Just a sect.

I used to live for years in Kathmnadu and read some used books from its excellent (and now mostly scrapped) bookshops.

I read a lot of texts on Eastern traditions and philosophies, mostly Buddhism and Upanishads, which are unparalleled in the sense that they put the mind “on the spot” of its own attention and awareness.

In short, they have proposed (and practiced what they preached) that in order to understand what is one has to understand the mind first. To know what lies “outside” know what lies within.

This is, of course, a universal maxim and could be even more generalized and extended to everything social, including socially constructed “memes”, beliefs and what they call “scientific consensus”.

To illustrate how real (and actual) these are, try to realize that all the efforts behind construction of the Great Pyramids were absolutely, beyond any doubt “justified”, “valid” and absolutely “right thing to do”.

Some day we would say the same about the LCH, which is trying to catch “statistical memes” (not unlike these LLMs).

This subject of how socially constructed memes infect our minds is quite complicated and subtle, and we have to go slowly and be as careful and accurate as humanly possible.

The memes can be as sophisticated as whole “scientific, (ok, humanitarian) theories”. It ranges from “Statistical Learning”, which is bullshit, to the books titled as “There Are No Facts” from no less than the MIT Press.

The proponents of so-called modern “smart” culture would argue, that every opinion has a merit and has to be taken into account and cannot just be dismissed as mere bullshit, largely because we cannot call bullshit anymore.

There is a guy named Vapnik, highly praised for his theories. He publishes books and lectures at MIT. What is his problem? Well, if we apply the old or even ancient standards for working with texts, we would call bullshit.

The technique of calling bullshit is indeed ancient, and it is about dismissing and discarding everything that follows after a logically or mathematically flawed statement, taken as a fact, and later built upon.

OK, just pure abstract logic and pure abstract mathematics are not enough to call bullshit. There is the only principle we have left with - there must not be a single contradiction with reality or “What Is”.

There are justification for this meta-statement. Everything What Is (at least within this planet) is a result of still ongoing process of “unfolding”, of which biological evolution is a set of sub-process (just like everything else).

When we “look back” we would see a tree-like trails or “footprints of causality”. The point is not in the metaphors but in the “abstract structures” – they are directed graphs.

The absence or even impossibility of having a contradiction to what is a subtle property of this planet being a literal closure under the what we call laws of physics or the physical constraints of the current shared environment.

Everything evolved (every single sub-process) out there is subject to these set of constraints, and nothing else. It is a closed environment. The minds, no batter what bullshitters talk about them, are no exceptions.

As long as we dismiss the constraints of our shared environment, the possibility of having a nice, comfy sect arises. One just have to choose an arbitrary set of beliefs and stick to it.

Another definitive attribute of a sect is that at some level the concepts involved become so abstract and complex, that they switched to “hand waving” and cease any efforts to validate statements at this level.

This threshold after which no one even try to validate anything is precisely what defines bullshit doctrines, “philosophies”, theories and any kind of sectarian bullshit in general - after a certain point it all becomes fugazi.

And, because of inability (too much effort) to validate this still sells exceptionally well.

Now, what about AI.

One would argue that every single model can be traced as a sequence of discrete steps (mathematical expressions and functions as transformations) and so it is possible to validate every single step and thus the whole, sort of mathematical an induction.

The problem is that if we do this we have to dismiss and discard the whole after a single “dubious” step. Yes, one can do hand-waving with the math (it is called “advanced statistics”) and with the code (when one do not understand the data anymore).

This is precisely the point. At some level, after another transformation, the data cease to make sense and cannot be validated anymore. After this threshold one is free to interpret the resulting nonsense anything he likes, and produce any bullshit he wishes.

And this is precisely how the modern sects are defined - they interpret nonsense and produces bullshit, just like any other sect since the beginning of time.

Where is this threshold in AI? When they throw “multi-head attention” and other transformations on the data This is what corresponds to “hand-waving” with the code. One performs operations which one do not understand or even follow.

And, of course, now they can brag, blog, publish about it - no one would spend an effort to validate and to call bullshit.

One more time. These are definitive, well-understood, recurrent social dynamics, based upon inability of the Mind to distinguish between what linguistic and conceptual “input” is valid and what is just socially constructed bullshit.

To do this requires an enormous effort, and most of people choose to give up and accept all the bullshit, especially labeled as “scientific” or “mathematical”.

And, yes, it all comes from the “advanced statistics cancer”, when transformations and the resulting data cease to make sense and just taken as a “belief”.

Now what about measured “performance” of a model, which employ statistical transformations one cannot understand and which makes no sense?

What they evaluate and measure is abstract distances between “points” in an abstract multi-dimentioanl space. Yes, that “abstract space” captures somethong about current language usage and does actual clustering based on “observed” texts (the trainign data).

This “clustering” and the whole resulting structure (after training) does not contain anything beyond what was feed into the process. All the talk about “emergent properties” is a delusion and misintepretation, exactly what people would say about “intelligence” of a parrot (a bird) .

Well, the transforamtions which has been “thrown in” in neural network “architectures” should be considered as added noise, which do not completely ruin the signal. One more time - they measure the abstract distances, so they have a better distances, and only them.

The signal there is the evolved use of a human language as a medium of describing (encoding for a verbalized communication or transmission) particular aspects of What Is. This is the signal. Transformations is added noise.

The results and measurements reflect the fact that there IS a signal in the original training data in the first place. And this is, basically, it.

I hope now one can, at least intuitively, understand how Vapnik is a bullshit, and most of statistical modeling are just sects. Notice that this is not just an opinion. This is literally how it works (at a social level) and why.