[Article] Harvard physicist's take on the VMS

[npcompl33t]

I'm literally doing that.. you can have access to the Voynich Archive via MCP on my signature.. hook it up to fable 5.5 or Astra and have fun... But... It's not gonna solve it...I don't believe it's a computational problem in the end.. I've even tried quantum (you can stand up a quantum computer on AWS) ... What it is very good for however is research.. plug in Exa and Consensus and the archive which covers everything I can find about Voynich including this forum live updated, and it will spot patterns...and will help you pull threads..  but the stats? So far a dead end like everyone else ...

Supposedly the millennium problem used an internal model that is better than astra, but regardless, the main issue is the scale. If openAI just dumped the equivalent of a single users, even 1,000 users entire monthly allowance, they probably would have gotten nothing. They spent several orders of magnitude beyond what I have access to. If we had $20 million in open AI credits to burn I think we could significantly advance the field. That is enough to have astra read, and transcribe every single non-transcribed digitized medieval manuscript on the web, with review checks, build custom models to classify all of the paleographic markers in all said documents, etc, and still have millions and millions left over. At the very least it could probably figure out where the manuscript was written.

[pjburkshire]

What it is very good for however is research.. plug in Exa and Consensus and the archive which covers everything I can find about Voynich including this forum live updated, and it will spot patterns...and will help you pull threads..  

Does that mean you will be feeding it my ideas, too?

[DG97EEB]

What it is very good for however is research.. plug in Exa and Consensus and the archive which covers everything I can find about Voynich including this forum live updated, and it will spot patterns...and will help you pull threads..  

Does that mean you will be feeding it my ideas, too?

it's an open forum already crawled/indexed by Google and Exa and most likely therefore part of all frontier models' datasets... All I've done is specifically add it to a private vector database to make it more easily searchable...

[pjburkshire]

it's an open forum already crawled/indexed by Google and Exa and most likely therefore part of all frontier models' datasets...  All I've done is specifically add it to a private vector database to make it more easily searchable...

What are you hoping to do that will make your work different from what is already being done?  What are you going to search for that we can't already search for?

[npcompl33t]

it's an open forum already crawled/indexed by Google and Exa and most likely therefore part of all frontier models' datasets...  All I've done is specifically add it to a private vector database to make it more easily searchable...

What are you hoping to do that will make your work different from what is already being done?  What are you going to search for that we can't already search for?

This forum is littered with leads and ideas people have posited without having the capacity to fully investigate it. The point would be to just give a hypothetical AI some leads to get started. The key distinction is that this wouldn’t just be the current single user chatGPT.

$20 million in compute is an insane amount, a scale that is hard to comprehend. Many very interesting leads have been found on this forum by looking at other mideval manuscripts, noticing similarities to drawings. D’imperio said she thought there was enough structure to map it to some medieval maniscript.

$20 million is enough for an ai that found those leads to literally meticulously catalogue every drawing in every digitized manuscript and compare all them to Voynich. It’s enough to develop custom medieval handwriting analysis to map writing styles of marginalia on every digitized manuscript. And still have money left over.

The point is most leads discussed on this forum have not been fully investigated due to the raw amount of work it entails, and $20 million in credits is enough to do that. It’s not about the direct solution, it’s about people discussing which leads are mechanically solid and then having it pursue them. Thats how it solved the millennium problem, the approach it used was a preexisting hypothetical angle of atrack, but actually proving it was orders of magnitude more effort.

Edit: just to be clear, I’m not arguing ethics about this, the millennium problem situation was a fiasco because of how uncertain credit for work was assigned. In just describing a hypothetical scenario where current day AI could solve Voynich.

[DG97EEB]

it's an open forum already crawled/indexed by Google and Exa and most likely therefore part of all frontier models' datasets...  All I've done is specifically add it to a private vector database to make it more easily searchable...

What are you hoping to do that will make your work different from what is already being done?  What are you going to search for that we can't already search for?

I built the website (Voynich archive) to help the community search for prior art. That is and remains its number 1 purposes..

[pjburkshire]

I hope it works.  I would loved to see the mystery solved as much as anyone.  But I don't think matching pictures is going to do it.

[Mark Knowles]

a plaintext hidden in a steganographic manner (it would be a massive overkill to use such a technique)

This is what I have argued, for some time. Why would it be a massive overkill to use such a technique?

[nablator]

This is what I have argued, for some time. Why would it be a massive overkill to use such a technique?

A huge waste of available space and writing time.

[eggyk]

I hope it works.  I would loved to see the mystery solved as much as anyone.  But I don't think matching pictures is going to do it.

"matching pictures" is basically how a vast amount of manuscripts are dated and localised. If we can work out where and when the manuscript may have been created (or where and when the content that inspired it was written) certain things may be learned that lead to a breakthrough. 

A genuinely successful image match to a certain specific written text may allow a comparison of voynichese to plaintext somehow. For example, the german manuscript "Coburg, Landesbibliothek Coburg, Ms 5" has a chapter that where the imagery undeniably matches the (decades earlier) Astrolabium planum in Latin. Lo' and behold, the german text is a translation of the original latin. If it had been ciphertext instead of german, you could probably work it out what it said, and how the cipher worked.