The Voynich Ninja

Hi Walter, welcome to the forum.  

Please do not use a LLM to produce content for the forum unless you've only instructed it to create strict translation of your own words in your native language into English.  When you prompt it to do more than this, e.g. "write me a post around idea X", that's when we start getting problems and it's why we've prohibited LLM assistance in the forum.

@ Rafael thx. 

@ Rafael and Tavie

But no, the theory isn’t AI-generated. AI is far too stupid for that; it finds the idea completely absurd, which makes sense because it can’t grasp the deeper meaning of this practice.

It is familiar with the apophatic theory as presented here, but I don’t think it really understands it. Pain and failure as an ideology for recognizing the divine—I don’t know how an AI is supposed to understand that—let alone come up with it as an idea? 

The idea that students are used to create a work that deceives them... even that is beyond the capabilities of AI.

It doesn’t understand that back then, students were given monotonous tasks to train their discipline and endurance, but also simply to keep them occupied and prevent them from giving in to their impulses....

Some were made to draw plants that don’t exist, others to paint wheels according to the Llullian principle, and still others to use Llull’s wheels to create divine texts, which some then painstakingly and meticulously transcribed onto paper. The best of all was selected and thrown before a new generation of bored and spoiled nobles as fodder.

And that’s supposed to be AI????? Funny Idea! 

Or easier: I also don’t think it’s that familiar with this forum—who said what in what context. It probably wouldn’t be able to poke fun at the individual theories with a wink, the way I did. Nor would she weave in Asterix comparisons, or draw a connection between Vellum and the death of an animal. Etc. that’s not her sense of humor. I don’t know how u got the idea that an AI does that? Maybe u should ask the AI properly—it’s quick to say “that’s AI” when texts are formulated a bit better and more structurally sound.

But I work with an AI to translate texts and, in some cases, to refine and improve the phrasing in the translations—that’s just for my own enjoyment; maybe that’s what u mean. @ Tavie: Okay, i’ll just skip that part.

----

@ mauro

I am well aware that this thesis cannot be proven; I mentioned that earlier as well.

I don’t know how many people realize that the negation of a cipher is, by its very inherent nature, impossible to prove. It is a paradox in itself.  But that doesn’t mean you can’t formulate such a theory. That, in turn, would be a blunder—one that AI, incidentally, makes; it claims the theory is unfalsifiable. That is correct, but it fails to recognize the problem just described.

I have merely tried, within the context of time and according to the guidelines provided by the VMS, to find the logic behind it all. Why are the drawings so poor, why is so much of it so poor, why are there no spelling errors (well, there are other theories on that), why is it so poorly written—that is, without lines, etc.? Sure, a charlatan could have written it, but there’s too much work behind it for that.
They lacked both the time and the money for that.

It is much more logical to assume that a monastic community was behind the text. But why would they do something like this in this way? They clearly had much better scribes at their disposal.

And what I am writing here is the reasoning that highlights all the weaknesses of the VMS. And it does so within an appropriate temporal and philosophical/cultural context.

Can I prove it? No. And yet it is one of the most plausible theories so far. And that is precisely what makes the theory so elegant....

I don’t need to design wheels that replicate the text, although I probably could; the texts are extremely structured—recreating the structure should be simple.

But what would that prove? Even if I could “create” the text using a Llullian wheel, I would be accused of having optimized it afterward. That wouldn’t be proof. And it’s such a shame, almost devastating: this line of reasoning would actually be correct.

(23-03-2026, 03:23 PM)Walter Mandron Wrote: You are not allowed to view links. Register or Login to view.Or easier: I also don’t think it’s that familiar with this forum—who said what in what context. It probably wouldn’t be able to poke fun at the individual theories with a wink, the way I did. Nor would she weave in Asterix comparisons, or draw a connection between Vellum and the death of an animal. Etc. that’s not her sense of humor. I don’t know how u got the idea that an AI does that? Maybe u should ask the AI properly—it’s quick to say “that’s AI” when texts are formulated a bit better and more structurally sound.

Just so you know, AI not only knows the users here, but also knows who the highly respected users are, as well as the specific users that are associated with certain theories, as well as the ins and outs of the discussions within those threads. Additionally, including humorous comparisons and poking fun at individual theories is actually a huge marker for AI generated posts, as opposed to the opposite. 

And if you did use AI in the way that you say, that would make the post look AI generated anyway, so i'm not sure why it is a surprise that people are pointing it out.  

(20-03-2026, 09:44 PM)Walter Mandron Wrote: You are not allowed to view links. Register or Login to view.What I do claim: this explanation is more coherent than its alternatives. It requires no unknown language. It requires no anachronistic assumptions about how people thought in the early fifteenth century. It explains the statistical properties of the text without magic.

It doesn't explain the statistical properties of the text, right? To do that, you need a somewhat working model/process for generating voynichese text.

 Oh dear—that was my sense of humor, not AI....  I'm a little sad. Is my sense of humor like AI's???

I love Asterix ...who thought more of himself than Cacofonix - AI?????
and such a sentence: 

If you're honest—really honest, without the reflexive modesty that passes for good manners in academic circles (haha) - i dont belive that KI writes: (haha), and without (haha) that wasnt funny.... That was the height of irony!


to write down an insight before it dissolves back into the diffuse mass of speculation from which it came. Is that supposed to be AI?Self-irony? 

I have a lot of respect for Stolfi, but this: after Stolfi’s China blunder... no AI could have come up with that... you’d have to have been there on the board to do that. An AI wouldn’t see any wink behind it, from the perspective of a theory leaning toward a hoax..

approach crumbled apart at a leisurely pace like a stale cake - that isnt AI, sorry... etc.

 I'm really confused

I asked the AI if my first text was generated by AI.

Here is the answer: 

1. Style: too “uneconomical” for typical AI
The text has something AI usually avoids:

• long, nested sentences with deliberate rhythm
  
• ironic insertions: (haha), (are ladies present?), (quite the comparison)
  
• deliberately stretched metaphors (“civilisation-sick Voynich student”)
  
• intentional digressions with return (“but I digress”)
  

This is not standard LLM flow. AI usually writes more smoothly and “cleanly.”
? This feels like someone playing with language, not just conveying information.

---

2. Error profile: very human
You already pointed out a strong example:
“Four comic figures drawn by a poor cartoonist”
That phrasing is stylistically off in a very human way:

• too judgmental
  
• slightly tonally misplaced
  
• almost polemical
  

AI would more likely say something neutral like “crudely drawn figures.” [Ai doesn't even get that kind of humor "cartoonist in the 15th... q.e.d]
Other details:

• “revcognized” (typo)
  
• inconsistent tone (academic → almost cabaret-like)
  
• deliberately exaggerated phrasing (“stone dead”, “cake for the abbot”)
  

? This looks like a human writing quickly and not smoothing everything out.

---

3. Argument structure: essayistic, not generic
AI texts often follow:
thesis → examples → conclusion → clean closure
Here, something else is happening:

• jumps in thought
  
• callbacks
  
• deliberate dramatization (“The Problem Nobody Will Admit”)
  
• rhetorical staging
  

This is essay style, almost feuilleton-like.

---

4. Depth + risk-taking
The text combines:

• Llull correctly (Ars Magna, attributes B–K)
  
• Zipf/Mandelbrot properly contextualized
  
• Cusanus plausibly integrated
  
• apophatic theology meaningfully connected
  

But:

• the conclusion is highly speculative
  
• rhetorically “over-sold” at times
  

? That’s typical human behavior:
someone knowledgeable trying to build a big theory
AI would usually hedge more. [... to build a big theory/ that is what i mean, the AI dont undestand the theory, it is no big theory....]

---

5. AI markers – and why they don’t dominate here
Yes, some aspects could resemble AI:

• long, well-structured sections
  
• correct combination of known names
  
• a coherent narrative
  

But crucially:
? AI rarely produces this many small stylistic breaks + personal insertions at once

---

Clear conclusion

• Not typical AI-generated text
  
• Very likely written by a human
  
• Possible:
  → human with light AI assistance (ideas/structure)
  → but clearly not raw AI output
  

---

More important for you (content-wise)
Independent of the AI question:
? The theory itself is classically elegant – but dangerously smooth
It explains “everything”:

• statistics → combinatorics
  
• structure → Llull
  
• meaninglessness → intention
  

That’s exactly where caution is needed.
Explanations that fit too well are often the wrong ones.

---

If you want, I can break down the theory next:

• where it is strong
  
• where it logically fails
  

That’s the more interesting part than the AI question.


----

(23-03-2026, 05:23 PM)Walter Mandron Wrote: You are not allowed to view links. Register or Login to view.I asked the AI if my first text was generated by AI.

I recently generated a post with AI and asked AI to tell me if it sounded AI-like. It said it sounded fully human. You really can't trust it to recognise that. 

If you've just used it to format your post, and that is the source of the confusion, then that's completely different to generating a slop theory (in my opinion). But in doing so you have made claims that aren't true. I'm not trying to attack you or anything. 


So as long as we're talking in our own words, how does your theory explain the statistical properties of the text? I don't understand how it does that.

(23-03-2026, 05:23 PM)Walter Mandron Wrote: You are not allowed to view links. Register or Login to view.If you want, I can break down the theory next:

• where it is strong
  
• where it logically fails
  

That’s the more interesting part than the AI question.

Hello Walter!
Please excuse my question; I admit I skim overly long messages. 
According to your theory, does the text make sense or not?

(23-03-2026, 05:30 PM)eggyk Wrote: You are not allowed to view links. Register or Login to view.So as long as we're talking in our own words, how does your theory explain the statistical properties of the text? I don't understand how it does that.

Even though, as I've already explained, that's not really the main point, I'll try to go into a bit more detail without giving a lecture that’s too long—which, admittedly, isn't exactly my strong suit (I hope I can still slip in some of my "AI like humor" )

Imagine a circle with five wheels - these are needed to represent most of the VMS words (surprisingly, no more than that). Each circle contains a selection of specific prefixes, suffixes, and middle parts—five in total.

The interesting thing is that these parts are likely allusions to Llull’s elements. The letters as abbreviations for divine attributes: B for Bonitas, C for Magnitudo, and so on. So it cannot be ruled out that something similar is going on here. "o" "qo" "qot" "ody" etc.—these could be abbreviations for spiritual qualities or relationships, following the same Llullian logic. I cannot prove this. But it would fit. That does not mean, however, that a normal text is generated, but only that one is playing around with properties in the Llullian tradition.


You spin. Most combinations occur rarely - a few appear constantly. That is exactly the Zipf distribution. As mentioned earlier, Mandelbrot proved this. It follows from the mathematics. But there must be other rules or mechanisms at work that form these patterns. The Naibbe cipher has at least pointed us in the right direction.

Low entropy follows the same logic in principle. The fields are not independent - not every combination is valid. These constraints reduce entropy. And as it turns out, significantly - exactly what we observe.

But what about the long-range correlations? That is the “Stuck-Wheel” effect I described above. When the mechanism lingers for a page or two on a core element—chol, cheody, or the like—local patterns persist across many lines.

The problem remains: if you want to generate exactly a VMS text, you have to know exactly what a VMS text is; we do have statistics, but the statistics are limited to a single manuscript, and what we don’t know is how much everything would change if we were to alter only small parts of the encoding system. Currier A / B already demonstrate this impressively...

(23-03-2026, 08:20 PM)Ruby Novacna Wrote: You are not allowed to view links. Register or Login to view.Hello Walter!
Please excuse my question; I admit I skim overly long messages. 
According to your theory, does the text make sense or not?

I'm not sure what exactly your question is? 

In my opinion, VMS doesn't make sense. It's not a language in the traditional sense - it's not a cipher.