• Math AI Vegans hate the Druids Cauldron [MakazhanAlpamys / Soup] (Re: So it was a race against Buzzard. Well done!)

    From Mild Shock@janburse@fastmail.fm to sci.logic,comp.lang.prolog,sci.math on Wed Nov 11 00:17:13 2026
    From Newsgroup: sci.logic

    Hi,

    How it started,

    Mathematics of the Future
    5.6 AI-vegan mathematicians
    https://math.mit.edu/~etingof/futuremath.pdf

    How its going:

    Soup Layer Streaming for LLM fine-tuning
    #4 Repository of the Day
    https://github.com/MakazhanAlpamys/Soup

    I fear one or two kids might teach it
    math (for example via Sutskever/Megill
    approach) and soon become new Einsteins,

    developing wormhole warpspeed math.

    Have Fun

    Bye

    Mild Shock schrieb:
    Hi,

    There's a larger project called FALSE (Formalising
    Arithmetic in the Lean Software Environment)
    explicitly aimed at "making progress towards
    -aformalising the Langlands program" . And a

    Twitter thread claims Claude autoformalized
    the Fermat proof "in just 11 days," with Kevin
    Buzzard now wanting to formalize the Langlands
    program because he's "concerned it might
    contain subtle errors"

    Now look at this:

    https://github.com/fukubillueda-web/epsilon_FML

    https://github.com/fukubillueda-web/epsilon_SML

    According to the related arvix paper:

    So the provenance chain is: Dwork (attributed,
    unpublished) raA Langlands (used it) raA Lakkis (partial
    proof, sign issue) raA Ueda (completes it, 2026). That's
    the "mixtape" you're looking for rCo the mathematical
    content passed through at least three hands before
    Ueda's Lean formalization.

    And then: "OpenAI ChatGPT was used extensively in
    the development of this proof." The paper says ChatGPT
    was used "to check calculations and compare them with
    the arguments of Dwork, Langlands, and Lakkis, and
    detect inconsistencies in intermediate versions of
    the proof."

    The author "checked the final mathematical
    arguments and assumes responsibility."

    Bye

    Mild Shock schrieb:
    Hi,

    OpenAI has announced enhancing their Math Swarms
    by an interpretability accelerator prompt recently
    discovered in a Terrence Tao workshop.

    So they are now takling:

    Mathematician W. Hugh Woodin Explains Continuum Hypothesis
    https://www.youtube.com/watch?v=ZC7wglkBWMM

    Infinite fun with infinite worlds
    https://ethz.ch/en/news-and-events/eth-news/news/2020/08/infinite-fun-with-the-infinite-worlds.html


    A partial Lean 4 formalization is already in
    the making. But the hurdles are high. Some
    speaker of the effort said:

    The computer might find a new tamagochi knottheory
    lemma, which will then allow SpaceX to fly to
    workholes, but the hurdle will be hight that

    a single human mathematician mind might
    ingest, understand and interpret it.

    Bye


    Mild Shock schrieb:
    Hi,

    Hyperspectral measurements for inter-calibration,
    reached a significant milestone this summer.

    The naval observatory only uses fidget spinners
    to measure time, so forget Colorado.

    Bye

    Ross Finlayson schrieb:

    So, there's at least three accounts of what can define
    "measurement of
    time", in the micro, meso, and macro.

    Then these days pretty much the world relies on "the naval
    observatory
    in Colorado" if not so much "the clock tower in Greenwich", as with
    regards to ye olde UTC, CUT, "Zulu", and so on, time.

    About relativity theory and "measuring rods and clocks", then, it's
    similar that "metric and norm" for "length and distance" also has
    about the Planckian what are "iota-values" as it would be, or, "Pauli >>> -a> Plancks", or "Dirac's Pauli Plancks", of measuring space, vis-a-vis, >>> -a> measuring time.


    So, there are at least three accounts of what defines measuring time, >>> -a> and at least three accounts of what defines measuring space.


    Then, since relativity theory is defined by what does that,
    it has the various ways, to so fulfill doing that.


    Yeah, I know, most "professional physicists" only have one of those
    on their little note-card with which they passed graduate school.

    Mild Shock schrieb:
    Hi,

    Ross Finlayson schrieb:
    Or shaves pennies.

    X-Machines Virtual Machine (XVMrao) is a neurosymbolic virtual AI >>>> -a>>> processor which combines neural processes with symbolic reasoning. >>>> -a>>> https://aws.amazon.com/marketplace/pp/prodview-6luxq22pgmehe

    Cost/hour: $200,000.00

    LoL

    Bye

    P.S.: If it could do some quant trading magic, one would
    possibly pay so much. But Logtalk is simply too lame:

    Version release notes
    XVM Engine v10.2.4 is the full engine capable of running
    all XVM and Logtalk programs, excluding for logtalk tools.

    Mild Shock schrieb:
    Hi,

    The 9122171.18175435 is a little offending, what if one
    keeps a federal secret after the 6 fraction digit?

    Bye

    Mild Shock schrieb:
    Hi,

    Ok, leaving the beaten path of my Prolog system
    probing, and look at some newer beast.

    This looks bad:

    ?- format('~6f', [pi**14]), nl.
    9122171.18175435

    Expected result:

    ?- format('~6f', [pi**14]), nl.
    9122171.181754

    Bye

    BTW: Tested using this test tester:

    X-Machines Virtual Machine (XVMrao) is a neurosymbolic virtual AI >>>>>> processor which combines neural processes with symbolic reasoning. >>>>>> https://aws.amazon.com/marketplace/pp/prodview-6luxq22pgmehe

    Mild Shock schrieb:
    See also:
    https://prolog-lang.org/ImprovementsForum/0110-format.html








    --- Synchronet 3.22a-Linux NewsLink 1.2