[arin-ppml] Slow summer....
Mike Burns
mike at iptrading.com
Fri Jul 24 08:43:23 EDT 2026
Hi Daryll,
“Autonomous” “Surprisingly elegant” “Groundbreaking”
I think a visit to the footnote links would be beneficial.
But it is besides the point.
We know LLMs are getting better.
Is it your position that it is okay for a poster to use an LLM to read the list and create an argument that the poster then proffers to the list?
Or that we should rely on LLM-detection outrunning LLM content generation to filter LLM-sourced content?
My underlying point remains. It does not matter the source of the argument made in a post.
Only the argument matters, so why use LLM to filter out arguments?
Regards,
Mike
From: Daryll Swer <contact at daryllswer.com>
Sent: Friday, July 24, 2026 8:29 AM
To: Mike Burns <mike at iptrading.com>
Cc: Mike Burns <mike at sum.net>; Fernando Frediani <fhfrediani at gmail.com>; arin-ppml at arin.net
Subject: Re: [arin-ppml] Slow summer....
<https://mailtrack.io/trace/mail/0e38cf9e4c22ec8565dda7470ced1608ba492eba.png?u=2153471>
Yes, I have read about using LLMs to detect LLMs, although at this point I think we humans can still do that effectively.
Long-term, LLM-based spam detection is the way.
Yes, AI has solved several significant unsolved math problems. Notable breakthroughs include:
The Planar Unit Distance Problem: An internal OpenAI model autonomously solved this 80-year-old problem posed by Paul Erdős. The AI disproved a central conjecture in discrete geometry, producing a 125-page "chain of thought" that top mathematicians—including Fields Medalist Timothy Gowers—verified as groundbreaking. [1, 2, 3, 4]
The Erdős Primitive Set Conjecture: An AI model solved this problem in about 80 minutes, using classical number theory tools to produce a proof celebrated by experts as surprisingly elegant. [1]
Various Erdős Problems: Google DeepMind and various researchers have used AI tools to solve dozens of lesser-known or partial unsolved problems and conjectures left behind by Paul Erdős. [1, 2]
Matrix Multiplication: AI systems like AlphaEvolve autonomously discovered faster matrix multiplication algorithms, achieving the first major improvement in 50 years. [1]
One can't say LLMs solved these problems with zero human involvement or intervention on any level; plenty of human work hours were involved in part of the process.
--
Best Regards
Daryll Swer
Website: daryllswer.com <https://l.shortlink.es/l/4e5ef9300a2918b334ac7bfa070d3caf4d0f3c81?u=2153471>
On Fri, 24 Jul 2026 at 17:51, Mike Burns <mike at iptrading.com <mailto:mike at iptrading.com> > wrote:
HI Daryll,
Yes, I have read about using LLMs to detect LLMs, although at this point I think we humans can still do that effectively.
For example, the text below above my salutation is obviously LLM generated and speaks to your point about the current limits of LLMs.
Yes, AI has solved several significant unsolved math problems. Notable breakthroughs include:
* The Planar Unit Distance Problem: An internal OpenAI model autonomously solved this 80-year-old problem posed by Paul Erdős. The AI disproved a central conjecture in discrete geometry, producing a 125-page "chain of thought" that top mathematicians—including Fields Medalist Timothy Gowers—verified as groundbreaking. [1 <https://l.shortlink.es/l/96bef505df831f771a42ef829ecfbd452520339c?u=2153471> , 2 <https://l.shortlink.es/l/c9b561e9daa96c46e6f2b1d7a83a23121bbacef1?u=2153471> , 3 <https://l.shortlink.es/l/051e49bbc625d94c55a72c151cca455850d487ca?u=2153471> , 4 <https://l.shortlink.es/l/3e037dbf3af7e7da0fb8d4fae0d0a00056a55f95?u=2153471> ]
* The Erdős Primitive Set Conjecture: An AI model solved this problem in about 80 minutes, using classical number theory tools to produce a proof celebrated by experts as surprisingly elegant. [1 <https://l.shortlink.es/l/eb26696ce8b221f7a75945b036e8dae1fbe700ad?u=2153471> ]
* Various Erdős Problems: Google DeepMind and various researchers have used AI tools to solve dozens of lesser-known or partial unsolved problems and conjectures left behind by Paul Erdős. [1 <https://l.shortlink.es/l/fe177c039a09f23d02a382313dfd19f0e6b468d5?u=2153471> , 2 <https://l.shortlink.es/l/b11cffceb1ec2e45a90ceb48ce999b779f33b9cc?u=2153471> ]
* Matrix Multiplication: AI systems like AlphaEvolve autonomously discovered faster matrix multiplication algorithms, achieving the first major improvement in 50 years. [1 <https://l.shortlink.es/l/7939d270707152cf18d12d3f066477a2e361e98f?u=2153471> ]
Regards,
Mike
From: Daryll Swer <contact at daryllswer.com <mailto:contact at daryllswer.com> >
Sent: Friday, July 24, 2026 8:14 AM
To: Mike Burns <mike at iptrading.com <mailto:mike at iptrading.com> >
Cc: Mike Burns <mike at sum.net <mailto:mike at sum.net> >; Fernando Frediani <fhfrediani at gmail.com <mailto:fhfrediani at gmail.com> >; arin-ppml at arin.net <mailto:arin-ppml at arin.net>
Subject: Re: [arin-ppml] Slow summer....
Fair enough, and I will refer to it as LLMs so as to avoid confusion.
Agreed, as of 2026, at least.
Surely you have heard about LLMs solving unsolved math problems already?
That's a misleading statement, to say the least. These LLMs are tools that constantly require human intervention to move forward. This means humans used LLMs to assist in solving some problems (including mathematics).
LLMs did not solve these problems "single-handedly" with zero human intervention. I know, sci-fi is cool and I'd like a time machine, but let's stick to reality in this forum/channel.
I am more concerned with the obvious path of these LLMs in the future.
My point stands. You don’t know if an LLM wrote this reply to you.
Agreed. Typically, some website hosting platforms use LLM itself to help spot or flag LLM-generated text content/spam.
So perhaps, for these mailing lists - the "spam filters" need an upgrade to support LLM-based detection/flagging with clear UI/UX signalling to subscribers so they know if an email body contains potential LLM-generated content.
--
Best Regards
Daryll Swer
Website: daryllswer.com <https://l.shortlink.es/l/edc6fb5ccdcfdb521745067f235d30b8e90add56?u=2153471>
On Fri, 24 Jul 2026 at 17:36, Mike Burns <mike at iptrading.com <mailto:mike at iptrading.com> > wrote:
Hi Daryll,
Fair enough, and I will refer to it as LLMs so as to avoid confusion.
Surely you have heard about LLMs solving unsolved math problems already?
I am more concerned with the obvious path of these LLMs in the future.
My point stands. You don’t know if an LLM wrote this reply to you.
Regards,
Mike
From: Daryll Swer <contact at daryllswer.com <mailto:contact at daryllswer.com> >
Sent: Friday, July 24, 2026 8:03 AM
To: Mike Burns <mike at sum.net <mailto:mike at sum.net> >
Cc: Mike Burns <mike at iptrading.com <mailto:mike at iptrading.com> >; Fernando Frediani <fhfrediani at gmail.com <mailto:fhfrediani at gmail.com> >; arin-ppml at arin.net <mailto:arin-ppml at arin.net>
Subject: Re: [arin-ppml] Slow summer....
In this peculiar type of forum how would you know whether you have seen it?
Which medical/neuroscience institute has verified that "AI" (something adjacent to, equivalent to, or surpassing the human brain) exists in the public domain?
I think you're conflating "AI" (which no authority has yet verified exists in the public domain) with LLMs and slop.
There's nothing to see regarding something that does not currently exist in the public domain.
--
Best Regards
Daryll Swer
Website: daryllswer.com <https://l.shortlink.es/l/38fb91dd4be6732199adbfa97771a871b07dbfeb?u=2153471>
On Fri, 24 Jul 2026 at 17:25, Mike Burns <mike at sum.net <mailto:mike at sum.net> > wrote:
Hi Daryll,
In this peculiar type of forum how would you know whether you have seen it?
Regards,
Mike
Sent from my Galaxy
-------- Original message --------
From: Daryll Swer <contact at daryllswer.com <mailto:contact at daryllswer.com> >
Date: 7/24/26 3:57 AM (GMT-05:00)
To: Mike Burns <mike at iptrading.com <mailto:mike at iptrading.com> >
Cc: Fernando Frediani <fhfrediani at gmail.com <mailto:fhfrediani at gmail.com> >, arin-ppml at arin.net <mailto:arin-ppml at arin.net>
Subject: Re: [arin-ppml] Slow summer....
> the best arguments if they come from AI
I've yet to see such an incident. And to my knowledge, "AI" isn't publicly available yet; we only have LLMs.
--
Best Regards
Daryll Swer
Website: daryllswer.com <https://l.shortlink.es/l/8ca45c7b170af55b883b29adc1a353d0ae727d00?u=2153471>
On Fri, 24 Jul 2026 at 02:33, Mike Burns via ARIN-PPML <arin-ppml at arin.net <mailto:arin-ppml at arin.net> > wrote:
Hi Fernando,
I welcome your thoughts.
It may get harder and harder to distinguish between polish and outright verbatim echoing of AI posts by a human interlocutor.
The open mailing list is designed to solicit the best arguments, should we ignore the best arguments if they come from AI?
It seems likely to be that we will be dealing with this situation at some point.
Regards,
Mike
From: ARIN-PPML <arin-ppml-bounces at arin.net <mailto:arin-ppml-bounces at arin.net> > On Behalf Of Fernando Frediani
Sent: Thursday, July 23, 2026 1:49 PM
To: arin-ppml at arin.net <mailto:arin-ppml at arin.net>
Subject: Re: [arin-ppml] Slow summer....
Surely no AI should post in a human discussion mail list. It doesn't make much sense
If people want to use AI to polish their text no problem, as long messages are reviewed and sent by humans.
Mail list should not be treated a destination for automated systems to send message, just for human discussions.
Fernando
On 7/23/2026 2:09 PM, Mike Burns via ARIN-PPML wrote:
Hello List,
Based on a current discussion on the AFRINIC policy development list, I wanted to discuss the use of AI to post to policy lists.
Because these are open forums handled by email, it is quite trivial for any AI to be echoed to the list by a real person, or even post directly in the guise of a real person.
I have used translators to help me post, I expect more and more will use AI to polish, if not create their posts.
So simply banning use of AI is not the answer, but it brings up some questions.
Is this the camel’s nose of AI into human government?
Should we empower an AI to post directly, or be callable like ask grok?
Should we treat the AI posts just like any other post?
I am not discussing abuse of the list through repeated AI posts restating the same argument and I assume lists have existing tools to exclude certain posters who do abuse.
Any thoughts on using AI on stakerholder governance lists?
Regards,
Mike
_______________________________________________
ARIN-PPML
You are receiving this message because you are subscribed to
the ARIN Public Policy Mailing List (ARIN-PPML at arin.net <mailto:ARIN-PPML at arin.net> ).
Unsubscribe or manage your mailing list subscription at:
https://lists.arin.net/mailman/listinfo/arin-ppml <https://l.shortlink.es/l/9590903c190250a43b5717a7174cc551403a5948?u=2153471>
Please contact info at arin.net <mailto:info at arin.net> if you experience any issues.
_______________________________________________
ARIN-PPML
You are receiving this message because you are subscribed to
the ARIN Public Policy Mailing List (ARIN-PPML at arin.net <mailto:ARIN-PPML at arin.net> ).
Unsubscribe or manage your mailing list subscription at:
https://lists.arin.net/mailman/listinfo/arin-ppml <https://l.shortlink.es/l/4603a5037ef898c97ea1be7f419c039296b20b8d?u=2153471>
Please contact info at arin.net <mailto:info at arin.net> if you experience any issues.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20260724/8c557f96/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.gif
Type: image/gif
Size: 42 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20260724/8c557f96/attachment-0001.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image002.jpg
Type: image/jpeg
Size: 27983 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20260724/8c557f96/attachment-0002.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image003.jpg
Type: image/jpeg
Size: 7008 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20260724/8c557f96/attachment-0003.jpg>
More information about the ARIN-PPML
mailing list