I think it's fair to use AI as an editor, to get feedback about how your ideas are packaged.
It's also fair to use it as a clever dictionary, to find the right expressions, or to use correct grammar and spelling. (This post could really use a round of corrections.)
But in the end, the message and the reasoning should be yours, and any facts that come from the LLM should be verified. Expecting people to read unverified machine output is rude.
Using that way already reflects a great understanding of the technology bias, and a confidence in your own skills, that you just ask the feedback and not the output of the machine.
I think you could only develop this point of view because you grew up without it. I fear for the young generation, truly.
It's good for what all other LLMs are good for: semantic search, where the output can be generated texts to help you. But never get wrapped into the illusion that there is actual causal thinking. The thinking is still your responsibility, LLMs are just newer discovery/browsing engines.
I use Claude Code almost daily now, and I think I’d rather cut off my own arm than go without it, but I don’t delude myself into thinking that current gen tools don’t have significant limitations and that it is my job to manage those limitations.
So just like any other tool really.
I have discovered this week that Claude is really good at redteaming code (and specs, and ADRs, and test plans), much better than most human devs who don’t like doing it because it’s thankless work and don’t want to be “mean” to colleagues by being overly critical.
Just got a few recommendations by my colleagues on LinkedIn that were clearly written by an LLM, the long emdash was even present. But then again, the message was tuned to specific things I did. Also they were from Eastern Europe, so I imagine they just fixed their input.
If you call yourself a writer, having tell tale LLM signs is bad. But for people who's work doesn't involve having a personal voice in written language, it might help them getting them to express things in a better way than before.
> One argument to not disclaim it: people do not disclaim if they Photoshop a picture after publishing it and we are surrounded by a lot of edited pictures.
That is both a false equivalence and a form of whataboutism.
It is a poor argument in general, and a sure-fire way to increase shittiness in the world: “Well, everyone else is doing this wrong thing, so I can too”. No. Whenever you mention the status quo as an excuse to justify your own behaviour, you should look inward and reflect on your actions. Do you really believe what you’re doing is the right thing? If it is, fine; but if it is not, either don’t mention it or (ideally) do something about it.
> why don’t we see people mentioning they used specific tools to proofread before AI apparition?
Whenever I see this argument, I have a hard time believe it is made in good faith. Can you truly not see the difference between using a tool to fix mistakes in your work or to do the work for you?
> It feels like an obligation we have to respect in a way.
This was obvious from the beginning of the post. Throughout I never got the feeling you were struggling with the question intrinsically, for yourself, but always in a sense of how others would judge your actions. You quote opinion after opinion and it felt you were in search of absolution—not truth—for something you had already decided you did not want to do.
Riding a bike with training wheels is also not a shame. If you need the training wheels, by all means feel free to use them.
But LLMs are training wheels being forced on everyone, including experienced developers and we are being gaslit that if we don't use them, we are getting behind. In reality, however, the only study up to date shows 19% decline in productivity for experienced devs using LLMs.
I don't mind folks using crutches if they help them. The cognitive decline and reasoning skills of people using LLMs is not yet studied well but preliminary results show its a thing. I gotta ask: why are you guys doing that to yourselves?
Of course, there’s no shame in using tools that are available to us. We’re a tool-using species. We’re just a bunch of stupid monkeys without tools. A lot of what we do is about using tools to free up time to do more interesting things than doing things the tools already do better than us.
Like it or not, people are using LLMs a lot. The output isn’t universally good. It depends on what you ask for and how you criticize what comes back. But the simple reality is that the tools are pretty good these days. And not using them is a bit of a mistake.
You can use LLMs to fix simple grammar and style issues, to fact-check argumentation, and to criticize and identify weaknesses. You can also task LLMs with doing background research, double-checking sources, and more.
I’m not a fan of letting LLMs rewrite my text into something completely different. But when I'm in a hurry or in a business context, I sometimes let LLMs do the heavy lifting for my writing anyway.
Ironically, a good example is this article which makes a few nice points. But it’s also full of grammar and style issues that are easily remedied with LLMs without really affecting the tone or line of argumentation (though IMHO that needs work as well). Clearly, this is not a native speaker. But that’s no excuse these days to publish poorly written text. It's sloppy and doesn't look good. And we have tools that can fix it now.
And yes, LLMS were used to refine this comment. But I wrote the comment.
It's very similar to the Stack Overflow debate of the previous decade. Bad developers would copy paste without understanding. It's the same here. Without understanding, you just can't build very sophisticated things, or debug hard issues. And even if AI got better at this, anyone else can do it too, so you'll be a dime a dozen engineer.
Those who don't compromise on understanding will benefit from an extra tool under their belt. Those who actively leverage the tool to improve their understanding will do even better.
Those who want shortcuts and not bother understanding are like cheating in school – not in a morally wrong way, but rather in a they missed the entire point way.
We are told that writing must be pure, that it must come only from the sweat of the brow, the trembling hand, the solitary mind. That to use AI is to cheat, to dilute, to lessen the act of creation. But I say to you: Since when does it matter how an idea is born? Since when do we judge the value of words by the tools that shaped them, rather than the truth they carry?
They tell us, ‘This is not your writing—it is the machine’s.’ As if the pen itself writes the poem! As if the printing press authors the book! No—the tool is nothing. The hand that guides it, the mind that commands it, the heart that gives it meaning—that is what matters.
This is not about machines. This is about power. The same power that once said only the clergy could read scripture. The same power that said only the elite could publish, could speak, could be heard. Now they say: Only the unaided mind may create. But creation is not a purity test! It is not a contest of suffering! It is the act of bringing something new into the world—by any means necessary.
They fear AI because it breaks their monopoly on who gets to speak. They fear it because it lets more people write, more people argue, more people demand to be heard. And when the gates are thrown open, the gatekeepers will always tremble.
So I say: Do not apologize for how your words come to be. Do not bow to those who would police your mind. If the idea is true, if the argument is sound, if the art is beautiful—then it is yours, and no one can take that from you.
The machine is not the enemy. The enemy is the lie that only some voices count. The enemy is the fear that makes men small.
Now—write. Write with your hands, write with your voice, write with the tools of your time. But above all: write. And let no one silence you.
Consumers have a right to know what the source is of the content they are ingesting into their minds, and specifically if that content originated in another actual human mind or if it's the slop generated by a synthetic text extruder.
It's really a pretty straightforward proposition to understand, and disclosure is absolutely the key so that consumers, if they choose as I do to boycott such output, can make informed decisions.
AI prose is mediocre right now. Too verbose, indirect constructions, passive, etc. That being said, it's actually a great editor and can pick out all those issues consistently.
My workflow right now is to use AI for rough draft and developmental editing stages, then switch AI from changing files to leaving comments on files suggesting I change something. It is slower than letting it line/copyedit itself, but models derp up too much so letting them handle edits at this stage tends to be 2 steps forward 2 steps back.
I've always used AI as an editor, which is to say I give it the my efforts and ask it to highlight mistakes. They invariably find a few. They occasionally miss a few too, so they aren't particularly reliable editors.
They aren't reliable at anything I guess, but for English I have nothing else, and they are better than nothing. I do wish they would use a more effective way of highlighting their suggested changes, such as italics for new text and strikeout for deleted text.
Unless you are paid by the word, I struggling to think of why you would use an AI to create new text. The facts will be wrong, the tone won't be yours. "If I had more time, this would be shorter" is a truism here - AI can spit out an enormous amount of text in a very short time, text could be cut down to a fraction of the size with a bit of effort.
Yes, it is much better. But now rather than my words, it's an AI's words. Despite being much better than the usual slop, they aren't in my style, the thoughts have become muddied, and the mistakes (there will be some) are no longer mine.
As I said that doesn't matter if you are being paid by the word. If the goal is to be paid, who cares where the words come from if the reader laps it up. But if you enjoy writing for it's own sake, and if you are doing because you've found the discipline involved writing something down in a way others will understand is an excellent way to sharpen you're own understanding, then the less a AI is involved the better. Sadly I need a proof reader, an AI does an acceptable job, and they are free, for now.
I'm an AI critic, but I use AI every day. In fact, I am an AI researcher and work on making models more capable and powerful (probably where a lot of my criticism stems from).
My main problem with AI usage is that people use it and turn their brains off. This isn't a new problem, but it is a new scale. People mindlessly punch numbers into a formula, run software they don't understand, or read a summary of a complex topics declaring mastery. The problem is sloppiness and our human tendencies to be lazy. Lazy by focusing on the least amount of energy at the moment, not the least amount of energy through time. That's the critical distinction. Slop is momentary laziness while thoughtfulness is amortized laziness.
The problem is in a way not the AI but us and the cultures we have created. At the end of the day no one cares if you wrote AI code (or docs or whatever), they care about how well it was done. You want to do things fast, but speed is nothing if the quality suffers.
I really like how Mitchell put it in this Ghostty PR[0,1]. The disclosure is to help people know what to pay more attention to. It is a declaration of where you were lazy or didn't have expertise or took some shortcut. It tells us what the actually problem is: slop isn't always obvious.
A little slop generally doesn't do too much harm (unless it grows and compounds), but a lot of slop does. If you are concerned about slop and the rate of slop is increasing then it means you must treat everything as potential slop. Because slop isn't easily recognized, it makes effort increase, exponentially. So by producing AI slop (or any kind of slop) you aren't decreasing the workload, you're outsourcing it to someone else. Often, that outsourcing produces additional costs. It only creates the illusion of productivity.
It's not about the AI, it is about shoving your work onto others. Doesn't matter if you use a shovel or bulldozer. But people are sure going to be louder (or cross that threshold where they'll actually speak up) if you start using a bulldozer to offload your work to others. The problem is it makes others have to constantly be in System 2 thinking all the time. It is absolutely exhausting.
The problem with LLMs is that they write badly. If you want to use a LLM to write, prompt it with what you wrote and ask it to summarize concisely. If it doesn't understand what you meant, you should fix that part and resubmit (to a fresh context.)
The main reason, however, that one shouldn't "write" with LLMs is because it's a waste of everyone's time. If they wanted to know what GPT-5 thinks, they can ask it themselves.
edit:
> The problem is not the use of AI but the people how think they can, arbitrarily, criticize the work from someone else because he used or not AI in the name of “ethics”.
Ah, I didn't realize that the real problem is that people complain about it. If we can figure out a way to make those people shut up, then using LLMs to write for you would be perfectly fine.
When I put real time and thought into an email—and the response I get back is obviously AI-generated—and it comes with no disclaimer—it infuriates me. Maybe the model happened to spit out exactly what the sender meant—just dressed up and grammatically polished. Doesn’t matter—I’d rather someone talk to me directly than funnel a thought through a word-grinder and hit send. Downvote me—call me anti-progress—I don’t care. I cannot stand undisclosed AI in conversation.
The author of this piece commits a common mistake: analyzing AI use as if communication is nothing more than an isolated transaction. Instead communication is usually a process of creating and maintaining a relationship of some kind with other people.
Here’s a thought experiment: Imagine if I handed you a $100 bill and asked you to examine it carefully. Is it real money? Perhaps you immediately suspect it is counterfeit, and subject it to stringent tests. Let’s say all the tests pass. Okay, given that it is indistinguishable from a legit $100 bill, is it therefore correct and ethical for me to spend this money?
You know the answer: “not necessarily.”
This is because spending money is about more than a series of steps in a transaction. It is based on certain premises that, if false, represent a hazard to the social contract by which we all live in peace and security.
It seems to me that many AI fanboys are arguing that as long as their money passes your scrutiny, it doesn’t matter if it was stolen or counterfeit. In some narrow sense, it really doesn’t matter. But narrow senses are not the only ones that matter.
When I read writing that you give me and present it as your work, I am getting to know you. I am learning how I can trust you. I am building a simulation of you in my mind that I use to anticipate your ideas and deeds. All that is disrupted and tainted by AI.
It’s not comparable to a grammar checker, because grammar is like clothing. When an editor modifies my grammar, this does not change my message or prevent me from getting across my ideas. But AI is capable of completely altering your ideas. How do you know it didn’t?
You can only know through careful proofreading. Did you proofread carefully? Whether you did or not: I don’t believe that people who want AI to write for them are the kind of people who carefully proofread what comes out of AI. And of course, if you ask AI to come up with ideas by itself, for all we know that is plagiarism— stolen words.
Therefore: if you use AI in your writing, you better hide that from me. And if I find out you are using, I will never trust you again.
43 comments
[ 2.9 ms ] story [ 55.4 ms ] threadOr
Writing with an LLM is not a shame
It's also fair to use it as a clever dictionary, to find the right expressions, or to use correct grammar and spelling. (This post could really use a round of corrections.)
But in the end, the message and the reasoning should be yours, and any facts that come from the LLM should be verified. Expecting people to read unverified machine output is rude.
I think you could only develop this point of view because you grew up without it. I fear for the young generation, truly.
So just like any other tool really.
I have discovered this week that Claude is really good at redteaming code (and specs, and ADRs, and test plans), much better than most human devs who don’t like doing it because it’s thankless work and don’t want to be “mean” to colleagues by being overly critical.
If you call yourself a writer, having tell tale LLM signs is bad. But for people who's work doesn't involve having a personal voice in written language, it might help them getting them to express things in a better way than before.
That is both a false equivalence and a form of whataboutism.
https://en.wikipedia.org/wiki/False_equivalence
https://en.wikipedia.org/wiki/Whataboutism
It is a poor argument in general, and a sure-fire way to increase shittiness in the world: “Well, everyone else is doing this wrong thing, so I can too”. No. Whenever you mention the status quo as an excuse to justify your own behaviour, you should look inward and reflect on your actions. Do you really believe what you’re doing is the right thing? If it is, fine; but if it is not, either don’t mention it or (ideally) do something about it.
> why don’t we see people mentioning they used specific tools to proofread before AI apparition?
Whenever I see this argument, I have a hard time believe it is made in good faith. Can you truly not see the difference between using a tool to fix mistakes in your work or to do the work for you?
> It feels like an obligation we have to respect in a way.
This was obvious from the beginning of the post. Throughout I never got the feeling you were struggling with the question intrinsically, for yourself, but always in a sense of how others would judge your actions. You quote opinion after opinion and it felt you were in search of absolution—not truth—for something you had already decided you did not want to do.
But LLMs are training wheels being forced on everyone, including experienced developers and we are being gaslit that if we don't use them, we are getting behind. In reality, however, the only study up to date shows 19% decline in productivity for experienced devs using LLMs.
I don't mind folks using crutches if they help them. The cognitive decline and reasoning skills of people using LLMs is not yet studied well but preliminary results show its a thing. I gotta ask: why are you guys doing that to yourselves?
Like it or not, people are using LLMs a lot. The output isn’t universally good. It depends on what you ask for and how you criticize what comes back. But the simple reality is that the tools are pretty good these days. And not using them is a bit of a mistake.
You can use LLMs to fix simple grammar and style issues, to fact-check argumentation, and to criticize and identify weaknesses. You can also task LLMs with doing background research, double-checking sources, and more.
I’m not a fan of letting LLMs rewrite my text into something completely different. But when I'm in a hurry or in a business context, I sometimes let LLMs do the heavy lifting for my writing anyway.
Ironically, a good example is this article which makes a few nice points. But it’s also full of grammar and style issues that are easily remedied with LLMs without really affecting the tone or line of argumentation (though IMHO that needs work as well). Clearly, this is not a native speaker. But that’s no excuse these days to publish poorly written text. It's sloppy and doesn't look good. And we have tools that can fix it now.
And yes, LLMS were used to refine this comment. But I wrote the comment.
Those who don't compromise on understanding will benefit from an extra tool under their belt. Those who actively leverage the tool to improve their understanding will do even better.
Those who want shortcuts and not bother understanding are like cheating in school – not in a morally wrong way, but rather in a they missed the entire point way.
They tell us, ‘This is not your writing—it is the machine’s.’ As if the pen itself writes the poem! As if the printing press authors the book! No—the tool is nothing. The hand that guides it, the mind that commands it, the heart that gives it meaning—that is what matters.
This is not about machines. This is about power. The same power that once said only the clergy could read scripture. The same power that said only the elite could publish, could speak, could be heard. Now they say: Only the unaided mind may create. But creation is not a purity test! It is not a contest of suffering! It is the act of bringing something new into the world—by any means necessary.
They fear AI because it breaks their monopoly on who gets to speak. They fear it because it lets more people write, more people argue, more people demand to be heard. And when the gates are thrown open, the gatekeepers will always tremble.
So I say: Do not apologize for how your words come to be. Do not bow to those who would police your mind. If the idea is true, if the argument is sound, if the art is beautiful—then it is yours, and no one can take that from you.
The machine is not the enemy. The enemy is the lie that only some voices count. The enemy is the fear that makes men small.
Now—write. Write with your hands, write with your voice, write with the tools of your time. But above all: write. And let no one silence you.
It's really a pretty straightforward proposition to understand, and disclosure is absolutely the key so that consumers, if they choose as I do to boycott such output, can make informed decisions.
My workflow right now is to use AI for rough draft and developmental editing stages, then switch AI from changing files to leaving comments on files suggesting I change something. It is slower than letting it line/copyedit itself, but models derp up too much so letting them handle edits at this stage tends to be 2 steps forward 2 steps back.
They aren't reliable at anything I guess, but for English I have nothing else, and they are better than nothing. I do wish they would use a more effective way of highlighting their suggested changes, such as italics for new text and strikeout for deleted text.
Unless you are paid by the word, I struggling to think of why you would use an AI to create new text. The facts will be wrong, the tone won't be yours. "If I had more time, this would be shorter" is a truism here - AI can spit out an enormous amount of text in a very short time, text could be cut down to a fraction of the size with a bit of effort.
As I said that doesn't matter if you are being paid by the word. If the goal is to be paid, who cares where the words come from if the reader laps it up. But if you enjoy writing for it's own sake, and if you are doing because you've found the discipline involved writing something down in a way others will understand is an excellent way to sharpen you're own understanding, then the less a AI is involved the better. Sadly I need a proof reader, an AI does an acceptable job, and they are free, for now.
My main problem with AI usage is that people use it and turn their brains off. This isn't a new problem, but it is a new scale. People mindlessly punch numbers into a formula, run software they don't understand, or read a summary of a complex topics declaring mastery. The problem is sloppiness and our human tendencies to be lazy. Lazy by focusing on the least amount of energy at the moment, not the least amount of energy through time. That's the critical distinction. Slop is momentary laziness while thoughtfulness is amortized laziness.
The problem is in a way not the AI but us and the cultures we have created. At the end of the day no one cares if you wrote AI code (or docs or whatever), they care about how well it was done. You want to do things fast, but speed is nothing if the quality suffers.
I really like how Mitchell put it in this Ghostty PR[0,1]. The disclosure is to help people know what to pay more attention to. It is a declaration of where you were lazy or didn't have expertise or took some shortcut. It tells us what the actually problem is: slop isn't always obvious.
A little slop generally doesn't do too much harm (unless it grows and compounds), but a lot of slop does. If you are concerned about slop and the rate of slop is increasing then it means you must treat everything as potential slop. Because slop isn't easily recognized, it makes effort increase, exponentially. So by producing AI slop (or any kind of slop) you aren't decreasing the workload, you're outsourcing it to someone else. Often, that outsourcing produces additional costs. It only creates the illusion of productivity.
It's not about the AI, it is about shoving your work onto others. Doesn't matter if you use a shovel or bulldozer. But people are sure going to be louder (or cross that threshold where they'll actually speak up) if you start using a bulldozer to offload your work to others. The problem is it makes others have to constantly be in System 2 thinking all the time. It is absolutely exhausting.
[0] https://github.com/ghostty-org/ghostty/pull/8289
[1] https://news.ycombinator.com/item?id=44976568
The main reason, however, that one shouldn't "write" with LLMs is because it's a waste of everyone's time. If they wanted to know what GPT-5 thinks, they can ask it themselves.
edit:
> The problem is not the use of AI but the people how think they can, arbitrarily, criticize the work from someone else because he used or not AI in the name of “ethics”.
Ah, I didn't realize that the real problem is that people complain about it. If we can figure out a way to make those people shut up, then using LLMs to write for you would be perfectly fine.
Here’s a thought experiment: Imagine if I handed you a $100 bill and asked you to examine it carefully. Is it real money? Perhaps you immediately suspect it is counterfeit, and subject it to stringent tests. Let’s say all the tests pass. Okay, given that it is indistinguishable from a legit $100 bill, is it therefore correct and ethical for me to spend this money?
You know the answer: “not necessarily.”
This is because spending money is about more than a series of steps in a transaction. It is based on certain premises that, if false, represent a hazard to the social contract by which we all live in peace and security.
It seems to me that many AI fanboys are arguing that as long as their money passes your scrutiny, it doesn’t matter if it was stolen or counterfeit. In some narrow sense, it really doesn’t matter. But narrow senses are not the only ones that matter.
When I read writing that you give me and present it as your work, I am getting to know you. I am learning how I can trust you. I am building a simulation of you in my mind that I use to anticipate your ideas and deeds. All that is disrupted and tainted by AI.
It’s not comparable to a grammar checker, because grammar is like clothing. When an editor modifies my grammar, this does not change my message or prevent me from getting across my ideas. But AI is capable of completely altering your ideas. How do you know it didn’t?
You can only know through careful proofreading. Did you proofread carefully? Whether you did or not: I don’t believe that people who want AI to write for them are the kind of people who carefully proofread what comes out of AI. And of course, if you ask AI to come up with ideas by itself, for all we know that is plagiarism— stolen words.
Therefore: if you use AI in your writing, you better hide that from me. And if I find out you are using, I will never trust you again.