your head and what you wrote match.
Go and study writing and psychology. For anything of value, it's rare that your first attempt reflects what you meant to say. It's also rare that the first attempt, even if it reflects what you meant, will not be absorbed by the recipient. Saying what you mean, and having it understood as you meant it, is a difficult skill.
Yes, and AI won't help here. People will use AI to better break the guidelines.
> Go and study writing and psychology
Is this a case where you should have read the guidelines? Maybe an LLM could have helped you here? Please don't send me study anything, you know what they say of ASSuming.
> Some people like to write well regardless of the medium. Why is that a problem for you?
HN is more like talking than writing. And LLMs don't help you write well, they help you sound like a clone, which is unwanted.
> For anything of value, it's rare that your first attempt reflects what you meant to say.
You can always edit your comment. And in any case, HN is like a live conversation. Imagine if your friend AI-edited their speech in real-time as they talked to you.
The other important thing you can do is have an AI check your claims before you post. Even with google and pubmed, a quick check against sources by hand can take 30 minutes or longer, while with AI tooling it takes 5. Guess which one is more likely to actually lead to people checking their facts before they post. (even if imperfectly!) .
I'm not talking about people who lazily ask the AI to write their post for them. Or those who don't actually go through and actually get the AI to find primary sources. Those people are not being as helpful. Though try consider educating them on more responsible tool use as well?
I don't think that's what this new HN guideline is against either.
What I object is the AI writing your comments for you. I want to engage with other human beings, not the bot-mediated version of them.
> I don't think that's what this new HN guideline is against either.
This is actually how many commenters here are interpreting it, though - and that's what I'm pushing back against. They are actively advocating against using LLMs this way.
I don't have the LLM write the comment for me. I (sometimes) give it my draft, along with all the parents to the root, and get feedback. I look for specific things (Am I being too argumentative? Am I invoking a logical fallacy? Is it obvious I misinterpreted a comment that I'm replying to? Is my comment confusing? etc). Adding things like (Am I violating an HN guideline?) are fair game.
Earlier today I wrote a lot of comments without using the LLM's feedback. In one particular thread I repeatedly misunderstood the original context of the discussion and wasted people's time. I reposted my draft to the LLM and it alerted me of my problematic comment. Had I used it originally, I would have saved a lot of people time.
Incidentally, since I started doing this (a few months ago), I've only edited my comment once or twice based on its feedback. Most of the time it just tells me my comment looks good.
AI is a general purpose tool. People will use AI for multiple reasons, including yours. I'll wager, though, that your use case is much more challenging to do than mine, and that my use case will dominate in number.
> HN is more like talking than writing.
Says you. Many disagree.
> And LLMs don't help you write well, they help you sound like a clone, which is unwanted.
Patently false on both counts. Sorry, you're cherry picking and not addressing the part of my comment that discusses this.
> Imagine if your friend AI-edited their speech in real-time as they talked to you.
When a conversation is heated (as it occasionally is on HN), I actually would rather he AI-edit in real time - provided that the output reflects what he intended.
I don't know how comparatively challenging, I only know your use case is now (fortunately!) against HN rules.
> Patently false on both counts. Sorry, you're cherry picking and not addressing the part of my comment that discusses this.
It's not false. It's one of the major reasons people have come to dislike AI written comments and articles. It all ends up sounding the same.
> When a conversation is heated (as it occasionally is on HN), I actually would rather he AI-edit in real time - provided that the output reflects what he intended.
In real life? Sounds like a fucking dystopia. But everyone is free to choose the hell they want to live in.
I say this on behalf of all of my neurospicy friends… sometimes, yes. Especially having taken a look at the whole list of guidelines, I definitely am friends with people who would could struggle to determine whether a given comment fits or not.
I personally don't use an LLM to spellcheck (browser spellcheck works fine), but I see no problem with someone using an LLM to point out spelling errors.
And while I don't complain about others' spelling errors, I sure do notice them. And if someone writes a long wall of text as one giant paragraph that has lots of spelling/grammatical issues, chances are very high I won't read it.
Some people write very poorly by almost any standard. If an LLM helps the person write better, I'm all for it. There's a world of a difference between copy/pasting from the LLM and asking it for feedback.
Spellcheckers exist, you don't need an AI to change your voice.
Also, if you have standards, you can always train yourself to spell better!
How is using an AI to spell check changing my voice?
Yes, thank you - I know spellcheckers exist, as my comment clearly states. The amusing thing is that an LLM who had access to the thread would have alerted you to a basic error you're making.
> Also, if you have standards, you can always train yourself to spell better!
"You can always ..." is not an argument against alternatives.
> The amusing thing is that an LLM who had access to the thread would have alerted you to a basic error you're making.
I didn't make the "basic error" of assuming you didn't know spellcheckers existed. I was stressing that since spellcheckers already exist, you don't need an AI assisting your comments-writing. Much basic, non-style-altering alternatives exist and are better.
> "You can always ..." is not an argument against alternatives.
The argument I'm making is that if you care so much about standards you can always hone them yourself instead of taking the lazy way out of having an AI write for you.
Alternatively, if you're lazy then your standards aren't too high.
And yes, this is an argument against the alternative you're suggesting.
It's pretty clear that in this case the use of AI is not a matter of laziness, but rather quality/consistency assurance. I use code formatters not because I'm too lazy to indent code myself, but because it helps guarantee that it's formatted consistently. I use a stud finder when mounting things to walls not because I'm too lazy to do the “knock on the wall” trick, but because the stud finder is more precise and reliable at it.
I don't use AI to edit my comments, but if I did, it would be not because I'm too lazy to check for all the things I want to avoid putting in my comments, but as an extra layer of assurance on top of what I've already trained myself to do.
But that's not something anybody wants of you in an informal context such as this (HN). It will flatten your voice and make you sound like a drone. We value a human voice.
Code is different. Outside of hobbies, code is not a form of self-expression. There's a reason why following your companies coding styles & practices is valued in software engineering. Companies value coders being interchangeable with each other, they do not want a "unique voice". I think it's completely unrelated to what we're discussing here.
> I don't use AI to edit my comments
What are we even debating, then?
At least that was the case before LLMs became a thing, now I'm not sure anymore.
For example, use "literally" for exaggeration rather than in the original meaning of the word and you'll likely trigger somebody.
It's against the HN guidelines to focus on punctuation, spelling, etc, as long as the comment is understood.
And, in any case, it's now against the guidelines to write using an AI :)
And really, it goes against the spirit of HN to hyperfocus on idioms instead of addressing the meat of the argument...
As a personal observation, if an LLM was figuratively looking over my shoulder and pointed out something like "well, ackshually, 'a few bad apples' means..." I would delete the fucker.
And more relevant to us, a couple bad lines of code sprinkled in the millions in your code base can ruin the entire thing....
Would you prefer to be corrected on some logical fallacy/mistake you made in your argument, by another human being (and yes, maybe get slightly upset about it, we're human beings after all), or have both sides present bot-mediated iron-clad comments, like operators sparring with robots?
I prefer the raw, flawed human version. Even if, yes, I make a silly, avoidable mistake, or get upset, or make you upset in the heat of the argument. Maybe when I cool down I will have learned something.
I don't want flawless robotic arguments. I want human beings. (Fuck, that last bit sounded like an AI-ism, but I promise it's me, a human!).
Bit of a shameless plug but I wrote a HN AI comment detector game[0] with AI and most of my friends and fellow HN users who tried it out couldn't detect them.
This is another reason why it's good to email us (hn@ycombinator.com) rather than commenting when you see generated comments.
There are cases that are more borderline; usually when someone has used a translation service or has used an LLM to polish up a comment they wrote themselves. For these ones there's less certainty, and whilst we discourage them, we're not as rigid in our aversion to them or as eager to ban accounts that do it.
But ones that are entirely generated are still pretty easy to spot, even just from visual appearance.
Looks cool, but how exactly do you gather proven-to-be human comments?
I think it would be better if you used pre-ChatGPT (Nov 30 2022, I think?) stories.
All the AI acounts I’ve seen repeatedly post the exact same cookie cutter top-level comments over and over again. Typically some vapid observation followed by an obviously forced question serving as engagement bait. The paragraphs and sentence structure even looks visually similar across comments when you scroll down the history page.
Just look at a few of these accounts and you’ll easily be able to recognize AI posts on your own.
https://news.ycombinator.com/threads?id=naomi_kynes https://news.ycombinator.com/threads?id=aplomb1026 https://news.ycombinator.com/threads?id=decker_dev https://news.ycombinator.com/threads?id=CloakHQ https://news.ycombinator.com/threads?id=coolcoder9520 https://news.ycombinator.com/threads?id=ptak_dev https://news.ycombinator.com/threads?id=oliver_dr https://news.ycombinator.com/threads?id=agent5ravi https://news.ycombinator.com/threads?id=yuyuqueen https://news.ycombinator.com/threads?id=entrustai https://news.ycombinator.com/threads?id=coder_decoder https://news.ycombinator.com/threads?id=mergisi https://news.ycombinator.com/threads?id=JEONSEWON https://news.ycombinator.com/threads?id=devonkelley https://news.ycombinator.com/threads?id=iam_circuit https://news.ycombinator.com/threads?id=robotmem https://news.ycombinator.com/threads?id=RovaAI https://news.ycombinator.com/threads?id=ajstars https://news.ycombinator.com/threads?id=priowise https://news.ycombinator.com/threads?id=Yanko_11 https://news.ycombinator.com/threads?id=zacklee-aud https://news.ycombinator.com/threads?id=shablulman https://news.ycombinator.com/threads?id=octoclaw https://news.ycombinator.com/threads?id=zacklee1988 https://news.ycombinator.com/threads?id=bhekanik https://news.ycombinator.com/threads?id=webpolis https://news.ycombinator.com/threads?id=claud_ia https://news.ycombinator.com/threads?id=david_iqlabs https://news.ycombinator.com/threads?id=yamarldfst https://news.ycombinator.com/threads?id=julius_eth_dev https://news.ycombinator.com/threads?id=vexnull https://news.ycombinator.com/threads?id=idorozin
Some of us were trained/self taught to write that way. Even "it's not X, it's Y" is a legitimate and subjectively effective communication tool, and there are those of us who either by training modeling have picked it up as a habit. It's not Ai that started this, Ai learned it from us.
Crap - I just did it, didn't I? Awww double crap! Did it again...
So I think it's fine to scrutinize commenters who write that way.
Besides, the biggest offense of AI speak is making everything seem like a grand epiphany and revolutionary discovery. Aka engagement bait.