Alex,

Re LLMs:  Of course, Arun and I use LLMs for what they do.  Please note our joint article:  Majumdar, Arun K., & John F, Sowa (2009) Two paradigms are better than one, and multiple paradigms are even better, Proceedings of ICCS 2009, edited by S. Rudolph, F. Dau, and S.O. Kuznetsov, LNAI 5662, Berlin: Springer, pp. 32-47. https://jfsowa.com/pubs/paradigm.pdf

Two paradigms are better than one, and multiple paradigms are even better.  Arun has certainly adopted LLMs as yet another very important paradigm.   They can support and extend the 60+ years of tools developed for AI, computer science, technology, etc.  But they don't replace them.

But what I do criticize are PEOPLE who claim that LLMs are a universal AI tool that can come close to or even rival human intelligence.   For that, there is ZERO evidence.  What I have been emphasizing is the immense complexity of the human brain (and the brains of other mammals, especially the great apes).  Compared to the human brain, LLMs are just beginning to scratch the surface.

Alex> GEG is a mental construct. I just thought that blob might be useful when implementing GEG. Mental constructions are usually done at a mathematical level of accuracy 

Re blobs:  Yes, they are very useful for computational purposes.  When GEGs are used to represent computational methods, blobs  would certainly be one of many kinds of implementations.  Other options would include every kind of digital and analog methods of recording anything.  There is no reason to exclude anything that anybody might find useful for any purpose.  We have no idea what might be invented in the future.

Re mental:  We have to distinguish multiple kinds of things:  MENTAL (human brain and brains of other animals); THEORETICAL (pure mathematics); DIGITAL computational; ANALOG computational; and various mixtures of them.  For mental, there is a huge amount of research, and every research question opens up a few answers and even more questions.  For theoretical, there are no limitations; mathematicians have proved many important theorems about infinite structures.  

Alex> You are using "an open-ended variety of methods". Perhaps mine will be useful to you too.

Of course, when I say "60+ years of AI, computer science, technology, etc,"  that includes all of them.  But there are some that are obsolete or based on mistaken theory.  We always need to compare any technology to new discoveries to see which is better.    More often than not, we find important features of the old technology that can be combined with the new methods.  That's one reason why I emphasize the DOL standard by the Object Management Group (OMG).  They emphasize methods of combining, relating, interoperating, and co-existence.

John
 

From: "Alex Shkotin" <alex.shkotin@gmail.com>

John,


You criticize LLM all the time, but this is not the Carthage of GenAI, but just a fragrant flower on the tip of the iceberg. Which everyone uses as a component, just look at our autumn session [1]. And you and Arun use it.

Now the most interesting question is how the Q* and the AI scientist of OpenAI work.

By the way, the main problem with AI is how quickly robots “get smart”, because... they act. Some hotheads might hook them up to an AI supercomputer. Moreover, there may not be an LLM at all. Well, maybe just talk to people.


GEG is a mental construct. I just thought that blob might be useful when implementing GEG. Mental constructions are usually done at a mathematical level of accuracy. This was done for the theory of categories from which the theory of topoi was born. This was done in the HoTT project [1.5]. Probably at some point you will give us a text to read in which GEG will be described as a kind of mathematical super structure. There is, for example, “Topoi: The Categorical Analysis of Logic” [2]. Why not be "GEG, the structure of everything." In addition, there is L. Horsten "Metaphysics and mathematics of arbitrary objects." [3]


You are using "an open-ended variety of methods". Perhaps mine will be useful to you too.

In "OWL 2 Functional Style operators from HOL point of view" [4] it is shown that HOL is sufficient for OWL2.

In "English is a HOL language message #1X" [5] it is shown that natural language is HOL.

The “Theory framework - knowledge hub message #1” [6] proposes a method for storing theoretical knowledge on a global scale in a concentrated, publicly accessible, and usable manner.


I am now looking at what will be the framework of the theory for genomics and what of this framework and in what form is contained in the GENO ontology [6].


Theoretical knowledge will necessarily include certain mathematical methods, including algorithms. This is the subject of specific research for each specific science. I hope to find out what the mathematics of genomics is next year.


I am looking forward to reading "GEG, the structure of everything." 🙂


Alex


[1] https://ontologforum.com/index.php/OntologySummit2024 

[1.5] https://homotopytypetheory.org/book/ 

[2] https://projecteuclid.org/ebooks/books-by-independent-authors/Topoi-The-Categorial-Analysis-of-Logic/toc/bia/1403013939

[3] https://assets.cambridge.org/97811087/06599/frontmatter/9781108706599_frontmatter.pdf 

[4] https://www.researchgate.net/publication/336363623_OWL_2_Functional_Style_operators_from_HOL_point_of_view 

[5] https://www.researchgate.net/publication/366216531_English_is_a_HOL_language_message_1X 

[6] https://obofoundry.org/ontology/geno.html