Path: ...!news.mixmin.net!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Mikko <mikko.levanto@iki.fi>
Newsgroups: comp.theory
Subject: Re: Even Google AI Overview understands me now --- different execution traces have different behavior !!!
Date: Fri, 4 Oct 2024 12:05:29 +0300
Organization: -
Lines: 135
Message-ID: <vdob4p$5sfp$1@dont-email.me>
References: <vdgpbs$2nmcm$1@dont-email.me> <vdgqhn$2nmcm$2@dont-email.me> <7c6cede5237e3eafee262c74dd1a1c90c6b2ffbb@i2pn2.org> <vdhblt$2qm1j$2@dont-email.me> <cafee8d7a14edd7b1d76bb706c36eef06ae82896@i2pn2.org> <vdi0f8$2u1aq$1@dont-email.me> <53a60609211a04a123adafa525bac39b5cbc6959@i2pn2.org> <vdjlum$38t86$4@dont-email.me> <bf681f4404a7df8e3ffc2059dcd7c5c302aeeff1@i2pn2.org> <vdkud3$3ipp4$1@dont-email.me> <vdm1tl$3npme$1@dont-email.me> <vdn0nv$3sa9k$1@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Injection-Date: Fri, 04 Oct 2024 11:05:29 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="b1c47a2304f6556c191fc9394816e07b";
	logging-data="193017"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX18bS+mvtSuGZe+BNgSbVGcQ"
User-Agent: Unison/2.2
Cancel-Lock: sha1:ULkaf6SdAr6+1+CssM9k4YemMtU=
Bytes: 7380

On 2024-10-03 21:01:51 +0000, olcott said:

> On 10/3/2024 7:15 AM, Mikko wrote:
>> On 2024-10-03 02:09:39 +0000, olcott said:
>> 
>>> On 10/2/2024 5:48 PM, Richard Damon wrote:
>>>> On 10/2/24 10:39 AM, olcott wrote:
>>>>> On 10/2/2024 6:08 AM, Richard Damon wrote:
>>>>>> On 10/1/24 7:26 PM, olcott wrote:
>>>>>>> On 10/1/2024 12:58 PM, joes wrote:
>>>>>>>> Am Tue, 01 Oct 2024 12:31:41 -0500 schrieb olcott:
>>>>>>>>> On 10/1/2024 8:09 AM, joes wrote:
>>>>>>>>>> Am Tue, 01 Oct 2024 07:39:18 -0500 schrieb olcott:
>>>>>>>>>>> On 10/1/2024 7:19 AM, olcott wrote:
>>>>>>>>>>>> https://www.google.com/search?
>>>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>> q=Termination+Analyzer+H+is+Not+Fooled+by+Pathological+Input+D&sca_esv=889093c5cb21af9e&sca_upv=1&source=hp&ei=Muf7ZpOyMZHfwN4PwYL2gAc&iflsig=AL9hbdgAAAAAZvv1Qg04jNg2ze170z3a8BSGu8pA29Fj&ved=0ahUKEwiTk7zkk-2IAxWRL9AFHUGBHXAQ4dUDCBg&uact=5&oq=Termination+Analyzer+H+is+Not+Fooled+by+Pathological+Input+D&gs_lp=Egdnd3Mtd2l6IjxUZXJtaW5hdGlvbiBBbmFseXplciBIIGlzIE5vdCBGb29sZWQgYnkgUGF0aG9sb2dpY2FsIElucHV0IERIAFAAWABwAHgAkAEAmAEAoAEAqgEAuAEDyAEA- 
>>>>>>>> 
>>>>>>>>>> AEC-AEBmAIAoAIAmAMAkgcAoAcA&sclient=gws-wiz
>>>>>>>>>>> https://chatgpt.com/share/66fbec5c-7b10-8011-9ce6-3c26424cb21c
>>>>>>>>>> It sounds like it’s trained on your spam. LLMs don’t know anything
>>>>>>>>>> anyway.
>>>>>>>> 
>>>>>>>>>> I asked: „What happens when HHH tries to simulate itself?”
>>>>>>>>>> ChatGPT: [my comments in brackets]
>>>>>>>> 
>>>>>>>>>> ### 1. **Self-Invocation in HHH**
>>>>>>>>>> If `HHH` calls itself (i.e., `HHH(HHH)`), it creates an infinite loop
>>>>>>>>>> of calls unless there is some form of termination condition built into
>>>>>>>>>> `HHH`.
>>>>>>>>>> Without such a condition, `HHH` would never return, resulting in non-
>>>>>>>>>> termination.
>>>>>>>>>> [So not a decider.]
>>>>>>>>>> 
>>>>>>>>>> ### 2. **Halting Criteria**
>>>>>>>>>> In a well-designed termination analyzer, `HHH` would need to have
>>>>>>>>>> specific logic to prevent endless recursion:
>>>>>>>> 
>>>>>>>>>> [Not present.]
>>>>>>>>>> 
>>>>>>>>>> ### 3. **Practical Consequences**
>>>>>>>>>> If `HHH` lacks the ability to handle self-referential calls correctly:
>>>>>>>>>> - **Stack Overflow**: Continuous self-invocation will eventually
>>>>>>>>>> exhaust the stack memory, leading to a stack overflow error.
>>>>>>>>>> [Not considered in theory.]
>>>>>>>>>> - **Incorrect Results**: If not designed carefully, it may lead to an
>>>>>>>>>> incorrect conclusion about halting, misrepresenting itself as either
>>>>>>>>>> halting or non-halting.
>>>>>>>>>> 
>>>>>>>>>> ### 4. **Best Practices for Termination Analyzers**
>>>>>>>>>> To avoid these pitfalls, a termination analyzer should implement:
>>>>>>>>>> - **Guard Clauses**: Checks at the beginning of the function to prevent
>>>>>>>>>> unnecessary self-simulation.
>>>>>>>>>> - **State Management**: A way to track the functions being analyzed,
>>>>>>>>>> allowing `HHH` to recognize if it's being invoked with itself and act
>>>>>>>>>> accordingly.
>>>>>>>>>> 
>>>>>>>>>> ### Conclusion In summary, if `HHH` tries to simulate itself without
>>>>>>>>>> proper safeguards,
>>>>>>>>>> it risks falling into an infinite recursion or causing a stack
>>>>>>>>>> overflow. A robust design would include checks to prevent
>>>>>>>>>> self-invocation, ensuring the termination analyzer can handle all
>>>>>>>>>> cases, including its own function,
>>>>>>>>>> gracefully.
>>>>>>>>>> 
>>>>>>>>> *You can continue this conversation with ChatGPT*
>>>>>>>> I’ll leave that to you.
>>>>>>>>> Click on the link and see how it answers this question:
>>>>>>>>> Is H a Halt Decider for D?
>>>>>>>> You should feed it our objections.
>>>>>>>> If you believe in it only when you prompt it, it is not suited as an
>>>>>>>> authority (fallacious anyway).
>>>>>>>> 
>>>>>>> 
>>>>>>> You feed it your objections.
>>>>>>> It will tell you how and why you are wrong.
>>>>>>> 
>>>>>> 
>>>>>> No, it will tell you something that matches the words you told it.
>>>>>> 
>>>>>> You don't seem to understand what Large Language Models are.
>>>>>> 
>>>>>> 
>>>>>> You seem to forget that LLM know nothing of the "truth", only what 
>>>>>> matches their training data.
>>>>>> 
>>>>>> They are know to be liars, just like you.
>>>>> 
>>>>> In the case of their evaluation of my work they are correct.
>>>>> 
>>>>> *Try this yourself*
>>>>> 
>>>>> https://chatgpt.com/share/66fbec5c-7b10-8011-9ce6-3c26424cb21c
>>>>> 
>>>>> Does HHH have to abort its emulation of DDD to prevent the infinite 
>>>>> execution of DDD?
>>>>> 
>>>>> 
>>>> 
>>>> Try asking it
>>>> 
>>>> Why is it ok for HHH to say that DDD doesn't halt when it will when run?
>>>> 
>>> 
>>> You guys try to pretend that the pathological relationship between
>>> HHH and DDD does not make any difference to the behavior of DDD
>>> knowing full well that it does make a difference.
>> 
>> The behaviour of DDD is what DDD does if executed. As DDD takes no input
>> its behavour is always the same. What does "make a difference" mean
>> in this context?
>> 
> 
> The behavior of the directly executed DDD is essentially the
> behavior of what would otherwise be infinite recursion except
> that the second recursive call has already been aborted.
> 
> This is not the same as the behavior of DDD correctly emulated
> by the same emulator that it calls where HHH cannot rely on DDD
> being aborted by any other process than itself.

So the behaviour of DDD "correctly" emulated is not the behaviour
of DDD.

> Do an entirely different thing and expecting the same results
> is just as insane as doing the same thing and expecting different results.

No, it is not. Different things can produce the same result.
although in this case they don't. Although HHH is not a halt decider,
for some inputs it does give the same answer as a halt decider would.

-- 
Mikko