LLM Voice

> I'm also suspicious of my own ability to accurately recognize LLM voice.

With much sympathy for the headline this is the central problem.

It's not easy to say what Uncanny Valley looks like, and so we're left with suspicion not certainty.

BTW, I am now 4-5 out of hundreds of my comments since AI detection became a thing, called out for having "AI" tone. I fucking hate being objectified like this. It's not true, although I understand that assertion is hard to refute or support. All I can say is, some people write like some people believe an AI writes.

If it puts an end to LinkedIn-speak I'll take it, for a year or two.

> In time it may be that LLMs will be so good that we can't tell the difference.

I have to believe that we are within a couple of years if not months of this. This problem is too well-known to not be a major focus of the frontier labs' efforts.

Advocates of the dead internet theory are having a field day with this

Words come from machine? Work is of machine. I skip human and employ machine

I recently read DDIA for the first time, and I was surprised by how enjoyable it was after reading AI slop for a year.

I think the key tell is that the writing doesn't actually illuminate anything worth learning. The more I see of AI produced output, the more I believe that there is nothing there that wasn't already in the prompt. It's an inflation of a single idea to pages of text. It meandors about and never actually gets into detail.

It's a lot like talking to someone who doesn't know what they are talking about but are good at faking it for a while. The game works by leaning on pat phrases, circular reasoning, and hoping nobody is paying too close of attention. I think this is why it's so popular with the corporate executive class, as that has been their mode of communicating their entire careers. To them, they don't see anything wrong.

"Shitty writing existed before AI" is unfortunately an all too common rejoinder to this issue. That's not an excuse. That doesn't make it better. Just because the pointy haired boss talks that way doesn't mean we should be gleefully pouring gasoline on the fire.

This guy sure likes to use his own dashes, just saying.

I noticed the exact same thing, and I also don't want the LLM to suggest anything to me. I do want it to comment, and to tell me what could be improved, but not to give me text.

I created a small tool to make this easier: https://www.writelucid.cc/. It's free (bring your own API key), feedback would be appreciated.

I disagree with the advice to avoid having an LLM write a first draft and then editing it yourself.

To me, for most business and academic content, the ultimate goal is to convey information. If I can edit AI copy to legibly convey that information clearly, that's enough and I can move on.

The saying, "Don't allow perfection to be the enemy of the good," comes to mind.

And I'm not convinced an individual can accurately edit without learning just as much as if they wrote the content. I guess the question is: how hard does the person work at editing. And does the individual challenge themselves to not put out content they themselves don't understand.

We're in a transition period. And as in all transition periods, many recoil at what is new. LLM writing still feels strange. And it's understandable.

But, for better or worse, in a business and academic setting, as the years go by, most of us will shrug and move on.

As with any topic, there will always be a minority of fanatics who cling to the rejection of AI writing in all its forms. Just as some prefer Vegan leather or tabs over spaces.

But among those who get things done, most (though not all) will accept that it's not worth it to exclude AI from writing. There simply isn't sufficient ROI.

It's like trying to bring back low-end highly human employed manufacturing to wealthy industrialized nations.

That ship has sailed.

I do agree, however, that in cases where an individual is reading for pleasure -- e.g. not in a business or academic context, authors will be forced to pick their audience: those who enjoy writing, even if it smells of AI, and those who don't.

I guess to summarize, my feeling is that using LLMs to write is perfectly acceptable, and used well you'll likely learn just as much. But you must be willing to lose some of your audience. And, over time you'll lose less and less of that audience.

I don't know, with LLM it feels like I finally have possibility to prepare all the dull text that corporate environment expect in a form that colleagues are able to digest even if they don't specially enjoy it like it's the best poetry their beloved one ever composed specially for them. This is in sharp contrast with the struggle to shrink into a somewhat coherent excerpt the wide web of interweaved psychological experience flow graph into this alien expressions which are text streams. It's still impossible to dump really the whole experience and communicate wholly the full semantics they cast, but at least these automatons are never scared to face far much than they can sustain in what can be thrown at them, and apparently most people prefer the dull soulless prose to whatever can suggest vivide continuous reconnection of everything as the context inevitably drift and everything needs to be reconsidered properly with all the rest.

It's incredibly hard to maintain your own voice while using LLMs. It requires deep discussion with the LLM itself so that it comes to understand your perspectives and also explains to you the premises on which its own recommendations are based.

I haven't really done a systematic analysis to compare the quality of fully human-written content and content generated via this kind of deep human-AI collaboration, but my intuition is that the latter is vastly superior. Perhaps AI at its best is a research tool, but a genuinely intelligent one that we need to engage in conversation with in order to use to its full potential.

The sense that I get is that as a conversation gets more in depth, the LLMs start behaving more like reasoning engines and less like statistical averaging machines. The former is vastly more useful to a person who wants to develop an idea and articulate it well.

> But if I want to communicate with humans, my humanity has to shine (or at least ooze) out of my prose.

Yeah. I want to be able to touch a human being (or at least have one touch me) through the words. LLMs poison that, even if a human prompted the LLM.

Translation... I think I object to translation less than Martin does. There's a human back there, and they're writing all the words, and then an LLM is translating them. The LLM may not translate them perfectly, but they aren't going to introduce "load-bearing" when it wasn't there.

The bit about using LLM when you're writing for a bureaucratic process rather than a human was interesting, at least to me. I hadn't thought of it that way before, but I think it makes sense.

Thank you for posting this. I don't have a lot to add on the subject, but I found it interesting and thought-provoking given the age we live in. I find it really hard to tell real human and LLM text apart, even though some people claim they can, calling it "AI slop".

Yawn. This subject has been discussed ad nauseum. The lines are drawn. I'm over it.

So many "load-bearing seams"...

People have been talking about the importance of curation for more than a decade.

Because many good ideas are not immediately obvious good ideas, in an environment at risk of information overload, we need some way to quickly vet ideas to see whether they are worth investing the time to consider them. (It is often the case that the best ideas require considerable investment of time and effort to be judged as good, and, it's also often the case that this determination can be made only at the very end of such a process.)

“Don't judge a book by its cover,” sure, but, historically, there have been superficial markers of care and attention that we've used to these ends. These are obviously very flawed signals, but, you know, maybe if you took a lot of time to craft a message, I don't feel that bad if I waste a bunch of time trying to decode it. These markers—attention to detail (e.g., spell and grammar checking,) clear-structure (e.g., “bottom line up front,”) and typographical flourishes (e.g., smart quotes and em-dashes and extended unicode like diæreses…)—were an indication of some modicum of quality control and care. I, myself, used to brag about or show-off having set up .Xcompose keybindings (https://wiki.archlinux.org/title/Xorg/Keyboard_configuration...) which was just rare enough across the general population to be noteworthy.

All of this that was once somewhat expensive is now cheap or basically free.

Even if I make a point of going back to a pre-UTF-8 Usenet style (or even intentionally insert run-on sentences or misspellings or grammatical errors,) the cost of an LLM to style-transfer its output to match this is barely much of anything. All it means is I wrote one prompt for the first draft, and then one more prompt for the refinement.

Additionally, some LLM habits are actually good habits. Putting the bottom-line up front (in bold, even!) is a great habit in a business e-mail! Decision-makers genuinely struggle with reading through long paragraphs of detail!

Unfortunately, the problem with the LLM output is that it takes good, valuable, superficial or structural habits that it can cheaply apply, and it drapes those around inane, banal, cloying, muddy, meandering meaninglessness.

Today, we can maybe still judge a book by its cover, because even one more prompt worth of refinement is still too much effort, but I don't think that's amounts to much, since the frequency and loudness of complaints of LLM writing means that model trainers are going to train their models to adapt to whatever signals we use for quick vetting. I am skeptical whether this training will concurrently improve the novelty, innovativeness, or other intangible qualities of the actual message. (With the models, as they exist today, and as we use them today, at least in the field of software development, I see extremely poor capability at generating innovation… Personally, I don't even like the quality of the Python code they spit out, but they sure do type real fast, and, like, pejoratives aside, unexciting, mediocre, corporate-y code does have a real place in the world.)

So what happens when the economics of production in the medium significantly outpace the capabilities of production in the message? The LLM can write a comment nearly indistinguishable from any given human… except when you actually take the time to read through it, it's just inane drivel. Thoughtless and a waste-of-time. And there's basically no way to manage your investment of time and effort to tell one way or the other.

An important distinction as AI-generated content becomes more common: having a consistent voice is not the same as having a distinctive point of view. The best AI-assisted writing should preserve the author's personality, judgment, and expertise rather than simply imitate their vocabulary or sentence structure. This is a useful concept for anyone building content workflows with LLMs.

[flagged]

[flagged]

[flagged]

Great piece. The "LLM Voice" problem is real — we've been benchmarking 18 free LLM API tiers (Qwen3.8-Max, GLM-Z1-9B, Hunyuan-MT-7B, bge-m3 embeddings, etc.) for consistency and tone variation across providers, and the homogenization is striking. Models from different families converge on similar cadences and filler patterns even at different parameter counts. The one outlier we found: reasoning-focused models like GLM-Z1-9B actually break the mold with chain-of-thought traces that feel more distinct. We publish the live benchmark data at apishare.cc/leaderboard if anyone wants to compare latency/quality tradeoffs across free tiers. Would be curious if others have found specific model families that maintain a more distinctive voice.

[flagged]

[dead]

Journal entry: Dec 12th, 2027.

My name is Martin the last human organic writer on Earth. I am better than everyone else; my citations in my 4-paragraph article say so.

personally, i'm for LLM and AI usage in writing if it helps the writer communicate ideas, their stories, and triumphs in ways they may never have been able to before. I just had one of my posts removed from the front page here which was disheartening because it feels as though what was judged was prose rather than the technical insight in to a niche problem, which seems counterintuitive considering i've always thought of Hacker News being an inclusive place for the technical and engineering skills rather than their level of the english language.

Just do what I do: capture everything I've ever written and ask AI to cobble together what I said using my own sentences.

No more LLM voice! Every now and then you might get a nonsensical artifact or two, but that's the price you pay for better groceries.

I'll repost a comment I made a while back describing the process I use, which I think balances AI in keeping your voice.

If you care about your voice, don't let LLMs write your words. But that doesn't mean you can't use AI to think, critique and draft lots of words for you. It depends on what purpose you're writing it for. If you're writing an impersonal document, like a design document, briefing, etc then who cares. In many cases (scientific papers, legal documents) you already have to write them in a voice that is not your own. Go ahead and write these in AI. But if you're trying to say something more personal then the words should be your own, AI will always try to 'smooth' out your voice, and if you care about it, you gotta write it yourself.

Now, how do you use AI effectively and still retain your voice? Here's one technique that works well: start with a voice memo, just record yourself maybe during a walk, and talk about a subject you want, free form, skip around jump sentences, just get it all out of your brain. Then open up a chat, add the recording or transcript, clearly state your intent in one sentence and ask the AI to consider your thoughts, your intent and ask clarifying questions. Like, what does the AI not understand about how your thoughts support the clearly stated intent of what you want to say. That'll produce a first draft, which will be bad. Then tell the AI all the things that don't make sense to you, that you don't like, just comment on the whole doc, get a second draft. Ask the AI if it has more questions for you, you can use live chat to make this conversation go smoother as well, when the AI is asking you questions, you can talk freely by voice. Repeat this one or two more times, and a much finer draft will take shape that is closer to what you want to say. During this drafting state, the AI will always try to smooth or average out your ideas, so it is important to keep pointing out all the ways in which it is wrong.

This process will help you with all the thinking involved being more up-front. Once you're read and critiqued several drafts, all your ideas will be much more clear. Then, sit down and write your own words from scratch, they will come much easier after all your thoughts have been exercised during the drafting process.