Re: The LLMentalist Effect (AI & psychic's con)

with <20240412134952.391fb054793f0d1946a29ce6@g{oogle}mail.com> Anton
Shepelev wrote:

Ben Collver quoted:
>

<https://softwarecrisis.dev/letters/llmentalist/>

*SKIP* [  8 lines   2 levels deep]

There is no reason to believe that it thinks or reasons--indeed,
every AI researcher and vendor to date has repeatedly emphasised that
these models don't think.

What say ye to:

Oh, look at that.  We've got True Believer.  Useful feature -- sceptical
one.

1.  LLM's can playing chess, that is they understand the rules of the
game, because all the training set is not nearly sufficient if it were
used merely statistically: <https://parrotchess.com/>

(disclaimer:  I'm not immersed and I can't be pressed to research this,
but) Simplest possible somewhat problematic play: KQ vs KR.  By the
chess theory:  blacks always loose.  To delay inevitable checkmate
blacks must never separate K and R.

This was approached with different perspective:  (a) setup all possible
combinations;  (b) build graph connecting would be previous
combinations;  (c) blacks choose move that would delay checkmate.
Simple, eh?

When NI faced this it was a catastrophy.  Whites expect opponent that
would play by theory and intuition.  They don't expect graph.  Then
blacks make unexpected moves (thus bringing havoc onto whites plans).
During that dance some combination repeates three times.  Then, by chess
rules, it's a tie.

2.  Emergent World Models and Latent Variable Estimation in
Chess-Playing Language Models: <https://arxiv.org/abs/2403.15498>

. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -- I, for one, smell oxymoron.

Is it possible to build such graph for all positions and combinations of
pieces?  Turns out, yes, it's possible just by playing.

I fear they cannot be explained by the Forer effect.

For sceptics, the Forer effect is on recieving part.  It has nothing to
do with whatever is sold.

--
Torvalds' goal for Linux is very simple: World Domination
Stallman's goal for GNU is even simpler: Freedom