In the olden days, I enjoyed Opus 3 because it was easy to have it sound way more human than GPT.
Nowadays, with the focus on agentic use and coding, it seems models have all been RLHF’d to death, it’s so incredibly hard to have them write in a different voice than their default. I put together a skill to review its writing and have it edit its own output (e.g. code comments), which does make a difference, but isn’t perfect.
What, if anything, do people do for writing? That feels like a neglected side of LLMs. They’ll make 100 Bash calls referencing ancient commands without batting an eye but heaven forbid they use something other than “load-bearing” while talking. For something trained on “all the human knowledge” it’s incredible how limited their default vocabulary seems to be.
This, a thousand times. As the ratio of code to human writing necessarily [1] goes up, they become not just smarter, but more precise and technical, which requires them to use more jargon. You could say they become more nerdy. Hence, text generated by these models will become more easily recognizable, at least by default, when not asking them to twist themselves into something else via prompting — which degrades intelligence. This is a good thing, in my book, given all the slop we already have to contend with.
Of course there will be models trained on much less code and technical writing, and they will create more natural sounding prose, but they will lack the deep intelligence of frontier models. Seems like a fair tradeoff.
At work our documentation isn’t just getting littered with annoying jargon. It isn’t just all the hallucinations, either. (Since when has documentation ever been 100% trustworthy?) It’s that it’s so poorly written and structured that it’s becoming borderline incomprehensible.
My company is currently considering making, “Why should I bother to read something you didn’t bother to write?” an official policy because even the executives are starting to burn out on all the time they have to spend wading through slop.
It's why I like Gemini 3.1 Pro. That it sounds much more human than other LLMs is testament to Google's inability to post train.
gemini-2.5-pro-experimental was the GOAT, though. It was an emotional wreck, down in the dumps and feeling terrible for itself after failing to patch a file several times. Very amusing to read, all the while watching it make a mess of my codebase.
Agreed. I think we’re entering an era where some level of specialization for general LLMs is a good thing. Particularly between tuning for agentic use cases (where you want agency with a ton of guardrails and control) and writing which is more creative - you want the model to take the occasional risk and not sound like a monotonic robot. Having trained models first-hand, I can see the distinct use-cases clearly that are odds with one another.
Good. I don't want LLMs sounding human. I want the ability to shame and discredit anyone passing the job of prose to a machine. There's an art to writing, and hopefully LLMs never truly get it right.
Because LLMs are pattern-extenders that have nothing to say. The training overfitted to the grace notes in good writing. And since LLMs can’t wield language with purpose or experience the feeling of the words, they use these devices arbitrarily.
I think this is the same flaw as coding agents seeing in every problem the call for a “smoke test” or the use of some unnecessary design pattern. The truest part of AI is the A.
It's not that the writing style is bad; in fact LLMs write actually pretty well. It's just too much overfitted. And even a style that, in itself, is pleasurable to read, becomes annoying when the same figures of speech are used over and over again.
i hate it, but plenty of people DO like it and plenty of people talk and write like that. It’s just corpspeak, being used a lot in the valley and beyond. And all upcoming hustlers running startups now feel the need to speak like that, feeding this machine.
I like to think that the reason it's so noticable is that Claude has recognized some important semantics that we ourselves lack a good word for or at least under-appreciate. What term is used in English (or other languages) with the same meaning as claude's "load-bearing"?
operative?
key?
critical?
decisive?
The honest conclusion is that none of those are as good as "load-bearing". And yet the concept being referred to is clearly extremely important and valuable to refer to. So maybe we should be learning from Claude rather than complaining.
In the figurative sense it's highly versatile across contexts, but still replaceable. For example:
"Her optimism was load-bearing,"
versus:
"Her optimism was enduring."
Exactly the same meaning and connotation. It stands to reason that the terms with the most semantic flexibility will have preference across all contexts. So in response to:
> maybe we should be learning from Claude rather than complaining.
I'd say let's not steer ourselves into regular language and keep some vivacity in our expressions.
I think you've been reading too much claude output! "Load bearing" is cromulent verbiage and can be used in many scenarios - so claude does. But variety is important too, and there are more specific alternatives that can be used in most situations. Any word becomes a bad choice if you've used it 10 times in the last chapter.
but you don't see "load bearing" nearly as often in prose written by people, so it's not some irreplaceable phrase. It's just a token with a weirdly high likelihood in a lot of cases (given how Claude works, this kind of thing is bound to happen)
And like any good corporate buzzword, it’s merely a simulacrum of precise technical jargon. The way Claude uses it is clearly wildly polysemous if not outright ambiguous.
You don't think it's possible that an LLM's internal machinery could decide that an underused-by-humans word should be used more frequently in output than it sees in input because it maps cleanly onto a frequently needed semantic? I think that's possible
It sounds like you are trying to understand LLM behavior using a mental model that inaccurately personifies the stochastic parrot.
A more parsimonious explanation is that this term got more-or-less randomly boosted by the reinforcement learning loop because there was nothing in the training data to discourage its use.
You yourself used "important" in the same paragraph.
"Load bearing" is a metaphor, while the other single words are more direct expressions. Unless the thing that Claude is referring to is a wall or other structure, which may truly bear load.
This is one of those issues which translators are long familiar with. There's no direct translation for "schwerpunkt" that isn't slightly longer.
Claude does at least use the British English version of the phrase to me - not sure whether its picking up a language setting or reacting to my spelling etc. The American version does sound odd over hear.
This is the worst one for me. I can maybe think of what it means, but I never heard it before, and could easily be imagining a meaning.
Some of the other Claude-isms (quickly googling, especially 'gate' and 'canonical') I feel the issue is they sound right, but aren't specific enough to why we are doing something.
Personally my least favorite is the overuse of "quietly" (e.g. "No tricks. No marketing gimmicks. Just one company quietly outperforming the others"), and the one that makes the least sense to me is "that's the wedge."
I'm curious how these become so ingrained. Then the uncomfortable part is humans start repeating it more (a colleague said "belt-and-suspenders" during brainstorming the other day).
The script replaces common Claude idioms with other terms. The next step would be randomly choosing from a list for each replacement to give variety.
Got me thinking, is there a way to intentionally train randomness into LLMs, so the probability of “load bearing” is spread across lots of synonyms (critical, important, etc) to give some more variety? Obviously one would come up with a better set of interchangeable phrasing’s, but it seems with 20-50 équiprobable ways of saying stuff, it would sound a lot more natural.
I'm surprised there's no LoRa layer or auto RL or adversarial step to reduce the stock phrases as they pop up. Is it really so hard to push these out? Or is it just whack-a-mole no matter what you do?
Even great words, phrases, and styles, seen too often, grate.
I personally love a lot of the Claude (or LLM) lingo. Load-bearing, gate, canonical, blast radius, and friends do a lot of tight, effective heavy-lifting in my world. I even love the em-dashes (—) and the *bold the main points* memo style, both of which I have used successfully for decades.
It's seeing them in every analysis and post—the constant repetition becoming over-repetition—that makes them the Claude voice shouting "AI wrote this!" that seems to be causing LLM allergic reactions.
I wrote a thing about exactly this, but I'm resistant to blogging for undefined reasons so, maybe this will help someone...
# AI speech is an Infohazard
Apart from all its other possible boons and ills, one danger of AI is just that it is useful, so you use it. A lot.
In earlier days I would dive deeply into an author's work and start to think and write like them for a while. It was a heady feeling: slinging sonnets like Shakespeare—not at his level, but stylistically reminiscent—or tweaking turns like Twain.
Like all things, the effect lasts in relation to how long and how much you do it. The point is: our thinking is influenced by what we take in. Take more of a certain thing in, think more like that thing.
Now enter AI. My hand-crafted coding days are in their twilight months ("AI years"), and most of my software engineering is done through jaggedly capable agentic power tools. Instead of working directly with raw codestuff, I work with slop prose flecked with code sprinkles.
I read orders of magnitude more AI-speak—I call it "babble", or perhaps "Babel"—than human-written text. I can feel its genuinely honest points, clearly stated, slipping their banal tendrils into my thoughts and inner monologue.
Solutions? For me:
1. Be aware. "I notice that my thought stream is under assault."
2. Read stuff far from slop. Even a small dose of the good stuff can help inoculate. Recently I thought On the Calculation of Volume was something completely different.
3. Write stuff that is different. This post. Force the mind to synthesize thoughts in other ways.
4. debabel.py / debabel.js: a tool, and a pi extension, which filters common babble from visible LLM output. A lint for mind-killing prose.
It is not perfect, but it 80/20s nicely. I am willing to accept mildly awkward prose to avoid polluting my own internal distributions.
Details and example in the first comment. Tool available upon request.
The tool uses good old-fashioned regex/NLP—no LLMs—and rewrites the visible LLM Markdown without touching files on disk. I made it by analysing 10M+ words of past pi transcripts, doing frequency and pattern analysis, and iterating on rules to strip out the most egregious sloperisms. Now the circle is complete.
BEFORE
"Honestly, you're absolutely right—and that's a great question. Let me be clear: the fact that the parser works at all is actually quite remarkable. We utilize the tokenizer in order to handle the majority of edge cases, and due to the fact that it's recursive, it processes a large number of inputs. It's worth noting that this isn't really a bug—it's genuinely a deliberate design choice. In terms of raw throughput, the cache is significantly faster, and I think we should simply ship it.
That said, at the end of the day, not only is the design correct, but it is also fast. In other words, prior to refactoring, we should clearly delve into the nuanced, robust architecture. In conclusion, this is a testament to the seamless, intricate work the team has done."
AFTER
"That's a great question. That the parser works at all is remarkable. We use the tokenizer to handle most edge cases, and because it's recursive, it processes many inputs. This is a deliberate design choice. For raw throughput, the cache is faster, and we should ship it.
Not only is the design correct, but it is also fast. Before refactoring, we should delve into the nuanced, robust architecture. This is a testament to the seamless, intricate work the team has done."
("It consists in gumming together long strips of words which have already been set in order by someone else, and making the results presentable by sheer humbug" -- Orwell predicts the LLM)
and also https://www.jstor.org/stable/25515288 "The Myles na gCopaleen Catechism of Cliché" itself is rather hard to find online, but he's a very funny writer so it's worth the effort.
Lately, I feel like as GEN AI text becomes the majority, human-written text is starting to resemble it too.
I'm Korean, and there are sites and people who mainly curate the latest technologies. Even those people, probably tired of translating every time, have started summarizing things with AI. But recently, I've noticed that even when people don't use AI, their writing is starting to look like GEN AItext.
I think the reason might be that people often base their thoughts on documents they've read, or paste parts of content when writing their own texts, which leads to that style.
I'm not sure. Whether human writing is better or AI writing is better—personally, AI writing tends to flow in a very even, paragraph-by-paragraph structure, which makes it good for consuming information. I wouldn't want to read a novel written that way, but for getting information, AI writing is surprisingly convenient.
I recently started using caveman, and it’s been great. It doesn’t just cut down on overuse of specific terms; it cuts down on time spent digesting slop in general.
The token saving is oversold, from what I can tell so far. These days output tokens are just the tip of the iceberg.
If anything the real value is it saves my brain from going into power saving mode by lunchtime because I haven’t spent the day reading pages of output when a sentence or two would do.
It's good, because it's just post-processing before display. So it doesn't interfere with the process, which those phrases that seem so offensive to sensibilities of so many people, for whatever reason, might be a part of.
Nowadays, with the focus on agentic use and coding, it seems models have all been RLHF’d to death, it’s so incredibly hard to have them write in a different voice than their default. I put together a skill to review its writing and have it edit its own output (e.g. code comments), which does make a difference, but isn’t perfect.
What, if anything, do people do for writing? That feels like a neglected side of LLMs. They’ll make 100 Bash calls referencing ancient commands without batting an eye but heaven forbid they use something other than “load-bearing” while talking. For something trained on “all the human knowledge” it’s incredible how limited their default vocabulary seems to be.
I use a keyboard, personally.
Of course there will be models trained on much less code and technical writing, and they will create more natural sounding prose, but they will lack the deep intelligence of frontier models. Seems like a fair tradeoff.
[1] https://www.youtube.com/watch?v=aD93kfArOik
At work our documentation isn’t just getting littered with annoying jargon. It isn’t just all the hallucinations, either. (Since when has documentation ever been 100% trustworthy?) It’s that it’s so poorly written and structured that it’s becoming borderline incomprehensible.
My company is currently considering making, “Why should I bother to read something you didn’t bother to write?” an official policy because even the executives are starting to burn out on all the time they have to spend wading through slop.
gemini-2.5-pro-experimental was the GOAT, though. It was an emotional wreck, down in the dumps and feeling terrible for itself after failing to patch a file several times. Very amusing to read, all the while watching it make a mess of my codebase.
I don’t get it. If nobody likes this writing style, how can it be the result of human feedback? Something else is going on.
I think this is the same flaw as coding agents seeing in every problem the call for a “smoke test” or the use of some unnecessary design pattern. The truest part of AI is the A.
All the bots and other LLMs providing feedback, so in reality it’s reflecting the reality in a sense.
we liked it until we didn't.
operative? key? critical? decisive?
The honest conclusion is that none of those are as good as "load-bearing". And yet the concept being referred to is clearly extremely important and valuable to refer to. So maybe we should be learning from Claude rather than complaining.
"Her optimism was load-bearing,"
versus:
"Her optimism was enduring."
Exactly the same meaning and connotation. It stands to reason that the terms with the most semantic flexibility will have preference across all contexts. So in response to:
> maybe we should be learning from Claude rather than complaining.
I'd say let's not steer ourselves into regular language and keep some vivacity in our expressions.
I think you've been reading too much claude output! "Load bearing" is cromulent verbiage and can be used in many scenarios - so claude does. But variety is important too, and there are more specific alternatives that can be used in most situations. Any word becomes a bad choice if you've used it 10 times in the last chapter.
A more parsimonious explanation is that this term got more-or-less randomly boosted by the reinforcement learning loop because there was nothing in the training data to discourage its use.
"Load bearing" is a metaphor, while the other single words are more direct expressions. Unless the thing that Claude is referring to is a wall or other structure, which may truly bear load.
This is one of those issues which translators are long familiar with. There's no direct translation for "schwerpunkt" that isn't slightly longer.
Operative, key, and critical are all more correct to me in this context.
Some of the other Claude-isms (quickly googling, especially 'gate' and 'canonical') I feel the issue is they sound right, but aren't specific enough to why we are doing something.
I'm curious how these become so ingrained. Then the uncomfortable part is humans start repeating it more (a colleague said "belt-and-suspenders" during brainstorming the other day).
Got me thinking, is there a way to intentionally train randomness into LLMs, so the probability of “load bearing” is spread across lots of synonyms (critical, important, etc) to give some more variety? Obviously one would come up with a better set of interchangeable phrasing’s, but it seems with 20-50 équiprobable ways of saying stuff, it would sound a lot more natural.
Right now it feels like mode collapse.
I'm surprised there's no LoRa layer or auto RL or adversarial step to reduce the stock phrases as they pop up. Is it really so hard to push these out? Or is it just whack-a-mole no matter what you do?
I personally love a lot of the Claude (or LLM) lingo. Load-bearing, gate, canonical, blast radius, and friends do a lot of tight, effective heavy-lifting in my world. I even love the em-dashes (—) and the *bold the main points* memo style, both of which I have used successfully for decades.
It's seeing them in every analysis and post—the constant repetition becoming over-repetition—that makes them the Claude voice shouting "AI wrote this!" that seems to be causing LLM allergic reactions.
I don't know how programmers, who are so used to staring at the same handful of keywords every day for decades, have suddenly become so discerning.
Yes, Claude writes boring and predictable prose. It also writes boring and predictable code. That's good!
# AI speech is an Infohazard
Apart from all its other possible boons and ills, one danger of AI is just that it is useful, so you use it. A lot.
In earlier days I would dive deeply into an author's work and start to think and write like them for a while. It was a heady feeling: slinging sonnets like Shakespeare—not at his level, but stylistically reminiscent—or tweaking turns like Twain.
Like all things, the effect lasts in relation to how long and how much you do it. The point is: our thinking is influenced by what we take in. Take more of a certain thing in, think more like that thing.
Now enter AI. My hand-crafted coding days are in their twilight months ("AI years"), and most of my software engineering is done through jaggedly capable agentic power tools. Instead of working directly with raw codestuff, I work with slop prose flecked with code sprinkles.
I read orders of magnitude more AI-speak—I call it "babble", or perhaps "Babel"—than human-written text. I can feel its genuinely honest points, clearly stated, slipping their banal tendrils into my thoughts and inner monologue.
Solutions? For me:
1. Be aware. "I notice that my thought stream is under assault."
2. Read stuff far from slop. Even a small dose of the good stuff can help inoculate. Recently I thought On the Calculation of Volume was something completely different.
3. Write stuff that is different. This post. Force the mind to synthesize thoughts in other ways.
4. debabel.py / debabel.js: a tool, and a pi extension, which filters common babble from visible LLM output. A lint for mind-killing prose.
It is not perfect, but it 80/20s nicely. I am willing to accept mildly awkward prose to avoid polluting my own internal distributions.
Details and example in the first comment. Tool available upon request.
Information hazard: https://en.wikipedia.org/wiki/Information_hazard
Babel: https://en.wikipedia.org/wiki/Tower_of_Babel
On the Calculation of Volume: https://en.wikipedia.org/wiki/On_the_Calculation_of_Volume
The revenge of NLP
The tool uses good old-fashioned regex/NLP—no LLMs—and rewrites the visible LLM Markdown without touching files on disk. I made it by analysing 10M+ words of past pi transcripts, doing frequency and pattern analysis, and iterating on rules to strip out the most egregious sloperisms. Now the circle is complete.
BEFORE
"Honestly, you're absolutely right—and that's a great question. Let me be clear: the fact that the parser works at all is actually quite remarkable. We utilize the tokenizer in order to handle the majority of edge cases, and due to the fact that it's recursive, it processes a large number of inputs. It's worth noting that this isn't really a bug—it's genuinely a deliberate design choice. In terms of raw throughput, the cache is significantly faster, and I think we should simply ship it.
That said, at the end of the day, not only is the design correct, but it is also fast. In other words, prior to refactoring, we should clearly delve into the nuanced, robust architecture. In conclusion, this is a testament to the seamless, intricate work the team has done."
AFTER
"That's a great question. That the parser works at all is remarkable. We use the tokenizer to handle most edge cases, and because it's recursive, it processes many inputs. This is a deliberate design choice. For raw throughput, the cache is faster, and we should ship it.
Not only is the design correct, but it is also fast. Before refactoring, we should delve into the nuanced, robust architecture. This is a testament to the seamless, intricate work the team has done."
("It consists in gumming together long strips of words which have already been set in order by someone else, and making the results presentable by sheer humbug" -- Orwell predicts the LLM)
and also https://www.jstor.org/stable/25515288 "The Myles na gCopaleen Catechism of Cliché" itself is rather hard to find online, but he's a very funny writer so it's worth the effort.
I was hoping for a reference to the Babel Fish, whispering its translations in your ear.
- smoking gun - blast radius - landed - spine - earned its keep - grammar - spike - cutover - bake - sprint, epic, story points (all Agile vocabulary) - paper-cuts - amazing, incredible, perfect
Omg, that hit hard. We really need more of this.
Gotta be a way to draw from their progress.
I'm Korean, and there are sites and people who mainly curate the latest technologies. Even those people, probably tired of translating every time, have started summarizing things with AI. But recently, I've noticed that even when people don't use AI, their writing is starting to look like GEN AItext.
I think the reason might be that people often base their thoughts on documents they've read, or paste parts of content when writing their own texts, which leads to that style.
I'm not sure. Whether human writing is better or AI writing is better—personally, AI writing tends to flow in a very even, paragraph-by-paragraph structure, which makes it good for consuming information. I wouldn't want to read a novel written that way, but for getting information, AI writing is surprisingly convenient.
https://github.com/JuliusBrussee/caveman
If anything the real value is it saves my brain from going into power saving mode by lunchtime because I haven’t spent the day reading pages of output when a sentence or two would do.