At least "wedged" is a thing i would say on a spinning out of control test that is stuck. "latch" though... not so much. I really wonder if this is the EU AI Act interfering with everything Claude does these days. As a non-EU citizen, I want a version without the rewriting of with watermarking in text.
I understand it's not active yet, and when it will be, it should only nudge the chances between choices that are anyway likely and are already randomized today via temperature.
Watermarking is not the reason Claude talks like that.
Oh, it is active, and a pre-release version went out with v5 models.
I understand there are other reasons (overdone RLHF, for example) why claude is wedged into this weird way of writing that takes the joy out of conversing with it, but this is one as well.
The meaning of the words it uses can be oh so close, but the popularity of the words are not, and not in that context -- but the use of these other words changes the context ever so slightly, and then it uses other words where those words would work better.
I had something in code that related to people over time periods, and once it switches to a vacuous word choice, i found it starting talking using all ERP terms. I had to google the whole sentence to understand that, individual words were fine they just didn't make any sense to me.
Im surprised by "provenance." I see it all the time. Except I've brought this up before and have never had it corroborated. I'm starting to think it's just my Claude.
I have a PhD and can confirm. Oftentimes, the stuff which comes out of Claude is just impenetrable because it invents jargon on the fly, and uses verbs in the most atrocious ways.
"The fibred side folded its capstone into the existing name, so the kinds are asymmetric."
What on earth does it mean to fold a capstone into a name‽
Is that an actual Claude output or hyperbole? It feels like I'm trying to parse an equation in a new math class which makes me want to take a stab at it regardless.
So there's a "fibred side".. the most likely candidate seems to be "fibred categories" which I hadn't heard of before, and it's talking about one side of some mapping between two sets such that if f is the primary function and f(x)=y then there exists an inverse function g(y)=x? Was it something that converted some data bidirectionally with a different algorithm on both sides?
The capstone of the inverse function would be the most important thing about it maybe?
My best guess is "In the process of working on the inverse function, the existing name (of the inverse function itself maybe?) was made to reflect the operation of the inverse function, so now the name does not follow the same naming convention as the name of the primary function (which does not contain its 'capstone')."
Its original wording is certainly dense and harder to follow for us, but it's fascinating how the model finds this the best fit for what it's trying to express IMO. Like it arrives at its own ways of overloading words/concepts, and things we would refer to in different ways in different contexts all get compressed to the same more-useful/complete idea.
Codex has never said anything nearly so alien as the Claude examples I've seen floating around, interestingly. I wonder if it just has a better training on choosing its words to present to the user or if it inherently arrived at a somewhat different mapping that favors 'plain language' more.
I am not really sure what Claude meant, but you are not too far off, from what I understand.
I have several similar folders with variants of a construction, but taking differently structured input. They are named “plain”, “fibred” and “indexed”. So the fibred variant is clear enough.
The Claude speak I struggle with is “the capstone” and what name it could be talking about. And what folding means here. I think it just means:
“I changed an important result of the construction in the fibred variant, but kept the name. so the fibred variant is now different from the others.”
My intuition is that Claude is trained to communicate to itself while coding. You see this in how bizarrely granular it is when explanation prior work, you also see this in the comments it leaves behinds.
don't forget LLMs are great at translating between languages, and within the same language. depending on the problem it works on, it will often reach for terminology that tend to be more common or familiar within that problem set. which appears inscrutable, but there's many different ways to skin a cat. just remind it to translate it back to the terminology and subject matter you're already an expert in.
1. Adds a small prompt to each turn with the agent[1].
2. Is like a band-aid on a bullet wound, properly solving it would mean retraining the model and they probably are already working on it.
My experiments with output styles have not been sufficient to keep the model in check. It still spits out incomprehensible gibberish and load-bearing-isms.
Does anyone have an output style nailed down that actually works? If so, please share!
Sometimes I can't even tell if what it's saying actually makes any sense to someone who understands all the terms its using, or if it's just throwing together words in a way that only make sense to its own model of language.
It's interesting that there must be a decision behind that, even if it's just appealing to the RLHF judges for some reason. Maybe there's an intention that if you cannot decipher what the chatbot is saying to you, you will have to ask and burn even more tokens.
Naively I would often expect it would talk to me about various niche topics like to a layman, which does occur about some topics an actual normal person would ask.
I used to assume it was the fault of average human annotators. That the people who are paid peanuts to rank chat outputs preferred the pretentious sounding ones. It wasn’t until quite a bit into the LLM boom that companies started to pay for domain experts. Overt watermarking is another possibility.
I’ve noticed that Sol is pretty good most of the time, but with long contexts it’ll start to devolve into Claudish.
I was pleasantly surprised when I attempted to scroll down and realized everything the author wanted to present fit on-screen. It's almost ironic that this site is able to make such an obvious, compelling presentation without being overly verbose or complicated (something which LLMs have a hard time doing). I wouldn't read TOO deeply into what is being presented, but the author has done a good job to not inject their own bias into the presentation which works well.
I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs, or at least certain language quirks that come from interacting with LLMs. Something I've noticed in my own writing is that I now present lists of examples in a consistent way: "... such as <example 1>, <example 2>, etc., ...". I started to notice I was using this pattern quite a bit somewhat recently, but I took a quick look at some of my social media posts and realized it's been occurring for a while. I had realized that I grown accustomed to this kind of language because, especially early on, LLMs would focus too much on the specific examples I'd provide when, really, I was just trying to give them a sense of what I was looking for. I just picked up that providing two examples then adding the "etc." worked to get the LLM to not focus so much on the specific examples and to understand that they need to consider more than what I explicitly presented. Of course, now I write like that in my social media comments, in Slack with my colleagues, etc. :>
I'd be interested to see if anyone can identify trends like this, since I think the human-language component of the adoption of LLMs is probably being somewhat neglected despite probably being surely dramatically affected.
Congratulations! I was going to comment on the scroll field in particular when I saw this. I didn't even realize you had to hand-craft the component, but it's such a nice UI idea in general, the way the scrolling works and how the content above changes.
Frankly, that's the one part I would have done differently. Mixing font sizes rarely looks good, but that's just my personal opinion. If you want to visually indicate intensity, I'd have used color, or rather saturation. Not the text itself, maybe a small pill next to it. But I don't think it's necessary, the order already conveys some sort of ranking, I don't think cardinal information adds much here.
You did a fantastic job; very to-the-point with effective visualisations. Kudos!
One very minor note: if your scroll device reports pixel-perfect deltas rather than discrete scroll-wheel ticks (e.g. logitech mx master, laptop touchpad, etc), the behaviour in the word search box is a bit weird. Arrow keys work fine though.
I did test it with my mac touchpad without issue. Maybe the problem is that I wanted to make discrete scroll-wheel ticks work. I just tried a fix by normalising deltaMode to pixels.
If you could just paste this in your console, I would have a better idea:
FWIW I got a bunch of "(-)10.00 native" in the console, "PAGE" never showed up no matter what I tried. Firefox, also a Logitech mouse, no trackpad though.
> Some believe what separates men from animals is our ability to reason. Others say it’s language or romantic love, or opposable thumbs. Living here in this lost world, I’ve come to believe it is more than our biology. What truly makes us human is our ability to hyperscroll!
Hyperscroll - special mode turned on with a button next to scroll that turns off scroll friction and let's you scroll into infinity (wheel keeps turning after you lift your finger).
Yeah, it takes some getting used to (the wheel is VERY sensitive to even the smallest finger movements), but when you get used to it, it's clearly superior to a regular scrollwheel. Much less finger fatique.
“non-trivial” is one of those words that seems like it should be a Claude-ism but is actually just the single best word for a concept that comes up extremely often in research and engineering discussions.
In the enterprise, we are already adopting them. “Seam”, I term I’d never heard used before, is now not only our defacto way to describe the boundary between systems/workflows/components, it is encoded into our organizational roles and responsibilities descriptions.
I am the proud owner of several seams, and am considering giving them person-names. The empty space between my inventory APIs and their clients might be “Karen”.
I can’t say that Claude invented this; the same type of terminology cycling happens every few quarters based on what leadership is reading/being told by Gartner.
I believe the term "seam" came from Michael Feathers' "Working Effectively with Legacy Code": “a seam is a place where you can alter behavior in your program without editing in that place”.
I've associated it with the concept of "carving nature at the seams", except that I just realized that in the standard phrasing it's actually "at the joints", not "seams".
I read that book right at the beginning of my career over 20 years ago, when I was the new guy who had actually studied software engineering among all my entirely self-taught older colleagues. Lots of big messes to clean up in that codebase!
"seam" dates back to refactoring & software engineering literature that pre-date LLM use by 20 years -- see Working Effectively With Legacy Code for one example (https://www.oreilly.com/library/view/working-effectively-wit...). Its a great book, still use techniques from it everyday.
"load-bearing" I have never heard used for programming before Opus, and its incredibly annoying and over-used.
Seam is used liberally throughout because it captures the idea well (i.e. a place where you can cleave ball of mud code apart to begin refactoring efforts in an isolated way) and that takes many different forms throughout the book whether that be via methods and classes, source files as a single unit, linker seams, on and on.
"seam" is a cop-out term for when someone (usually AI) lacks the imagination to come up with something more descriptive. It's like saying "middle thing". Just name it by what it does, and if you can't, remove it because it's not doing anything.
"load-bearing" on the other hand is just a weird way to say "required" or "prerequisite" without drawing any attention to the fact that one cannot articulate what something is a prerequisite for, probably because that fact has since been lost from context.
Saying "the language of LLMs", while technically correct, is not exactly precise. It should really be the language of "AI agents", or "chatbots". OG LLMs would just mimic the style of its context based off of (pre)training from text on the Internet.
The kind of quirks you see came from crowd-sourced human-in-the-loop fine-tuning, with not very good work conditions or level of qualification (so resulting in "what non-writers thought good writing looked like", before people had developed the flair to detect these patterns) as well as feedback loops during agentic reinforcement learning and RLVR.
100%. Literally just started writing a sentence in my previous comment and realized I was using a “It’s not X, it’s Y” (not verbatim) rhetorical structure and cringed internally.
The idiom "added color to" as an expression of adding detail or nuance to something goes back much further than the AI era. I've certainly used it at various times in my career spanning 4 decades now.
Yes, it is fun to play with different turns of phrase, even in a stale work environment. I do it to combat the banality, but not so much to sound pretentious hopefully ;)
The problem is once a fun phrase is, uh, coloured by "Claude" or some other LLM, it just turns to AI slop.
"Load-bearing" is fine to me. I've heard it regularly pre-Claude. It has a specific meaning that conveys in one stock phrase.
"Adds color to" says nothing that "Adds detail to" doesn't already say, and the latter is more straightforward English that is quickly readable and understandable by even relatively weak English speakers.
I get wanting to use big words or metaphorical phrasing; I'm guilty of saying "ameliorate" too much. I'm ok if it adds meaning. But when it's just a more colorful word (note that this use of "color" doesn't match the one I originally brought up, which is also a huge problem) and doesn't add any utility or detail, then it's best to avoid it, imo
> I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs, or at least certain language quirks that come from interacting with LLMs.
Or we'll do the opposite and even deliberately use some words that an LLM would never use to identify whose content is worth reading?
> I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs, or at least certain language quirks that come from interacting with LLMs. Something I've noticed in my own writing is that I now present lists of examples in a consistent way: "... such as <example 1>, <example 2>, etc., ...".
Why would you consider using lists an "LLM" language thing? I really don't understand this concept. Like even this project, the top words I can see are:
How are they LLM words? I used them all the time before ChatGPT was a thing.
Like if your claim is you use "etc." more because you talk to an LLM, I think you're just discovering human language, no? Like it's not a unique LLM thing that magically dropped and is now going to make people do it. It's a legit thing that most human writers have used for a long time -> which is why LLMs copy it. It's not backward.
A better claim than "people will start writing like LLMs" could be that people will become more literate in expressing themselves. But that literacy is not alien or AI to us, it is basic human language only used properly.
the issue with some of these quirks is that not that they are not part of the English language but rather the llms usage is unnatural. Also when you have that redundancy proliferated to millions of users it gets tiresome fast, and becomes a tell-tale sign for AI generation. It’s like an AI signature on some written content. Because many people do not want to consume AI generated content this becomes a sign of what to avoid.
Granted picking apart single words outside of context doesn’t give the full picture. (well except “load bearing” which is turning into a meme at this point)
Yes I understand that but I refuse to believe this idea that it influences how people talk. As you said, the typical AI-style of writing is very easily recognised and is not really great for everyday communications. Also it makes no sense you will start writing like an AI because you spend time communicating with it. There is also no benefit in typing like an AI to an AI. So i think its a good cover up for "i used ai to write this comment" but claim "i write like AI now lol because i talk to AI so my ai comment is actually human comment" which seems to happen a lot here lately as pushback against ai generated text becomes common
> I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs, or at least certain language quirks that come from interacting with LLMs.
What's somewhat interesting to me is that "load-bearing" was already a common thing to say in certain communities, like lesswrong.com. Whether it breaks out from that subculture with Claude as the vector, or disappears from there because nobody wants to sound like Claude, we'll have to see. How many children are called "Elvis" these days?
That confirms the recent spike of Claude calling everything I was recently working on a 'spike'. I still don't know what that term is supposed to represent (apparently.)
In some software development methodologies, "spike" is a task whose goal is figuring something out instead of delivering shippable code. https://agiledictionary.com/209/spike/
I was half-joking, of course I could've just asked Claude, but the linked site shows there has been actual recent spikes in the use of the word 'spike'. The term does match what I was recently doing, but hacking around legacy ERP software, blackboxes and other enterprise abominations isn't that out of the ordinary for me.
Thanks to the infinite well of human creativity I am able to read "load-bearing" both as the intended affectation (I won't call it meaning) as as well "being full of shit".
Author here, thank you so much! I really tried to make it nice to use, beyond the (quite original) modelling.
A prototype I did tried to detect some grammatical constructions, eg "it's not ..., it's ...", but I am not sure how to systematize that.
Also just a disclaimer: I am NOT tracking Claude tics, I am merely finding that a particular cluster of vocabulary increases. Tracking Claude requires labelled data IMO. I tried using model release dates in a structural model to constraint the clusters but the result was not compelling, so I ended up simplifying the model a lot!
I love this. I think you should be clearer about that - it's obvious to people who have experience doing this kind of data analysis, but a lot of people I've shown this were confused, as in "huh, load-bearing and genuinely makes sense but how'd he pick all these other words!?"
I think it should be made clear that the sorting is sort of Texas Sharpshooter-ish - the ones on top are on top because they sort that way.
The fact that load-bearing ends up on top is the proof that this works because we all know a priori that load-bearing is a Claudism.
I find Claude language often hard to process and having to wade through these words can be draining. Embarrassingly, I’ve recently caught myself using them in conversations! Do all models have the their own jargon?
I swear claude took a detour recently, its written output has been nearly incomprehensible to me. At first i thought i was getting AI-brained and just lost critical thinking but as i dug into response after response its was just the most obtuse language to explain what was going on. Really mentally taxing to wade through it all day.
It’s bad enough that I’ve told Claude never to write cleverly and never to use metaphors. Opus 5 mostly ignores those kinds of instructions, annoyingly enough.
I've noticed ChatGPT these days often starts a response with "Yes." when it doesn't actually make sense to do so, and sometimes says "and" where "but" seems more appropriate (or vice-versa).
Depends what you mean by "upstream". It would probably be inefficient to force the model to use more human-readable wording in its internal thinking traces, while translating a block of text at the end is a pretty trivial task for an LLM.
Nice work. I would suggest rewriting your README with Gemini, the most human of them, so it doesn't read exactly like the thing you are pointing at: https://github.com/louisabraham/load-bearing
Agree with rewriting with Gemini, but I'd characterize its output as very "neutral" and "encyclopedic", rather than "human" (contra Claude which is as of recent models "trying-too-hard-to-be-human"-sounding).
Very cool! I'm trying to understand the graph, so the bottom-most section seems to be the cluster identifying Claude written PRs. What are the other 7, any reason there are 8 in total?
I've been scraping instagram posts recently to identify AI misinformation accounts that all repost each other's carousels and get hundreds of thousands of likes in engagement. Thinking of ways to present it and your dashboard looks very helpful. Did you experiment with any other types of visualizations before deciding on a stacked area chart for the clusters?
The design is really impressive. Beautiful and dense, yet very understandable.
How did you come up with this particular design? It's really nice and everything fits, the colors, the layout, all of it.
Were you inspired by anything in particular? I feel like this reminds me of something, but I don't remember what exactly.
I agree with you, I really like OP's design choices, even more so amidst the flood of Claude-Generic these days. Although Claude just can't resist those little muted-toned ui-monospace small-caps...
It reminds me of Neue Grafik/Swiss Style, and Vignelli:
I asked Claude for a few original designs, then iterated a lot for the UI. It called this design "Rasterfeld", which is a word that means "field of grids", a term used in swiss design https://docs.mew.design/blog/swiss-design-style/ as @alwa said.
The 3 primary colors refer more to pop art.
I'm really eager for a nice book about "design" that would document all those styles, if someone has a recommendation.
I added the search bar, increased the data to 1000 PR a day (more than 50M words total) and added a feature to explore the other clusters as well. I hope the page is not getting cluttered.
I really love this. It’s comprehensive, it consolidates the data to the point where the argument effectively ‘makes itself’, and the way it’s presented respects the reader’s time. It also makes for an interesting challenge (for me at least) to try to characterise the subject matter of a language problem so narrowly.
No ream of slides. No narrative. Just a lovely big painful conclusion.
I had an idea for an experiment. Take a decent text, any one, and ask AI to rate it. Then patch that text by replacing words for ones that AI likes (honestly, load bearing etc) whenever possible and ask AI to rate it again to see how the evaluation changes.
@labo333 do u think soon we may need a dictionary? ive been playing with something u may like, but my approach has been to ask for definitions in-session so hard to do from outside.
I wonder to what extent this is the result of suboptimal RLHF versus the inherent intelligence of the model making its language more intricate and difficult for humans to easily parse? On the one hand, it's a common trope that highly educated people can talk in a way that's confusing and annoying to regular people who don't know all the jargon. But on the other hand, it's a mark of a skilled communicator to be able to efficiently distill complex information to its bare essentials in an easily-digestible way. Of course, that also seems to imply that these models are working at a higher level and need to talk down to us to an extent. Or maybe "Claudish" is just akin to stuff like "caveman", raw chain of thought, neuralese, etc., which are likewise much more dense/efficient but harder to interpret?
I don’t think they’re “talking down”. If anything - it’s way more difficult to distill something into a genuinely easy to digest format. I personally think that they aren’t immediately capable of this, and so we get word salad instead. Extra prompting required to strip extraneous prose out.
Maybe I am dumb and it IS talking down to me, but there have been many occasions where I’m reading AI generated docs / plans and it makes absolutely no sense, but looks really in depth at a glance.
It doesn't seem like word salad as such. There's normally a coherent point expressed, it's just obscured by circuitous sentence structures, unusual word choices, "verbing weirding nouns", metaphors, etc. Could be a result of training that rewards novel/surprising language, but it also feels like it could be an artifact of models imperfectly compressing high-level multidimensional reasoning into language that's easy for them to process but cognitively taxing for humans.
I've been thinking more about how 99.9% of us don't have the experience of someone significantly more intelligent, yet also subservient working under us, which is why I keep going crazy second guessing whether Claude is spouting RLHF'd bullshit that sort of resembles English, or is genuinely (pun not intended) just better at "intuiting" things I'm working on, leading to its language.
A notable exception would be people like CEOs and managers higher up in big tech, who might be used to skilled engineers and domain experts reporting to them in unfamiliar lingo. Maybe that's why we don't hear as much on the everyday annoyances of Claude's language from that camp?
I think it's model collapse - excessive feedback and excessive RL.
What RL does is narrow the variety generated by the model by steering the output towards the goal being rewarded. It's a bit like putting blinkers on a horse.
Of course RL is a very crude tool - it affects the entire model, even if you are just trying to make it better at some specific task(s), or trying to imbue a certain kind of personality (OpenAI's recent goblin problem).
The social graph proximity of Rationalists to Anthropic will be lost on no one who reads Astralcodexten. So guess which website has served as the thickest reservoir of 'Claude-isms'.
Imo their language is not precise enough for their intelligence to be the reason when it's difficult to understand. Maybe I'm prompting wrong, but when I don't understand, it's almost always because they just mash together words from context that don't form sentences with a clear meaning.
My unprovable pet theory is that, especially for writing about technical topics, the RL process has an open-ended way to weight things for quality: textbooks and first-party docs preferred to old stackoverflow answers and obscure blog/forum posts, and so on. The open-endedness of that quality gradient results in something in the RL process (maybe HF, maybe not) massively over-weighting some very small corpus of “quality = near infinite” content. The distribution of quality scores that inform the degree to which RL affects output has some extremely influential outliers, in other words.
Whatever that small corpus is, it contains some very specific grammatical tics, and that’s how we get Claudish.
Anyone who thinks a company/project as big as Anthropic/Claude wouldn’t make such a big mistake should take a look at how Azure cross-account federated login used to work.
> I wonder to what extent this is the result of suboptimal RLHF versus the inherent intelligence of the model making its language more intricate and difficult for humans to easily parse?
Its not the latter; its just excessively verbose wirh awkward word choices, the same as many poor writers. (And, like many such writers, the particular bad choices fall into recognizable, regularly recurring patterns.)
It's easy to think "it's not talking down, because I don't understand it, and I'm intelligent". But how is less intelligent being supposed to fully understand a more intelligent one, honestly speaking? All I know is that Claude understands Claude perfectly. I have the common session pause/resume setup that sometimes produces completely incomprehensible markdown files, but a new Claude session picks them up perfectly, down to the smallest details. What if what we consider excessive circular gibberish is actually highly precise set of instructions needed to minimize error cases for that unreliable human?
If Claude understands Claude, Claude understands human, and human doesn't understand Claude, that doesn't argue well for "Claude is a caveman".
I treat the Claude output that is hard to comprehend as an encoding/encryption. Claude knows how to decipher it, humans don't. I see this frequently in design docs from inexperienced engineers who used LLMs - they will contain terms (often two words hyphenated) that aren't obvious and should be defined, or simply replaced with simple language. If you prompt claude it is able to decipher and explain / replace this gibberish.
> But how is less intelligent being supposed to fully understand a more intelligent one, honestly speaking?
It is really simple. It is on the supposedly more intelligent person to be able to phrase things in simple way. Writing something incomprehensible and convoluted is easier then writing something simple to understand. Even for people.
> a new Claude session picks them up perfectly, down to the smallest details.
I genuinely doubt so.
> What if what we consider excessive circular gibberish is actually highly precise set of instructions needed to minimize error cases for that unreliable human?
It is not highly precise set of instructions and it is not following them in highly precise way.
> the inherent intelligence of the model making its language more intricate and difficult for humans to easily parse?
Humans who are actually well above average intelligence don't write like that simply to signal intelligence (although sometimes they're constrained by style expectations of their communication channels). So I hesitate to accept it as a sign of increasing model "intelligence", either.
Why are people getting so hung up on the "load-bearing assumption" turn of phrase that Claude uses? I get that it becomes cliche, but it is also a rather semantically dense way to communicate an idea that a lot of people run into.
It might be, when used sparingly, but when it's the first sentence in a wall of text as Claude goes way off the deep end on a two page description it becomes the easiest readily available tell that you're about to be frustrated.
Humans are very good at pattern recognition - Claude is _incredibly_ repetitive in the way it starts to struggle to communicate. I think there's also a ton of overlap in the Jargon instead of Usefulness that developers see in annoying middle management/salespeople. Circle back, synergy blah blah.
I don't think the individual turns of phrase are inherently problematic - but the process is triggering.
ItMs because Claude sprinkles these words as flavoring without aiding understanding. It feels like Claude thinks of metaphors that don’t actually mean anything (or maybe only makes sense to itself).
Generally the claude jargon is valid jargon I’ve seen real people use, it just uses jargon so much more often than a human would. And because it has particular jargon it likes, and claude is widely used, you see so much of that particular jargon that you get sick of it.
It’s like having one coworker with a very particular writing style which is mildly annoying, but then it suddenly feels like half the internet was written by that one person and it becomes a lot more annoying.
That's the problem for me: "semantically dense". Sometimes a single paragraph is filled with several of this expression, one after the other, and there's a real mental effort to get what Claude is saying.
A lot of these “Claudeisms” are simply jargon I’ve seen or heard firsthand myself while working at tech companies. I don’t think it’s limited to Claude either; I’ve seen Codex use load-bearing and many of these phrases as well.
I think using agents is just like speedrunning the whole experience of working with technical coworkers. Whereas you might have had a few coworkers at your company who used some of these phrases regularly, you now have a “coworker” who uses all of them regularly at a much faster pace.
Everyone has their own distinct mix of jargon and uses some words more or less than others, but now it's as if you have 1000 coworkers who are all the same person, and talk the exact same way, and it just so happens that their writing is grating to read. It also doesn't stop when you log off for the day. The same guy follows you home in ad copy, message board comments, spam, and in some cases even the correspondence of your loved ones.
The jargon isnt even the worst part. The phrasing is just downright incomprehensible sometimes. And its not even "academic". It's just extremely awkward, bad writing. And it is much worse than say Opus 4.6-8.
Actually I have found the copy that Claude Design spits out is way better than using the same model directly. I have no idea why. It has its flaws but it sounds like it's written by a human who uses derivative language. But usually the models just soudn incoherent.
> A lot of these “Claudeisms” are simply jargon I’ve seen or heard firsthand myself while working at tech companies.
---
The difference in distribution between how often LLMs use them and how often people use them in PR descriptions is almost the entire content of the link.
Claude is all we have at $work, if you don’t count MS Copilot (and you shouldn’t count MS Copilot).
Imagine being “incentivized” to aggressively use a tool for your job, and that tool produces thousands of lines of text in Olde English which you need. You’d be griping too, methinks.
I think you are referring to GitHub Copilot, which is fine (even though their context windows are artificially truncated and their dev tooling leaves a lot to be desired). I was making fun of the Copilot that’s shipped with MS stack. “Hey summarize my Teams meetings and try not to fuck that up too badly” - that one.
Anthropic has done their utmost to put Claude front and centre for professionals in multiple industries. This includes extremely tedious marketing campaigns. I'm sure they can handle a few negative comments from people they've snared, by hook or by crook.
Things like seam, fold, and load-bearing are useful concepts, they are everywhere, and they are more descriptive and more concise than alternatives. Over-usage can definitely be irritating (e.g. these should NOT appear in documentation) but they are almost unavoidable for humans engaged in code review or colab on complex stuff.
I don't want to use more words or letters than "seam" to actually pinpoint boundary conditions and the mechanical details of joinery when the context is understood by all. Too much effort for people! Easy for robots though.. so why are they abbreviating, and why would we want to allow it? A phrase like that permits a human who wants to educate a human to do so quickly with minimal time/effort. But it allows a robot a chance to not mention a filename, function-name, or to not reinforce/clarify it's own understanding or to state specific intentions.
It's bad for human-to-human comms if we just accept "ok, all technical terms are slop now, we have rephrase everything". Now YOU must cite details and sources, and the robot doesn't? Fuck that noise. Seam and fold are fine! Humans can be lazy! Robots should do the real work of explaining themselves without hiding behind tactical ambiguities.
I tend to agree. Claude's language doesn't bother me that much, because even the cliches are load-bearing to an extent (so to speak.) But some of the examples I've seen others post are well worth complaining about.
Everyone talks about Claude, but I'd like to bitch about Sol. "Unusually" is its absolute favorite word in the chat interface (less so in Codex), and I fucking hate it. Every single thing is unusually something. "Unusually good", "unusually efficient", "unusually inexpensive", "unusually attractive", "unusually difficult", "unusually nasty", "unusually cacheable", "unusually interesting", "unusually decisive", "unusually narrow", I'm surprised I haven't seen "unusually unusual". If I search chat history for "unusually" it brings up every single fucking chat I have from before I added it to an illegal words instruction. I can't believe I haven't seen anyone else complaining about this, it's as pervasive if not moreso than anything I've seen from Claude.
It's trained with excessive weighting on LessWrong , folks. Don't believe me, just take your favorite Claudeism that is otherwise obscure English and see how many hits you get within site:lesswrong.com .
302 comments
[ 0.22 ms ] story [ 61.0 ms ] threadThe word selection and way of writing has taken the joy out of using Claude.
I understand it's not active yet, and when it will be, it should only nudge the chances between choices that are anyway likely and are already randomized today via temperature.
Watermarking is not the reason Claude talks like that.
I understand there are other reasons (overdone RLHF, for example) why claude is wedged into this weird way of writing that takes the joy out of conversing with it, but this is one as well.
The meaning of the words it uses can be oh so close, but the popularity of the words are not, and not in that context -- but the use of these other words changes the context ever so slightly, and then it uses other words where those words would work better.
I had something in code that related to people over time periods, and once it switches to a vacuous word choice, i found it starting talking using all ERP terms. I had to google the whole sentence to understand that, individual words were fine they just didn't make any sense to me.
I feel like I need a PhD to understand its explanation of a code snippet.
"The fibred side folded its capstone into the existing name, so the kinds are asymmetric."
What on earth does it mean to fold a capstone into a name‽
So there's a "fibred side".. the most likely candidate seems to be "fibred categories" which I hadn't heard of before, and it's talking about one side of some mapping between two sets such that if f is the primary function and f(x)=y then there exists an inverse function g(y)=x? Was it something that converted some data bidirectionally with a different algorithm on both sides?
The capstone of the inverse function would be the most important thing about it maybe?
My best guess is "In the process of working on the inverse function, the existing name (of the inverse function itself maybe?) was made to reflect the operation of the inverse function, so now the name does not follow the same naming convention as the name of the primary function (which does not contain its 'capstone')."
Its original wording is certainly dense and harder to follow for us, but it's fascinating how the model finds this the best fit for what it's trying to express IMO. Like it arrives at its own ways of overloading words/concepts, and things we would refer to in different ways in different contexts all get compressed to the same more-useful/complete idea.
Codex has never said anything nearly so alien as the Claude examples I've seen floating around, interestingly. I wonder if it just has a better training on choosing its words to present to the user or if it inherently arrived at a somewhat different mapping that favors 'plain language' more.
As far as I can tell "the capstone" is what Claude usually calls my current goal if it thinks it is a satisfying result.
I have several similar folders with variants of a construction, but taking differently structured input. They are named “plain”, “fibred” and “indexed”. So the fibred variant is clear enough.
The Claude speak I struggle with is “the capstone” and what name it could be talking about. And what folding means here. I think it just means:
“I changed an important result of the construction in the fibred variant, but kept the name. so the fibred variant is now different from the others.”
My intuition is that Claude is trained to communicate to itself while coding. You see this in how bizarrely granular it is when explanation prior work, you also see this in the comments it leaves behinds.
https://www.bullshitgenerator.com/
https://code.claude.com/docs/en/output-styles
1. Adds a small prompt to each turn with the agent[1]. 2. Is like a band-aid on a bullet wound, properly solving it would mean retraining the model and they probably are already working on it.
[1] https://x.com/_can1357/status/2090360068529111530
Does anyone have an output style nailed down that actually works? If so, please share!
alwayshasbeen.jpg
Naively I would often expect it would talk to me about various niche topics like to a layman, which does occur about some topics an actual normal person would ask.
I’ve noticed that Sol is pretty good most of the time, but with long contexts it’ll start to devolve into Claudish.
I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs, or at least certain language quirks that come from interacting with LLMs. Something I've noticed in my own writing is that I now present lists of examples in a consistent way: "... such as <example 1>, <example 2>, etc., ...". I started to notice I was using this pattern quite a bit somewhat recently, but I took a quick look at some of my social media posts and realized it's been occurring for a while. I had realized that I grown accustomed to this kind of language because, especially early on, LLMs would focus too much on the specific examples I'd provide when, really, I was just trying to give them a sense of what I was looking for. I just picked up that providing two examples then adding the "etc." worked to get the LLM to not focus so much on the specific examples and to understand that they need to consider more than what I explicitly presented. Of course, now I write like that in my social media comments, in Slack with my colleagues, etc. :>
I'd be interested to see if anyone can identify trends like this, since I think the human-language component of the adoption of LLMs is probably being somewhat neglected despite probably being surely dramatically affected.
Basically when scrolling, the frame around words will have a stable top edge to avoid flickering, and it took me a few iterations to make it work.
I tested multiple designs, some of them inspired by word clouds and the size variation stuck afterwards.
One very minor note: if your scroll device reports pixel-perfect deltas rather than discrete scroll-wheel ticks (e.g. logitech mx master, laptop touchpad, etc), the behaviour in the word search box is a bit weird. Arrow keys work fine though.
I did test it with my mac touchpad without issue. Maybe the problem is that I wanted to make discrete scroll-wheel ticks work. I just tried a fix by normalising deltaMode to pixels.
If you could just paste this in your console, I would have a better idea:
> Arthur Conan Doyle, The Lost World
Hyperscroll - special mode turned on with a button next to scroll that turns off scroll friction and let's you scroll into infinity (wheel keeps turning after you lift your finger).
I was focused on the data initially scrolling through until I suddenly realised, wow, this is really nice!
I am the proud owner of several seams, and am considering giving them person-names. The empty space between my inventory APIs and their clients might be “Karen”.
I can’t say that Claude invented this; the same type of terminology cycling happens every few quarters based on what leadership is reading/being told by Gartner.
https://martinfowler.com/bliki/LegacySeam.html
Claude is using it a bit liberally, but not totally incorrectly.
"load-bearing" I have never heard used for programming before Opus, and its incredibly annoying and over-used.
http://www.hrwiki.org/wiki/Ye_Flask
Seam is used liberally throughout because it captures the idea well (i.e. a place where you can cleave ball of mud code apart to begin refactoring efforts in an isolated way) and that takes many different forms throughout the book whether that be via methods and classes, source files as a single unit, linker seams, on and on.
"load-bearing" on the other hand is just a weird way to say "required" or "prerequisite" without drawing any attention to the fact that one cannot articulate what something is a prerequisite for, probably because that fact has since been lost from context.
The kind of quirks you see came from crowd-sourced human-in-the-loop fine-tuning, with not very good work conditions or level of qualification (so resulting in "what non-writers thought good writing looked like", before people had developed the flair to detect these patterns) as well as feedback loops during agentic reinforcement learning and RLVR.
I'm already seeing it. A coworker said something like "<person> added the color to the ticket here" meaning that someone added details to a ticket.
I've started seeing Opus 5 talk about "hermetic testing" when it just means "unit testing", so I hope that doesn't catch on...
"Load bearing", not so much. Ick.
The problem is once a fun phrase is, uh, coloured by "Claude" or some other LLM, it just turns to AI slop.
"Adds color to" says nothing that "Adds detail to" doesn't already say, and the latter is more straightforward English that is quickly readable and understandable by even relatively weak English speakers.
I get wanting to use big words or metaphorical phrasing; I'm guilty of saying "ameliorate" too much. I'm ok if it adds meaning. But when it's just a more colorful word (note that this use of "color" doesn't match the one I originally brought up, which is also a huge problem) and doesn't add any utility or detail, then it's best to avoid it, imo
Or we'll do the opposite and even deliberately use some words that an LLM would never use to identify whose content is worth reading?
Why would you consider using lists an "LLM" language thing? I really don't understand this concept. Like even this project, the top words I can see are:
1. load-bearing 2. plainly 3. quietly 4. refusal 5. survived
How are they LLM words? I used them all the time before ChatGPT was a thing.
Like if your claim is you use "etc." more because you talk to an LLM, I think you're just discovering human language, no? Like it's not a unique LLM thing that magically dropped and is now going to make people do it. It's a legit thing that most human writers have used for a long time -> which is why LLMs copy it. It's not backward.
A better claim than "people will start writing like LLMs" could be that people will become more literate in expressing themselves. But that literacy is not alien or AI to us, it is basic human language only used properly.
Granted picking apart single words outside of context doesn’t give the full picture. (well except “load bearing” which is turning into a meme at this point)
What's somewhat interesting to me is that "load-bearing" was already a common thing to say in certain communities, like lesswrong.com. Whether it breaks out from that subculture with Claude as the vector, or disappears from there because nobody wants to sound like Claude, we'll have to see. How many children are called "Elvis" these days?
> "The name Elvis was not among the top 1,000 US baby names in 2010, the first year it had not made the list since 1954, the US government said." - https://www.bbc.co.uk/news/world-us-canada-13302517
Most recent famous Elvis on Wikipedia is Kosovan footballer Elvis Letaj, born 2003 - https://en.wikipedia.org/wiki/Elvis_(name)#People_with_the_n...
I think it was subtly dissing you.
I was half-joking, of course I could've just asked Claude, but the linked site shows there has been actual recent spikes in the use of the word 'spike'. The term does match what I was recently doing, but hacking around legacy ERP software, blackboxes and other enterprise abominations isn't that out of the ordinary for me.
Is it possible to expand this analysis beyond words to other Claude ticks? Contrastive framings, sentence length, caveating, for instance.
A prototype I did tried to detect some grammatical constructions, eg "it's not ..., it's ...", but I am not sure how to systematize that.
Also just a disclaimer: I am NOT tracking Claude tics, I am merely finding that a particular cluster of vocabulary increases. Tracking Claude requires labelled data IMO. I tried using model release dates in a structural model to constraint the clusters but the result was not compelling, so I ended up simplifying the model a lot!
I think it should be made clear that the sorting is sort of Texas Sharpshooter-ish - the ones on top are on top because they sort that way.
The fact that load-bearing ends up on top is the proof that this works because we all know a priori that load-bearing is a Claudism.
I've been scraping instagram posts recently to identify AI misinformation accounts that all repost each other's carousels and get hundreds of thousands of likes in engagement. Thinking of ways to present it and your dashboard looks very helpful. Did you experiment with any other types of visualizations before deciding on a stacked area chart for the clusters?
I did experiment and a stacked chart seemed the most clear, with the important cluster at the bottom.
The design is really impressive. Beautiful and dense, yet very understandable. How did you come up with this particular design? It's really nice and everything fits, the colors, the layout, all of it.
Were you inspired by anything in particular? I feel like this reminds me of something, but I don't remember what exactly.
It reminds me of Neue Grafik/Swiss Style, and Vignelli:
https://designreviewed.com/series/neue-grafik-new-graphic-de...
https://www.printmag.com/featured/swiss-style-principles-typ...
https://www.designculture.it/interview/massimo-vignelli.html
https://andrewclark.co.uk/all-media/the-vignelli-canon
I asked Claude for a few original designs, then iterated a lot for the UI. It called this design "Rasterfeld", which is a word that means "field of grids", a term used in swiss design https://docs.mew.design/blog/swiss-design-style/ as @alwa said.
The 3 primary colors refer more to pop art.
I'm really eager for a nice book about "design" that would document all those styles, if someone has a recommendation.
I added the search bar, increased the data to 1000 PR a day (more than 50M words total) and added a feature to explore the other clusters as well. I hope the page is not getting cluttered.
No ream of slides. No narrative. Just a lovely big painful conclusion.
https://www.themachinevernacular.net/
Maybe I am dumb and it IS talking down to me, but there have been many occasions where I’m reading AI generated docs / plans and it makes absolutely no sense, but looks really in depth at a glance.
A notable exception would be people like CEOs and managers higher up in big tech, who might be used to skilled engineers and domain experts reporting to them in unfamiliar lingo. Maybe that's why we don't hear as much on the everyday annoyances of Claude's language from that camp?
What RL does is narrow the variety generated by the model by steering the output towards the goal being rewarded. It's a bit like putting blinkers on a horse.
Of course RL is a very crude tool - it affects the entire model, even if you are just trying to make it better at some specific task(s), or trying to imbue a certain kind of personality (OpenAI's recent goblin problem).
Whatever that small corpus is, it contains some very specific grammatical tics, and that’s how we get Claudish.
Anyone who thinks a company/project as big as Anthropic/Claude wouldn’t make such a big mistake should take a look at how Azure cross-account federated login used to work.
Its not the latter; its just excessively verbose wirh awkward word choices, the same as many poor writers. (And, like many such writers, the particular bad choices fall into recognizable, regularly recurring patterns.)
If Claude understands Claude, Claude understands human, and human doesn't understand Claude, that doesn't argue well for "Claude is a caveman".
It is really simple. It is on the supposedly more intelligent person to be able to phrase things in simple way. Writing something incomprehensible and convoluted is easier then writing something simple to understand. Even for people.
> a new Claude session picks them up perfectly, down to the smallest details.
I genuinely doubt so.
> What if what we consider excessive circular gibberish is actually highly precise set of instructions needed to minimize error cases for that unreliable human?
It is not highly precise set of instructions and it is not following them in highly precise way.
Humans who are actually well above average intelligence don't write like that simply to signal intelligence (although sometimes they're constrained by style expectations of their communication channels). So I hesitate to accept it as a sign of increasing model "intelligence", either.
Humans are very good at pattern recognition - Claude is _incredibly_ repetitive in the way it starts to struggle to communicate. I think there's also a ton of overlap in the Jargon instead of Usefulness that developers see in annoying middle management/salespeople. Circle back, synergy blah blah.
I don't think the individual turns of phrase are inherently problematic - but the process is triggering.
It’s like having one coworker with a very particular writing style which is mildly annoying, but then it suddenly feels like half the internet was written by that one person and it becomes a lot more annoying.
I think using agents is just like speedrunning the whole experience of working with technical coworkers. Whereas you might have had a few coworkers at your company who used some of these phrases regularly, you now have a “coworker” who uses all of them regularly at a much faster pace.
Actually I have found the copy that Claude Design spits out is way better than using the same model directly. I have no idea why. It has its flaws but it sounds like it's written by a human who uses derivative language. But usually the models just soudn incoherent.
Imagine being “incentivized” to aggressively use a tool for your job, and that tool produces thousands of lines of text in Olde English which you need. You’d be griping too, methinks.
Having said that, I just subscribed to ChatGPT yesterday, as I've become impatient with Claude for a text-dense project I'm working on.
I don't want to use more words or letters than "seam" to actually pinpoint boundary conditions and the mechanical details of joinery when the context is understood by all. Too much effort for people! Easy for robots though.. so why are they abbreviating, and why would we want to allow it? A phrase like that permits a human who wants to educate a human to do so quickly with minimal time/effort. But it allows a robot a chance to not mention a filename, function-name, or to not reinforce/clarify it's own understanding or to state specific intentions.
It's bad for human-to-human comms if we just accept "ok, all technical terms are slop now, we have rephrase everything". Now YOU must cite details and sources, and the robot doesn't? Fuck that noise. Seam and fold are fine! Humans can be lazy! Robots should do the real work of explaining themselves without hiding behind tactical ambiguities.
https://internals.rust-lang.org/t/random-thought-of-the-day-...