Posted by usefulposter 23 hours ago
This feels like don't buy at Walmart, support the local small shop. We passed the no return sign miles ago.
Gemini's:
This is like advocating for artisanal blacksmithing in the age of industrial steel. It sounds great in theory, but we passed the point of no return miles back.
Yeah, we can tell the difference :)
Unrolling a metaphor into its literal meaning is one of the most annoying features of the "AI voice", IMO
I acknowledge this is partly just my personal bias, in some cases really not fair, and unenforceable anyway, but someone relying on llms just makes me feel like they have... bad taste in information curation, or something, and I'd rather just not interact with them at all.
I am one of those folks, and I’m strongly against AI writing for that use case as well.
The only reason I can communicate in English with some fluency is that I used it awkwardly on the internet for years. Don’t rob yourself of that learning process out of shyness, the AI crutch will make you progressively less capable.
Why do you need to communicate in English with us native English speakers? Why don't we need to learn your language to communicate with you?
The way I'm looking at it is that you're putting all this effort towards learning how to communicate with people who would never without an outside pressure do the same for you.
If language learning is intrinsically a positive thing what can we do to encourage it in native speakers of English, specifically Americans who are monolingual (as they dominate this website)?
Imagine a scenario where Dang announced that we're only allowed to post in English one day week -- every day is dedicated to another language, like Spanish, Russian, Mandarin and the system auto deleted posts that weren't in those languages. Would that be a good thing? Would we see American users start to learn Spanish to post on HN on Tuesdays?
A century ago it was French or Latin, and a century from now it might be Mandarin or something else. The existence of a standard is what matters.
The only complain I have about Americans and language is that most tech companies fail spectacularly at supporting multilingualism, from keyboards struggling with completion to youtube and reddit forcing translations on users.
I don't care if they use an LLM to ask questions about grammar or whatever, as long as they write their own text after figuring out whatever it was they were struggling with.
I'm an English speaker with some Spanish education and practice. My experience is that reading, writing, listening, and speaking can be quite uneven. Uneven enough to matter.
In the long-run, yes, learning a language is better, assuming your goal is to learn the language. I'm not trying to be snarky: sometimes people simply want to communicate an idea quickly in the short-run and/or don't prioritize deepening a language skill.
I would rephrase the comment above as a question: "Given the set of tools available (in person tutoring, online tutoring, AI-tooling, etc) and what we know about learning from cognitive science, for a given budget and time investment, what combination of techniques work better and worse for deepening various language skills?"
We've all pasted news articles into 2022 Google Translate and a modern LLM, right, and there was no comparison? LLMs even crushed DeepL. Satya had this little story his PR folks helped him with (j/k) even, via Wired June '23:
---
STEVEN LEVY: "Was there a single eureka moment that led you to go all in?"
SATYA NADELLA: "It was that ability to code, which led to our creating Copilot. But the first time I saw what is now called GPT-4, in the summer of 2022, was a mind-blowing experience. There is one query I always sort of use as a reference. Machine translation has been with us for a long time, and it's achieved a lot of great benchmarks, but it doesn't have the subtlety of capturing deep meaning in poetry. Growing up in Hyderabad, India, I'd dreamt about being able to read Persian poetry—in particular the work of Rumi, which has been translated into Urdu and then into English. GPT-4 did it, in one shot. It was not just a machine translation, but something that preserved the sovereignty of poetry across two language boundaries. And that's pretty cool."
---
edit: this comment has some comparisons incl. w/the old Google Translate I'm referring to:
https://news.ycombinator.com/item?id=40243219
Today Google Translate is Gemini, though maybe that's not the "traditional translation tool" you were referencing... but hope there's enough here to discuss any aspect that might be interesting!
edit2: March 2025 comparison-
https://lokalise.com/blog/what-is-the-best-llm-for-translati...
"falling behind LLM-based solutions", "consistently outperformed by LLMs", "Not matching top LLMs"
Telling an LLM to "refine" your writing is just lazy and it doesn't help you learn to express yourself better. Asking it for various ways of conveying something, and picking one that suits you when writing a comment is OK in my book.
The way I see it, people will repeat the same grammar and pronunciation mistakes, and use restricted vocabulary their whole lives, just because learning requires effort, and they can't be bothered.
I can accept that nobody is perfect, as long as they have the will to improve.
To me those are the same thing excepting the number of options given to the human...
Then you should have no issue with people using LLMs to communicate more clearly.
My raw thought: I wonder how many people are really objecting to the loss of exclusivity of their status derived from their relative eloquence in internet forums. When everyone can effectively communicate their ideas, those who had the exclusive skill lose their advantage. Now their core ideas have to improve.
Same idea, LLM-assisted: I wonder how many objections to LLM-assisted writing really stem from protecting the status that comes with relative eloquence. When everyone can express their ideas clearly, those who relied on polished prose as a differentiator lose that edge. The conversation shifts to the quality of the underlying ideas — and not everyone wants that scrutiny.
Same ideas. Same person. One reads better. Which version do you actually object to?
AI polished writing shaves away all those weird and charming edges until it's just boring.
First, what "loophole" is the comment above referring to? Spell-checking and grammar checking? They seem both common and reasonable to me.
Second, I'm concerned the comment above is uncharitable. (The word 'loophole' is itself a strong tell of that.)
In my view, humanity is at its best when we leverage tools and technology to think better. Let's be careful what policies we put in place. If we insist comments have no "traces of LLM" we might inadvertently lower the quality of discussion.
Unfortunately (a) is more common, and the backlash against has been removing the communinity incentive to provide (b).
But the "This is what ChatGPT said..." stuff feels almost like "Well I put it into a calculator and it said X." We can all trivially do that, so it really doesn't add anything to the conversation. And we never see the prompting, so any mistakes made in the prompting approach are hidden.
You don't possess an AI, you are using someone's AI
I'm reasonably sure the instance of Olmo 3.1 running locally on this very machine via ollama/Alpaca is very much in my possession, and not someone else's.
No? Then it's not "your" AI, it's an AI that you are using.
An alternative I tried was sharing links my LLM prompts/responses. That failed badly.
I like the parallel with linking to a Google/DuckDuckGo search term which is useful when done judiciously.
Creating a good prompt takes intelligence, just as crafting good search keywords does (+operators).
I felt that the resulting downvotes reflected an antipathy towards LLMs and the lack of taste of using an LLM.
The problem was that the messengers got shot (me and the LLM), even though the message of obscure facts was useful and interesting.
I've now noticed that the links to the published LLM results have rotted. It isn't a permanent record of the prompt or the response. Disclaimer: I avoid using AI, except for smarter search.
If we want human "on the other end" we gotta get to ground truth. We're fighting a losing battle thinking that text-based forums can survive without some additional identity components.
Look at Reddit… abundance of rules do not save that place at all. It’s all about curating what kind of people your site attracts. Reddit of course is a business so they don’t care about anything other than max number of ad views.
Small non profit forums should consciously design a site to deter group(s) of people that they do not want.
I don’t think most people read any sort of TOS, site rules, end license agreements, when was the last time you ever did?
Besides, sometimes it’s worth it to keep a rule breaking user if they are interesting and have worthwhile things to say despite their… theoretical conflict with the site intended use. Rules are too crude of a tool. Especially in case of AI they are quite nebulous even in a world where detection would be perfect (it isn’t).
What you want is to design a site that pulls people that value genuine human interaction. Niche sites are already immune to commerce and adversary bots because no one cares/knows about them. Well this site isn’t that niche I guess, some corporate astroturfing happens.
I am on one niche subculture social media and it has suprisingly well made design that is paramount to who it caters and who it dissuades. The result is lack of text ai content even though it isn’t obvious at first glance. LGBT flags are everywhere to dissuade the chuds. Israel flags are present to dissuade the annoying politics ppl from reddit. Lots of artsy stuff to speak to the genuine creativity.
It looks stupid but it isn’t stupid. It’s actually quite ingenious.
HN is probably already dead as it is too high profile in certain circles to avoid mainstream adversarial AI content.
Once LLM generated speech or content start getting into the live answers of Q&A sessions, that would be sad. I know some people try to get through interviews, but I think that might be a bit harder to not detect.
That's just marketing-speak. LLMs sound like that because LLMs were trained on marketing-speak.
Whether a company/business uses an LLM or a real human to write a particular piece of text, that piece of text is entitled to free speech protections on the basis of the company signing off on it. Not on the basis of how that piece of writing was produced.
That said, I believe that LLMs' "unique" writing style may be useful ability to protect anonymity against stylometric attacks, although that still ought to be checked. If true, that would be a case where LLMism would be desirable by the author.
But here's where it gets tricky: Do I prefer low-effort, off-the-top-of-my-head reactions, as long as it is human? Or do I want an insightful, well-thought-out response, even if it is LLM-enhanced?
Am I here to read authentic humans because I value authenticity for its own sake (like preferring Champagne instead of sparkling wine)? Or do I value authentic human output because I expect it to be of higher quality?
I confess that it is a little of both. But it wouldn't surprise me if someday LLM-enhanced output becomes sufficiently superior to average human output that the choice to stick with authentic human output will be more painful.
This is an artificial dichotomy. HN’s guidelines specify thoughtful, curious discussion as a specific goal. One-off / pithy / sarcastic throwaway comments are generally unwelcome, however popular they are. Insightful responses can be three words, ten seconds to write and submit, and still be absolutely invaluable. Well-thought-out responses are also always appreciated, even if they tend to attract fewer upvotes than a generic rabble-rousing sentiment about DRM or GPL or Apple that’s been copy-pasted to the past hundred posts about that topic. But LLM-enhanced responses are not only unwelcome but now outright prohibited.
Better an HN with fewer words than an HN with more AI writing words. We’ve been drowned in Show HN by quantity as proof of why already.
That's the dichotomy: Do we prefer text with the right "provenance" over higher quality text?
[Perhaps you'll say that human+LLM text will never be as high-quality as human alone. But I'm pretty sure we've seen that movie before and we know how it ends.]
That said, you're right that because human+LLM is so much more efficient, we'll be drowning in material--and the average quality might even go down, even if the absolute quantity of high-quality content goes up.
I think, in the long term, we will have to come up with more sophisticated criteria for posting rather than just "must be unenhanced human".
HN need not offer itself up as a Petri dish for AI writing experimentation. There are startups in that space, and at least one must be YC-funded, statistically speaking. Come back with the outcomes of the experiment you describe and make a case that they should change the rule. Maybe they will! As of today, though, they are apparently unconvinced.
> the average quality might even go down
We have a recent concrete analysis of Show HN indicating support for this possibility, resulting in the mods banning new users for posting to Show HN (something they’ve probably been resisting for close to twenty years, I imagine, given how frequent a spam vector that must be).
> Perhaps you’ll say that human+LLM text will never be as high-quality as human alone
Please don’t put words in my mouth, insinuating the tone my reply before I’ve made it, and then use that rhetorical device to introduce a flamebait tangent to discredit me with. I’ve made no claims about future capabilities here and I’m not going to address this irrelevance further.
> in the long term, we will have to come up with more sophisticated criteria
Our current criteria seem sophisticated already. Perhaps you could make a case that AI-assisted writing helps avoid guideline violations. This one tends to be especially difficult for us all today:
”Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith. Eschew flamebait. Avoid generic tangents.”
I apologize--the "you" I meant was the person currently reading my post, not the person I was replying to. I was merely trying to answer a common objection that I've heard.
> HN need not offer itself up as a Petri dish for AI writing experimentation.
I'm not sure HN has a choice. I don't think we can prevent posters from experimenting with LLMs to post on HN--even if they adhere to the guidelines. For example, can I ask the LLM to come up with the strongest argument it can and then re-write it in my own words? That seems to be allowed by the guidelines. Would someone even be able to tell that's what I did? [NOTE: I did not do that.]
I think you're arguing that we should not encourage even more use of LLMs on HN. I get that. But I feel like that this community is uniquely qualified to search for better solutions.
> Our current criteria seem sophisticated already.
I hope you're right! That implies that you believe the current guidelines are sufficient to keep HN as the place we all love despite the assault from LLMs. I'm skeptical, but I've been wrong plenty of times!
And yet, she persisted, we will still set guidelines; so that people know they’re unwelcome to do so when they do, so that they can’t argue that they didn’t know, so that we as a social club can strive towards the standards we argue about and accept from the organizers. The point of guidelines is not that they prevent malicious intent; the point is that they inhibit those behaviors that exceed the defined boundaries, however vague or precise they may be. Prevention of malice is an impossibility in all human social affairs, whether guidelines are defined or not; one must find other reasons for rules than prevention to understand why rules are at all.
I'm not sure if you're including or excluding me from the "we". If you're excluding me, then I feel our conversation has come to an end.
But if you're including me, then I think the guidelines need to evolve to deal with LLMs. Maybe not right now--maybe the current guidelines are sufficient for the next year or two or three. But I think we as a community are uniquely qualified to design and influence the future of internet social clubs in the face of LLMs.
“We” here refers to individual human beings that are members of the human social-entity constructs (‘social clubs’) that precipitate naturally out of human groups, both in general to all such groups and in specific to the group under discussion here today, HN participants.
Whether or not you’re a member of “we” HN participants is conditional on whether or not you are honoring the policy of no AI-assisted writing at HN that is in effect as of whenever you saw this post or the new guidelines. I have no judgment to offer you in that regard, and in any case you’re readily able to decide that for yourself. Separately, I’m not engaging with discussion about future policy; perhaps you should start a top-level thread about it, or write a blog post and submit it (after a few days have passed, so it doesn’t get topic-duped and so that passions have cooled somewhat).
The guidelines are perfectly clear, no matter the outcome of your thought experiment. Hacker News wants intelligent conversation between human beings, and that's the beginning and the end of it.
If you want LLM-enhanced conversation then I'm sure you will find places to have that desire met, and then some. Hacker News is not that place, and I pray that it will never become that place. In short, and in answer to "Do we prefer text with the right "provenance" over higher quality text?".
Yes. Yes, we do.
For me it's the first one every time. If only because LLM don't learn from responses to it (much less so when the response is to a paste of their output). It's just not communication. From that perspective, the quality of even the most brilliant LLM output is zero, because it's (whatever high value) multiplied by zero.
Even a real person saying something really horrible and too dense to learn from any response at least gives me a signal about what humans exist. An LLM doesn't tell me anything, and if wanted the reply of an LLM, I would simply feed my own posts into an LLM. A human doing that "for me" is very creepy and, to my sensibilities, boundary violating. Okay, that may be too strong a word, but it feels gross in a way I can't quite put my finger on, but reject wholeheartedly.
I'd argue that anything insightful or well-though-out doesn't use LLMs at all. We can quibble over whether discussions with an LLM lead to insightful responses, but that still isn't your own personal thought. Just type what's on your mind, it's not that hard and nitpicking over this is just looking for ways to open up unnecessary opportunities for abuse.
The more you use an LLM to write for you, the worse you will become at writing yourself. There is simply no other possible outcome. It's even true of spellcheck - the more you use a spellcheck the worse you become at spelling. I know this for a fact because I can no longer spell for shit. However, spelling is to writing as arithmetic is to mathematics. I also can't add up, but I have a degree in pure mathematics.
LLMs are a cancer on human thought and expression.
LLMs help to express what many people dont have the energy or ability to express. It also has a broader scoped view of protocol...It does not have emotions, which often leads to less than optimal discourse.
In many ways, it help those who are challenged in discourse to better express themselves...rather than keeping silent or being misunderstood.
This seems especially relevant for non-English-fluent commenters, who are increasingly using LLMs to be able to communicate more effectively on an English-only site like Hacker News than they'd otherwise be able to do.
It's still daunting posting in a second language, and LLMs are an attractive solution to that (depending on your definition of 'solution').
In any case, I don't think it's a bad thing to want to communicate as clearly as possible, and if an LLM helps you do that, I ain't one to judge. Sure, ideally I'd want to read folks' thoughts without the LLM-induced layer of vaseline smoothing them over, but even that's better than not reading them at all :)
Use them to get better, like how reading good writing directly (not summarized) will also make you a much better writer. Learn from the before and after so next time there isn't a need to reach for Ai.
Anyone learning the language and some people with learning disabilities, for example, may communicate better via an LLM.
The tension is that as insightful discussion becomes easier/better with LLMs, there is less need to read HN. All I'm left with is provenance: reading because a human wrote it, not because it is uniquely insightful.
LLMs, as we know them, express things using the patterns they've been developed to prefer. There's a flattening, genericizing effect built in.
If there are people who find an LLM filter to be an enhancement, they can run everything through their favorite LLM themselves.
1. I enter "Describe the C++ language" at an LLM and post the response in HN. This is obviously useless--I might as well just talk to an LLM directly.
2. I enter "Why did Stroustrup allow diamond inheritance? What scenario was he trying to solve" then I distill the response into my own words so that it's relevant to the specific post. This may or may not be insightful, but it's hardly worse than consulting Google before posting.
3. I spend a week with creating a test language with a different trade-off for multiple-inheritance. Then I ask an LLM to summarize the unique features of the language into a couple of paragraphs, and then I post that into HN. This could be a genuinely novel idea and the fact that it is summarized by an LLM does not diminish the novelty.
My point is that human+LLM can sometimes be better than human alone, just as human+hammer, human+calculator, human+Wikipedia can be better than human alone. Using a tool doesn't guarantee better results, but claiming that LLMs never help seems silly at this point.
I think where you are getting hung up is the idea of "better results". We as a community don't need to strive for "better results" we can easily say, hey we just want HN to be between people, if you have the LLM generate this hypothetical test, just tell people in your own words. Maybe forcing yourself to go through that exercise is better in the long run for your own understanding.
But my point is that I read HN partly because people here are insightful in a way I can't get in other places. If LLMs turn out to ultimately be just as insightful, then my incentive to read HN is reduced to just, "read what other people like me are thinking." That's not nothing, but I can get that by just talking with my friends.
Unless, of course, we could get human+LLM insightfulness in HN and then I'd get the best of both worlds.
And what motivated you to make it -- probably the most interesting thing to readers, and not something an LLM would know.
Believe me, I don't care what an LLM has to say about your thing. I care about what you have to say about your thing.
The value proposition is that someone who is a lousy writer (perhaps only in English) with deep domain knowledge is going back and forth with the LLM to express some insight or communicate some information that the LLM would not produce on its own.
Wouldn't it work better to just write the thing in whatever language they can actually write in and then do a straightforward translation in a single pass?
> someone who is a lousy writer with deep domain knowledge going back and forth with the LLM to express some insight or communicate some information that the LLM would not produce on its own
This sounds reasonable on its face, but how often does it actually come up that somebody can't clearly express an idea in writing on their own but can somehow get an LLM to clearly express it by writing a series of prompts to the LLM?
And, if it does come up, why don't they just have that conversation with me, instead?
Nontrivial translation tools are AI(neural net)-based tools (although not necessary LLM). Whole transformer neural net architecture was originally designed for translation.
Because (the royal) you will be argumentative and shitty, and sour this person on their desire to communicate their knowledge at all.
Just as Google-enhanced output and Wikipedia-enhanced output has helped my writing/thinking, I believe LLM-enhanced output also helps me.
Plus, I personally gain more benefit from using an LLM as a researcher than as a writer.
Neither. I want insightful, well-thought-out, human comments.
It's a little sad that this might be too much to ask sometimes...
If your definition of "superior" includes some amount of "provides a meaningful connection to another living being", then LLM output will rarely be superior even when it's factually and grammatically correct.
Pretty sure this comment is AI
and
> Or do I want an insightful, well-thought-out response, even if it is LLM-enhanced?
What is the difference? What's the line between these two?
The prompt: "Analyze <opinion> and respond" is pretty clearly "I would just ask it." and, the prompt: "here's my comment, please ONLY the check the grammar and spelling" would probably be ok.
What about prompt:"I disagree with using LLMs for commenting at all for <reasons>. Please expound on this and provide references and examples". That would explode the word count for this site.
1. "Here is my answer to a comment. Give me the strongest argument against it."
2. "I think xyz. What are some arguments for and against that I may not have thought of."
3. "Is it defensible for me to say that xyz happened because of abc?"
All of these would help me to think through an issue. Is there a difference between asking a friend the above vs. an LLM? Do we care about provenance or do we care about quality?
And I find the decision to "ban" AI slightly ironic, when HN has a disdain (unlike its predecessor Slashdot) for funny or sarcastic comments, which require the reader to think more, rather than having a clear argument handed on a silver platter. I mean, it is what truly human communication is like - deliberately not always crystal clear.
I suspect that HN will eventually be replaced by an AI-moderated site, because it will have more quality content.
I believe banning AI is a temporary solution. Even today it is very hard to tell human from AI. In the future it will be impossible. We are in the Philip Dick future of "Do Androids Dream" (the book, not the movie). Does it matter if we can't tell human from AI? The book proposes that how we feel about the piece we're reading is the only thing that matter. How the piece got created is irrelevant.
1. Pre-moderation - AI looks at your comment before you submit it, and suggests changes for clarity, factuality and argumentative strength. You can decide whether to accept these (individual) changes or not. It will also automatically flag if it breaks moderation guidelines too much.
2. Discussion summary - AI will periodically edit main debate points and supporting sources into a comprehensive document, which you can further add to with your comment. This will help to steer the discussion and make it easier to consume in the future. It can also make discussions less ephemeral, which is a huge problem.
As humans, we have directives (genetic, cultural, societal, etc.) to prioritize humanistic endeavors (and output) above all else.
History has shown that humans are overwhelmingly chauvinistic in regards to their relationship to other animals in the animal kingdom, even to the point of structuring our moral/ethical/legal systems to prioritize human wellbeing over that of other animals (however correct/ethical that may ultimately be, e.g., given recent findings in animal cognition, such as recent attempts to outlaw boiling lobsters alive as per culinary tradition).
But, it seems that some parties/actors are willing (i.e., benefiting) from subverting this long-standing convention (of prioritizing human interests) in the face of AI (even to the point of the now-farcical quote by Sam Altman that humans take far more nurturing than LLMs...)
So: should we be neglecting our historical and genetic directives, to instead prioritize AI over human interests? Or should we be unashamedly anthropic (pun intended), even at the cost of creating arbitrary barriers (i.e., the equivalent of guilds) intended to protect human interests over those of AI actors?
I strongly recommend the latter, particularly if the disruptions to human-centric conventions/culture/output are indeed as significant (and catastrophic) as they will likely be if unchecked.
There's no insight nor well-thought-out response once a person decides to "LLM-enhance" their response. The only insight that the person using the LLM is too limited to have a decent conversation with.
My ideal vision is that instead of outsourcing indefinitely, we learn from the enhanced versions and become better independent writers.
Mate, Champagne is a sparkling wine. In French you can even at times hear people asking for "un vin mousseux de Champagne" meaning "a sparkling wine from Champagne" instead of the short form (just saying "un Champagne" or "du Champagne").
Now, granted, not all sparkling wine are Champagne.
The Wikipedia entry begins with: "Champagne is a sparkling wine originated and produced in the Champagne wine region of France...".
I drank enough of it to be stating my case, of which I'm certain!
P.S: and btw, yup, authentic humans content only here, even if it's of "low quality". If I want LLM, I've got my LLMs.
So just like Armanac's are like Cognac's for lower price, good Cremant will be cheaper and more enjoyable that cheaper Champagne (I've not had any really expensive Champagne).
Then you have Cava from Spain which is similar process to Cremants and Champagne. The difference would be in type of grapes used. A friend of mine swears by Cavas just like I swear by Cremants from Loire region. However my wife hates Cava.
Then Proseccos from Italy again are similar, but quality varies more.
After that we get into more questionable cheaper sparkling wines which usually means some sort of out of bottle insertion of CO2 and even worse version include some other modifications such as sugar.
In general to avoid literal headaches you want BRUTs. Anything semi-sweet or sweet is suspicous.
Again I am not a full wine expert but this is mostly years of ahem experience.
And no, I wouldn't think an HN post is it either.. I'm just saying, there should be a good place to post the output of good questions asked iteratively.
Claude is a bit better but still prone to rambling.
I think that if you actually try reading someone else's conversation with LLM, you'll find out that it's less exciting than it seems.
For the one who has the conversation the excitement comes mostly from the ability to steer it the way you want. Reader doesn't have this ability, so they are just forced to endure the excessive wordiness, that is so typical for most LLMs.
If you learned something interesting, then why not express this knowledge in a normal article/blogpost? What advantage does a conversation between you and LLM has over just a normal text or, perhaps, text with pictures, diagrams, maybe some interactive illustrations etc
If you can’t even be arsed doing that how much value is there, really?
Personally the only thing less interesting to me than someone else’s conversations with an LLM is hearing about someone else’s dream they had last night but you never know, some people may be interested.
But I was thinking less blog and more like an LLM research notebook, à la Jupyter. Jupyter for LLM prompts, outputs, refinements.
Where to post it? Any blog site, probably a good few Show HN too. Will anyone read it? I haven't read anyone else's, I'm more inclined to dock them reputation for suggesting I read their Ai session. Snippets of weird things shared on socials were interesting to me early on, but I'm over that now too.
[1] https://simonwillison.net/2025/Dec/25/claude-code-transcript...
I think "generated comments" is a pretty hard line in the sand, but "AI-edited" is anything but clear-cut.
PS - I think the idea behind these policies is positive and needed. I'm simply clarifying where it begins and ends.
All this stuff is in flux. I thought a lot about whether to add the "edited" bit - but it may change. What I deliberately left out was anything about the articles and projects that get submitted here. There's a lot of turbulence in that area too, but we don't yet have clarity, or even an inkling, of how to settle that one.
Edit: what I mean is this: while most of those submissions aren't very interesting, some really are. Here's an example from earlier today:
Show HN: Vanilla JavaScript refinery simulator built to explain job to my kids - https://news.ycombinator.com/item?id=47338091
How do we close the aperture for the lame stuff while opening wider for the good stuff? That is far from clear.
If you're going to say that the AI said X, Y, Z, provide a rationale on why it is relevant. If you merely found X, Y and Z compelling, feel free to talk about it without mentioning AI.
> If you merely found X, Y and Z compelling, feel free to talk about it without mentioning AI.
I think you're seeing this as too black-and-white, and missing the heart of the issue.
The purpose of mentioning AI is to convey the level of (un)certainty as accurately as possible. The most accurate way to do that would often be to mention any use of AI, rather than hiding it.
If AI tells me that it believes X is true because of links A and B that it cites, and I find those links compelling, then I absolutely want to mention that AI gave me those links because I have no clue whether the model had any reason to bias itself toward those sources, or whether alternate links may have existed that stated otherwise.
Whereas if a normal web search just gives links that mention terms from my query, then I get a chance to see the other links too, and I end up being the one who actually compare the contents of the different pages and figure out which one is most convincing.
Depending on various factors, such as the nature of the question and the level of background knowledge I have on the topic myself, one of these can provide a more useful response than the other -- but only if I convey the uncertainty around it accurately.
In my experience, LLMs hallucinate citations like crazy. Over 50% of the times I've checked, the citation either didn't exist, or it did but didn't support the LLM's assertions.
This is true not just from the chat, but for Google AI summaries.
When the references are more often wrong than not, you can understand why many will simply downvote you for bringing LLM citations into the conversation. Why quote a habitual liar?
(If you look at my other comments, I'm actually in favor of using LLMs in some capacity for HN comments. Just not in this case.)
> In my experience, LLMs hallucinate citations like crazy. Over 50% of the times I've checked, the citation either didn't exist, or it did but didn't support the LLM's assertions.
Note that those are specifically not the cases where the AI is citing "sources that I feel appear plausible."
(I also don't find over 50% hallucination to be accurate for Google AI summaries in my experience, but that depends on your queries, and in any case, I digress...)
> When the references are more often wrong than not, you can understand why many will simply downvote you for bringing LLM citations into the conversation. Why quote a habitual liar?
To be clear, I do understand both sides of the argument, and I don't think either side is unreasonable. I've also had the experience of being on both sides of this myself, and I don't think there's a clear-cut answer. I'm just hoping to get clarity on what the new policy is as far as this goes. I'm sure it'll be reevaluated either way as time goes on.
I should point out that I'm not saying 50% of the AI summaries have an error. Merely that the references it provides me don't state what the summary is claiming. The summary may still be accurate, while the references incorrect.
However, that's probably not critical enough to formally add to the explicit guidelines, so it's probably fine to leave it in the "case law" realm—especially because downvoters tend to go after such comments.
The comments thing is a lot more intimate in the sense that anyone posting comments is inside the house.
I have a kid with severe written language issues, and the utilisation of speech to text with a LLM-powered edit has unlocked a whole world that was previously inaccessible.
I would hate to see a culture that discourages AI assistance.
These rules are always fuzzy and there's always a long tail of exceptions. All the more so under turbulent conditions like right now. I wrote more about this elsewhere in the thread, in case it's useful: https://news.ycombinator.com/item?id=47342616.
> I would hate to see a culture that discourages AI assistance.
Mostly I think the push back is about ai assistance in its current form. It can get in the way of communicating rather than assisting. The cost though is mostly borne by the readers and those not using the AI for assistance. I have seen this happen when the ai adds info and thoughts that were tangental to the original author and I think, but I can not verify times where an author seems to try to dig down on the details but seemingly can not.
https://news.ycombinator.com/item?id=47326351
Yes, please at least have a carveout for accessibility. I definitely have dictated HN comments in the past, and my flow uses LLMs to clean it up. It works, and is awesome when you're in pain.
It's better to communicate as an individual, warts and all, than to replace your expression with a sanitized one just because it seems "better." Language is an incredibly nuanced thing, it's best for people's own thoughts to come through exactly as they have written them.
So yeah, it can change the character of your writing, even if it's just relatively subtle nudges here or there.
edit: we suggested that he disable that feature to help him learn to write independently, and he happily agreed.
1. A system that suggests words, the child learns the word, determines whether it matches their intent, and proceeds if they like the result.
2. A system that suggests words, and the child almost-blindly accepts them to get the task over with ASAP.
The end-results may look the same for any single short document, but in the long run... Well, I fear #2 is going to be way more common.
The phenomenon was observed in religious philosophy over a millennium ago (https://terebess.hu/zen/qingyuan.html).
Now that it is, I just turn tab completion off totally when I write code by hand. It's almost never right.
I have mixed feeling about it. On the one hand, you're right: carefully considering suggestions can be a learning opportunity. On the other hand, approval is easier than generation, and I suspect that without flexing the "come up with it from scratch" muscle frequently, that his mind won't develop as much.
A "click to see more about why this answer fits" crossword, on the other hand...
Shoal, Hasty, Lobby, Vogue, Gunky,
Sheep, Theft, Linen, Slime, Fluke,
Hydra, Dizzy, Lance, Shred, Buyer,
Attic, Guava, Awake, Stank, Hoist,
Mogul, Squad, Roost, Skull, Bloom,
Mooch, Surge, Vegan, Scene, Cello,
If I had to guess at a riskier 20%... Guava, a fruit some people may not have had; Gunky because it's slang; Mogul, Vogue, and Mooch were borrowed from other languages; Cello is something people may have heard more than read; Hoist.
That's a good point and could very well be true. I just know I've played plenty of games where I was mad that they didn't show the meaning. So let's say its 5% for native speakers, and up to 20% for non-native speakers - that's still a golden opportunity to expand vocabularies. And honestly it can't be a lot of work to add a couple lines of static text. At worst it would be ignored, and at most, help people learn more interesting words.
A certain amount of friction is necessary, at least if the goal is to help the person learn or make something original.
As an adult, I do too. As a middle schooler, we absolutely used word processors’ thesaurus features to add big words to our essays because the teachers liked them.
Anyway before that she HATED the thesaurus. And she could tell when students were using it to make their writing more fancy pants.
I had two teachers who called us out on this, and actually coached us on our writing, and I remember them fondly. (They were also fans of in-class essaying.)
The others wanted to count big words.
It is definitely not true that it is better for a poster to communicate like an individual when it comes to spelling and grammar. People ignore posts that have poor grammar or spelling mistakes, and communications that have poor grammar are seen as unprofessional. Even I do it at a semi-subconscious level. The more difficult or the more amount of attention someone has to pay to understand your post, the less people will be willing to put in that effort to do so.
[It looks like MS Word 97 had the ability to detect passive voice as well, so we're talking 30 year old technology there that predates LLMs -- how far down the Butlerian Jihad are we going with this?]
There is no need for that here beyond maybe spellcheck. Use your own thoughts, voice, and words.
> HN is for conversation between humans.
If it is enhancing that instead of detracting and wasting peoples time it does not seem to be against the spirt of the rules.
That is from dang's post in: https://news.ycombinator.com/item?id=47342616
That whole post is clarifying for the intent of the new rule(s).
"Don't post generated comments or AI-edited comments."
What about non-native speakers? Can they not use translation software like google translate any more?
"Don't post generated comments or AI-edited comments, except for translating to english"
What about cases of disabilities?
"Don't post generated comments or AI-edited comments, except for translating to english and when used as assistive technologies."
Some translation tools and assistive technologies are still going to case the same issues that we have right now so maybe limit the technologies used
"Don't post generated comments or AI-edited comments, except for translating to english and when used as assistive technologies. Technologies x, y, z are not allowed a and b and similar can be used for translation c and d as assistive technologies"
But we do not want to spend time/effort on filtering technologies and/or people into the above categories.
In the long run we likely will come up with technologies that most everyone is satisfied with using in different use cases, spelling grammar, assistive, maybe even tone, and others.
In the mean time we can not let the perfect be the enemy of the good. If there are clear standards that achieve the goals, great, if not we have to do something until everything shakes out.
Nobody is going to stop using grammarly extensions to post to HN, nobody is going to be able to detect its usage.
This thread just lets a certain kind of people put on their best condescending hall-monitor voice and lecture other people about how they should behave.
And the rule is arguably less useful than speed limits and will be broken about as often (at least speed limits have a very real link to physical safety via kinetic energy).
"Your unique human voice is more valuable than a thousand prompt-driven LLM doggerels."
Edit: I already got downvoted. :-) Sure, no one can tell exactly why. Maybe the combination of bad English _and_ talking sh*ce isn't ideal at all. :-D Anyways, I have enough karma, so I can last quite a while..
The quality of my writing varies (based on my mood as much as anything else, I suppose), but when it is particularly good and error-free then I often get accused of being a bot.
Which is absurd, since I don't use the bot for writing at all.
How do you know? Is it possible the downvoters just didn't like what you said?
It suggests a bias in writers to assume that people would agree with them if only they could express their thoughts accurately.
There are people here who sit at a desk all day banging out multipage emails for work who decide to write posts of a similar linguistic calibre for funsies.
Meanwhile you have someone in a developing country who just got off a brutal twelve hour shift doing manual labour in the sun who wants to participate in the conversation with an insightful message that they bang-out on a shitty little cellphone onscreen keyboard while riding on bumpy public transit.
You could have a great idea and express it poorly and be penalized for doing so here while someone could have a blah idea expressed excellently and it's showered in replies despite being in some metrics (the ones I think are most important) worse than the other post.
What's the solution for that?
Remember that you're on a message board and you're not actually 'competing' for anything?
I knew someone was going to comment on my use of the word there despite me putting it in quotes which was intended to let the reader know that I meant that word as an approximation of what I was meaning.
When I say competing I mean competing in the space of ideas here. There is a ranking system here that raises or lowers the visibility and prominance of your comments and it's based on upvotes by other uses. For better or worse people penalize comments with grammatical errors over ones that don't and that affects how much exposure other users have to the ideas that people write and how much interaction they get from them.
If that's the case why would somebody who has good ideas but poor expressive capability bother posting here if their comments are just going to get ignored over relatively vapid comments that are grammatically correct?
The main problem is that ai consistently is seeing making things worse. Take a look at the examples in Dang's link in their comment: https://news.ycombinator.com/item?id=47342616
In the ones I read the AI editing is either hurting or needs to be much, much better to help.
In English. You have to put your best foot forward in English. And in your environment with the resources you have at your disposal.
For example, I'm currently engaging with you between steps in a chemistry process that's happening under the fumehood next to me while wearing a respirator, a muggy plastic chemical resistant gown and disposable gloves nitrile globes.
I am absolutely certain that these conditions are different than the ones I would need to 'put my best food forward' in this discussion. I'm also certain that quite certain that you and I would both absolutely stumble if we were obligated to particpate in this forum in a language that we're not proficient in as many users often attempt to do and are unfairly penalized for by other members of the community.
I'm with you on the LLM usage for grammatical issues for non-native speakers. I bet more in this community would feel the same way if Dang whimsically mandated that people had to use a language other than English on certain days of the week.
I absolutely do not understand this comment. Are you saying that posting is competitive and that comments have "metrics"?
The guidelines state:
> Be kind. Don't be snarky. Converse > Edit out swipes. > Don't be curmudgeonly.
On the best of days I manage to follow the rules, but I'm only human. If I run my comment through ChatGPT to try and help me edit out swipes on the bad days, that's not ok?
I'm not using ChatGPT to generate comments, but I've got the -4 comments to show that my "thoughts exactly as they have written them" isn't a winning move.
This is the opposite of how language works. You want people to understand the idea you're trying to communicate, not fixate on the semantics of how you communicated. Language is like fashion - you only want to break the rules deliberately. If AI or an editor or whatever changes your writing to be more clear and correct, and you don't look at it and say "no, I chose that phrasing for a reason" then the editor's version is much more likely to be understood correctly by the recipient.
I just want clean, easy-to-read content and I don't care about the person who wrote it. A tool like Grammarly is the difference between readable and unreadable (or understandable and understandable) for many people.
You could even write a plugin for your favorite web browser to do that to every site you visit.
It seems hard to achieve the inverse that is (would you rather I use i.e.?) rewrite this paragraph as the original author did before they had an AI re--write it to make it clean, (--do you like oxford commas, and em/en dashes! Just prompt your AI) and easier to read
For those coming from a language other than English, you are more likely to lose information by using a tool to “reconstruct” meaning from poorly phrased English as an input, as opposed to the poster using a tool to generate meaningful English from their (presumably) well-written native language.
But that creates a private version of the text which the original poster didn't sign off on. You could have fixed something contrary to their intent.
I personally don't see a problem with someone using a grammar checker as long as they aren't just blindly accepting its suggestions. That said, if someone actually is using it in that way, it shouldn't be detectable anyway, so it probably doesn't matter all that much whether or not it's included in the letter of the rule.
For me, the line is precisely at the point where a human has something they want to say. IMO - use the tools you need to say the thing you want to say; it's fine. The thing I, and many others here, object to is being asked to read reams of text that no-one could be bothered to write.
This is probably ok:
>> On a technical level, you can really only guard against software that changes your semantics or voice. If you're letting it alter the meaning (or meanings) you intend, or if it starts using words you would never normally use, then it's gone too far.
This is probably too far:
>>> On a technical level, it's important to recogn1ize that the only robust guardrail we can realistically implement is one that prevents modifications to core semantics or authorial voice. If you're comfortable allowing the system to refine or rephrase the precise meanings you originally intended — or if it begins incorporating vocabulary that doesn't align with your typical linguistic patterns — then you've likely crossed a meaningful threshold where the output no longer fully represents your authentic intent.
Something to consider is that you can analyze your own stylometric patterns over a large collection of your writing, and distill that into a system of rules and patterns to follow which AI can readily handle. It is technically possible, albeit tedious, to clone your style such that it's indistinguishable from your actual human writing, and can even icnlude spelling mistakes you've made before at a rate matching your actual writing.
AI editing is weird, though. Not seeing a need, unless English isn't your native language.
To be clear, I also think you shouldn't rely on auto-correction or LLMs for correctness (they are great for identifying your mistakes, but I think you should then fix the mistakes yourself, to develop your brain). It's just that "assisted" correctness isn't misleading/harmful in the way that "assisted" tone/character/semantics are.
When a policy is introduced to seemingly guard against new problems, but happens to be inadvertently targeting preexisting and common technology, I don't feel like it is "lawyering" it to want clarity on that line.
For example, it could be argued this forbids all spellcheckers. I don't think that is the implied intent, but the spectrum is huge in the spellchecker space. From simple substitutions + rule-based grammar engines through to n-grams, edit-distance algorithms, statistical machine translation, and transformer-based NLP models.
Ultimately, this comes down to people making a good-faith judgment about how much AI was involved, whether it was just minor grammatical fixes or something more substantial. The reality is that there isn’t really a shared consensus on exactly where that line should be drawn.
You forgot the /s ?
Then, I considered whether HN would appreciate posts/comments by a human where they’d had a PR team or a hired editor come in and review/modify/distort their original words in order to make them more whatever. I think that this probably is most likely to have occurred on the HN jobs posts, and I’ve pointed out especially egregious instances to the mods over the years — but in general, the people who post on HN tend to do so from their own voice’s viewpoint, as reaffirmed by the no-AI-writing guideline above. So I decided instead to say “pay a proofreader” because, bluntly, if the community found out that someone was paying a wage to a worker to proofread their HN comments, the response would plausibly be the same mob of laughing mockery, disgusted outrage, and blatant dismissal that we see today towards AI writing here. “You hired someone to tone-edit your HN comments?!” is no different than “You used Grammarly to tone-edit your HN comments?!” to me, and so it passed the veracity test and I posted it.
It was asked that if "AI Generated Code" is just code suggested to you by a computer program, where does using the code that your IDE suggests in a dropdown? That's been around for decades. Is it LLM or "Gen AI" specific? If so, what specific aspect of that makes one use case good and one use case bad and what exactly separates them?
It's one of those situations where it seems easy to point at examples and say "this one's good and this one's bad", but when you need to write policy you start drowning in minutia.
IDE code suggestions come from the database of information built about your code base, like what classes have what methods. Each such suggestion is a derived work of the thing being worked on.
By the same token, what if I have a human editor help me out? What if we go back and forth on how to write something, including spelling, grammar, tone, etc. For example, my wife occasionally asks me to review her messages before sending them because she thinks I speak well and wants to be understood correctly.
The problem is that we are punishing the technology, not the result. Whether it's a human or an LLM that acts as your editor should be irrelevant; what matters is that you are posting your own work and not someone else's. My wife having me write all of her messages for her would be just as dishonest as her having an LLM write all of her messages for her if she always presented them as her own writing. But if she writes the copy and I provide suggests for changes, what's the harm in that? And why should it matter if it's a human or an LLM that provides that assistance?
i type my comments without capitalization like i'm typing into some terminal because i'm lazy and people might hate it but i'm sure they prefer this to if i asked an LLM to rewrite what i type
your writing style is your personality, don't let a robot take it away from you
In fact, I'd argue that lazy commenting is the real problem, which has now been supercharged by LLMs.
I benefit from my phone flagging spelling errors/typos for me. Maybe it uses AI or maybe it uses a simple dictionary for me. Maybe it might even catch a string of words when the conjunction isn't correct. That's all fair game, IMO. But it shouldn't be rewriting the sentence for me. And it shouldn't be automatically cleaning up my typos for me after I've hit "reply". That's on me.
But when I argue on the internet, it's always a 100% me.
And if I get a wiff of LLM-speak from whoever I'm wrestling in the mud with at the moment, they'll instantly get an entry in my plonk-file. I can talk with ChatGPT on my own thank you very much, I don't need a human in between.
"But my <language> is bad... that's why I use LLMs"
So was mine when I started arguing with strangers on the internet. It's better now. Now I can argue in 3 different languages, almost 4 =)
Also low quality wine[0]
That takes (much) time, though. I took about a decade to be comfortable about that.
For instance, if a non-native speaker translates their own writing using machine translation or an AI, is that problematic—provided they personally review and vet the content before posting? I don't think the people calling out AI use on this board are taking issue with that. Ultimately, it’s not about the method; it’s about the author's attitude.
The reason LLMs are so disruptive now is that while "shitposts" used to be obvious, we're now seeing "plausible" low-effort content generated without any human oversight. Irresponsible people have always been around, but LLMs have given them the tools to scale that irresponsibility to an unprecedented level.
The biggest current social problem with AI content is our collective lack of transparency into how much human responsibility was taken.
Give a <100% reliable/accurate AI tool, the same post/code may have had {every line vetted by a human} or {no lines vetted by a human}... and readers have no way of telling which it is!
Because even if no edits needed to be made, the former carries a lot more signal than the latter, because it reduces risk of AI slop and therefore makes the content more valuable.
At the same time, it also costs more time to produce, so in any competitive marketplace (YouTube, paid comments, startup code, etc.) the unvetted AI content will dominate.
AI is a tool. You can use it constructively, like Grammarly, or spellcheck. You don't need to be afraid of it.
Are you learning something in the process? does ti have your full emotional context, beside the full conversation context? There are probably many bade side-effects if people would actually start doing what you mention at scale.
One thing is computer code, which is an intermediate product to an end (instruct the computer what it needs to do) and another is YOUR direct output to some other human being, which is the end game in human-to-human communication.
It can't. It will rewrite anything you give it.
> it can verify your claims before posting
It can't.
> You don't need to be afraid of it
Nobody is afraid of it. It's annoying. General population cannot be trusted to use it in whatever idealistic way you are imagining.
Elon said it well, there must be some disincentive to do this.
After all, no one knows I'm a dog.
When someone posts:
> You could use Redis for that, sure, I've run it and it wasn't as hard as some people seem to fear, but in hindsight I'd prefer some good hardware and a Postgres server: that can scale to several million daily users with your workload, and is much easier to design around at this stage of your site.
then the beholder is trusting not just the correctness of that one sentence but all of the experiences and insights from the author. You can't know whether that's good advice or not without being the author, and if that's posted by someone you trust it has value.
An LLM could be prompted to pretend they're an experienced DBA and to comment on a thread, and might produce that sentence, or if the temperature is a little different it might just say that you should start with Redis because then you don't have to redesign your whole business when Postgres won't scale anymore.
This implies they know the author and can trust them. If they don't know the author then there is no trust to break and they are only relying on the collective intelligence which could be reflected by the AI.
That is to say that trusting a known human author is very different from trusting any human author and trusting any human author is not that much different from trusting an AI.
This is my point.
There is no sane endgame here that doesn't end up with each user effectively declaring who they do and don't care to hear, and possibly transitively extend that relationship n steps into the graph. For example you might trust all humans vetted by the German government but distrust HN commenters.
For now HN and others are free to do as they will (and the current AI situation has been intolerable), however, I suspect in the near future governments will attempt to impose their own version of it on to ever less significant forums, and as a tech community we need to be thinking more clearly about where this goes before we lose all choice in the matter.
This already falls apart though. There are while categories of things which I find "incorrect" and would take up as an argument with a fellow human. But trying to change the mind of an LLM just feels like a waste of my time.
It often is with humans as well.
Look, I'll give you a loose example: It's not uncommon to see a post making an "error" I know from experience. I might take the time to help someone more quickly learn what I felt I learnt to help me get out of that mistaken line of thought. If it's an LLM why would I care? There's thousands of other people, even other LLMs, that I could be talking to instead.
You've set up a framework here where "mutual understanding" is the end goal but that's just not always what's on the line.
Arguing for the sake of convincing onlookers reading the conversation is more likely to be effective, and in that case it doesn't matter if the other person is an LLM.
(naturally "birds aren't real" is a correct vs not correct thing, but the same can be applied to many less-objective things like the best mechanical keyboard or the morality of a war)
It also points out the need for AI writing tools that very strictly just:
1. Point out misspellings and typos.
2. Point our grammar mistakes, if they confuse the point.
3. Point out weaknesses of argument, without injecting their own reasoning.
I.e. help "prompt" humans to improve their writing, without doing the improvement for them.
In fact, I would like a reliable version of that approach for many types of tasks where my creativity or thought processes are the point, and quality-control feedback (but not assistance), is helpful.
This is a mode where models could push humans to work harder, think deeper, without enabling us to slack off.
I don’t mind when non-native speakers use it to express themselves, especially if disclaimed (but I give a pass even if not). Does it bother you?
Personally I would like people to try learning other languages more (it's hard but rewarding) but you can't learn every language ever, and it is really hard to learn a language to fluency.
Not all, but some machine translators can be comically (if not horrifically) bad sometimes. Search Twitter-become-X for examples. Native writers can't pick a working machine translator unless they are explicitly allowed to do so themselves.
But that a site might still want to discourage it, to avoid general degradation. It is a tradeoff.
If someone can write in the target language, just not well, a model could be asked to point out problems for the writer to fix. Rewrite a difficult sentence.
Ideally, I want the speaker’s words translated “verbatim” to English, to the extent possible.