• Re: Five different LLM systems agree with my rebuttal of the HP proofs.

    From Mr Flibble@21:1/5 to olcott on Tue Sep 2 18:31:01 2025
    On Tue, 02 Sep 2025 13:18:51 -0500, olcott wrote:

    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own without prompting.
    They also correctly determined that the HP proof decider would be
    correct to reject its input as non-halting.

    <Input to LLM systems>
    Simulating Termination Analyzer HHH correctly simulates its input until:
    (a) Detects a non-terminating behavior pattern:
    abort simulation and return 0.
    (b) Simulated input reaches its simulated "return" statement:
    return 1.

    typedef int (*ptr)();
    int HHH(ptr P);

    int DD()
    {
    int Halt_Status = HHH(DD);
    if (Halt_Status)
    HERE: goto HERE;
    return Halt_Status;
    }

    What value should HHH(DD) correctly return?
    <Input to LLM systems>

    https://claude.ai/share/da9e56ba-f4e9-45ee-9f2c-dc5ffe10f00c

    https://chatgpt.com/share/68939ee5-e2f8-8011-837d-438fe8e98b9c

    https://grok.com/share/c2hhcmQtMg%3D%3D_810120bb-5ab5-4bf8-af21-
    eedd0f09e141


    Gemini had to be forced into do not guess mode https://g.co/gemini/share/4f44c883b348

    ChatGPT 5.0 had to be forced into do not guess mode https://chatgpt.com/share/68abcbd5-cee4-8011-80d7-93e8385d90d8

    Make it clear to the LLMs that HHH is a Halting Problem halt decider
    rather than just a partial decider masquearading as a termination analyser
    and you will get very different responses. But of course you know this
    already because, just like Richard Damon, you are a lying c-nt.

    Pink is not a physical colour.

    /Flibble

    --
    meet ever shorter deadlines, known as "beat the clock"

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to olcott on Tue Sep 2 19:52:03 2025
    On 02/09/2025 19:18, olcott wrote:
    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own  without prompting.

    ... except for a little bit of prompting.

    They also correctly determined that the HP proof decider would
    be correct to reject its input as non-halting.

    HP proof deciders do not have the option to reject their input.
    They *must* decide. Non-deciders aren't deciders. Therefore your
    program was incorrect, and your LLM systems were wrong to say
    otherwise.

    <Input to LLM systems>
    Simulating Termination Analyzer HHH correctly simulates its input
    ^^^^^^^^^^^^^^^^^^^

    Begs the question. (The wrong question, as it happens, but it
    still begs it.)

    <snip>

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to olcott on Tue Sep 2 20:16:40 2025
    On 02/09/2025 20:00, olcott wrote:
    On 9/2/2025 1:52 PM, Richard Heathfield wrote:
    On 02/09/2025 19:18, olcott wrote:
    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own  without prompting.

    ... except for a little bit of prompting.

    They also correctly determined that the HP proof decider would
    be correct to reject its input as non-halting.

    HP proof deciders do not have the option to reject their input.
    That is exactly what a every decider does.

    No, it isn't.

    Only two options accept or reject for every decider.

    No, only one option - to correctly report the halting status of a
    given program with given input.

    A program that can't do that is not an HP proof decider.

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Kaz Kylheku@21:1/5 to olcott on Tue Sep 2 19:50:09 2025
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on
    that network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token prediction
    is going to go down a path that traverses the space of your own texts.

    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?

    --
    TXR Programming Language: http://nongnu.org/txr
    Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
    Mastodon: @[email protected]

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to Kaz Kylheku on Tue Sep 2 21:11:36 2025
    On 02/09/2025 20:50, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on
    that network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token prediction
    is going to go down a path that traverses the space of your own texts.

    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?

    Bear in mind that Olcott's "I have proved" appears to be
    synonymous with "I have asserted".

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Mr Flibble@21:1/5 to olcott on Tue Sep 2 20:20:05 2025
    On Tue, 02 Sep 2025 15:18:06 -0500, olcott wrote:

    On 9/2/2025 2:50 PM, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on that
    network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token prediction
    is going to go down a path that traverses the space of your own texts.

    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?


    It is an objective matter of fact that I have shown that the HP proof
    decider H would be correct to reject the HP proof input D on the basis
    of a D correctly simulated by H.

    Your H says D doesn't halt, D halts.

    Pink isn't a physical colour.

    /Flibble



    --
    meet ever shorter deadlines, known as "beat the clock"

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Mr Flibble@21:1/5 to olcott on Tue Sep 2 20:26:46 2025
    On Tue, 02 Sep 2025 15:23:29 -0500, olcott wrote:

    On 9/2/2025 3:20 PM, Mr Flibble wrote:
    On Tue, 02 Sep 2025 15:18:06 -0500, olcott wrote:

    On 9/2/2025 2:50 PM, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on that
    network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token
    prediction is going to go down a path that traverses the space of
    your own texts.

    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?


    It is an objective matter of fact that I have shown that the HP proof
    decider H would be correct to reject the HP proof input D on the basis
    of a D correctly simulated by H.

    Your H says D doesn't halt, D halts.

    Pink isn't a physical colour.

    /Flibble


    My H says that its input D specifies a sequence of moves that cannot
    possibly reach their own final halt state as proven by D correctly
    simulated by H.

    Your H says D doesn't halt, D halts.

    Pink isn't a physical colour.

    /Flibble





    --
    meet ever shorter deadlines, known as "beat the clock"

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Kaz Kylheku@21:1/5 to olcott on Tue Sep 2 21:27:23 2025
    On 2025-09-02, olcott <[email protected]> wrote:
    On 9/2/2025 2:50 PM, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on
    that network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token prediction
    is going to go down a path that traverses the space of your own texts.

    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?


    It is an objective matter of fact that I have
    shown that the HP proof decider H would be correct
    to reject the HP proof input D on the basis of a
    D correctly simulated by H.

    There is no "would be" operator in mathematics, only "is'.

    "An ellipse would be a rectangle, if it only grew corners." --- What is
    that? If serious, it is a puerile abuse of language, devoid of thought; otherwise it is just math-related jest.

    --
    TXR Programming Language: http://nongnu.org/txr
    Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
    Mastodon: @[email protected]

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to olcott on Tue Sep 2 22:19:59 2025
    On 02/09/2025 21:23, olcott wrote:

    <snip>

    My H says that its input D specifies a sequence of
    moves that cannot possibly reach their own final
    halt state as proven by D correctly simulated by H.

    DD halts.

    If HHH (as written) doesn't claim DD halts, HHH is mistaken.

    If HHH is modified to claim that DD halts, DD will no longer
    halt, so HHH is mistaken.

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Kaz Kylheku@21:1/5 to olcott on Tue Sep 2 21:29:13 2025
    On 2025-09-02, olcott <[email protected]> wrote:
    On 9/2/2025 3:20 PM, Mr Flibble wrote:
    Your H says D doesn't halt, D halts.

    Pink isn't a physical colour.

    /Flibble


    My H says that its input D specifies a sequence of
    moves that cannot possibly reach their own final
    halt state as proven by D correctly simulated by H.

    Your H does not say this when the use of static data in making the abort decision is disabled; it simply fails to return.

    When you introduce the static fuse which splits HHH into two, the
    simulated DD is not the original DD; HHH wrongly reporets that DD halts
    by considering the simulated DD, but the original DD that it's supposed
    to be analyzing actually halts.



    --
    TXR Programming Language: http://nongnu.org/txr
    Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
    Mastodon: @[email protected]

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Mr Flibble@21:1/5 to olcott on Tue Sep 2 21:48:37 2025
    On Tue, 02 Sep 2025 16:34:14 -0500, olcott wrote:

    On 9/2/2025 4:19 PM, Richard Heathfield wrote:
    On 02/09/2025 21:23, olcott wrote:

    <snip>

    My H says that its input D specifies a sequence of moves that cannot
    possibly reach their own final halt state as proven by D correctly
    simulated by H.

    DD halts.

    If HHH (as written) doesn't claim DD halts, HHH is mistaken.

    If HHH is modified to claim that DD halts, DD will no longer halt, so
    HHH is mistaken.


    As I have repeated many times no Turing machine based halt decider can possibly see its actual caller thus cannot possibly report on the
    behavior of its actual caller.

    A halt decider doesn't have to see its actual caller to be able to report
    on its actual caller if a description of its actual caller is passed to it
    as an input -- which is what the Halting Problem proofs do.

    Pink is not a physical colour.

    /Flibble

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to olcott on Tue Sep 2 23:07:18 2025
    On 02/09/2025 22:34, olcott wrote:
    On 9/2/2025 4:19 PM, Richard Heathfield wrote:
    On 02/09/2025 21:23, olcott wrote:

    <snip>

    My H says that its input D specifies a sequence of
    moves that cannot possibly reach their own final
    halt state as proven by D correctly simulated by H.

    DD halts.

    If HHH (as written) doesn't claim DD halts, HHH is mistaken.

    If HHH is modified to claim that DD halts, DD will no longer
    halt, so HHH is mistaken.


    As I have repeated many times no Turing machine
    based halt decider can possibly see its actual
    caller thus cannot possibly report on the behavior
    of its actual caller.

    I've explained to you several times how you can do this. You
    never listen.

    I understand that you don't want to, but "don't want to" is not
    "can't".

    HHH(DD) can't tell if its called from main() or
    called from DD().

    Who cares? You do the job that's in front of you, no matter what
    called you.

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Mr Flibble@21:1/5 to olcott on Tue Sep 2 22:03:35 2025
    On Tue, 02 Sep 2025 16:53:56 -0500, olcott wrote:

    On 9/2/2025 4:48 PM, Mr Flibble wrote:
    On Tue, 02 Sep 2025 16:34:14 -0500, olcott wrote:

    On 9/2/2025 4:19 PM, Richard Heathfield wrote:
    On 02/09/2025 21:23, olcott wrote:

    <snip>

    My H says that its input D specifies a sequence of moves that cannot >>>>> possibly reach their own final halt state as proven by D correctly
    simulated by H.

    DD halts.

    If HHH (as written) doesn't claim DD halts, HHH is mistaken.

    If HHH is modified to claim that DD halts, DD will no longer halt, so
    HHH is mistaken.


    As I have repeated many times no Turing machine based halt decider can
    possibly see its actual caller thus cannot possibly report on the
    behavior of its actual caller.

    A halt decider doesn't have to see its actual caller to be able to
    report on its actual caller if a description of its actual caller is
    passed to it as an input -- which is what the Halting Problem proofs
    do.


    Yet none of these proofs ever considered a simulating halt decider. Thus
    they never noticed the recursive simulation non-halting behavior
    pattern.

    Any infinite recursion manifesting due to the use of an SHD simply
    confirms that the problem is undecidable, well done, you have confirmed
    the extant Halting Problem proofs are correct!

    Pink is not a physical colour.

    /Flibble


    --
    meet ever shorter deadlines, known as "beat the clock"

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to olcott on Tue Sep 2 23:11:48 2025
    On 02/09/2025 22:53, olcott wrote:
    Yet none of these proofs ever considered a simulating
    halt decider.

    Wrong. They considered all deciders.

    Thus they never noticed the recursive
    simulation non-halting behavior pattern.

    They never cared, because it doesn't matter.

    No matter how your decider "works", the proof uses it to break
    itself.

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Kaz Kylheku@21:1/5 to olcott on Tue Sep 2 23:08:21 2025
    On 2025-09-02, olcott <[email protected]> wrote:
    As I have repeated many times no Turing machine
    based halt decider can possibly see its actual
    caller thus cannot possibly report on the behavior
    of its actual caller.

    HHH(DD) can't tell if its called from main() or
    called from DD().

    That's absolutely correct. Expressions in math formulas do not have a relationship to parent expressions in which they are embedded (beyond
    needing certain scoping indications about their free variables).

    So, yes, HHH(DD) cannot tell in what situation it is being evaluated;
    that information must not be available to HHH(DD) if it is to be a
    purely functional evaluation.

    But, that doesn't mean that DD must not contain a functional clone of
    HHH, or else must not apply it to DD.

    --
    TXR Programming Language: http://nongnu.org/txr
    Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
    Mastodon: @[email protected]

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Kaz Kylheku@21:1/5 to olcott on Tue Sep 2 23:15:13 2025
    On 2025-09-02, olcott <[email protected]> wrote:
    On 9/2/2025 4:48 PM, Mr Flibble wrote:
    On Tue, 02 Sep 2025 16:34:14 -0500, olcott wrote:
    As I have repeated many times no Turing machine based halt decider can
    possibly see its actual caller thus cannot possibly report on the
    behavior of its actual caller.

    A halt decider doesn't have to see its actual caller to be able to report
    on its actual caller if a description of its actual caller is passed to it >> as an input -- which is what the Halting Problem proofs do.


    Yet none of these proofs ever considered a simulating
    halt decider.

    False; I just posted today about 1972 paper in which, Tony Hoare and his
    buddy deeply considered the role of simulation in the halting problem.
    (They refer to it as interpretation, which is the same; they envision a
    higher level language being interpreted at the AST level rather than
    compiled to byte code or machine code which is then interpreted.)

    Hoare and his co-author identified a theorem between interpretation and halting: namely that no language powerful enough to encode its own
    interpreter can express an algorithm to decide halting for that
    language. And vice versa: if halting can be decided in a given language,
    it is not possible to express an interpreter for that language in that language.

    Computer scientists have been thinking about simulation/interpretation
    in the context of halting probably since probably not long after
    the problem was identified.

    The most *painfully obvious* way to explore whether a program halts is a combination of running it and looking for patterns in what it's doing:
    e.g. identifying some run-away recursion or infinite loop.

    The odds that yau are the first to be aware of something *painfully
    obvious* are vanishingly small.

    You might as well claim that you invented decimal to binary conversion,
    or Bubble Sort.

    --
    TXR Programming Language: http://nongnu.org/txr
    Cygnal: Cygwin Native Application Library: http://kylheku.com/cygnal
    Mastodon: @[email protected]

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Damon@21:1/5 to olcott on Tue Sep 2 22:40:15 2025
    On 9/2/25 2:18 PM, olcott wrote:
    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own  without prompting.
    They also correctly determined that the HP proof decider would
    be correct to reject its input as non-halting.

    <Input to LLM systems>
    Simulating Termination Analyzer HHH correctly simulates its input until:
    (a) Detects a non-terminating behavior pattern:
        abort simulation and return 0.
    (b) Simulated input reaches its simulated "return" statement:
        return 1.

    being a misleading statement by ignoring the actual case of (unless it
    makes a mistake)

    (c) or continues running forever unless it can find an ACTUAL
    non-halting pattern, that follows from the behavior of the direct
    exectution of the input not halting, or reaches a final state.

    This is what happens since if HHH every thinks it finds a pattern, it
    will also return a 0 to DD, so that DD will halt.


    typedef int (*ptr)();
    int HHH(ptr P);

    int DD()
    {
      int Halt_Status = HHH(DD);
      if (Halt_Status)
        HERE: goto HERE;
      return Halt_Status;
    }

    What value should HHH(DD) correctly return?
    <Input to LLM systems>

    Making the error of presuming that an answer exists, the questions
    should be what answer CAN HHH(DD) return to be correct.


    https://claude.ai/share/da9e56ba-f4e9-45ee-9f2c-dc5ffe10f00c

    https://chatgpt.com/share/68939ee5-e2f8-8011-837d-438fe8e98b9c

    https://grok.com/share/c2hhcmQtMg%3D%3D_810120bb-5ab5-4bf8-af21-
    eedd0f09e141

    Gemini had to be forced into do not guess mode https://g.co/gemini/share/4f44c883b348

    ChatGPT 5.0 had to be forced into do not guess mode https://chatgpt.com/share/68abcbd5-cee4-8011-80d7-93e8385d90d8



    When we correct the prompt to not hide the possibility of HHH not
    answering, they all find the rignt answer, that HHH can never return a
    correct answer.

    Your ignoring this just proves your natural stupidity.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Richard Heathfield@21:1/5 to Fred. Zwarts on Wed Sep 3 08:53:12 2025
    On 03/09/2025 08:49, Fred. Zwarts wrote:
    Op 02.sep.2025 om 20:18 schreef olcott:
    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own  without prompting.

    Incorrect. You feed the AI with the incorrect information that
    HHH is capable of detecting a non-termination pattern.
    We know that cannot be true, it is exactly the claim that needs a
    proof.
    Proving something true with the assumption that it is true is an
    invalid circular reasoning.

    All true, but in my experience ChatGPT is pretty sharp about
    this. Tell it that HHH returns 1 if DD halts and 0 if it doesn't,
    show it DD's source, and watch it laugh in your face.

    --
    Richard Heathfield
    Email: rjh at cpax dot org dot uk
    "Usenet is a strange place" - dmr 29 July 1999
    Sig line 4 vacant - apply within

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Fred. Zwarts@21:1/5 to All on Wed Sep 3 09:57:08 2025
    Op 02.sep.2025 om 23:37 schreef olcott:
    On 9/2/2025 4:27 PM, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    On 9/2/2025 2:50 PM, Kaz Kylheku wrote:
    On 2025-09-02, olcott <[email protected]> wrote:
    Five different LLM systems figured out the recursive simulation

    You've written a lot of incorrect texts and published them on
    that network.

    LLMs are trained on texts scraped from the Internet. They are trained
    by repeatedly adjusting their parameters until they predict the next
    word in a given piece of text, which could easily be a fragment of a
    Peter Olcott posting from comp.theory.

    When you prompt LLMs with your usual terminology, the token prediction >>>> is going to go down a path that traverses the space of your own texts. >>>>
    Can you /prove/ that Claude, Gemini and others are /not/ predicting
    tokens based on your own internet texts?


    It is an objective matter of fact that I have
    shown that the HP proof decider H would be correct
    to reject the HP proof input D on the basis of a
    D correctly simulated by H.

    There is no "would be" operator in mathematics, only "is'.


    Still this is much more advanced than both return
    values are the wrong answer.

    Although in both cases there is a correct answer: the opposite one.
    Note that the input for both deciders is different.
    This proves the failure of trying to create a decider that always gives
    the correct answer. No such decider exists. There is always an input for
    which it returns an incorrect answer.

    We now know the correct
    answer and the architecture of the design to find it.

    There is a correct answer, but that is not the one given by the decider
    for the input based on this decider.
    Sometimes we can construct another decider that finds the correct answer
    for this input, but that we can construct another input for which the
    new decider fails as well.


    "An ellipse would be a rectangle, if it only grew corners." --- What is
    that? If serious, it is a puerile abuse of language, devoid of thought;
    otherwise it is just math-related jest.




    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Fred. Zwarts@21:1/5 to All on Wed Sep 3 09:49:35 2025
    Op 02.sep.2025 om 20:18 schreef olcott:
    Five different LLM systems figured out the recursive simulation
    non-halting behavior pattern on their own  without prompting.

    Incorrect. You feed the AI with the incorrect information that HHH is
    capable of detecting a non-termination pattern.
    We know that cannot be true, it is exactly the claim that needs a proof. Proving something true with the assumption that it is true is an invalid circular reasoning.

    They also correctly determined that the HP proof decider would
    be correct to reject its input as non-halting.

    <Input to LLM systems>
    Simulating Termination Analyzer HHH correctly simulates its input until:
    (a) Detects a non-terminating behavior pattern:
        abort simulation and return 0.

    Here olcott injects the incorrect information that a HHH is able to
    detect non-termination behaviour pattern.
    It cannot be true, as proven in the halting theorem.
    So, the set of HHH that meets this requirement is empty. This makes the conclusion of the AI vacuous.
    The example of the HHH that is presented by olcott in Halt7.c, does not
    meet the requirement. It pretends to detect a non-termination pattern,
    but it fails, because no such pattern is present in the input. Only a
    finite recursion.
    HHH fails to analyse the conditional branch instructions encountered
    during the simulation When it stops the simulation, it has not yet
    proven that the other branches would not have been followed if the
    simulation would be continued correctly.
    This bug makes that HHH incorrectly sees a finite recursion as a non-termination pattern.

    (b) Simulated input reaches its simulated "return" statement:
        return 1.

    typedef int (*ptr)();
    int HHH(ptr P);

    int DD()
    {
      int Halt_Status = HHH(DD);
      if (Halt_Status)
        HERE: goto HERE;
      return Halt_Status;
    }

    What value should HHH(DD) correctly return?

    Irrelevant question. No HHH exists that correctly detects any
    non-termination pattern. The return value of a non-existent HHH is not relevant.

    <Input to LLM systems>

    https://claude.ai/share/da9e56ba-f4e9-45ee-9f2c-dc5ffe10f00c

    https://chatgpt.com/share/68939ee5-e2f8-8011-837d-438fe8e98b9c

    https://grok.com/share/c2hhcmQtMg%3D%3D_810120bb-5ab5-4bf8-af21-
    eedd0f09e141

    Gemini had to be forced into do not guess mode https://g.co/gemini/share/4f44c883b348

    ChatGPT 5.0 had to be forced into do not guess mode https://chatgpt.com/share/68abcbd5-cee4-8011-80d7-93e8385d90d8



    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)