"Those rethinks are load-bearing parts of your voice."

Loved the (possibly not deliberately ironic) use of the LLM'S favourite analogy, load-bearing.

Nice wee article though. I particularly agree with "Don’t take all of the model’s copyediting advice." I've been using LLMs to review teaching material and often it just doesn't get the point of teaching. It's good for prompting reflection on your writing but it's still an idiot savant.

This article was so much better than what I though it would be

Does anyone know about "Style: Lessons in Clarity and Grace" ? Seems like an interesting book

I fear in the future people will read even less because of how often people cheap out and use AI to write. I know the author of this article tells you never to use a single word it suggests, however I still, while reading, am checking for any amount of AI used. This makes reading less enjoyable, and even a bit stressful.

If you can't spend the time to write it, why should anyone read it?

Everything I post is from my own finger tips, only to get a reply from someone using a bot. I'm tired.

If you can't spend the time to write it, why should anyone read it?

That's actually a legitimate question, and one I've been thinking about, as I've been using LLM tools to compile factual information based on various search tools it uses. The workflow gets sources, fact checks them, filters them according to my needs, and gives me clean urls for source information to hand off to an editor for writing a short listing. It's very hard for the editor to do a much better job at this type of writing; the real benefit of the work was in the research, which was handled by the LLM. The rest is basically just putting the facts in the right order, with some guidance on style, preferred word choices, etc. I'm not comfortable with this at all, as giving work to an LLM that a human could do strikes me as wrong. But using LLM tools gives me a big competitive advantage (for now), in the space I'm doing this in.

I'm not comfortable with this at all, as giving work to an LLM that a human could do strikes me as wrong

Where does this feeling come from? Is it just because the work is complex? We’re all (excepting maybe the amish) comfortable with using at least some automation, be it mechanical or digital, elsewhere in our lives

compile factual information based on various search tools it uses. The workflow gets sources, fact checks them, filters them according to my needs, and gives me clean urls for source information to hand off to an editor for writing a short listing.

What’s preventing the editor from using one of your prompts/tooling and replacing you entirely?

There’s zero value in sharing ai-generated output unless your ai has access to something nobody else’s does on which to base things. If all it’s doing is web research you should share your prompt instead. That way people don’t have to deal with ai-written walls of text.

Excessive generative AI output devalues all communication media unfortunately. As a greater and greater portion of text, imagery, audio, and video come to be synthetic, I don't see why anybody would be motivated to do deep reading, or listening, or analysis of said media. It's frightening to think what that might do to our civilization, almost like putting us back into a weird version of the ancient past when nobody was literate, save the priest.

Maybe humans can adapt to care about generated content and treat it with the same emotional weight that human communication has. I don't know if that's good or not.

There is a recent PISA research that students slowly decline in understanding the long reads.

E.g. in one country, 21% of high school students do not even achieve “sufficient reading literacy”…

So the question should be - what are we going to do with this? LLM can produce texts but we as people will not be able to understand anyway…

About PISA research for example here: https://www.edweek.org/leadership/reading-ability-craters-on...

You can't even watch funny animal compilations anymore!

I don't ever watch TV or streaming shows anymore. It's been a huge improvement in my life. It's the same with reading. Reading is mostly a waste of my time, with some notable exceptions. I'm glad I spend way less time doing each of these!

It's sad that reading has become that way for you.

I love reading, but I mostly read non-fiction and I'm a bit of an amateur history buff. Last time I read a fiction book I was left feeling like I was preached to by an alien about how life doesn't really matter (sci-fi authors will often do that). The state of literature is pretty dire.

Just read people you can trust. Why are you giving yourself all this busy work?

Just make your own tree of material to read. You should have a couple dozen writers/journalists you admire, look at who these people read then start reading those people, then you repeat the cycle.

Don't waste your time reading anything new, most new things are bad. Most ideas we struggle with, are ideas we have been discussing since the written record was invented out of necessity.

Anything new worth reading will assuredly make your way to you if you are at all connected.

If everyone followed this advice it would turn influencer culture to 11, rich get richer, with no "social mobility"

No it really wouldn't. I like the authors Jeff VanderMeer and Peter Watts. They have had multiple interviews where they talk about their peers, who they respect from the past, and who they like in our current age. I have found maybe subsequent authors + journalists to read from these two.

Try to think a little more critically here. If you can't trust the opinion of people you supposedly like, already established in their craft, who can you trust?

Have no idea why you think this would increase influencer culture, but thanks for telling us all publicly how you interact with social media: in an sloppy uncritical way.

My advice for you is to pull the plug and read books that are older than 125 years.

If you can't spend the time to write it, why should anyone read it?

100%. We're producing more content, faster, than ever before but ... most of it is just click-producing garbage that won't stand the test of time.

If you can't spend the time to write it, why should anyone read it?

Because the information in the post could be valuable regardless of human or LLM involvement.

But how much slop are you willing to wade through, hoping to find a nugget of insight?

People on my FOSS project couldn't see the issue with their AI slopped documentation contributions.

On one hand I'm like, yeah fair enough it needed writing and something is better than nothing.

On the other I know nobody wants to read it now because it's just word vomit rather than how a human would write something they'd like another human to read.

I hate it so much.

On Debian we recently voted whether to permit or forbid llm usage (spoiler, AI usage at will won).

One person was arguing that they need llm because english isn't their first language and then it helps them write better and be understood. I was incapable of reading their (extremely long) email of arguments; I tried but my mind just went elsewhere as my eyes were moving through the lines.

At the very end there was a disclaimer that llm had been used to "aid" in writing that very email.

Poor guy thinks he's communicating better but all he's doing is making his emails completely unreadable. It's a sad state of affairs.

If you can't spend the time to write it, why should anyone read it?

Because the value of writing are not the words on a piece of paper but the idea they convey.

Politicians speech are not worth listening too because they have a copywriter polish them?

Teacher assistant homeworks are not worth to do, because they are made by a TA and not by the course professor?

Of course there are writing as: "Hey ChatGPT write me a piece around coding with LLMs" and yeah, those are not worth reading.

But most of the writing is usually, these are the ideas, this is how the idea are structured together, now let's review them and then let's get a nice prose out of it.

But most of the writing is usually, these are the ideas, this is how the idea are structured together, now let's review them and then let's get a nice prose out of it.

I know lots of AI-enabled writers believe this, but I've never actually seen it be true.

If you can't spend the time to write it, why should anyone read it?

This isn’t your thought; you are merely regurgitating a sentence we have all heard many times before, practically word for word.

Presumably you think there is value in posting yet another copy of that other person’s thoughts in this discussion, despite having spent no time or effort in thinking it up yourself.

Why is that? Is it because you think there is inherent value in the sentence you transcribed, in spite of the near zero effort you spent creating it? If you think anybody here should read it, then you have undermined your own point. If you don’t think anybody should read it, then you shouldn’t post it.

How do you know it's not his/her thought? Just because others have had it does not mean he/she did not put effort into considering it.

despite having spent no time or effort in thinking it up yourself.

I mean, how can you say that he didn't think it himself? I thought this same thing and realised apparently I'm not the first. You have no idea how many of my blog posts ideas have come in HN first before I could publish. People can have the same thoughts. And have em all the time as well.

Psss.. See how @sm-silversight who commented on your post said the same thing I did? That's realtime example right there.

Coming back to the topic at hand. I agree with the OP. If you don't care to write it, why should anyone read it? AI has good usecases. But am not sure writing is one.

Writing is not just about the end result. It's also about the thought process. I learn about the topic more as I write. I correct my incorrect understandings as I write. When am pissed, my article has a different tone. Am rude, too direct. When am content/happy, different one. All of this is sterilised into something robotic where the author won't even remember what he/she wrote? What's the to like? For brownie points?

Selecting between alternatives is also a very valuable skill. Also known as "taste".

You are a senior software architect, you "prompt" 10 junior architects to write a software architecture proposal and then pick one which you send the client.

Would you say it's worthless, because the senior architect didn't care to write it?

There are plenty of industries where "taste" is a real job.

Or you think that for some reason prompting a LLM and forwarding it's output after vetting is different than prompting a human and forwarding the output after vetting.

Would you say it's worthless, because the senior architect didn't care to write it?

Lemme ask you a counter question, the senior architect writes the proposal and the junior writes it. Which one do you think client likes better?

Moreover, Your focus is on the end-result. Sure, in corporate environment, most people don't care about what they do as long as the end results checks some boxes. That wasn't the discussion here. The discussion is about writing and reading in general. Whether it's worth my time or not if you don't mean (forget mean, don't even remember) what you wrote.

Also, in your own example, the client don't care as well. As long the client's checks all the boxes his/her/their manager said should satisfy. Nobody really gives a damn about the quality.

Also, are you saying you'll be fine paying for Financial Times, WSJ and NYTimes KNOWING all they do is shove the story through an LLM? Neither will there be a soul, nor will there be any material effect on people who reads it.

Or you think that for some reason prompting a LLM and forwarding it's output after vetting is different than prompting a human and forwarding the output after vetting.

There is difference no? I forward the output to get good feedback and correction from someone BETTER. And there is a process involved in it which showcases the "TASTE" you mentioned from the person whom you are forwarding your output to be vetted and that person comes up to you with feedback. And guess what? You learn from the person's feedback. You don't go back again and again forever to learn the same thing. That's exactly what am trying to point out.

how can you say that he didn't think it himself?

Because the same sentence has been repeated practically word-for-word ad nauseam all over social media for years. What do you think is plausible - that they repeated something that is damn near omnipresent in the discourse, or that they never saw any of the many, many times it has come up and independently came up with the exact same formulation? They are perfectly able to step in and claim that it was their original thought if they want to.

What do you think is plausible - that they repeated something that is damn near omnipresent in the discourse, or that they never saw any of the many, many times it has come up and independently came up with the exact same formulation?

Both are fully possible.

We're talking about a simple sentence that sounds a lot like something a primary school teacher or parent of young children would say. "If you won't do/don't/haven't done X, why should I Y?" is very normal moral/ethical phrasing.

I suppose I heard someone say it before I spelled it out in that way, but I am pretty confident I would ultimately have used a sufficiently similar expression, because I have been on the receiving end of the problem.

And that is putting aside the communicative compression of aphorisms. People don't use shared aphorisms to avoid ever doing the thinking themselves; they use them to shorthand more complex arguments for a listener who has already heard them. They are shared macros for agreed concepts. (Using them correctly requires having done the thinking they package up at least once).

I don’t care what’s possible, I asked what was plausible.

Do you think that they came up with it independently? If you aren’t willing to say you believe this, you are just asking me to waste my time arguing with a phantom opinion nobody holds.

“Just do it” is an incredibly common, generic phrase, but if a new shoe manufacturer chose it as their slogan, it isn’t plausible that they came up with it independently of Nike.

I don’t care what’s possible, I asked what was plausible.

OK then. BOTH ARE PLAUSIBLE. Clear enough?

It's absolutely not up to me to decide whether they came up with it independently.

But then again, it isn't up to you, Jim.

If you aren’t willing to say you believe this, you are just asking me to waste my time arguing with a phantom opinion nobody holds.

This is the weirdest thing to put in a comment you could simply not have made. If your time is so valuable why are you here commenting on HN?

Again: aphorisms have value regardless of how many times they are used. People can use them without you having the right to demand proof they thought them up for themselves like some prep school debate bully picking on a younger child.

> If you aren’t willing to say you believe this, you are just asking me to waste my time arguing with a phantom opinion nobody holds.

This is the weirdest thing to put in a comment you could simply not have made. If your time is so valuable why are you here commenting on HN?

It’s basic intellectual honesty. If nobody believes that they came up with it independently, putting it forward as if it’s a real argument is just a form of trolling. If I don’t believe it, you don’t believe it, and even the person who posted it doesn’t claim it’s original, you’re just baiting me to respond to a position nobody holds.

Do you believe that they actually came up with it independently? If you do actually believe this, then a discussion is not a waste of time. But if you aren’t willing to claim that position, it’s pointless me arguing against it because it’s just you deliberately wasting my time on something you don’t believe.

aphorisms have value regardless of how many times they are used.

This is a cornerstone of the point I was making. You’re literally supporting the point I was making.

The way I see it is that a clearly AI generated article is the result of some prompts. So the "author" wrote tens of words, which then got inflated to hundreds and maybe thousands by a computer program. This is a very inefficient way to convey an idea, especially if the reader has access to the same program. I'd rather the author shared their prompts, because maybe there's some originality in it, and I can then decide if I want to explore the idea further, maybe even with an LLM.

What if the prompt for the 1500 word article you're reading is 10k words of research/data collection, thesis testing, etc?

I've seen this in action as well. The biggest value by far is seeing the overused phrases and tics in your writing style. The prompt I like is to tell it to highlight and flag passive voice and run ons since those are my major pain points. I can also tell when its getting out of "real" advice since it'll offer the same "trim 10-15%". Also important to tell it to highlight specifics otherwise it'll give vague editorial advice that isn't anchored. Another fun exercise is to submit other authors writing as well. Say a short story from a magazine. Some of the edits are legit, but do it enough times and you'll start to see the same generic writing advice gets applies to just about any input.

another iteresting thing models can do is to highlight and notice themes and motifs in a piece. I find that helpful when thinking about a piece of work. Might be obvious ones, sometimes it can be subtle themes. Good for stirring your own brain.

overused phrases and tics in your writing style

In the extreme these can be annoying, but more often I find them kinda charming, as an expression of the author's personality.

Simple example from this piece: I scrubbed 7 instances of the word “just” from it. Obviously, Emacs can do that on their own, but you have to know to look for that word, and every other one.

You are erasing yourself in doing this, not improving your writing.

If you want to improve your writing, read more.

use them like a copyeditor rather than a ghostwriter

This is what you need to know. Seriously, after 25 years in news (writing), you realize that you need “second eyes”, not the first hands.

It's darkly amusing that while the author advises us not to let LLMs pick our words, their post includes this chunk of AI slop:

"Those rethinks are load-bearing parts of your voice. Readers won’t put their fingers on what’s wrong, but they’ll sense that you’ve become artificially-flavored."

The metaphor at the start is AI-like weirdness too: "without pasteurizing and jacking it with corn syrup." Pasteurising is good, corn syrup is bad - what the heck am I supposed to make of it? It's nonsense.

It may even have been handwritten by the author, but it's slop nonetheless.

You may be telling yourself that you're disciplined and you're just using the LLMs for review. But while you're reading their output, the slop makes its way directly into your brain anyway - and then it drips out into your writing.

I saw "load-bearing" there and instantly assumed Thomas was making a deliberate point that it's OK to use terms that LLMs have poisoned provided you use them effectively and in a way that adds to the piece.

I'm personally furious that "load-bearing" has become an LLM tell, it's a really useful metaphor!

It was a joke.

Recent comments from the simonw account suggest it isn't operated directly by a human anymore /sigh/

How to Write with an LLM

TLDR: Don’t.

That's what the piece says.

Even if you don't want style advice from an LLM (I generally don't, not for blogging), for a technical blogger "please check this post for factual accuracy" is a good way to not look like an idiot. Often it complains about hyperbole, which I tend to ignore since there's no point blogging unless you have some firm opinions, but on at least one occasion it pointed me towards some documentation that undermined my entire post - and led to a much better one a week later once I'd processed and applied that new information. Thanks, LLM.

I do this all the time. It's shocking how well it works. A year or two ago suggesting that someone use LLMs for fact checking would get you laughed out of a room, but today the good ones (Fable, GPT-5.6/6) with a search tool enabled are genuinely excellent at this.

They can even retrieve articles you link to and check that you aren't misrepresenting them.

This is bad advice for anything serious that can be subtly wrong. I’ve had Fable being consequentially wrong about what a certain, rather small, code block does in our codebase. Not even on something very complex which wasted a lot of time and effort from the team until we checked the code ourselves. I think these tools are still useful and one should learn to use them but blind trust for anything of importance is misguided IMO.

To paraphrase, "if a junior programmer doesn't understand what your code does, it's bad". If Fable doesn't understand, well...

Hyperbole = strong opinions? If you're stating your 'strong opinions' as fact, then you deserve the push back. If you're stating your strong opinions as strong opinions, then fair game.

If you're stating your 'strong opinions' as fact

Is there anybody on this site who isn't?

For commit messages and other technical writing, I use a "narrative coherence" checklist to help ensure that a change is clear and understandable:

- Establish a single thesis an external reviewer could recover from the diff.

- Unify vocabulary across code, comments, tests, and commit message.

- Use the same vocabulary consistently to refer to the same concepts.

- Keep every hunk that serves the thesis; consider the removal or deferral of the rest.

- Introduce abstractions at the point of need.

- Align tests to narrate the same story as the implementation.

- Reconcile the commit message and the diff.

- Order changes expositorily, not chronologically.

- Explain what the change does and why. Do not explain the details of the development process.

- Prefer to edit subtractively.

- Recompile and run all relevant tests after edits.

- Iterate.

That checklist sounds like at least as much work as writing a commit message.

Ironically, "load-bearing" appears in the main text of the article that says never to use a single word the LLM suggests. Womp womp.

Present the original and new writing to the model and ask it which is better.

LLMs have absolutely terrible taste when it comes to writing. I don't find their feedback useful at all, beyond trivial spelling/grammar mistakes, which you don't really need an LLM for in the first place.

Proofreading is all you need.

Edit: I do sometimes ask an LLM for a fact-check, though.

That’s what this is: it’s mechanical proofing.

As a "proof" of their poor judgement, take a paragraph you like. Ask the LLM to rewrite it to make it better (which I think we both agree will not make it better), and then in a fresh session ask it which it thinks is best. It'll almost always pick its own writing, even when it sucks.

Right, I agree: that’s exactly not what I’m recommending.

You are specifically recommending asking the model which of two versions is better (the quote in my top-level comment).

We both agree that they are poor "make it better" machines, but I also believe they are bad A/B testers and I'm using the former to demonstrate the latter.

You are writing both paragraphs. You’re specifically not asking a model to make a better paragraph. That would be a load-bearing debacle.

It doesn't matter who writes what, what matters is that LLMs have a preference for LLM-shaped writing. By A/B testing against an LLMs opinion, you are optimizing in the direction of LLM prose even if the LLM never writes any of the prose itself.

Till today, I was doing the complete opposite of what the article suggests. Definitely going to try this approach out, pretty helpful.

I have a specific ChatGPT chat that rewrites my posts really well every time; I usually just give it a rough draft and let it clean things up. But it was pretty interesting to realize how good these LLMs are at actually finding mistakes rather than just rewriting things. The point about not letting the model influence your voice was also very, very relevant.

I mostly use LLMs for coding, so this is an interesting use case. I’m curious how long an article it can handle in practice before the context starts becoming an issue.

The frontier live voice models (especially GPT Voice/Live or whatever they call it this week) do not have the typical AI-tells. So I wonder about a few things:

Why is this?

Why couldn’t whatever went into post-training these voice models be applied to the text-gen models to get rid of AI smells?

Could we set up a pipeline that leverages the voice models to clean up text that reeks of AI ?

I think it's good advice. I find it a bit funny the author said to not use a single word coming out of the LLM, but they probably did just that because I never heard a human being say "load-bearing" before but now it's all over the place, also used here. Or I don't get the irony.

If you’re familiar with tptacek’s writing you would be confident this is deliberate and recognize it as an ‘I see what he did there’ moment.

If not, it does risk planting the seed of doubt - have I been duped once again into reading tokenslop again?

Risky move. Surprised the LLM copy editor didn’t flag it.

Writing is thinking. You are outsourcing the process of thinking through your thoughts so you can communicate and understand them.

Those rethinks are load-bearing parts of your voice.

I really hope this was intentional.

In context, he's just saying this self-scrutiny is what makes your words authentic.

I was referring to the use of the word "load-bearing". It was most likely an intentional joke, but it's one of the most grating current AI-tells to the point it's a word you'll at this point generally want to avoid (unless it's a joke).

I think style is overrated. Yeah if you write commercially, you are a journalist or a copywriter, sure, you need to write in certain style. There’s rules to this shit. I think it was Hemingway who said that. Then again, the good ones break the rules.

Today even if you don’t use LLM’s following certain formal style you end up sounding like ad copy or wannabe Atlantic -hack. The. You might read some kid on Tumblr writing beautiful prose with typos and errors and all.

First of all you should have something to say. I write a blog just to make tutorials for myself on stuff I need to regularly check out. I don’t really care if anyone reads it. I don’t use LLM to make the posts since, well, pride, but also because I learn better when I need to write the stuff down and check my facts several times. I don’t much care if I make a typo or grammar error here and there. If I want to write it as Jeeves from Wodehouse book, fuck it, I will.

If you have a story you have something to write about. It doesn’t have to be fiction, it can be a story on how multicast works. And it needs to be your story. Then the words come by themselves and writing is relatively easy. To have a story you have to live and think. LLM’s don’t help in that.

You can just not copyedit. That's fine. More than fine, lately: rough edges are a signifier of authenticity. If that's the way you want to go, writing raw, I think that can be a good strategy.

But if you've generally been doing drafts and want to keep writing like that, if part of your process is going to continue to be refining and sharpening, then I think you should know that we're at a point where automation can accelerate that process for you. You can be getting the same amount of work done with less effort, or allowing yourself to concentrate more on the creative thinking part and less on the tedium.

That's the point of the post.

My point was, however, is that it depends on what -- and how -- you write.

Style is kind of a pretense. Communication in general is a sort of a mucks game of guessing how to get your point across across the medum, whether that of air, FM-radio, writing, internet, video... In academia, copywriting and so-on we create these patterns we call writing styles to agree on the rules on the game. But honestly, we should use sá lot more fuckin swear words when disproving the fuckin conjectures.

Yes I know automation makes things quicker. Sometimes even close to fast. But htere's things you should do slow.

Writing and thinking are amongst them. And hey, LLM's can help you here as well. But they should not speed you up, they should slow you down. Why did you say this? WHat is your premise? What is the source of this? KIM by Moonshot is really easy to take a system prompt that makes the models output a biggest asshole editor in the world. It doesn't help me at all, on the contrary, it creates friction.

I don't think more speed is what we need at this point of evolution.

I've found the only good way to write with LLMs is to absolutely not use it for the first few drafts, the using it as a tool to list specific kinds of problems in text (word repetition, unnatural hard to read sentences etc) and approaching those points with caution and skepticism. Even copy pasting a single sentence, I've found, is a slippery slope.

Just write, write your own words, then create an effective communication persona within your LLM's non-reasoning (non-Chain-of-Thought) capabilities and have it critique your writing after describing your target audience. Then, write your own revision following the critique. Using this, it is not possible to write "AI slop" unless your own writing is already in the style of LLM output. Which I get accused of when I write using my formal voice, but fuck that sorry for being well educated.

So give the options to a model that doesn’t have the context of your editing process.

I conjured a bit of software to manage this for me[...]

I don't understand how the annotation application shown helps with the stated goal of preventing the model from praising you. What's the actual software assisted workflow?

"LLM paragraph will register to much of your audience not as writing but as output"

The best advice for writing for other humans is: don't use LLMs.

If you're writing for machines, liking coding, then fine go for it. If you're writing for processes with formal highly structured content like manuals, specifications, form content, procedures, information, that sort of thing, then also ok to use LLMS. But if you're writing for a human mind to ingest and extract meaning from, then LLMs are poison.

Agreed. If I’m spending my precious time reading something thing, I want the author to have invested their personal effort to write it.

Or to put it another way: Why should I be interested in reading something you weren't interested in writing?

the author can ask the same question - why should I write something if you're just going to use a LLM to summarize it ?

This is an unnecessary argument both ways. Whether you use a LLM or write it yourself or read it yourself (and then comment on it), you're putting your signature beneath it. I don't care whether you use your natural brain or an artifical brain (LLM)

Those of us using our human minds to extract meaning from specifications and manuals don't want to look at LLM slop either.

Interesting way to put it because it means in most professional software development contexts, it is also NOT appropriate for coding since other human will work on the same codebase, at least review the submitted code. So... according to this (IMHO pragmatic) heuristics it's only acceptable for coding alone, e.g. weekend project that won't be pushed online. Unless code has no meaning beside the output it produces, which does exist but I would wager is rare.

But reviewing isn't the same as reading.

When you review code, you're not taking it all in end to end like an email or memo or blog, you're looking for important symbols, recognizing structures and patterns, jumping between different files at different levels of reference and abstraction, and building a model in your head of the principles that make it work and their consequences for the overall procedure.

Code is "meant to be read" in the same way as a law, blueprint, policy document, or manual is, ie: jumping between relevant sections to get small pieces of information relevant to some task. The act of "reading" never starts at the beginning, it starts with a ctrl-f or flipping through an index or table of contents for a specific relevant symbol that is the only reason you are looking at the document.

Code is invention, not art (and should be covered by patent law, not copyright law, I will die on this hill)

It is not the same yet it requires reading which in turns requires interpretability, trying to make sense of what was read within a very specific, maybe explicit context, e.g. specifications, and that can be optimized for an audience, or not.

For example the code I would write for somebody who just started to learn programming will not be the same as the code who is an expert even though both will pass the specifications.

Consequently taking into account who (if anyone) will actually review, and thus read, the code changes the code itself.

If you're writing for processes with formal highly structured content like manuals, specifications, form content, procedures, information, that sort of thing,

Yeah... no. The people doing this lack the communications training, see the output has the necessary information, and regurgitate it with no effort or care. This needs to stop.

We spent decades format building to make it easy quick and easy to get through something like a runbook. If your commands are bulleted instead of numbered and code blocked, it's wrong. If you didn't crawl through the playbook, it's immoral to hand that to me, you're wasting my time with untested slop.

This is a hill I will die on or absolutely start slaughtering people on. I just refuse to deal with this crap.

I used an LLM a year or more ago to generate a description of our SDLC for a compliance certification. Perfect application of LLMs IMO. Do you want to die on that hill as well?

A lot of documentation generated in corporations has marginal real value.

If your commands are bulleted instead of numbered and code blocked, it's wrong.

That’s easy to instruct an agent to do. Put it in a skill.

If you didn't crawl through the playbook, it's immoral to hand that to me, you're wasting my time with untested slop.

Using an AI doesn’t absolve the user of their responsibilities. This is an easy problem to solve, just make sure teams know what’s expected of them and encourage everyone to push back (professionally) against offenders.

If someone sends me AI slop, they're getting chewed out and told I'm not doing it and I'll even tell my boss "no" and why. If I got fired over something like that, then it tells me everything I need to know about company and the leadership's priorities. I'll die on that hill.

To be completely fair though, I'm against the behaviors in information transfer I've been seeing. I've pass along AI generated runbooks, but they look nothing like the default outputs of these models. It's because I took time to apply all the writing knowledge I like to see in my curation. If people are doing this, I can't even tell it's AI writing. My work is done in minutes instead of deciphering so BS pseudo language they developed in their AI workspace (people really need to turn off those memory features).

---

Edit: Also if I'm the guy receiving a security report and it's AI generated and poorly formatted, I'm failing you short of producing something for a human to parse. Simple as that.

If a security report is written for any particular human at all, I’d consider it to be a failure as an enterprise policy document. It should be written for The System, not for the boss; and LLMs are perfect for producing ritual boilerplate.

Also if I'm the guy receiving a security report and it's AI generated and poorly formatted, I'm failing you short of producing something for a human to parse. Simple as that.

It’s clear that you’ve never worked for a compliance company and probably have never been involved in a compliance project. As such, you don’t have the context needed to participate usefully in this discussion.

In fact, post-LLM quality monitoring is a huge time trap.

I can't write anything for humans using LLMs, because there is so much to change that it ends up like the Ship of Theseus, I would have been faster writing it myself than delegating the task.

I've had some success writing certain explicitly technical documents, with a style guide provided to the LLM that it can match, and then going over it by basically iterating on every sentence in the document and asking "can this sentence be removed?".

Style guides are a big tools people are missing out on. It isn't enough to say "write concisely and technically". Give them a sample. Yea verily these many 5 or 6 years ago, "style transfer" was a big thing that early LLM tech was doing. It's still very good at it.

That said I still tend to cut out at least 25% of the resulting verbiage and adding back another 10% or so more of my own original content even under those circumstances.

Where it has been really helpful is that I tend to want to write in a conversational style that doesn't seem to match most people's expectations of a technical document. The LLMs let me write my way and style-shift it into something closer to what people expect. And LLMs, since let's be honest they're the primary audience nowadays. Which I am not even upset about; I'd rather 5 LLMs read the architecture document than the amortized .2 or so humans I could expect in the same circumstance 5 years ago.

Which also implies, in many cases, I am feeding the AI as much text as I expect to come out, or in some cases, even more, as I am describing context, reasoning, and other things that may impact the writing but are not necessarily repeated in the final text. I factored out a lot of the context into my user-level CLAUDE.md which has helped cut that down a bit.

Having the style guide and proofreading it and linking it is more effort than the lazy slopulists want to put in, so your writing will reach far better than theirs will

You have to assume that in many situations a human is asking an LLM about the content you've provided, even if they or another human reads some or all of it themselves too. So it's valuable to proactively ask your own LLM in the same way, so you can see what the other person will get out of it. And tune the verbiage until the LLM tells them what you want it to :)

Sorry if this comes off as critical, but are you seriously doing that? I can't imagine optimizing my writing for people choosing to mangle it through a LLM rather than for the people reading it as-is.

Also the word "verbiage" ticks me off as a LLM-ism on the level of "delve" or "smoking gun". Are its users aware that in addition to its usage as somewhat of a synonym of "idiolect", it more commonly means "a profusion of words usually of little or obscure content", i.e. slop?

Too simple, too dogmatic at the start and at the end (“don’t use LLMs” and “then they are poison”). The middle section is all good.

Writing with a LLM is a boon if you are trying to write while mastering a new field. Imagine a neuroscientist entering robotics or an electrical engineer trying to enter the field of neuromorphic systems. In these cases a human will need help from an LLM to think and write more deeply AND broadly. I sure do.

You could object that this form of human-LLM collaborative writing is just for one reader; the writer. True at the start but not true toward the end.

I do agree that even the frontier models do not touch a really strong human writer. Way too much gimmickry. They do not yet match individuals such as Leslie Lamport, Terry Winograd, John MacArthur, Richard Rorty, Humberto Maturana, William James, Rodney Brooks. They don’t have the depth of understanding to simplify. But give them a few more years and excursions in the real world and a concept if “now” and the situation may improve.

I often ask Fable and Opus to explain difficult topics to me as if I were the canonical dog Rover. It works.

Maybe don't write about something you don't know?

Absolutely correct. No matter what the detractors say, your writing will get glazed over and you will lose the focus of readers if your AI output sounds like all the AI output your readers have read before.

don't use LLMs

Don't you think they're useful as fact checkers?

Sure, if you don't mind your fact checker to be confidently wrong often.

This would sound insane to me from two years ago but I have recently started insisting on writing all my own commit messages and pull request descriptions. I do usually have an agent review them for factual accuracy, but not rephrase them.

It slows things down a bit, but in the best possible way. It has helped immensely to improve the depth of my understanding of the agent-generated code. When agents are doing everything its way too easy to “skim” diffs and not really absorb them.

I always prided myself on my technical writing, and commit messages and PRs were a great place to hone that skill. I found that I missed it and my work is better now I’ve reclaimed that part of my old job back.

A big win of LLMs in my book is the absolute reduction of commit messages with just “fixes” or “updates”. Commit messages have become more meaningful and useful, even if far from perfect.

We have one dev who uses LLMs to write the code, but still commits by hand. Most of his messages are of the type above, and none of them are useful.

Why do you want to understand the agent generated code though? That would make you the bottleneck. Shouldn’t you just concentrate on checking for optimal outcomes, exhaustively trying to find/prompt edge cases and profiling performance?

“Rule Number One: You may not use a single word an LLM suggests to you.”

So far so good…

“Those rethinks are load-bearing parts of your voice.”

Oops.

This article should just be one word:

Don't.

Readers can detect LLM words in the parts per trillion. However much work you put into scuffing up and humanizing it, an LLM paragraph will register to much of your audience not as writing but as output. So, first the bad news: you have to write for yourself.

---

and then

Those rethinks are load-bearing parts of your voice. Readers won’t put their fingers on what’s wrong, but they’ll sense that you’ve become artificially-flavored.

---

Are they wrong in the first paragraph? Or are they simply not following the advice they dispense to others?

I thought that was a joke.. Like how I used "delve into" back in 2024 or so lol

Step one: open up the LLM

Step two: close the LLM

Step three: pick up a pen and write

Writing with LLMs is a great way to make yourself stupid: https://lambdaland.org/posts/2026-08-07-ai-writing-stupid/

I'm glad the OP acknowledges that you have to have to do the writing yourself—though they say this more as a means to avoid getting pegged as having used an LLM, whereas I see the problem as being cognitive surrender.

LLMs seem overpowered for the use the author suggests, finding and marking problems in prose. Actual human-written programs exist for that purpose, things you don't have to bludgeon into not praising you or semantically ablating your work. Start with langtool or the like and start adding others. Even art-bollocks mode for Emacs can be useful.

I sort of agree, but I also think the advice here is circular. To look at the style suggestions produced by an LLM and decide that you agree with some of them in spirit, but want to put it in your own words, you already need how to know how to write well.

In other words, you need to read a style manual or two, and you need to read and have opinions about other people's work. You must have taste. Otherwise, you can't really distinguish between good and bad advice. You'll anchor to the suggestions made by the LLM and it will turn your writing into LLMese. I guess that works - it sure gets clicks on HN - but it won't be your voice anymore.

There's still value in getting a second pair of eyes on what you write. But here's another cool trick that gives you a similar benefit: just let it stew. Go for a walk, grab a lunch. Re-read what you wrote. You'll be surprised.