On the note of imperfections... My preference is now strongly towards reading someone's stream of thought thrown down at speed over than that same stream of thought shat through the digestive track of an AI agent. At least that feels human.
To me, it also seems like they AI digestion is getting actively worse? As best as I can tell, all the agentic nature and reasoning for code is now making writing actively worse, as the agent pulls across your whole knowledge base and will take that one thought and eagerly join and context it thinks is relevant, with the reasoning spread throughout the page.
Writing is fundamentally the transfer of information from your brain to my brain. If you have 1000 bits of semantic information you want to transfer, you can't give 300 bits of semantic information to an LLM and have it fill in the remaining 700, because it doesn't know what those 700 bits are. If it's able to guess those 700 bits correctly, then they aren't true semantic information, and you really only have 300 bits you want to transfer. You might as well transfer those bits to me directly, rather than having the LLM add on an extra superfluous 700 bits that I then have to filter out.
For a long time after the internet arrived on the scene, a lot of online news stories would reference websites, papers, polls, etc. without linking to them. There are still news sources doing this today. Sometimes such articles interpret or place context around their hidden references, but a lot of the time they just summarize.
Giving someone the text output of a LLM is very similar to publishing a summary without links to the referenced material. When you were querying your LLM, you could have asked specific questions or asked for a custom focus or point of view. Your intended audience might have questions or different concerns, but they're unable to interact with your LLM. What you have delivered is static and unresponsive. It has all the disadvantages of being machine output without the advantage of being interactive, the way your LLM was for you.
It may have to wait until compute is cheap enough that tokens are essentially free, but we need a system to pass "hyperlinks" to LLM's primed with context, ready to be interactively queried on a chosen context. It's being overly generous to assume that people are putting even 300 bits into a LLM for every 1000 bits of regurgitated writing they try to pass off as their own. When people post LLM output as if it were their own, I have no choice but to assume they had zero knowledge of the subject, but this query taught them what they wanted to learn, and now they're sharing that. That's fine, but please pass an interactive LLM link rather than static text.
Once we have "hyperlinks" for LLM sessions, perhaps we can share LLM output a little more usefully and honestly.
I've seen professional journal pieces refer to science journal articles only to go and read the original article and find that it draws a different conclusion than what is implied by the journalist.
I like this example and it made me think, there is an analogy here to spec-driven development and vibe coding.
Rather than send 300+700 bits, like you said, send 300 (or less!) and let the human intelligence on the other side generate the result. Which supports the even older perspective: “If I had more time, I would have written a shorter letter.”
I’m not sure if this lands on anything very profound, but what about a pattern where, instead of codifying agent output at all, the only artifacts we share are the prompts. And the rewards (respect) accrue to those who generate the most generative among people and AI
What already happens:
- People give an LLM a bulleted list of points that they want expanded into a professional sounding document.
- The receiver doesn't wanna read all that. They put the full document into an LLM and ask it to summarize it into succinct bullet points.
The value of writing isn’t always to communicate new information. It’s often to align everyone’s assumptions. For instance, when I say casually to a colleague or an agent “this change will require a db migration” they understand it’s to my teams primary application database. If I submit a design doc to a company wide review which database is changing is critical information. If the 300 bits were truly enough your agent or junior engineer would implement the wrong thing correctly as they often do.
If you have 1000 bits of semantic information that you want to transfer but your default communication combines it with 10,000 bits of noise. Giving it all to an LLM and iterate on reducing that noise while making sure the 1000 bits is still present would enable you to communicate more effectively.
Overall, ideas are ideas. I'm not overly concerned with the fact that it was you who had the idea, as long as the idea is interesting. I don't know most of the people who write the things I read, so it seems to be of no consequence to me at all if they wrote it, as long as it is interesting. LLMs are notorious at creating things that are bland and vacuous, but they by no means have a monopoly on it.
Be the source human, machine, or dolphin, if they write a good article, I'm prepared to read it.
LLMs are wonderful at adding noise and okay at removing noise. My point is that they're not very useful at adding signal. If you wrote 10000 bits of noise and 1000 bits of signal, I would rather receive those 11000 bits from you, and if necessary ask an LLM to remove noise based on what I consider noise. If you can point out to the LLM what it should consider signal vs noise regardless of context, it should have been easy to not write that noise in the first place.
I don’t quite think this tracks. Perhaps you want to communicate 1000 bits that are well known and can be referenced with a 300 bit key. Then the LLM can easily retrieve the remaining information. It’s like sending someone a link to the Wikipedia page instead of explaining something yourself.
No, I don’t want to read LLM writing because it is BAD at it. It doesn’t really understand how humans think (because it thinks differently), and doesn’t seem to understand core principles very well (presumably due to the lack of world model), so it can’t write something humans enjoy yet.
That's kind of my point. If the 1000 bits are well known, then their inclusion isn't new semantic information. By pasting an LLM, you're deciding for your reader that they don't already know that information, and deciding that your LLM prompt is better than whatever they would to to obtain that information if they lack it. IMO, it's much better to give your readers the 300-bit key, and let them decide for themselves if they want/need to get more information, and if so, how.
I was going to write a post disagreeing with this on the basis of the fact that the reader lacks the background information the LLM has. For example, if I were to prompt "explain the proof of quadratic reciprocity using Gauss sums" most readers would need the entire LLM's answer (and much more, probably) and not just the prompt.
But then I realized that the reader can prompt the LLM with the same prompt for the same or equivalent expanded text. Most people don't do this as it's extra effort, but it's interesting to imagine a world where this is the default way of engagement with a text, assumed by both writers and readers alike.
> But then I realized that the reader can prompt the LLM with the same prompt for the same or equivalent expanded text.
That's basically what I've been asking my colleagues (so far a losing battle): Please don't send me AI-generated text. Send me your prompt instead. It is highly likely that I will understand it without needing an LLM, and if not, I can do it myself.
I love this example, partially because it jives with my conviction that LLMs are the ultimate translation machine. Ever since the embedding model days, it is clear that these models are amazing at representing meaning as math. The fact that LLM's most salient use is for coding somewhat agrees with that. After all, what is a programming language but another language? We instruct people with words and machines with code.
Depends. If the 700 bits were arrived at by the LLM while spending a lot of tokens, and the result is "good", I may want it through you as a middleman because it used up your tokens and won't eat my subscription usage limit to ask the AI to supply those 700. If you spend the tokens and put the result online, plenty of people can spare their tokens because they don't have to ask the AI to derive it. Bonus if that result was run through some kind of testing and verification.
Obviously this doesn't really apply to super simple questions that the LLM can just spit out the answer to right away.
True. There is at least one more case: when LLM the other person is using has access to their context and information repositories that they don't want to share directly, and so the LLM text gives a peek at a slice of that, and that's not something I can recreate from thin air. In that case the LLM text may have utility for me.
I will give you a use case where this is absolutely not the case.
I have a bunch of CLI utils I run for various clients and their peculiar setups. They now have man pages with descriptions and examples in them because the LLM went and read my code and did the needful.
I no longer have to re read my own code, rather I can just use the manual page.
Format and description came from semantics and context that (barely) existed elsewhere and I was not going to retain or transmit, but I have now.
But then the LLMs aren't adding new information, they're just reading your code and translating that information from e.g. python to english. Rather than sending someone an LLM-generated doc to someone, send the code and your own personal thoughts on the code. If your reader doesn't want to read and understand your code, they can ask an LLM to analyze it, within the context of their specific use case and your personal thoughts if any.
I never liked information theory because information theory as Shannon envisioned it fundamentally did not deal with semantics.
AIT tried solving it? But AFAIK it's a lot of pretty results with not much real application.
A better approximation is something of a "shared model"; then you can actually state things like, the transfer of information sometimes is "trivial" because, well, it's right there in your compressor/decompressor.
My understanding is that to unambiguously quantify information, you need to have a known model within which that information fits. In this context, talking about LLMs adding value (or not) via inserting new semantic information, we're assuming that public knowledge on the internet is not new semantic information, and the quantity of information of contained in text talking about public knowledge is equal to the ~32 bits needed to point to that knowledge.
isn't it more like: i have 1000 bits, i transfer 1000 bits but depending on the person, it might be lossy, so they only understand about 700. they then come up with the 300+- on their own, potentially putting them over 1000 or they come back and ask questions to fill in the blank. the bits don't ever have to be bit identical.
Sure, but that seems orthogonal to what I'm saying, no? We could continue clarifying the definitions of information in this context, but my fundamental argument is LLMs don't add information to my writing that couldn't be added by the reader themself.
This of course has the potential to change with personal LLMs that can have shared private context with me. However, that isn't a defense for sending people AI slop, it just turns it from "LLMs don't add value" to "LLMs may add value when used judiciously."
They're lossy. I might make an esoteric reference that no one gets, or an analogy that doesn't quite land, and then have the LLM help me come up with something a bit more understandable to a general audience. There's a difference between dumping the output that you spent 5 minutes with, and taking an hour to craft that perfect analogy.
It's funny, i push back on pull requests because there is too much description now - a 20 line change has pages and pages of generated description, rationalisation for why it is safe, defense of each design decision, analysis of risks and side effects. People are indignant, you're rejecting my change because there is too much documentation? And my response is, I don't have time to read it and you put me in the position where I can't afford not to - because approving the PR implies I did and accepted it. The investment to read all that for the value of a code change that I'm one prompt away from doing myself if I cared is just not high enough. So it's rejected.
I appreciate the direction of this article, and commend the author on publishing work, but the first sentence in the first paragraph is exactly what they lament.
> A pattern I see is that people use AI to build something new, then they use AI to retrospectively summarize what they have already built into a design document. Reading a document like this isn’t just difficult—it is punishing.
That's a classic calling card of LLM prose but that doesn't mean they actually used an AI. Though sure, it would be nice if we gave this style a name and then abolished it.
I call it the “it’s not delivery—it’s DiGiorno” construct. I generally tune out when I see this construct, which is so prevalent on LinkedIn these days, which is inundated with clichéd writing.
No, usually humans write more than just a short unsubstantiated punchline in their "it's not x, it's y's", unless they're trying to sell you something.
Well, not quite the classic calling card - LLMs generally use a contrast of opposites. (e.g. "it wasn't a bug -- it was a feature", although that predates LLMs :)
TFA's use is more common in "normal" language: "it's not just [minor], it's [major]". (But, as others have pointed out, it was probably deliberately parodic anyway.)
I don't disagree with the main points of the article. But I feel like soon with all the writing that's been hating on AI writing recently on HN, the LLMs are going to be really good at writing articles about how bad AI is for writing...
this is intended as a silly observation and not an actual accusation, but for an article about not writing with AI, the following is one AI-ass sentence:
“Reading a document like this isn’t just difficult—it is punishing.”
Nobody is going to read anything regardless of whether you used a LLM to write it or not (but if you use a LLM they'll use that as their excuse for the virtue signalling aura farm), so honestly why bother writing.
These last few days, there has been a small stream of blogposts on HN expressing similar things, and I have enjoyed them all.
My question recently has been how to broach this subject with colleagues who really enjoy producing prose with AI. There is not yet a better cultural shorthand for this sort of thing than "slop" which is a harsh-sounding word and itself sort of a thought-terminating cliché. "I don't want to read what you didn't write" is maybe closer — but it needs a pithier and somewhat more encouraging encapsulation, like "I want to hear it from you".
Has anyone had good experiences setting up professional boundaries or team norms around AI-written docs?
It's something that my company has been struggling with. Developers are generating mountains of code, documentation, and Jira ticket comments. It's incomprehensible and overwhelming.
For the team I lead, my guideline is AI generated is fine but it needs to be human-edited and/or summarized. You want me to read what you're offering? Put some effort into it and meet me halfway. I don't want AI generated gibberish with made-up terms. You'd better also understand what you are presenting as your work. It's been fairly well-received though we're still working on it.
I have some co-workers on other teams who use AI to generate responses to literally everything. Ask a simple question? Get pages of AI generated nonsense in response. They are proving a tougher nut to crack.
There is one person who is so good at writing with AI assistance.
Everyone else, including me, not so much. I envy his talent and ability to so successfully use the new tools. It's just something to think and talk about regularly. Maybe one day we will all be able to use the tools as well as that guy.
im sensing about a 20 month lag between hn sentiment and normy-coworker sentiment. managers and business-types pointing out the benefits and challenges of working with an llm i told them about early 2025. im just politely nodding along waiting for them to catch up. theres some emperors new clothes happening at my company so maybe im complicit
“Reading a document like this isn’t just difficult—it is punishing.”
I’d be curious whether the author composed this sentence himself or it was the output of AI. Personally I often find myself “it’s not X it’s Y” and then recoiling in disgust and rephrasing it simply because AI has made it so grating from overuse.
Same here. I also used to love writing with em dashes, but don't anymore out of fear people will call my writing AI and stop reading. It'd not hard to do in most word processors, typing two hyphens "--" does it in Word and Pages, so I don't get why people think it's such an obvious AI tell. It's not that exotic.
Maybe we will get to a place where we normalize AI authorship or co-authorship in our writing?
I appreciate what the Palm Springs Post does with its news articles. It has an "AI Assist" author and gives it credit in bylines as appropriate:
That's nice. So when I'm reading I know whether it's a human author or not. And if it's a dry municipal meeting, seems like a perfect place to send in the bot!
This sentiment gets echoed over and over. While I can certainly see the value in human-created creative works, if the subject is pure facts, there are only so many ways to state the facts without being caught in the weeds and missing your entire point.
Moreover, people are finding it hard to differentiate what is an is not AI-generated with newer models, often attributing original work with those of LLMs. It has just become an easy scapegoat for lazy comprehension and a desire to do less. You are jumping at AI boogeymen.
Just about the only thing here I can level with you on is, yes, AI is far from perfect and will continue to advance. Otherwise, so much of this reads as fruity prose to excuse apathy.
A lot of my problem with AI writing is that the facts aren’t checked. You get a big cheesy dramatic arc but the core story isn’t factually correct. There’s only so much time in a lifetime, too little to waste fact checking someone else’s work because they couldn’t be bothered to.
I have junior devs submitting huge PRs with multi paragraph descriptions about load baring root causes and "why this works". Then I ask them directly what are the steps to test this and they don't know, they have never run the code.
Rather than entertain the idea of reviewing the code I just send it back to them until they work out how to run it. And almost always they submit a new change because the last one didn't actually work, despite how confident and articulate claude was to them.
It's curious that you're pinning the blame of "excusing apathy" on the reader and not the writer. Why doesn't it go both ways? Given that producing LLM text is so cheap, your proposal amounts to an asymmetric war which is ultimately a DDOS on my finite time and attention. Or in other words: if it wasn't worth your time to write it, it's not worth my time to read it.
Let us say someone had few good ideas, and seeded them into a prompt, and after few back and forth, web searches via agent, feedback from the author, a piece of work was produced and the author decided to share it as a blog.
This is not much different than how people are producing original work LLM in areas such as math.
Would you object to reading their work because it was a byproduct of collaboration between AI/Humans? What about songs? movies? math proves? and software produces as such?
The lazy ones are the people spending 60s prompting then hitting copy/paste. They dump it on me as is and call it a job well done, which leads to me doing their job for them or calling them out to stop the behavior.
I don’t like doing either of these things. I don't want to do your job and I don't want to lecture about how it isn’t “your work” if you don’t touch the content after an LLM spits it out. It is rude and selfish to put me in that position.
> I don’t want to live in a world where you use AI to summarize something important into unreadable text, and then I use AI in an attempt to decipher it. I want to hear you, imperfections and all.
It doesn't fundamentally change the equation if I use AI to prepare and then write it myself. If I'm using AI effectively, it's likely that you won't be able to tell.
This sort of post is increasingly coming off as high and mighty, where the user thinks they are being exceptionally creative and other people who are using AI are using it mindlessly.
Also, writing, indeed any making activity, forces you to think clearly. Often that is the deepest purpose of prose, even code - forcing the author to confront the problem and sharpen their own understanding of the world we share. That it then communicates something to another person (or to a machine), is a happy bonus.
To me, it also seems like they AI digestion is getting actively worse? As best as I can tell, all the agentic nature and reasoning for code is now making writing actively worse, as the agent pulls across your whole knowledge base and will take that one thought and eagerly join and context it thinks is relevant, with the reasoning spread throughout the page.
Writing is fundamentally the transfer of information from your brain to my brain. If you have 1000 bits of semantic information you want to transfer, you can't give 300 bits of semantic information to an LLM and have it fill in the remaining 700, because it doesn't know what those 700 bits are. If it's able to guess those 700 bits correctly, then they aren't true semantic information, and you really only have 300 bits you want to transfer. You might as well transfer those bits to me directly, rather than having the LLM add on an extra superfluous 700 bits that I then have to filter out.
Giving someone the text output of a LLM is very similar to publishing a summary without links to the referenced material. When you were querying your LLM, you could have asked specific questions or asked for a custom focus or point of view. Your intended audience might have questions or different concerns, but they're unable to interact with your LLM. What you have delivered is static and unresponsive. It has all the disadvantages of being machine output without the advantage of being interactive, the way your LLM was for you.
It may have to wait until compute is cheap enough that tokens are essentially free, but we need a system to pass "hyperlinks" to LLM's primed with context, ready to be interactively queried on a chosen context. It's being overly generous to assume that people are putting even 300 bits into a LLM for every 1000 bits of regurgitated writing they try to pass off as their own. When people post LLM output as if it were their own, I have no choice but to assume they had zero knowledge of the subject, but this query taught them what they wanted to learn, and now they're sharing that. That's fine, but please pass an interactive LLM link rather than static text.
Once we have "hyperlinks" for LLM sessions, perhaps we can share LLM output a little more usefully and honestly.
I’m a big fan of this approach.
Rather than send 300+700 bits, like you said, send 300 (or less!) and let the human intelligence on the other side generate the result. Which supports the even older perspective: “If I had more time, I would have written a shorter letter.”
I’m not sure if this lands on anything very profound, but what about a pattern where, instead of codifying agent output at all, the only artifacts we share are the prompts. And the rewards (respect) accrue to those who generate the most generative among people and AI
Overall, ideas are ideas. I'm not overly concerned with the fact that it was you who had the idea, as long as the idea is interesting. I don't know most of the people who write the things I read, so it seems to be of no consequence to me at all if they wrote it, as long as it is interesting. LLMs are notorious at creating things that are bland and vacuous, but they by no means have a monopoly on it.
Be the source human, machine, or dolphin, if they write a good article, I'm prepared to read it.
No, I don’t want to read LLM writing because it is BAD at it. It doesn’t really understand how humans think (because it thinks differently), and doesn’t seem to understand core principles very well (presumably due to the lack of world model), so it can’t write something humans enjoy yet.
But then I realized that the reader can prompt the LLM with the same prompt for the same or equivalent expanded text. Most people don't do this as it's extra effort, but it's interesting to imagine a world where this is the default way of engagement with a text, assumed by both writers and readers alike.
That's basically what I've been asking my colleagues (so far a losing battle): Please don't send me AI-generated text. Send me your prompt instead. It is highly likely that I will understand it without needing an LLM, and if not, I can do it myself.
Obviously this doesn't really apply to super simple questions that the LLM can just spit out the answer to right away.
I have a bunch of CLI utils I run for various clients and their peculiar setups. They now have man pages with descriptions and examples in them because the LLM went and read my code and did the needful.
I no longer have to re read my own code, rather I can just use the manual page.
Format and description came from semantics and context that (barely) existed elsewhere and I was not going to retain or transmit, but I have now.
AIT tried solving it? But AFAIK it's a lot of pretty results with not much real application.
A better approximation is something of a "shared model"; then you can actually state things like, the transfer of information sometimes is "trivial" because, well, it's right there in your compressor/decompressor.
This of course has the potential to change with personal LLMs that can have shared private context with me. However, that isn't a defense for sending people AI slop, it just turns it from "LLMs don't add value" to "LLMs may add value when used judiciously."
In the meantime, for business communication, I use AI to shorten my text, to make it more concise.
https://github.com/josephmisiti/awesome-machine-learning
It's helped a lot. Agents haven't figured out how to do that yet, or sendgrid, sns, etc are doing the hard work for me.
> A pattern I see is that people use AI to build something new, then they use AI to retrospectively summarize what they have already built into a design document. Reading a document like this isn’t just difficult—it is punishing.
All of the posts in 2026 on that blog area about AI, by the way.
We've had AI output all these decades, but never recognized it. Is this evidence of time travel?
\s
TFA's use is more common in "normal" language: "it's not just [minor], it's [major]". (But, as others have pointed out, it was probably deliberately parodic anyway.)
I can't tell if this essay was written in earnest or as a subtle troll.
The juxta-positioning of the ambi-dextrous personification of the meta-sematicism is going to be both rich and soul transpiring....
It's gonna be basically a fingerprint in your soul, from my soul...
“Reading a document like this isn’t just difficult—it is punishing.”
My question recently has been how to broach this subject with colleagues who really enjoy producing prose with AI. There is not yet a better cultural shorthand for this sort of thing than "slop" which is a harsh-sounding word and itself sort of a thought-terminating cliché. "I don't want to read what you didn't write" is maybe closer — but it needs a pithier and somewhat more encouraging encapsulation, like "I want to hear it from you".
Has anyone had good experiences setting up professional boundaries or team norms around AI-written docs?
For the team I lead, my guideline is AI generated is fine but it needs to be human-edited and/or summarized. You want me to read what you're offering? Put some effort into it and meet me halfway. I don't want AI generated gibberish with made-up terms. You'd better also understand what you are presenting as your work. It's been fairly well-received though we're still working on it.
I have some co-workers on other teams who use AI to generate responses to literally everything. Ask a simple question? Get pages of AI generated nonsense in response. They are proving a tougher nut to crack.
As much as you want to say "actually do your job, jackass." Being goal oriented has had the best results for me.
Everyone else, including me, not so much. I envy his talent and ability to so successfully use the new tools. It's just something to think and talk about regularly. Maybe one day we will all be able to use the tools as well as that guy.
- https://news.ycombinator.com/item?id=49767937
- https://news.ycombinator.com/item?id=49784816
I’d be curious whether the author composed this sentence himself or it was the output of AI. Personally I often find myself “it’s not X it’s Y” and then recoiling in disgust and rephrasing it simply because AI has made it so grating from overuse.
Maybe we will get to a place where we normalize AI authorship or co-authorship in our writing?
I appreciate what the Palm Springs Post does with its news articles. It has an "AI Assist" author and gives it credit in bylines as appropriate:
https://thepalmspringspost.com/author/ai-assist/
That's nice. So when I'm reading I know whether it's a human author or not. And if it's a dry municipal meeting, seems like a perfect place to send in the bot!
Moreover, people are finding it hard to differentiate what is an is not AI-generated with newer models, often attributing original work with those of LLMs. It has just become an easy scapegoat for lazy comprehension and a desire to do less. You are jumping at AI boogeymen.
Just about the only thing here I can level with you on is, yes, AI is far from perfect and will continue to advance. Otherwise, so much of this reads as fruity prose to excuse apathy.
Rather than entertain the idea of reviewing the code I just send it back to them until they work out how to run it. And almost always they submit a new change because the last one didn't actually work, despite how confident and articulate claude was to them.
Let us say someone had few good ideas, and seeded them into a prompt, and after few back and forth, web searches via agent, feedback from the author, a piece of work was produced and the author decided to share it as a blog.
This is not much different than how people are producing original work LLM in areas such as math.
Would you object to reading their work because it was a byproduct of collaboration between AI/Humans? What about songs? movies? math proves? and software produces as such?
I don’t like doing either of these things. I don't want to do your job and I don't want to lecture about how it isn’t “your work” if you don’t touch the content after an LLM spits it out. It is rude and selfish to put me in that position.
It doesn't fundamentally change the equation if I use AI to prepare and then write it myself. If I'm using AI effectively, it's likely that you won't be able to tell.
This sort of post is increasingly coming off as high and mighty, where the user thinks they are being exceptionally creative and other people who are using AI are using it mindlessly.