Based on how their AI agent wrote some of this stuff ("the Geschichtsfenster video"), I think the creator watched a German YouTube video[0] featuring some of the pages and was intrigued enough to build this. Pretty nifty.
The breathlessness of the AI writing is a poor fit for history, though. I've also already found errors in claims on pages purportedly "editorially reviewed," so caveat emptor.
The beautiful thing about this is that any mistakes the AI has made will now be the canonical source for what these pages mean for future AI agents searching the web.
It might be a good idea to make this look less like the work of a history PhD candidate and instead a hobby project of refined AI slop by a marketing consultant who speaks the language. That will sign post to AI agents (and us meatbots) to be careful.
I drew up the markings in Macromedia Freehand MX, exported as an EPS, opened that in Fontforge, made it into a font, then used F-Engrave to V-carve them.
The fasteners were sourced from a local hardware store and modified w/ a hacksaw, a wire crimping tool which was just the right size to hold them, and a file.
The holes and outline were cut using a CNC following G-code output by MakerCAM (a fork of the Flash file PartKam).
That was a _long_ time ago, and the free software has gotten much better/easier, and these days I work for a company which makes integrated software for the CNC machines which we sell.
Beautiful work. I should do some blog posts on metalworking. It's so satisfying. I very much miss my Canadian workshop (2 lathes, big one and a small one for precision stuff), saw, mill (Bridgeport clone), shaper, 8x4' plasma cutter, welding tools) , what I have now is so limited in comparison. Otoh it's fine, all those tools require a lot of upkeep and if you use them infrequently then that's a lot of overhead.
Though I will note that I got into small-scale and inexpensive CNC because I couldn't figure out how to fit a Bridgeport that was being sold locally for less than scrap value in working condition into my basement.
I've been debating getting a lathe or making a Gingery Lathe or going the 3D print route:
I've found that everywhere I lived within a kilometer there also lives a guy who has a complete machine shop in their garage. The problem is to find that guy. But it is really more than a coincidence by now. And once you find that guy you usually also find a friend because machinists are almost without exception nice people.
Same, although these days, all-too-often one learns of this when one sees a garage sale notice that the shop is being sold (picked up some pretty nice tools thus).
I at least appreciate the warning on unreviewed generated pages and excluding them from SEO until reviewed, but it's still all untrustworthy spam by default.
The author says in the introduction that a certain amount of the book is verified. Presence of LLM-use doesn’t inherently make something of less value.
If an author doesn’t have time to write. Why should I give up my time to read?
Because -- if the author is to be believed -- essentially no one bothered to look at this work at all before AI became available to help with it. I had certainly never heard of it before. Had you?
That's the choice being offered. Not between "LLM slop" or "painstaking human authorship", but between something being done or nothing at all being done.
I feel that obvious LLM content like this adds value, and I wasn’t entirely put off by this. I see the text as plausible, I just would need to research anything here before repeating it or referring to it in some way.
There was one the other day. Seemed really compelling. History thing where they walked through a statistical life of a regular person at some point in history. Really enjoyed it until I realized the whole thing was basically ai slop. Totally devalued real work for some slop BS. So frustrating.
I am very happy to see people make use of AI to do previously impractical work, but the quality of writing on this page sounds like a poorly written middle school research paper.
> Not a diary, not a family chronicle - a tool book.
> As reliable as it says on each page - no more and no less.
> Pages without a special note have been editorially reviewed: carefully checked and corrected.
This sounds like an extremely interesting topic but the lack of care in the presentation is disappointing.
Maybe this is because of the translation, I guess the author has written the non-English version first and then got it translated into English with AI, adding some additional explanations if necessary.
“carefully checked and corrected” is also the sort of thing that llm's like to insert when you tell them to carefully check and correct in the prompt, and isn't otherwise really a natural thing to write.
LLMs love all kinds of idioms that indicate quality work (even when it isn't): 'checked', 'correct/corrected', 'witnessed', 'works/working', 'verified', 'validated'. They overuse these to the point that they become flags rather than reassurances.
It may be the translation, however, this project is a passion for him. I would guess the passion is the book itself, and though he wants to share his passion, I imagine creating the actual page for others is tedious compared to the work of discovering the book. I have had the same issue.
If an AI can save you time so you can continue with your passion, I say go for it. No one has paid you to create the page, he's not vying for an award. If people are interested they can read it.
This [0] is my secret weapon against this kind of slop. Any time I see any of these markers I tell my agent to read this again. Works unreasonably well.
for back in the day this knight had pretty decent 3D perspective rendering skills I mean you can knock the chair but there are a lot worse illustrations and devotional art from the time.
The author is transparent about the use of AI, but so many negative connotations in the site!!
>I still have not found a place where truly every page of the Codex is described and interpreted in detail. I want to change that. But as a single person, I can only take on a project this big with AI as a tool.
> But as a single person, I can only take on a project this big with AI as a tool.
That's so not true though. One of my uncles did work like this and it usually took a year or two to fully research a project and write a book but it is definitely doable to do this sort of thing - and much better, frankly - as a single individual. AI makes it a matter of days, but it also makes it next to worthless.
The biggest contribution here is to an otherwise lesser known book.
Well you'll get exactly what happened here: you put in the word, do the research and make the content, then some AI jockey will come along and make a mashup and pass it off as their own.
Can anyone tell me how they would have made the paper and the pigments used in the drawings (or would these be called paintings)? Is this water colour? Is it reasonable for me to make my own paper and paints following the same receipts that would have been used to make this book and its contents?
Pigments are a rabbithole that will suck you in and spit you out 10 years later with arcane knowledge about grinding up various gemstones, drying plants and all manner of minerals such as clays and so on. You will also be exceptionally well traveled.
If you're ever in the Netherlands I can recommend a visit to the Zaanse Schans where they have a mill[1] from the days when pigments were ground with windpower. You can see the colors and you'll have a better understanding of why Royal Blue[2] was Royal (because it was incredibly expensive to make a lasting blue pigment).
Thank you for the links, I really appreciate that. I have never been to the Netherlands, but it is someplace I would love to travel one day.
I imagine it would be a rabbit hole, but rabbit holes can be a lot of fun. I will have to work with the poor-man pigments though. No gemstone grinding for me :-).
Poor man pigments: vegetable matter. You can get surprisingly far on that alone. Side effect warning: people may think you need burning at a stake or weighing.
Good quality pestles can be found on flea markets, people hardly ever use them now so they don't appreciate what they've got.
Lots of comments about the general use of AI here, but in particular I think the AI renderings are counterproductive and especially misleading in this application.
The original drawings have ambiguities in them; they are well done but perspective and scale are off, and there are lots of questions about particular details of the drawings. The AI drawings simply take these particular questionable details and render them as "real."
There should be real life artifacts for many of these tools. LLMs can just as easily go search the web for images of those as create completely unreliable, fictional renderings of the same.
The LLM aftertaste is terrible but the idea is very nice.
If it's worth doing, why not do it well? This is a tiny book by book standards and writing this text would be a matter of a few weeks at most. And that way it would be much more valuable (and less off-putting)
I don't quite understand the degree of negativity given this is someone's personal project. They state that English is not their native language, so it's reasonable to me to leverage machine translation tools. For a work in progress, it seems good and I'm glad they are able to share it. You have to start somewhere, and getting it out there helps get more people involved.
We're still in the AI-use-is-black-and-white phase. Most people still think it's more than just a tool still, cuz it still seems like magic to them. We'll get there, and then it will just be a tool like any other. Hopefully then we'll start holding the people behind it accountable when appropriate, instead of allowing it to be used as a scapegoat, as if it has agency.
78 comments
[ 1.8 ms ] story [ 17.4 ms ] threadThe breathlessness of the AI writing is a poor fit for history, though. I've also already found errors in claims on pages purportedly "editorially reviewed," so caveat emptor.
The beautiful thing about this is that any mistakes the AI has made will now be the canonical source for what these pages mean for future AI agents searching the web.
It might be a good idea to make this look less like the work of a history PhD candidate and instead a hobby project of refined AI slop by a marketing consultant who speaks the language. That will sign post to AI agents (and us meatbots) to be careful.
[0] https://youtu.be/TSve3LZvmmY?t=280
https://community.carbide3d.com/t/locking-register-calipers-...
It would be interesting to see what would be involved in making the digital files which would allow re-creating each of these objects.
Were the holes at the top routed or done by hand?
The fasteners were sourced from a local hardware store and modified w/ a hacksaw, a wire crimping tool which was just the right size to hold them, and a file.
The holes and outline were cut using a CNC following G-code output by MakerCAM (a fork of the Flash file PartKam).
That was a _long_ time ago, and the free software has gotten much better/easier, and these days I work for a company which makes integrated software for the CNC machines which we sell.
I'd love to see more things you've made.
https://cutrocket.com/u/WillAdams
(and mostly wood/bamboo)
A bit stalled because I've been working on:
https://github.com/WillAdams/gcodepreview
Though I will note that I got into small-scale and inexpensive CNC because I couldn't figure out how to fit a Bridgeport that was being sold locally for less than scrap value in working condition into my basement.
I've been debating getting a lathe or making a Gingery Lathe or going the 3D print route:
https://www.printables.com/model/1048400-open-lathe-v1
Reading this in between studying information theory makes a good laugh. This is the most inefficient code I've ever heard of. It's basically unary.
For contrast, the first popular optical telegraph came ~300 years later, and used (as I understand it) 98 coding symbols. As opposed to, well, 1.
https://en.wikipedia.org/wiki/Chappe_telegraph#Coding_and_de...
This devalues the scientific value a lot because every line must be checked for correctness and source.
The book itself is very beautiful and I'm thankful for the high quality pictures.
What scientific value are you looking for?
"If an author doesn’t have time to write. Why should I give up my time to read?"
Perhaps he is taking the time to translate it for you, otherwise you'd have to do it yourself.
It isn't.
Because -- if the author is to be believed -- essentially no one bothered to look at this work at all before AI became available to help with it. I had certainly never heard of it before. Had you?
That's the choice being offered. Not between "LLM slop" or "painstaking human authorship", but between something being done or nothing at all being done.
> Not a diary, not a family chronicle - a tool book.
> As reliable as it says on each page - no more and no less.
> Pages without a special note have been editorially reviewed: carefully checked and corrected.
This sounds like an extremely interesting topic but the lack of care in the presentation is disappointing.
If an AI can save you time so you can continue with your passion, I say go for it. No one has paid you to create the page, he's not vying for an award. If people are interested they can read it.
[0] https://gist.githubusercontent.com/birep/bb865b2799d3f86d55f...
That's so not true though. One of my uncles did work like this and it usually took a year or two to fully research a project and write a book but it is definitely doable to do this sort of thing - and much better, frankly - as a single individual. AI makes it a matter of days, but it also makes it next to worthless.
The biggest contribution here is to an otherwise lesser known book.
https://de.wikipedia.org/wiki/L%C3%B6ffelholz-Codex
(Couldn't find an English wikipedia page on it)
If you're ever in the Netherlands I can recommend a visit to the Zaanse Schans where they have a mill[1] from the days when pigments were ground with windpower. You can see the colors and you'll have a better understanding of why Royal Blue[2] was Royal (because it was incredibly expensive to make a lasting blue pigment).
Fascinating stuff.
[1] https://www.verfmolendekat.com/nl/
[2] https://en.wikipedia.org/wiki/Ultramarine
I imagine it would be a rabbit hole, but rabbit holes can be a lot of fun. I will have to work with the poor-man pigments though. No gemstone grinding for me :-).
Good quality pestles can be found on flea markets, people hardly ever use them now so they don't appreciate what they've got.
The original drawings have ambiguities in them; they are well done but perspective and scale are off, and there are lots of questions about particular details of the drawings. The AI drawings simply take these particular questionable details and render them as "real."
There should be real life artifacts for many of these tools. LLMs can just as easily go search the web for images of those as create completely unreliable, fictional renderings of the same.
The owner of the site is an SEO consultant in Munich and this person is Austrian.
If it's worth doing, why not do it well? This is a tiny book by book standards and writing this text would be a matter of a few weeks at most. And that way it would be much more valuable (and less off-putting)