LLMs as Revenge of Absurdist Behavioralism

"Behavioralism" is a bullshit pseudo-science from last century, and LLMs are a Behavioralist's wet dream. Quoting from Wikipedia:

Behavioralism attempts to explain human behavior from an unbiased, neutral point of view, focusing only on what can be verified by direct observation, preferably using statistical and quantitative methods.^[2][3] In doing so, it rejects attempts to study internal human phenomena such as thoughts, subjective experiences, or human well-being.^[4].

As much pseudo-science does, Behavioralism likes to dress itself with legitimate tools, like statistics or mathematics more generally, but those are a sham, and are always misused. Mathematics does not "prove" anything about the real world, theories can only be disproved by counter-evidence.

A philosophy of science as espoused by Behavioralism is necessarily limited to “phenomenological descriptions”. There can be no explanations whatsoever, only “I saw it did a this or that”, which is incredibly convenient for the grifters, they don’t have to explain anything because you can’t.

LLMs match this exactly. Statistical properties are extracted from a bunch of words. The model is "words in, words out" (WIWO), and nothing in-between is considered. Sure, people like to hand-wave this or that "deep structure" bullshit, but not as an explanation, merely as a defense against having to explain anything rigorously in way that could be refuted by evidence.

In reality, the same collection of words can mean different things, and different collections of words can mean the same thing. Some collections of words are thus ambiguous, and without reference to the actual meaning and not the arrangement of words, it's impossible to disambiguate.

Impossible. Not, inconvenient, not sometimes off, not might work for you. It's impossible for the tool. So that necessitates some other tool. If it's a program, you can try to run it, but that's still not a guarantee (go ask Microsoft how many severe security vulnerabilities they fixed last month). Most of the text shat out by LLMs is merely browsed (if at all) by an extremely gullible and easily tricked human. Weee...

As I mentioned in The LLM Grift Playbook:

The genius of the LLM grift is two-fold: To an extent never seen before, they...

1. Create a truly personalized experience. It's most likely that no two people see the same thing; and
2. They hijack the gullibility of a human's anthropomorphizing machinery.

If you're new to the idea of scientific explanations, David Deutsch has a great introduction: A New Way To Explain Explanation

Another good option is reading all of these books: The Art of Thinking Clearly, Thinking Fast and Slow, and Predictably Irrational: The Hidden Forces That Shape Our Decisions.

Anyway, core take-away is this: LLMs are utter bullshit, always have been and always will be. Behavioralism had its day, far too long of one, and so thankfully, we can be pretty certain that the LLM-hysteria will also go the way that all pseudo-science does.