[ih] Comments re the packet radio discussion

Tue Apr 21 19:14:24 PDT 2026

Glad to see that AI is on the problem!

My first encounter with network technology involved AI.  I had a lot to 
learn about networking.  Somewhat long story...sorry, just hit delete 
now if you aren't interested.

It started in 1963.  In high school, I was interested in STEM (acronym 
not yet in existence though), had joined the school's Radio Club, gotten 
a amateur radio ("ham") license, and somehow gotten involved in 
something (in the US) called the National Traffic System (NTS - that 
acronym did exist).

NTS was a network.  Actually an Internet.   Actually a multi-technology 
Internet of interconnected networks.  It had been evolving since shortly 
after World War II.  Packets (think telegrams or emails) were 
transported around the world.  There were LANs covering cities, Regional 
nets over the US, and gateways to nets covering other parts of the 
world.  Even a TransContinentalCorps (TCC) that acted as a WAN.

I learned about all sorts of the technical mechanisms that operated 
inside the network: routing, flow control, addresses, error control, 
protocols for retransmissions, compression techniques, etc.

NTS typically carried messages such as Birthday Greetings, e.g., from a 
soldier stationed in Guantanamo Bay (address: "GITMO") to a family 
member back home somewhere in the US.  Although most of the distance was 
covered by radio, using Morse Code, the final link might use a different 
"LAN", e.g., a telephone call or even simply driving to the house and 
delivering a piece of paper.  NTS still operates today.  More details at 
https://en.wikipedia.org/wiki/National_Traffic_System if you're curious.

When I was in Lick's orbit at MIT, that experience made it easy for me 
to understand the ARPANET.  Later also the Internet.   Techniques were 
different but the problems remained the same.

At one point circa 1975, ARPA asked for our help with a critical DoD 
need.  They wanted computers to be able decode hand-sent Morse code.  No 
one ever explained exactly why to us tech grunts, but we knew Morse was 
being used by both the good and bad guys.  The AI technology of the day 
was "Expert Systems".   We were tasked to use whatever we thought would 
help solve the problem.

That started a project which lasted several years, and produced a system 
which did about as well, possibly even better, than a human 
operator.  From my high school experience, I acted as the "Expert", and 
we had a series of staff and student projects to make the PDP-10 do the 
same things I had learned to do while participating in the radio 
network.  ARPA was happy.  DoD was happy.  No ham operator could afford 
the millions of dollars to have their own PDP-10 system, but the 
government didn't have that constraint.

The project even appeared in a 1990 Mitre report I think of as a 
compendium of "ARPA's Greatest Hits".  See Chapter 22 of 
https://apps.dtic.mil/sti/tr/pdf/ADA239925.pdf  (FYI, ARPANET is Chapter 
20 of same report).

The reason I mention this is because Gemini's interpretation reminded me 
of the behavior we noted in that AI system -- in 1975. The radio 
environment was pretty nasty, with conditions such as interference or 
lightning strikes randomly causing "dropped packets", which the network 
had to mitigate.  The software we built had parameters which we could 
change to modify how confident it had to be before issuing a translation 
of the "Morse" it was hearing. It was in effect a "knob" that you could 
use to tell the system how hard to work to do its job.

One day, just for the hell of it, we set the confidence knob as low as 
it could go.  Then we tuned the radio to a frequency with no one 
transmitting at all, just atmospheric noise.

The system tried real hard.  The processor was no doubt overheating.  
The memory cores were melting.  But it produced words.  It created 
sentences.  It created paragraphs.  It described things and the content 
actually made sense.  All of it was nonsense of course, pure gibberish.

Unless... perhaps we were actually listening to some extraterrestrial 
communications?  We should have informed SETI.  Klingonese or Cardassian 
-- didn't matter, it is still proof - "they do exist!"

Today's AI is apparently even more creative in interpreting gibberish.  
It even explains how it's reaching its conclusions!

/Jack

On 4/21/26 13:39, vinton cerf wrote:
> gemini:
>
> It looks like you’ve found the "right-side up" version of the text you 
> shared earlier! Seeing it this way makes the patterns much clearer, 
> though it’s still definitely dancing on the edge of "uncanny valley" 
> English.
>
> This isn't a standard language; it’s a form of *pseudo-English* or 
> *Asemic writing*. It’s designed to /look/ like a coherent paragraph to 
> your brain at a distance, but when you zoom in, the words dissolve.
>
> ------------------------------------------------------------------------
>
>
>       The Linguistic Breakdown
>
> While it’s mostly gibberish, we can "reverse engineer" the intended 
> words based on the letter structures. It feels like a corrupted 
> digital file or a transcript from an AI that’s having a minor stroke.
>
> *Key "Deciphered" Fragments:*
>
>  *
>
>     *"medete ont to emoe ‘entmaxe o bear ed outs"*
>
>      o
>
>         /Translation:/ "Meditate on to come, examine to bear the outs."
>
>  *
>
>     *"eae onkwsxs oF beats nood nat Iebow"*
>
>      o
>
>         /Translation:/ "Are knows of beats down that I bow."
>
>  *
>
>     *"edt of paatectire ant Moold"*
>
>      o
>
>         /Translation:/ "The architecture and mold" (or "world").
>
>  *
>
>     *"guivkscer his note a buat cence?"*
>
>      o
>
>         /Translation:/ "Receiver his note a [about?] sense?"
>
> ------------------------------------------------------------------------
>
>
>       What is this, exactly?
>
> Since you aren't sure where it came from, here are the three most 
> likely suspects:
>
> 1.
>
>     *A "Markov Chain" Generator:* This is a classic result of an
>     algorithm being fed a lot of old literature (like Poe or
>     Lovecraft) and trying to predict the next letter. It gets the
>     "rhythm" of the words right, but the spelling fails.
>
> 2.
>
>     *Optical Character Recognition (OCR) Gore:* If someone scanned a
>     very old, physically damaged book using cheap software, the
>     computer might have guessed the letters incorrectly, resulting in
>     this "alphabet soup."
>
> 3.
>
>     *Weirdcore / Digital Art:* There is a massive online subculture
>     that creates "cryptic artifacts"—text that looks haunted or
>     forbidden—just to spark curiosity.
>
>
>       My "Best Guess" at the Meaning
>
> If I had to summarize the "vibe," it reads like a *distorted manual 
> for a ritual or a meditation guide.* It uses words like /meditate, 
> examine, architecture, slumberous,/ and /note./ It’s trying to tell 
> you to look at something ("examine to bear the outs"), but the 
> "signal" is breaking up.
>
>     *Technical Note:* Notice how it uses "u" and "n" or "p" and "b"
>     interchangeably? That’s a hallmark of text that was meant to be
>     read upside down (the "Flip Text" from before). It was likely
>     written specifically to be readable—albeit weirdly—from both
>     directions.
>
> *Where did you encounter this?* Was it in a text file, a hidden 
> comment in code, or maybe a strange social media "bot" account? 
> Knowing the medium might help me figure out if there's a hidden key to 
> the gibberish.
>
>
> On Tue, Apr 21, 2026 at 4:29 PM Carsten Bormann via Internet-history 
> <internet-history at elists.isoc.org> wrote:
>
>     Hmm, I really wanted to get-bunter-ebay already, and I could
>     definitely use a few Denknuxe (*).
>
>     So I asked deepseek, and it suggested this might have been scanned
>     upside down (or maybe as ink leaked through the page?).
>     It couldn’t really help any further though except suggesting a
>     service like https://www.upsidedowntext.com/, which gets me (not
>     quite what I was looking for):
>
>     ɯǝpǝʇǝ ouʇ ʇo ǝɯoǝ ‘ǝuʇɯɐxǝ o qǝɐɹ ǝp onʇs
>     ǝɐǝ ouʞʍsxs oℲ qǝɐʇs uoop uɐʇ Iǝqoʍ
>     ┴p ‘ ǝpʇ qǝǝ ɐɯǝ ʇɐʌǝ uǝɹuɐbǝʇ Ԁ∀ƃɥ¿ ˥ᴉǝs
>     | ɐlqɐɥǝɹǝs ɐɐpʇo >ɐǝʞ ┴onɐɹs slǝɐᴉuᴉɐs ǝnɹos˙ poʇᴉʇʎ uq(
>     ǝǝʇ ʇo ʇooʇʇǝ ǝp¿ ‘s)ɐoʇzɐzɐ ǝpʇ oɟ dɐɐʇǝɔʇᴉɹǝ ɐuʇ Woolp «
>     ʍɥǝ oʇ pǝuʞunxǝ פǝsp ouᴉx ɐǝp ǝɟouʇǝɹɹs¿ opʎ sǝ ǝʌoʇʇǝp |
>     ǝɐǝ ʍǝǝ ɥǝuʇɐƃ( ƃǝʇqnuʇǝɹǝqɐʎ ǝps ┴Hɹǝɯǝlƃbns' oʇ qoʍ ƃɐʇǝp oʇɐ ƃᴉɐo
>     ‾ ʇoɐʇoʍƃ ɟǝsɔoɯʇsǝƃzǝ ǝpʇ ʇopʎ˙ o8˙ ǝɯoɹǝxǝ oʇɥɐ ʇǝxᴉɐb ʇǝ )ǝp
>     Iʍɐs’: ɟǝuʇᴉʎ ɾ
>     op Sᴉʞʍ ʇɐls sonɹʇɐsʇ ǝqɐɹobɹoǝɐʇ ɯǝu ɯsʇoʎǝ ƃnᴉʌʞsɔǝɹ ɥᴉs uoʇǝ ɐ
>     qnɐʇ ɔǝuɔǝ¿
>
>     Maybe some other AI can help here…
>     Need to find a student.
>
>     Grüße, Carsten
>
>     (*) The first sentence of my message is easier to understand if
>     you speak German.
>
>     > On 2026-04-21, at 22:10, Jack Haverty via Internet-history
>     <internet-history at elists.isoc.org> wrote:
>     >>
>     >>    medete ont to emoe ‘entmaxe o bear ed outs
>     >>    eae onkwsxs oF beats nood nat Iebow
>     >>    TD ‘ edt bee ame tave nernaqet PAgh? Lies
>     >>    | albaheres aadto <aek TouaRs sleainias euros. dotity nb)
>     >>    eet to tootte ed? ‘s(aotzaza edt of paatectire ant Moold «
>     >>    whe ot Denknuxe Gesd onix aed efonterrs? ody se evotted |
>     >>    eae wee hentag) getbunterebay eds THremelgqus, ot bow gated
>     ota giao
>     >>    _ toatowg fescomtsegze edt tody. o8. emorexe otha texiaq te
>     (ed Iwas’: fentiy j
>     >>    od Sikw tals sourtast ebaroqroeat men mstoye guivkscer his
>     note a buat cence?
>     >>
>     >>
>     >>    I guess I'm still not allowed to read this PRTN....
>     >>
>     >>    Medete ont to you too!
>     >>
>     >>    /Jack
>
>     -- 
>     Internet-history mailing list
>     Internet-history at elists.isoc.org
>     https://elists.isoc.org/mailman/listinfo/internet-history
>     -
>     Unsubscribe:
>     https://app.smartsheet.com/b/form/9b6ef0621638436ab0a9b23cb0668b0b?The%20list%20to%20be%20unsubscribed%20from=Internet-history
>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 665 bytes
Desc: OpenPGP digital signature
URL: <http://elists.isoc.org/pipermail/internet-history/attachments/20260421/5f0a85f2/attachment.asc>