• Re: The future Numa Brains will be gorgeous (Re: Train yourself to become a nosomatic AI chirurgeon)

    From Mild Shock@janburse@fastmail.fm to sci.physics.relativity,sci.physics on Mon Sep 28 14:15:02 2026
    From Newsgroup: sci.physics

    Hi,

    Yesterday I had the feeling P(doom E-graphs) > 10%,
    since they have E-boost intiatives where they use
    parallism and more heuristics, also tapping into GPUs,

    TPUs, NPUs, etc.. you name it. So Prolog might sooner
    or later feel the wind of E-boost Schinkansen? Now
    I have the feeling the E-graph community is still

    in the nowhere of la la land. Reading:

    recording out of egraphs
    https://www.philipzucker.com/

    WTF is "recording out" ? Look at that beauty of
    SWI-Prolog unifiable/3 predicate. It can even deal
    with cyclic terms:

    ?- unifiable(f(X,X,X),f(Z,b,Y), L).
    L = [Y=b, X=b, Z=X].

    ?- unifiable(f(X,Y),f(f(A,Y),f(B,X)), L).
    L = [Y=f(B, X), X=f(A, Y)].

    You can even code unifiable/3 in 100% Prolog itself,
    BTW I am not saying it would be pure. But it wont need
    any extensions that go beyond ISO core standard.

    You could even experiment with a unification that
    does minimization (*), but most Prolog systems don't
    do that for their unification:

    ?- unifiable(X, f(f(X)), L).
    L = [X=f(f(X))].

    Bye

    (*)
    Gerhard Huet was all in for minimizations, but the sad
    news is while terms have minimal forms, equations don't
    really have unique minimal forms, this transpired only later.

    RESOLUTION D'EQUATIONS DANS DES LANGAGES D'ORDRE 1,2,.rCorCo, https://www.researchgate.net/publication/213879499

    Algorithms for equivalence and reduction to minimal form https://hal.science/hal-01239749v1/document

    Mild Shock schrieb:
    Hi,

    Non-uniform memory access (NUMA) is a computer
    memory design used in multiprocessing, where the
    memory access time depends on the memory location
    relative to the processor.

    The future Numa Brains will be gorgeous. In like
    3-4 years from now, they might have 3.2 TB RAM
    on a SOC, by not using HBM but HBF:

    High Bandwidth Flash: What Is It Good For? https://www.youtube.com/watch?v=3nTpW52nioI

    There might be new profession like nosomatic
    AI chirurgeon (I made that up):

    Tiny engine, immense model
    https://github.com/JustVugg/colibri

    The Brain page, Explore: the measured expert
    atlas of GLM-5.2 drawn as a cortex. 13,260
    characterised experts in ten regions (Python,
    SQL, mathematics, poetry, law, ChineserCa);

    position is measured routing affinity, not a
    learned embedding. We might want a brain to
    amputate from the Python expert sections, so

    that we can write code without a slow
    CPython runtime.

    Bye


    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Mild Shock@janburse@fastmail.fm to sci.physics.relativity,sci.physics on Mon Sep 28 14:15:22 2026
    From Newsgroup: sci.physics



    Hi,

    Yesterday I had the feeling P(doom E-graphs) > 10%,
    since they have E-boost intiatives where they use
    parallism and more heuristics, also tapping into GPUs,

    TPUs, NPUs, etc.. you name it. So Prolog might sooner
    or later feel the wind of E-boost Schinkansen? Now
    I have the feeling the E-graph community is still

    in the nowhere of la la land. Reading:

    recording out of egraphs
    https://www.philipzucker.com/

    WTF is "recording out" ? Look at that beauty of
    SWI-Prolog unifiable/3 predicate. It can even deal
    with cyclic terms:

    ?- unifiable(f(X,X,X),f(Z,b,Y), L).
    L = [Y=b, X=b, Z=X].

    ?- unifiable(f(X,Y),f(f(A,Y),f(B,X)), L).
    L = [Y=f(B, X), X=f(A, Y)].

    You can even code unifiable/3 in 100% Prolog itself,
    BTW I am not saying it would be pure. But it wont need
    any extensions that go beyond ISO core standard.

    You could even experiment with a unification that
    does minimization (*), but most Prolog systems don't
    do that for their unification:

    ?- unifiable(X, f(f(X)), L).
    L = [X=f(f(X))].

    Bye

    (*)
    Gerhard Huet was all in for minimizations, but the sad
    news is while terms have minimal forms, equations don't
    really have unique minimal forms, this transpired only later.

    RESOLUTION D'EQUATIONS DANS DES LANGAGES D'ORDRE 1,2,.rCorCo, https://www.researchgate.net/publication/213879499

    Algorithms for equivalence and reduction to minimal form https://hal.science/hal-01239749v1/document

    Mild Shock schrieb:
    Hi,

    Non-uniform memory access (NUMA) is a computer
    memory design used in multiprocessing, where the
    memory access time depends on the memory location
    relative to the processor.

    The future Numa Brains will be gorgeous. In like
    3-4 years from now, they might have 3.2 TB RAM
    on a SOC, by not using HBM but HBF:

    High Bandwidth Flash: What Is It Good For? https://www.youtube.com/watch?v=3nTpW52nioI

    There might be new profession like nosomatic
    AI chirurgeon (I made that up):

    Tiny engine, immense model
    https://github.com/JustVugg/colibri

    The Brain page, Explore: the measured expert
    atlas of GLM-5.2 drawn as a cortex. 13,260
    characterised experts in ten regions (Python,
    SQL, mathematics, poetry, law, ChineserCa);

    position is measured routing affinity, not a
    learned embedding. We might want a brain to
    amputate from the Python expert sections, so

    that we can write code without a slow
    CPython runtime.

    Bye


    --- Synchronet 3.22a-Linux NewsLink 1.2