jeremyjh
These are all important points and I love the analogy. But there is an even bigger issue with having LLMs write for you:
Writing is thinking. Thinking and deciding. There have been many times when I start out writing something substantial - could be an email, a blog post, a software design document, anything - when my own views substantially changed during the writing process. Writing forces you to serialize your thoughts - and you can't always trust the gestalt.
Reviewing gives you the chance to ensure the arguments connect solidly, that references are accurate (even informal references) and gives you the time to consider counter-arguments you aren't addressing.
None of this matters much on LinkedIn, but it matters a lot in our work. You cannot outsource your understanding to AI. They are powerful tools but they do not have any human understanding - that isn't their optimization target.
jampekkareply
> Writing is thinking. Thinking and deciding. There have been many times when I start out writing something substantial - could be an email, a blog post, a software design document, anything - when my own views substantially changed during the writing process. Writing forces you to serialize your thoughts - and you can't always trust the gestalt.
I don't disagree, but I think it's often not appreciated how much there's other work to writing too.
The biggest one is that you have to communicate non-interactively to an unknown audience. Having to (literally) put it in someone else's assumed terms does help giving different perspectives into the matter, but doesn't necessarily help one's own thinking that much. Instead you have to do some of the reader's thinking for them.
You also have to spend time on textual matters like grammar and style and a lot of "unspoken rules", which aren't really about linearizing your thinking about the contents.
Not all writing is thinking and not all thinking is writing.
iterateoftenreply
> You also have to spend time on textual matters like grammar and style and a lot of "unspoken rules", which aren't really about linearizing your thinking about the contents.
Hard disagree. Constraint is the driver of creativity. Also rewording sentences to sound better or make sense can make you reconceptualize the whole concept you are expressing
jampekkareply
Sure, but doing an interpretative dance or an abstract painting can make you reconceptualize the whole concept too. But we're not really pushing those tools as much as writing.
Take for example a non-native writer of the language. I'm sure having to check up words from a dictionary may help to reconceptualize things, but I'm sure also that it's not often very efficient. And I think similar is going on for natives too for many types of writing.
27183reply
The number of times the friction of writing has saved me from prematurely communicating a poorly understood idea must be in the hundreds or maybe thousands. For me when something is difficult to write about, it's a very good signal that I don't understand it well enough. So I think all the things you label as aspects of writing that aren't necessarily "thinking" are nonetheless good for thinking because they provide some necessary friction.
hi_im_greg_hreply
This to me is the key reason people are so obsessed with using an LLM for everything.
They’re completely opposed to experiencing any type of friction.
fwlrreply
Strange! I would quite specifically highlight “having to do the reader’s thinking for them”, as well as more generally developing the skill to “communicate non-interactively with an unknown audience”, as extremely valuable upgrades to my thinking.
card_zeroreply
It's a kind of shared fiction of an imaginary person's thinking, really. The author writes "now, I know what you must be thinking", and hopes the reader will agree "OK, close enough".
DenisMreply
Most valuable thinking is done at the margins, where you don’t have much capacity to emphasize with a diverse and unknown audience.
That said, abdicating to an LLM is the worst of all worlds - you’re not thinking and the product is not tailored.
The solution is obvious - write as much detail as you need and allow readers to interrogate the virtual you with an LLM, maybe not even reading what you write.
agile-gift0262reply
I completely agree. And I'd say that at least in my case, it applies to both writing prose and code
jeremyjhreply
Agreed, but a lot of code doesn't need to be understood. Knowing which is which is the art right now.
ripereply
> a lot of code doesn't need to be understood
Really? I am having difficulty thinking of any examples of code that doesn't need to be understood. If it isn't understood by someone, then how is it even working?
If you mean like a library you are using, where you aren't even reading the internals or might not even have access to it, OK, but that code is stull understood by its authors, surely?
27183reply
Uhh... what? You'll instantaneously feel very differently when the service you're responsible for is down at 3:15am and your logs are full of stack traces that end somewhere in that code that "doesn't need to be understood". At that point, you will need to understand it well enough to fix it stat.
btrettelreply
I've thought exactly that "writing is thinking" before as a reason to not let a LLM write for me.
Then again, I've seen a counterargument [1] by someone who clearly heavily uses LLMs for writing (going by both their LLMy writing style and their own admission). The person I'm citing describes a process where they get a LLM to write something, they check over it and provide feedback to the LLM, the LLM rewrites, and the process repeats iteratively. So clearly he is putting thought into the process.
I think there is something valuable missing, even if it's hard to clearly express. I'll try. The threshold for what I'm willing to accept if I'm simply approving something is likely different from what I'll get if I write something myself, for instance. Saying "LGTM" is too tempting. It seems to me like he's outsourcing his selection of topics to cover as well. If you're not thinking yourself about what to cover then it would be very easy to miss a critical subject. There also an asymmetry between checking and generating something with constraints placed on it. Checks can't catch everything, and a constrained generating process can reduce the amount that needs to be checked, avoid issues that can't be checked so easily, and focus your attention on areas that you know historically have had issues with this generating process. I've thought about this quite a bit in terms of whether to write new code or use an existing library. Sometimes "the devil you know" (my code) is better than an existing library simply because I understand its flaws better.
[1] https://www.nature.com/articles/d44148-026-00236-3
jeremyjhreply
I've tried doing it that way, and thought it was even acceptable for the reasons you said. I did learn a lot through that process and clarified my ideas. But later I rewrote the whole thing from scratch and then had the LLM review it. It made some good suggestions but no substantial changes. The difference was night and day. That final product had my voice, and I understood it better. LLMs are powerful tools and can improve quite a lot of the writing process, but using them to do all the writing leaves a lot on table along with your fly open.
dynm
I'm a bit skeptical of arguments of the form, "You should not use LLMs without disclosure because LLMs at bad at writing."
My reasoning is: If LLMs get better at writing—which I think is extremely likely—will you switch positions and say that now using LLMs without disclosure is A-OK?
Surely some people are willing to bite that bullet and say yes. But for most people, my guess is that the answer will remain no. Thus, I tend to think that the "real" reason most of us don't like it when people use LLMs to write without disclosure is that it's misleading: It's a sort of a claim that certain thoughts can be attributed to a human being when in fact they can't.
(The em dashes in this message were rendered using keyboard shortcuts.)
visargareply
Some people type 100x more words than the final article just digging into a topic and turning it on all sides. The final article being generated tells you nothing of the size of the effort going in.
masswerkreply
I think, there's a systematic flaw in RLHF: constructs reinforced as effective or sophisticated style will always suffer overuse and significant overexposure. Which will be also very obvious when used out of the intended context, it's copied from. There will be always a "smell", something that causes us to turn our backs to these texts, and possible the supposed authors, as well.
There's also the problem that certain types of phrasing are perceived as effective, because they mark a pivotal point in the progress of the text and are, as such, used sparingly, but are now becoming everyday templates that incorporate whatever is available in the context. There is no way this passes the Turing test of a competent reader.
And there's yet another issue: in social research, there has been the concept of semantic position, indicated by deviation from the mean (or median). If you don't deviate from the mean, your semantic position is zero. There's simply no expression. In this sense, next token prediction really amounts to a desemantificiation of the context. There's really no sense in uttering any of these productions, no plausible motivation, other than for the purpose of raising you hand to be seen.
zahlmanreply
> There is no way this passes the Turing test of a competent reader.
Every day I get more depressed thinking of what's happened to the supply of competent readers.
IshKebabreply
> If LLMs get better at writing—which I think is extremely likely—will you switch positions and say that now using LLMs without disclosure is A-OK?
I would say it is certainly significantly more ok. There are two reasons reading AI-generated text sucks now:
1. It's usually low value and not trustworth - I could have just asked the AI myself.
2. The prose style is horrible to read.
If we eliminate the second reason then it's definitely an improvement. (Although on the other hand the terrible prose can be quite a helpful indication that you're wasting your time reading slop, so maybe we shouldn't complain about it!)
CuriouslyCreply
> 1. It's usually low value and not trustworth - I could have just asked the AI myself.
You have no way of knowing what the input to the writing was. You're making the erroneous assumption that the person posting the article was a mouthbreathing spammer who used the prompt "write me an article on subject XYZ, make no mistakes!" and you could have supplied the same prompt.
fn-motereply
Alternative uses for reading:
1. Facts and processes. I just want to know something. I do not care if some Nerds for Nginx article is AI written.
2. Opinions and experiences. I want to know a human is writing because emotionally engaging with an AI isn’t building a stronger society - it’s increasing isolation.
3.
In another front, I feel like the value of an AI story cannot be greater than the value of the inputs. If you wrote a two paragraph prompt and an LLM produced a 10 page story, the truth is, it’s only worth two paragraphs. That’s just a feeling, but as of now I don’t think the LLMs have any additional life experiences to draw on to increase the value of their storytelling. (I understand this is debatable, but still- what they offer is available to everyone for now; it is a baseline.)
NathanielKreply
Yeah that's the point. Spammers have an infinite desire to make more spam, so statistically most LLM writing is garbage.
Readers are making this connection from their own experience. All these tells just associate it with garbage.
In the before times, sending a formal document full of typos and errors would show you didn't bother to proofread, now having a doc full of lazy "LLMisms" also looks like you were too lazy to proofread.
qltereply
> You're making the erroneous assumption that the person posting the article was a mouthbreathing spammer
Yes, that's the risk of using an LLM to "clean up" your writing. It's human nature to think you're unique and everyone else should somehow instinctively realize your output is unlike all the LLM spam garbage and worth their time but that doesn't work in practice.
It's unreasonable to expect other people to suppress their intuitive heuristics formed from the bait and switch of being subjected to endless LLM spam every day and blame them for not giving you a fair chance.
lsofzz
Please read past the LLM-ness of any given post.
The message of the post is what _should_ hopefully matter to a reader.
And, not the fact that it was written by an LLM.
I am a technical guy at heart; also an introverted extrovert. I hate writing docs that are to be written to satisfy someone else's metrics. For it to be a tickbox'ed item.
I delegate that to an LLM. I want to spend my time solving interesting challenges instead.
So, Mr. Cantrill, you got a problem with that? So be it.
Jtariiiiireply
Some people actually want to enjoy their moment to moment existence on this planet and not read garbage someone shat out.
It's just digital pollution.
lsofzzreply
I hear ya. However, that ship has sailed a long time ago and succumbing to this very idea is exactly why puritans in any form will not survive the next 50 years.
Jtariiiiireply
I don't think my survival hinges on my reluctance to read AI slop.
The ship has not sailed, I can just find a human who wrote words, and read them. And if they aren't human words, I will just, not read them.
bigstrat2003reply
> Please read past the LLM-ness of any given post. The message of the post is what _should_ hopefully matter to a reader. And, not the fact that it was written by an LLM.
No. I am interested in what a human has to say, not a clanker. If you don't wish to write that is fine, but don't hand it off to the slop machine and present it as though you did anything of value.
simonwreply
> Please read past the LLM-ness of any given post.
I just can't. LLM assisted posts are often so long, and the hints are usually obvious right at the start.
They're so unpleasant to read. "It's not X, it's Y" etc are bad because the comparisons rarely add anything at all to the message. It's filler that wastes my time. At that point I'd rather see your original prompt.
Why should I invest several minutes of my day reading something if I've already seen evidence that the author doesn't respect my time?
It's not that an LLM wrote it so much as the author couldn't be bothered to clean it up and remove the garbage cliches before publishing it.
CrimsonRain
LLMs write better than most journalists. It also writes better than majority people in the world when it comes to English as majority are really bad in writing.
And of course, LLMs write as you prompt.
Personally, I'm happy to read broken English from non English people or good English from English people. But I'd rather read LLM writing than read typical long form perfect English journalist crap.
wavemodereply
> LLMs write better than most journalists
In my personal experience I strongly disagree. Curious what publications you read wherein this is the case.
anon84873628reply
I suspect the commenter prefers the concise, factual summary of an LLM.
It is true that modern journalism on the web is trying to get ad impressions. After all, this is how we ended up with clickbait. Sure you still have the Economist and Atlantic (which aren't beyond criticism), but your local county paper is an absolute mess. The content is stretched, meandering, and designed to keep you scrolling through more impressions.
zahlmanreply
> concise, factual summary of an LLM.
Pardon; the what?
jampekkareply
> LLMs write better than most journalists.
I used to think this, but now that I've been trying to use them to help with writing, it turns out they are not necessarily that great on the deeper level. The grammar and the surface style are impeccable, but they often struggle with continuity and carrying on a point or an argument.
CrimsonRainreply
With current LLMs, the continuity/carrying/context management is _your_ task. When AGI is achieved etc...
riskablereply
I've come to trust their grammar corrections implicitly. So much that I've setup a "Fix Grammar" command with Obsidian Copilot that I'll just let it run on a whole chapter (YOLO mode):
Proofread the following text for grammar, punctuation, and typos while
strictly preserving the author's voice, pacing, and intentional stylistic
choices (such as sentence fragments or character dialogue quirks).
Rules:
- Correct objective spelling errors, misused words, and unintended punctuation mistakes.
- Smooth out unintentional syntactic snags without homogenizing unique phrasing.
- Retain all Markdown formatting.
- Return ONLY the corrected text with no intro, summary, or explanations.
Text to edit:
{selection}
About a year ago, all three big AI vendors could end up with strange "corrections" that would throw you off, but not anymore. As far as I'm concerned, they've solved "the grammar checking problem."
The complexity in this command is there for phi4:14b and qwen3:14b which I run locally via ollama. If the text doesn't have any Obsidian callouts or similar, fancy stuff I just use either of those and they do a fantastic job in seconds. For Big AI (e.g. Gemini, GPT-whatevs, Claude) you can literally just tell it, "fix the grammar." No need for the lengthy command.
NOTE: I am decent with English grammar so most of what needs fixing is typos I didn't spot or misplaced commas and periods inside/outside of quotes (I always screw that up without thinking—even though I know the rules! LOL). Occasionally, Big AI (Gemini, specifically) have disagreements about whether a comma is necessary in a particular spot but it's always of no real consequence.
zahlmanreply
> About a year ago, all three big AI vendors could end up with strange "corrections" that would throw you off, but not anymore.
Without a history of diffs created by this method, I don't believe you.
> I am decent with English grammar so most of what needs fixing is typos I didn't spot or misplaced commas and periods inside/outside of quotes (I always screw that up without thinking—even though I know the rules! LOL).
The rules there vary by style guide and are not objective.
LLMs seem like massive overkill for something the red and green squiggles can already accomplish.
krupanreply
Good writers break grammar rules when it helps convey a point, emphasize an idea, or just makes the writing more interesting. You are completely missing out on that.
anon84873628reply
I would almost never transmit LLM output without a decent amount of editing. But I also don't understand people who paint it as this horrible offensive unreadable slop.
Calling it better than most journalists is extreme (and probably the reason for down votes), but it must certainly be better than the average person.
We're hearing this criticism from the highly educated professional class people who bother to have their own blog or otherwise spend their time talking about technology online. I mean come on.
To most people, the LLM must feel incredibly empowering, like us wearing a mecha suit. Would we always show restraint and only apply our newfound enhanced physical strength in carefully considered situations?
antonvsreply
> But I also don't understand people who paint it as this horrible offensive unreadable slop.
When a piece of writing is full of LLM tells, I find it as offensive as any formulaic writing, except that the LLM style has quickly become pervasive, much more so than any other type of formulaic writing.
I’m not reacting out of some general dislike of LLMs - I use them daily. I’m reacting because I don’t like terrible writing.
qltereply
Well said. When I realize I've been reading LLM filler and leave the page it's the same reaction I would have pre-2021 thinking I found a good source for something and realizing I'm actually reading blogspam.
The difference being this variant of blogspam is produced with basically zero cost and thus the infection has spread far beyond SEO into every UI surface with textbox. Plus there are now passionate defenders who insist finding their blogspam unpleasant to read is disrespectful or anti-progress somehow.
