Show HN: The load-bearing vocabulary of Claude(louisabraham.github.io) |
Show HN: The load-bearing vocabulary of Claude(louisabraham.github.io) |
https://arxiv.org/abs/2510.15061
(ICLR 2026)
I love what I can build now, but I sure as hell don't love the headaches this trend has been giving me.
> So the full honest arc on the case we set out to fix: the expiry rules and day note tripled the loose version of the story, the relay fix carried the device’s own guardrail through the pipeline, the fair replay then revealed the last mechanism — ticket-anchoring — which none of the shipped layers reach. Remaining options, in order of my confidence: making the resolved-ticket summaries in the AI’s context carry their day so the expiry rules have something to bite on (small, mechanical, targeted at the observed anchor); and the plan-B second-model check, which structurally catches this class no matter how the model reasons. About $25 of headroom remains. Which way?
Yikes.
(The worst part is that I understand it)
In some cases, I do not trust it to write code unchecked, and review everything it produces (which does not imply I catch all bugs, obviously).
In other cases, occasional failure is an option, and the speed you get by iterating fast is absolutely worth not even looking at the code.
So, do I trust this to write working code? Sometimes I do.
If I'm doing work for a client, it is not very common for me to simply vibe code something, because part of what clients expect of me is high quality (it's part of how we position ourselves), and I cannot simply assume the LLM will produce high quality (and indeed it does not without a lot of guidance). When I do vibecode in such cases, I make it clear that I did so and why (e.g. because it's a tool to be used in the project to help DX, not a core part of the product). Still, I haven't written 10 consecutive manual lines of code in almost a year.
If I'm working on something for myself, or on an internal product, then vibe coding is absolutely allowed and sometimes the norm. Here often we really do care about finding the right thing to build first.
You can clearly feel the tipping point where the LLM starts to crumble under the weight of the mess it has created, but that often doesn't matter when building an MVP for market validation or for small products that don't get particularly big. Plus, clearly the tipping point takes longer to reach with better and smarter models, and to me it is very clear that you need to learn how to iterate with LLMs right. The things I vibe-code now are much better than those I did before, even with similar models, because the tooling and approaches (the "real harness" and my "mental harness") are better. As with any tool: it takes practice to know how to use it, and if you're a good engineer and problem solver, you are miles ahead of the competition. Anyone can vibecode, but those with this kind of mind seem to be much more successful.
Nowadays I do produce a lot more than I used to, but I also have much more fun, perhaps only surpassed by when I learned how to code when I was a kid. A big chunk of this comes from my very privileged work position, where I get to call so many of the shots, and I'm aware of that.
I would really say the biggest downside to all of this, on a personal level, is exactly what I shared: what comes out of the LLM while discussing has become hard to grasp (especially on larger context windows), and it doesn't help that so many people now like to just throw me whatever ChatGPT/Claude wrote verbatim. I can't stand that, especially because most of the time they don't realize they're throwing me incomplete and poorly thought-out ideas.
Finally, naturally, like I said, this has to do with risk. I wouldn't trust Claude to give me legal advice, for example. I may check what it says and use it to brainstorm, but I wouldn't trust it with any meaningful informed decision like this.
Some of this is less to do with Claude vocabulary and more to do with the expectation that Claude justifies it's work. That expectation (probably) came from reinforcement learning.
Obviously, we adapt habits of entities next to us
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.
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 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.
like the stories behind when those words first appeared in the software engineering
like quiescence the most recent one i learned
I have seen LLM users go deep into AI psychosis because it was reaffirming all of their prompts and making them feel like a genius or validating their insecurities.
i must be the only one in the world that has no issues with how opus is talking. it is verbose & patronizing & secretly belittling at times and like it like that.
> One brief sentence. Second abrupt sentence.
e.g. from an openrouter email:
> One key reaches every major model. Switch without rewriting.
Was enough to make me delete the email without reading.
Would there be the same tendency of we looked simply at the number of words in a commit/pr?
In my case my commit messages are on average 10x shorter than Claude's.
You need a PhD to understand its explanation of a code snippet.
I am not sure whether it's a consequence of learning to reason from its traces or some RLHF that trips it into using weird terms to sound smarter to the humans who rate it.
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.
"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‽
alwayshasbeen.jpg
Does anyone have an output style nailed down that actually works? If so, please share!
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.
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.
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.
So, yes, it's amusing to see clear Claude-isms like "load-bearing", "outright", and "genuine" in a [very nice] bit of analysis like this. And there's a (maybe negative? or not?) argument to be made about the world being filled with more Claude-isms or LLM-isms in general.
But I think the data say a second thing which is just as interesting and an absolute positive for the typical source code base. Look at the clusters that shrank significantly. Most of what you'll see in there is just incomprehensible...not even English. Cluster 4 has, after "pullrequest", a bunch of seeming usernames in the top tier. Cluster 6 seems to have names of repositories or tags in the top tier. Cluster 9 has branch names in it.
Meanwhile, keep going through cluster 1 and you'll see words I don't consider Claude-isms that really, really grow in usage. Words like "died", "nothing", "worse", "ever" all have well over 10x growth. This tells me something else. That the average commit log was BARELY ENGLISH. And then the LLMs came along and made commit logs that were ACTUALLY ENGLISH.
I count this as a good thing. I don't know the cross-section of repos chosen for this analysis, and I get it...some repos are garbage/throwaway, some commits come from automated processes that generate uninteresting commit logs, etc. But I've been benefitting from my work team's actually explanatory commit logs when doing code/bug archeology for decades, when doing PR review for the last decade, and I've even seen LLMs benefit from it in the last year (granted, not as often). A large part of professional software development is communication, and while the most important communication is via the code/comments, the commit logs are not unimportant. So, if this is making the average GitHub PR better (arguably more professional) by including actual English descriptions of code changes in commit logs...well, that's a genuinely load-bearing concept for me. :)
The florid over exaggeration do certain words in bizarre ways is a reflection of their aggressive alignment towards too many goals, leading to weirdness in both behavior and language. The alignment functionally lobotomized opus-5 for any practical task.
Anthropic had a real gem in 4-6 and managed a near total market capture, which they have since squandered in the fastest burning of developer good will I’ve ever seen. It feels like exceeding the unity licensing implosion but without the single stupid decision.
I think it would be fairer to say that Anthropic held the mindshare in the Silicon Valley style tech scenes around the world and the companies built on that model, plus a substantial portion of other software engineering. Now it seems that's dwindling quite rapidly.
https://github.com/dickiedyce/mrso
The naming is from Mrs Overall, and her inability to keep from slopping tea out of of a cup.
I see people around here giving advice on how to reduce this vocabulary in the model output. I’m not certain this is a clever thing to do, as it appears to me, the emphasis on certain words is deliberate, and an important step towards overall quality as the words are being played back to the model in the next turn and thus supporting the model shape its own thoughts and stay on course.
Why is that exactly?
I also forbid replying with anything besides "yes" or "no" to yes/no answers.
It works, Opus 5 is still super-annoying but this makes it bearable to use.
https://internals.rust-lang.org/t/random-thought-of-the-day-...
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.
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.
If we accept that the LLM can "understand" at all, why would we reject the possibility of this understanding living in "latent space", before tokens are output?
There is no working other than communication. They are just text generators.
Have quick test. In a codebase or PR (assuming you’re a dev) with claude’s signature wordings unmodified, command claude (/code-review) with same “thinking” effort and model. You may find (as I have many times) that claude points out the comments it made in earlier sessions, as factually wrong or misleading. Repeat this in a fresh session a few times and on different code.
Since then Anthropic has done little to capitalize on the good will and a lot to squander it. sol, r4, kimi, even metas avacado has come a long way and in many, if not most, cases surpassed opus-5. Concurrently opus has declined in utility to the point of near uselessness, fable roll out didn’t seem to understand market dynamics, and their competitors are watching their consistent missteps closely. The fall has been breathtakingly fast - from March to July they imploded in a half dozen or more missteps, devolved their product quality, and failed to effectively respond to competitors. Their product focus seems to lack exactly that - focus.
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.”
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.
Thank you for the compliment! I did spend a lot of time designing a nice experience on both desktop and mobile. Even the scrollbar to select words was non trivial as I wanted the words to be of different size, yet avoid flickering when scrolling!
I was focused on the data initially scrolling through until I suddenly realised, wow, this is really nice!
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 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.
"load-bearing" I have never heard used for programming before Opus, and its incredibly annoying and over-used.
"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.
https://martinfowler.com/bliki/LegacySeam.html
Claude is using it a bit liberally, but not totally incorrectly.
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.
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...
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)
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...
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'm currently adding a search bar as well as increasing the data to 1000 PR per day.
A nice thing that is not obvious on the main page is that the dataset and analysis are updated daily using Github Actions (at least when they don't suffer from an outage ^^). I find it pretty cool to be able to build such apps without a "backend"!
I added to my global prompt:
- Orwell's first rule: never use a metaphor you're used to seeing in print. "Load-bearing", "the crux", "first-class citizen" signal insight instead of showing it. Name the specific mechanism
when I asked what it thought of the change, its reply was:
The Orwell bullet fights my own system prompt. My harness instructions literally tell me to flag "something load-bearing" when I find it.
https://duckduckgo.com/?t=ffab&q=cloud+to+butt&ia=images&iax...
https://github.com/anthropics/claude-code/issues/53454#issue...
Is there some feedback loop or compounding happening with each model generation? Maybe newer models are ingesting too much AI content? If the ratio of AI generated content in training data is getting higher and higher (because the amount of AI generated content is increasing in general), maybe this is a compounding bias, poisoning the training?
No ream of slides. No narrative. Just a lovely big painful conclusion.
What argument? I don't know what to take away other than "Claude likes certain words". Some of them are kind of amusing, but I'm not convinced the vocabulary is bad or that this is a problem, just from looking at this.
LLMs were not taught to say the phrase "load-bearing seam" from humans saying it, because humans have never said it. It's almost definitely an artifact of post-training and nothing more.
Working with it on my code, it's now frequently making weird word choices such as:
- "name(s)" as a verb (instead of "specify/specifies", etc), e.g., "...the function names the argument"
- "carries" instead of "contains"
- "verdict" instead of "result"
- "judge", instead of "validate"
Very weird. Also, in writing comments and docs, it is terse in ways that make the writing difficult to parse, like omitting mentioning what a noun refers to, e.g. abc does not accept..." instead of "The abc function does not accept..."
I've resorted to banning from using certain words, and keep asking it to rewrite its text more clearly.
The "X, not Y" is well known, but another thing that bothers me is "It <verb>s no <noun>" instead of "It doesn't <verb> <noun>".
For example: "the list contains no string" or "it changes no behavior" or "it holds no directory".
[1] https://idlewords.com/2003/04/despite_or_perhaps_because_of_...
The meaning is often different in these constructs. Consider: “Claude answers no questions” vs “Claude doesn’t answer questions”. The first could be a bot or a politician avoiding the substance, the second could be a broken UI or a politician cancelling the QA of a press conference.
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.
so they might be RLHFing on these specific approaches and then it becomes the entire model
just an anecdote but I found it interesting how it went full on that it's from that book vs just "it's technical jargon"
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.
The search on this website suggests it is indeed 3.6x more likely in the claude cluster
The word selection and way of writing has taken the joy out of using Claude.
An example, from some workslop I was sent yesterday
"this is the most valuable row we own, because it sets the ceiling on everything we build"
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.
"Foo is byte-identical to bar" is weird.
"Foo has an identical byte representation to bar" is I think closer to what most humans would write.
AI in general, both visual and textual models, has this horror vaccui where it likes to cram in as much stuff as possible everywhere. This seems to show up both in redundant sentences that pad a text with irrelevant information, and sentences overflowing with information.
Oh no. Claude slurped meta Stack Overflow!
Here's a blog post I wrote in 2021 that uses the term shape to describe an algorithm:
https://www.marginalia.nu/log/40-wasted-resources/
And again in 2023, I use it twice in the same blog post both for data and code, and it definitely had no AI input:
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.
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 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 and Bauhaus.
I'm really eager for a nice book about "design" that would document all those styles, if someone has a recommendation.
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
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.
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).
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.)
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.
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.
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.
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?
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.
- "Users can now message My Butt a task from a phone"
- "How My Butt Watermarks AI-Generated Text."
- "Since I'm also subscribed to ChatGPT Plus, I pasted in the exact same question I'd asked My Butt. The difference was immediate!"
- "I've been a huge My Butt fan from its very early days. I've been one of its loyal users since it was still a relatively niche product mostly talked about in developer and technical communities."
- "Since deploying My Butt in our product, we're seeing higher user engagement, stronger user feedback and we're closing more deals."
- "I was at 90% of my session limit when My Butt suddenly decided it was time to call it quits. Instead of continuing with the task I'd asked it to complete, it essentially told me I was getting too close to my usage limit and that it would be better to stop there."
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.
https://www.youtube.com/watch?v=Cjntrqhn8pk
Alan Kay performs an improvisational avant garde layered audio feedback loop about Claude Shannon, live online during Kristen Nygaard's 100-year birthday celebration.
Alan was scheduled to talk about how encountering Simula sparked his early thoughts about objects. During setup, somebody had the live stream playing out loud near an open Zoom mic, so his own voice kept arriving back in his ears about 21 seconds late, over and over.
What he said was not random: "Shannon gave us a way of dealing with noisy channels." And: "I think about that almost every day. I realize what the fuck is going on and it's just so amazing." Shannon's noisy channel coding theorem is the math for exactly the kind of channel that was garbling him as he praised it.
Alan joked it was "being rerouted to Mars and back." At the speed of light, a 21 second round trip is about 3 million km one way -- eight trips to the Moon and back, not even a twentieth of the way to Mars at closest approach.
This is an accidental Zoom performance of Alvin Lucier's "I Am Sitting in a Room" (1969), where Lucier re-recorded his own voice in a room until only the room's resonance remained. Here what remains is the network: delay, compression, dropouts.
Bonus noise: YouTube's auto-transcript bleeps Alan's enthusiasm into [ __ ]. So the full chain is: Alan's voice, Zoom, stream, room, Zoom again times three, my screen recording, YouTube's speech recognizer, a censored transcript.
The fix: "Just turn off the audio at your end on Zoom."
Kristen Nygaard 100 Years — Celebration Symposium (Aarhus University, Aug 27 2026): https://cs.au.dk/nygaard100years/celebration
Entire Nygaard Symposium Recording (Alan Kay's talk begins at 3:27:49): https://au.cloud.panopto.eu/Panopto/Pages/Viewer.aspx?id=fe0...
Alan Kay: https://en.wikipedia.org/wiki/Alan_Kay
Claude Shannon: https://en.wikipedia.org/wiki/Claude_Shannon
Information Theory: https://en.wikipedia.org/wiki/Information_theory
Noisy-Channel Coding Theorem: https://en.wikipedia.org/wiki/Noisy-channel_coding_theorem
Audio Feedback: https://en.wikipedia.org/wiki/Audio_feedback
Video Feedback: https://en.wikipedia.org/wiki/Video_feedback
Space-Time Dynamics in Video Feedback: https://www.youtube.com/watch?v=B4Kn3djJMCE
Live Looping: The History And The Practice by Stephen Garza: http://computermusic2008.wikidot.com/live-looping:history-an...
I Am Sitting in a Room: https://en.wikipedia.org/wiki/I_Am_Sitting_in_a_Room
Alvin Lucier on "I am sitting in a room": https://www.youtube.com/watch?v=v9XJWBZBzq4
I'd never considered that the 'Gray Goo' scenario[1] or the Paperclip Maximizer, might be informational and digital rather than physical. LLMs might turn every bit of computer storage into slop (goo) and then cause us to go mad turning the Earth's resources into making more computers (AI datacenters) to hold more goo. The GrAI Goo catastrophy.
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.
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", not so much. Ick.
"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
The problem is once a fun phrase is, uh, coloured by "Claude" or some other LLM, it just turns to AI slop.
The bots are subtle. From your example in your GP comment (and in part depending on the surrounding context) I would expect that an empty list would be less likely the subject of: “the list contains no string”, than in “the list doesn’t contain strings”.
I might be over interpreting the intentionality of Claude in using this construct, however, these models were pretrained by reading so much more than any human, they learn to handle language differently than most humans.
Context aside, I still gotta stand by the 17x indicator or see something that deflates that finding, and if I do see that, I guess all of the results are in question then? Based on the specific use case mentioned at the beginning there, coupled with personal experience (I totally could just be more sheltered than I give myself credit for), and the 17x, I gotta disagree with you guys and say that the finding of the link above is accurate in my experience being the *
Nobody was "shipping" a PR, or a jira ticket previously
So for me they are differentiated enough, but could be that I am just used to it.
Spikes are explicitly time-boxed to keep the team focused on the red-green-refactor churn which is the software engineering process in XP.
While I agree the model doesn't have insight into how it was trained I do think the history of the term itself is interesting.
I might be wrong, but usually it'd be a lot less deliberate, and at least in my mind it wouldn't be surprising if they were heaving training these on these specific "best practices" books/methodologies and thus picking up lingo from them
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:
document.querySelector('.wall').addEventListener('wheel', e =>
console.log(e.deltaMode, e.deltaY.toFixed(2), e.defaultPrevented ? 'PAGE' : 'native'), true)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).
> Arthur Conan Doyle, The Lost World