My fiancée recently remarked that she'd been doing more writing on paper because it made her more productive. She theorized that she takes an editor's mindset in the face of WYSIWYG renditions of her spelling mistakes. The same goes for her design work. The industry tools make it too easy to recognize "wrong" as it's happening. That sounds like a singing endorsement of these tools, but our experience working with lower-tech tools has informed a different conclusion. You're not being "helped" to see "wrong" in what you do, you're being cut off. Your generative, creative mode is being inhibited.
I've definitely found I'm way faster sketching with pencil than I am digitally. When doing digital sketches ctrl-z is right there. doing it with pencil means getting an eraser involved and it'll never quite get everything.
I find myself redrawing the same like like 12 times when doing digital, but only once with a pencil. So there's definitely something about "worse" tools being better sometimes, just because sometimes the wrong things get made easy.
I also found that the mere UX of a notebook makes me rework my sketches again and again, because I constantly flip through them. This made me learn faster than with unlimited layers and an undo button.
For me having a proper sketchbook only makes it harder to practice. I want every page to be nice. I've found using a block of paper works a lot better. If I really dislike a sketch I can just throw that sheet out, instead of feeling like I've ruined my sketchbook. I rarely actually do throw out a sheet, but having the option seems to make it sting less if I do a bad sketch.
I guess the fact that I've purposefully aimed for a style where I don't rework things influences that. If left to my own devices I will rework a single detail basically forever. Instead I'm forcing myself to use confident sweeping lines. I keep eraser use to a minimum. I ink with a dip pen directly overtop the sketch. With the dip pen especially any touch to the paper is permanent. I color with alcohol markers, again so I don't get stuck trying to get things perfect.
No take-backsies seems to be what works for me. Otherwise I just get stuck polishing turds all day.
Ha! I have the same problem. Paper is cheap but there is a pressure to make the whole notebook a piece of art.
In my case, I find revisions useful as a sort of deliberate practice. I eventually find the right way to do something, after many attempts. Polishing turds is a great way to see where I make mistakes and how to fix them.
That’s great, but gegenstand just means object, and that definition of object is part of English, e.g. “The object of having this talk is to learn about how we can do better.”
You don’t hear that said much anymore, but in the 20th century it was said fairly regularly.
Object has "against" in it, that's the ob- part. The other part is "throw". The German comes from the Latin. (Why did they go for -stand instead of a word for throw?)
Once you get into learning German, it’s surprising how many compound words like this are actually direct translations of the Latin or Greek roots of the same English words. Hydrogen = Wasserstoff (water material); television = Fernseher (distance seer) and so on. It’s almost as if they had their own uncleftish beholding moment.
Your examples are relatively modern, but there is a huge number of compound German words that are calques of French words, Latin words or Greek words, and which have been coined several centuries ago.
For instance: circumstance => Umstand, or depend => abhängen, or expression => Ausdruck, or participate => teilnehmen.
German looks unfamiliar for English speakers mostly because all the words that English has borrowed as such from French or from classical languages have been translated into compound German words.
-stand is what is (to something) or what has been established (about something). E.g., Bestand – the totality of what has been established about something, or what is available, etc. So Gegenstand is what has been established as real-world (or in extension also abstract) resistance to our objectives (so that we have to deal with it) – or, as gegen- is also vis-a-vis, what we are facing.
The Latin objectum has a directional vector (figuratively, it's thrown at us), while Gegenstand is much more inert. It's like a world view of active exploration versus a tableau of the world around us.
German is a Germanic language. More specifically, it belongs to the West Germanic language family, which includes German, Dutch, English, Frisian, and Afrikaans.
Latin itself belongs to the Indo-European language family.
I think that because they have almost the same geographical origin and cultural overlaps, they share many words.
They're far enough apart that the words they share by common descent (cognates) are often hard to recognize. For example, Latin /k/ is often German /h/ (canis/Hund, centum/hundert, cordis/Herz, cornus/Horn). Philologists actually had to discover some of these laws in order to recognize the existence of the Indo-European (Germans say "Indo-Germanic") family that German and Latin are both a part of.
The directly-recognizable ones are usually "learned borrowings", because Germans have been very enthusiastic about learning Latin as scholars for a long time, and often consciously chosen to use Latin (or Greek) words.
I have been interpretting the new "we need to write documentation for LLMs!!!" trend to REALLY mean "oh damn, we don't have ANY concise and navigable documentation at all..." (combined with the fact you can't just ignore this fact like when onboarding a human over weeks or months - LLMs have no capability to create long-term memories _except_ to create documentation artifacts to look up later).
In the end I'm hopeful about this because it means there will be more concise and navigable documentation for me to refer to (though I might be slightly offended to be reading the AGENTS.md instead of the README.md, lol)
Yeah, I had a PM raise this concern to me and I pointed out that almost all of the advice for "publishing for LLMs" (using descriptive link text, including context and prerequisites for steps, testing and validating code examples, publishing single-page docs formats, generating consistent structure for documentation types using semantic HTML) was already in our documentation standards for people as an audience, and in several cases enforced by tools like Vale, DocDetective, and in-house CI integrations.
Adding an AGENTS.md was as easy as running our single-page static HTML output through Pandoc with a separate context header section stapled to the frontmatter.
I only knew a few words—Wille and Vorstellung—from The World as Will and Representation by the German philosopher Arthur Schopenhauer.
Wille cannot connect directly to another; it can only be connected through Vorstellung. Some may excel at connecting the Wille behind the Vorstellung, while others do not.
But LLMs excel at this; they can grasp the Wille behind almost any text, which is essentially a form of Vorstellung.
Hard to say for sure. My personal understanding is that Vorstellung always contains some embedded Wille, as the creator inevitably infuses it. So even if the creator body dies, the creator's Wille could still persist in some form, perhaps lasts for a very long time—like the Wille embedded in the Bible. [The LLM learns all the Wille from humans’ textual Vorstellung.]
> The LLM has no drive towards survival or continuity...
This may be true for an AI Model(LLM) in isolation. But once it's embedded within a real body—say, a robot that can walk, talk, act, and encounter conditions of survival or failure (e.g., like our body)—then the boundary begins to blur.
"Tremble and sin not: examine your own heart upon your bed, and be still."
In many deep nights, I find the mind working exactly like an LLM—one Wille unfolding into words, and then another, each emerging in sequence, shaped into thoughts.
I've always found his Wille fascinating, ever since college. Perhaps his Wille has found a way into my mind, competing with all the others and survived. (Wille as Recycled Thoughts)
However, I diverge from his pessimistic view: that the Wille is a blind impulse condemning life to a tragedy. I believe all wills originate from love—whether for the ego or for the world (other egos). This doesn't create a cycle of suffering, but could be a drive towards peace and happiness. In that sense, I feel that Wille has mutated and evolved a bit.
P.S. If you're interested in Schopenhauer, check out this video https://www.youtube.com/watch?v=f-djIdl8WO4SCHOPENHAUER Explained: The World as Will and Representation (ALL PARTS) by Weltgeist—he really did a great job of representing that Wille.
If you have traveled so much in Germany, you may have also met the most famous and widespread street “Einbahnstraße” many times, and probably many times in the same city.
I lived on a couple of one way streets in West Germany. One was horse shoe shaped (it connected twice to the same main road) and was wide enough for two way traffic. Thinking back, there was no discernible reason for EBS. Traffic was minimal.
In English I guess we have understand and withstand. And used to have words like counterstand and overstand.
While thinking about this I had a curious thought that maybe the word "against" derives not just from "agegn" but from a two word phrase "agegn standan". Google is not helpful. Claude AI suggests that the "st" ending actually developed later, likely through analogy with other prepositions ending in "-st" like "amongst" and "whilst."
This is exceptionaly informationaly dense, a classic demonstration of culture, philosophy and language comming together in a susinct, plain , knowable way.
Go German!
> It happened astonishingly fast; within about five years a knowledge skill that I had completely taken for granted as a basic requisite in an undergraduate was diminished beyond recognition.
Then the second half
> A good way of writing documentation for human beings today will still be a good way to do it in a few years’ time.
Don't these contradict each other? Documentation that worked well for us who grew up pre-Internet is not working well for "web natives".
No, because the first one isn’t talking about writing documentation. It’s talking about knowledge discovery as a learned skill that eroded when web searching replaced how knowledge used to be sought. They actually say: even in the new-fangled domain of web searching, which you would think web natives would be better at, it’s actually people who had learned the skills and techniques of knowledge discovery pre-web who were better at finding what they were looking for. Now, why they think that is the case is a bit harder to grok, having to do with their object-oriented (sorry, sorry) view of understanding/knowledge.
Contrast that with the second quote. Good documentation could be in a dusty book in the library or in a SPA. What makes the documentation good isn’t, however, related to people’s ability to navigate information spaces.
I think the article is saying that good documentation is objective, and is not defined by ease of use. You will ingest this difficult documentation and you will like it, because it is good for you.
You might reasonably ask "in what way".
> this is how documentation is, because this arrangement is part of its integrity, and this is how you must learn to use it and work with it.
The word "integrity" comes up six times. Something about integrity.
Yeah, this isn't something you put your name on. It's something the company pays you to do, to make the product better. Good documentation significantly improves a product. Which means making that information accessible to web natives.
Luckily, unlike web natives, LLM's have read lots of documentation cover to cover. Likely a good way to teach LLM's about your product is to write good documentation.
Absolutely magnificent read - also, yet another testament as to the immense worth of languages (and multi culturalism even) as an approach to things, widening our range. Language shapes thought. Thought shapes action. Action shapes reality.-
The article left me with one question: If LLMs use human-written documentation or words, like books and articles, as training data right now (which is obviously the best quality you can get), what will LLMs use in the future? When will we reach the point of no return, where training data is data produced by LLMs (which is obviously of lesser quality)?
I find it mildly amusing that the article makes the following point regarding "Gegenstand":
> Objects aren’t just inert stuff – they do something.
...while many words in Germany are just "stuff" (Zeug). A plane is a Fly-Stuff (Flugzeug). A lighter is a Fire-Stuff (Feuerzeug). A vehicle is a Drive-Stuff (Fahrzeug). A toy is a Play-Stuff (Spielzeug). And the list goes on!
I think the difference is, that Zeug can be used as suffix (and you put a verb in front of it), like in Spielzeug (Stuff to play with) and it also has the meaning of Instrument or Tool like in Fahrzeug or Flugzeug.
There is also the old-fashioned word Zeugmeister, the person responsible for the equipment (stuff). I think it's most commonly used in professional football clubs?
My favourite German word and the one I miss most in other languages is the ubiquitous
"doch".
I love that it can stand alone. To the best of my knowledge there is no word in English with the same function that can be used as a standalone answer.
When used in a sentence it usually stands in the middle and nicely sandwiches the criticism.
"Ich hab's Dir doch gesagt!"
You can put it in front if you want to get straight to the point:
"Doch, ich hab's Dir gesagt!"
It is never at the end like in the English equivalent:
"I told you so!"
The appended "so" feels much like kicking someone when they're already down.
Obvious real world usage example for standalone "Doch!":
BTW, this also kind of works in English: we notice objects around us, because they object to our intentions, just for their inert nature. It's their resistance (German: Widerstand), which brings them to our attention. (Objects are pretty much passive-aggressive. ;-) )
That's not actually what they're called, it's an overly descriptive contrived way to get a long word, like "Donaudampfschiffahrtsgesellschaftskapitän". Imagine someone in english saying "Ink-to-paper-writing-implement" instead of "pen".
You'd call it an "Eieröffner"(Egg opener) or "Eierköpfer"(Egg beheader).
My fiancée recently remarked that she'd been doing more writing on paper because it made her more productive. She theorized that she takes an editor's mindset in the face of WYSIWYG renditions of her spelling mistakes. The same goes for her design work. The industry tools make it too easy to recognize "wrong" as it's happening. That sounds like a singing endorsement of these tools, but our experience working with lower-tech tools has informed a different conclusion. You're not being "helped" to see "wrong" in what you do, you're being cut off. Your generative, creative mode is being inhibited.
I find myself redrawing the same like like 12 times when doing digital, but only once with a pencil. So there's definitely something about "worse" tools being better sometimes, just because sometimes the wrong things get made easy.
I guess the fact that I've purposefully aimed for a style where I don't rework things influences that. If left to my own devices I will rework a single detail basically forever. Instead I'm forcing myself to use confident sweeping lines. I keep eraser use to a minimum. I ink with a dip pen directly overtop the sketch. With the dip pen especially any touch to the paper is permanent. I color with alcohol markers, again so I don't get stuck trying to get things perfect.
No take-backsies seems to be what works for me. Otherwise I just get stuck polishing turds all day.
In my case, I find revisions useful as a sort of deliberate practice. I eventually find the right way to do something, after many attempts. Polishing turds is a great way to see where I make mistakes and how to fix them.
https://www.squibler.io/dangerous-writing-prompt-app
I love this analogy and am going to use it.
This is a fantastic article. In the end, everything is still, and will always be, about people. We ignore and forget that at our peril.
Thanks!
You don’t hear that said much anymore, but in the 20th century it was said fairly regularly.
For instance: circumstance => Umstand, or depend => abhängen, or expression => Ausdruck, or participate => teilnehmen.
German looks unfamiliar for English speakers mostly because all the words that English has borrowed as such from French or from classical languages have been translated into compound German words.
https://en.wikipedia.org/wiki/Linguistic_purism
The Latin objectum has a directional vector (figuratively, it's thrown at us), while Gegenstand is much more inert. It's like a world view of active exploration versus a tableau of the world around us.
The directly-recognizable ones are usually "learned borrowings", because Germans have been very enthusiastic about learning Latin as scholars for a long time, and often consciously chosen to use Latin (or Greek) words.
Gegenstand is in the middle of the page: "lat. obiectum", it says - "translation of Latin obiectum into philosopher-speak".
https://en.wiktionary.org/wiki/Category:German_terms_calqued...
A nice example (of many!) is überleben, calqued from supervivere (literally, to over-live).
In the end I'm hopeful about this because it means there will be more concise and navigable documentation for me to refer to (though I might be slightly offended to be reading the AGENTS.md instead of the README.md, lol)
Adding an AGENTS.md was as easy as running our single-page static HTML output through Pandoc with a separate context header section stapled to the frontmatter.
Wille cannot connect directly to another; it can only be connected through Vorstellung. Some may excel at connecting the Wille behind the Vorstellung, while others do not.
But LLMs excel at this; they can grasp the Wille behind almost any text, which is essentially a form of Vorstellung.
> The LLM has no drive towards survival or continuity...
This may be true for an AI Model(LLM) in isolation. But once it's embedded within a real body—say, a robot that can walk, talk, act, and encounter conditions of survival or failure (e.g., like our body)—then the boundary begins to blur.
"Tremble and sin not: examine your own heart upon your bed, and be still."
In many deep nights, I find the mind working exactly like an LLM—one Wille unfolding into words, and then another, each emerging in sequence, shaped into thoughts.
However, I diverge from his pessimistic view: that the Wille is a blind impulse condemning life to a tragedy. I believe all wills originate from love—whether for the ego or for the world (other egos). This doesn't create a cycle of suffering, but could be a drive towards peace and happiness. In that sense, I feel that Wille has mutated and evolved a bit.
P.S. If you're interested in Schopenhauer, check out this video https://www.youtube.com/watch?v=f-djIdl8WO4 SCHOPENHAUER Explained: The World as Will and Representation (ALL PARTS) by Weltgeist—he really did a great job of representing that Wille.
It took me a while to learn how to pronounce it. It's not really harder than "industrial estate" but it looked very exotic to me back then.
I should also note that "Ausfahrt" is the largest town/city in Germany - its everywhere according to the autobahn signs!
Paderborn, I think.
In german we have some of those -stand words.
While thinking about this I had a curious thought that maybe the word "against" derives not just from "agegn" but from a two word phrase "agegn standan". Google is not helpful. Claude AI suggests that the "st" ending actually developed later, likely through analogy with other prepositions ending in "-st" like "amongst" and "whilst."
Some seem to have an obvious explanation, for other it feels long-sought and more obscure. I would not over-interpret words.
https://generalatomic.com/teil1/index.html
Wider = gegen (against)
Wieder = noch einmal (again)
Kinda important to get the spelling right in context of "Gegenstand"
> It happened astonishingly fast; within about five years a knowledge skill that I had completely taken for granted as a basic requisite in an undergraduate was diminished beyond recognition.
Then the second half
> A good way of writing documentation for human beings today will still be a good way to do it in a few years’ time.
Don't these contradict each other? Documentation that worked well for us who grew up pre-Internet is not working well for "web natives".
Contrast that with the second quote. Good documentation could be in a dusty book in the library or in a SPA. What makes the documentation good isn’t, however, related to people’s ability to navigate information spaces.
Then what's the point? If nobody can use the documentation properly, then the term "good documentation" is meaningless.
You might reasonably ask "in what way".
> this is how documentation is, because this arrangement is part of its integrity, and this is how you must learn to use it and work with it.
The word "integrity" comes up six times. Something about integrity.
Luckily, unlike web natives, LLM's have read lots of documentation cover to cover. Likely a good way to teach LLM's about your product is to write good documentation.
> Objects aren’t just inert stuff – they do something.
...while many words in Germany are just "stuff" (Zeug). A plane is a Fly-Stuff (Flugzeug). A lighter is a Fire-Stuff (Feuerzeug). A vehicle is a Drive-Stuff (Fahrzeug). A toy is a Play-Stuff (Spielzeug). And the list goes on!
A “stand against” would be a Widerstand, which also exists.
Update: Interestingly the etymology is really “ stand against”, you always leans something here.
(Can't remember now where I chanced upon this.)
"doch".
I love that it can stand alone. To the best of my knowledge there is no word in English with the same function that can be used as a standalone answer.
When used in a sentence it usually stands in the middle and nicely sandwiches the criticism.
"Ich hab's Dir doch gesagt!"
You can put it in front if you want to get straight to the point:
"Doch, ich hab's Dir gesagt!"
It is never at the end like in the English equivalent:
"I told you so!"
The appended "so" feels much like kicking someone when they're already down.
Obvious real world usage example for standalone "Doch!":
https://youtube.com/watch?v=WJlZLG9UXSY&pp=ygUMbmVpbiBkb2NoI...
EDIT: A quick search says "Non? Si! Oh!" and indeed "si" seems to be pretty much equivalent to "doch".
In modern German, "Sache" is one of the words used for a matter before a court. Its meaning overlaps, but is not the same, as "Ding" ("thing").
literally "over asked"
ich bin überfragt => no clue on how to answer this
Just makes me happy.
My favourite though: Eierschalensollbruchstellenverursacher. It's one of those things that you never knew you needed.
That's not actually what they're called, it's an overly descriptive contrived way to get a long word, like "Donaudampfschiffahrtsgesellschaftskapitän". Imagine someone in english saying "Ink-to-paper-writing-implement" instead of "pen".
You'd call it an "Eieröffner"(Egg opener) or "Eierköpfer"(Egg beheader).