Dear alt.usage.english,
comp.ai.philosophy,
The L.L.Ms. are stupid. Never forget that. I was wondering, when the
same king ruled both Iceland, and Sweden. I made the naive question to
the search engine as follows.
when was iceland part of sweden
And the /search assist/ answer that came back was
Iceland was never part of Sweden; it was under Norwegian rule from
1262/64 and then Danish rule from 1380 until it became independent in
1944.
which may or may not be correct, I haven't checked yet. Then I asked
when was the same king ruling both denmark and sweden
and the /search assist/ answer that came back was
The same monarch ruled both Denmark and Sweden during the Kalmar Union
period, such as King John, who reigned over Denmark (1481–1513),
Norway (1483–1513), and Sweden (1497–1501).
which also may or may not be correct, I haven't checked either.
Anyway, a human insufferable know-it-all could have corrected my ques-
tion, and followed up with an /achtually/ stating that we are merely
talking about four years, from fourteen hundred and ninety seven to fif-
teen hundred and one [1].
Now, the members of comp.ai.philosophy, who believe the L.L.Ms. are not
only insufferable know-it-alls, but also stochastically correct, will
have to explain to me, in details, how this can be corrected once and
for all. That is, this mistake must not happen again, with /any/ L.L.M. home trained or not. It stands to reason, that once something is poin-
ted out to be wrong to an L.L.M., it has to be possible to teach all of
them forever. No matter who is doing the training, and with what bud-
get.
Because once this is corrected, we also know how to correct for all fu-
ture L.L.Ms. any misunderstanding, forever. And we can be safe to ass-
ume that the people training L.L.Ms. are all reasonable and well meaning people who will make sure all past mistakes are not forgotten, and no L.
L.M. will be trained on incorrect data, forever.
Best wishes, and happy L.L.M. training!
[1] Members of alt.usage.english are welcome to correct my counting of
the years.
when was the same king ruling both denmark and sweden
and the /search assist/ answer that came back was
The same monarch ruled both Denmark and Sweden during the Kalmar Union
period, such as King John, who reigned over Denmark (1481–1513),
Norway (1483–1513), and Sweden (1497–1501).
Den 30.09.2026 kl. 22.51 skrev Johann 'Myrkraverk' Oskarsson:
when was the same king ruling both denmark and sweden
and the /search assist/ answer that came back was
The same monarch ruled both Denmark and Sweden during the Kalmar
Union period, such as King John, who reigned over Denmark
(1481–1513), Norway (1483–1513), and Sweden (1497–1501).
We have never had a king named "John" or any other name that might be
represented as "John". The first king of the Kalmar Union was
formally Erik af Pommern who was son of Margrete 1st, and it was in
reality her that ruled.
We have never had a king named "John" or any other name that might be
represented as "John". The first king of the Kalmar Union was
formally Erik af Pommern who was son of Margrete 1st, and it was in
reality her that ruled.
Google calls him Charles XIV John, and he was a Bernadotte. But that's
the wrong historical period. Delete that guess.
Ah, here's the right one. Hans, King of Denmark, sometimes called John
or Johannes.
tion, and followed up with an /achtually/ stating that we are merely
talking about four years, from fourteen hundred and ninety seven to fif-
teen hundred and one [1].
Dear alt.usage.english,Johann, you do realise that LLMs are populated with human knowledge,
comp.ai.philosophy,
The L.L.Ms. are stupid. Never forget that. I was wondering, when the
same king ruled both Iceland, and Sweden. I made the naive question to
the search engine as follows.
right? It's text ('knowledge') generated by humans. So, LLMs are no
stupider than their input. (I am side-stepping the delicate issue of
humans who ask these questions of AIs, rather than researching their questions the long-hand method.)
My point was that when the "A.I." /knows/ two different points of a
factiod [1] it still cannot /connect the dots/ to make an overall ass- >essment to answer a larger question.
Unlike a human insufferable know-it-all.
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
Anyway, a human insufferable know-it-all could have corrected my ques-
tion, and followed up with an /achtually/ stating that we are merely
talking about four years, from fourteen hundred and ninety seven to fif-
teen hundred and one [1].
OTOH, more than once I got answers where the model apparently
tried to guess my intend, and I had to write in the next prompt,
|Please answer my question literally.
Similarly, Ancient Egypt had a hereditary class of scribes that
administered a civil-service bureaucracy.[18]
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
My point was that when the "A.I." /knows/ two different points of a
factiod [1] it still cannot /connect the dots/ to make an overall ass-
essment to answer a larger question.
Unlike a human insufferable know-it-all.
On December 14, 2023, Google DeepMind's FunSearch paired a large
language model with an automated evaluator to solve an open
challenge in extremal combinatorics by discovering the largest known
mathematical constructions for the cap set problem. - It bypassed
traditional brute-force limits by connecting the dots between
algorithmic computer science and high-dimensional vector space
geometry, mapping code generation outputs directly onto abstract
geometric spaces that human intuitive leaps had failed to bridge.
In September 2026, Anthropic's Claude solved a decades-old
"holy grail" problem in probability theory by producing a
formally verified Lean proof that resolves the dying percolation
conjecture (\theta(p_c) = 0) for dimensions 3 through 10. - The
model connected the dots by using truths from algebraic topology,
specifically persistent homology, to analyze infinite path spaces
in statistical mechanics, revealing macro-level phase geometric
invariants that probability experts could not mathematically link.
On September 8, 2026, OpenAI's unreleased next-generation multi-agent
system tackled the 90-year-old Navier-Stokes existence and smoothness
Millennium Problem by generating a 166-page analytical proof that
demonstrates how a singularity can form in an initially smooth fluid
flow. - It connected the dots by applying discrete number theory
and arithmetic dynamics to complex fluid mechanics, recognizing a
hidden structural symmetry in prime-factor bounds that governs the
exact chaotic behavior of non-linear differential equations, dots
which human experts failed to connect before.
We are also not told -- at least I have never seen such texts -- how
long it took the L.L.M. to /solve/ these puzzles, nor how many false
starts it took the human team behind the machine, before they got some-
thing they could use to gaslight regular people with in the news.
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
We are also not told -- at least I have never seen such texts -- how
long it took the L.L.M. to /solve/ these puzzles, nor how many false
starts it took the human team behind the machine, before they got some-
thing they could use to gaslight regular people with in the news.
Here's one example:
Claude was prompted:
|The problem is to compute the Six-particle (hexagon)
|amplitude in planar N=4 SYM at nine loops.
. It did the computation during several days for about USD 1500 using
two different approaches and making sure the results agree. So far
humans were only able to do eight loops, which already took years.
So this is the first time this nine loop calculation was ever done.
Human experts later confirmed Claude's results to be correct.
Newsgroups: alt.usage.english,comp.ai.philosophy,corel.coreldrawx3 Followup-To: alt.usage.english,comp.ai.philosophy
I knew that intelligent people have high reactance. Therefore, I
used a little psychological trick to get you to discredit yourself
by posting in wrong newsgroups repeatedly. Ha, ha, you fell for it!
So let me get this straight, some people coded up some specific particle
loop into an L.L.M., then used the verbal directions as a /front end/ to
make people believe Claude did it all on its own?
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
So let me get this straight, some people coded up some specific particle
loop into an L.L.M., then used the verbal directions as a /front end/ to
make people believe Claude did it all on its own?
What the reports said is they did not do any such preparation.
A physicist issued a challenge, saying, he'd be really impressed
if Claude could do A or B. They asked Claude, which it was more
likely to tackle and then gave the prompt I quoted before. After
that only management prompts followed, like
|I'm going to sleep and won't be available for another several
|hours. Keep working on this until I tell you to stop. Give me
|updates every 4-6 hours.
In July 2026, mathematician and startup co-founder Dmitry Rybin used
OpenAI's ChatGPT (specifically the GPT-5.6 Pro model) to disprove
the 30-year-old Dinitz-Garg-Goemans conjecture in graph theory.
After chatting with the model about the conjecture he used four
prompts.
|Do a breakthrough and find a structured counterexample.
|Continue the search.
|Continue searching.
|Enough of partial results, let's finish.
(These quotes are paraphrased; the exact wording may vary slightly.)
ChatGPT falsified the conjecture after five hours.
Another example is the journalist Matthew Sparkes asking ChatGPT
about a dice problem, how to design five dices so that each player
only has to throw his die once to always get a fair clear order
for the five players (no one has to throw twice and everyone has
the same chance for each position). ChatGPT found a new solution
after 15 minutes.
So, today many reports claim that they only gave the system a
small prompt and any following prompts were motivational prompts
like "Continue! You can do it!".
Some prompts need to provide more information to the chatbot if the
query comes from a very special field where the chatbot cannot find
the required information on his own. Like Ioannis Tsiokos who said,
"I gave Astra the theory, I gave it the laws, I gave it the theorems,
the Lean code, the papers". He gave it all this information about
nonperiodic tilings of the /plane/, and the chatbot then found the
first ever monotile to cover the /space/ aperiodically.
To be clear, that's five hours of actual datacentre time, and not just >someone sitting in front of a laptop for five hours, getting at most ten >seconds of datacentre time?
Which only means they're asking the machine for more randomized output
of gibberish. It means nothing.
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
To be clear, that's five hours of actual datacentre time, and not just
someone sitting in front of a laptop for five hours, getting at most ten
seconds of datacentre time?
I assume it's realtime, and I guess approximately equivalent
to data center time during which often one main chatbot uses
several agents which run in parallel.
Which only means they're asking the machine for more randomized output
of gibberish. It means nothing.
What Claude just did, proving the Kozma-Nitzan conjecture to
then prove the Dying percolation conjecture (see Wikipedia)
for all dimensions is something a human mathematician probably
would get the Fields medal for.
jectures into a machine, and got randomized gibberish out, that just
happens to be a valid proof. How is that impressive?
If the machine could trivially solve simple conjectures, it would have
solved Collatz' Conjecture years ago. Yet nobody has written about that
to my knowledge.
So how many centuries do you think it will take for a machine to gene-
rate a valid proof of Collatz' Conjecture? Can a human do it in a num-
ber of years instead? And what about the power bill for the datacentre
in the mean time? Who pays that?
On 10/1/2026 2:27 PM, occam wrote:
Dear alt.usage.english,Johann, you do realise that LLMs are populated with human knowledge,
comp.ai.philosophy,
The L.L.Ms. are stupid. Never forget that. I was wondering, when the >>> same king ruled both Iceland, and Sweden. I made the naive question to >>> the search engine as follows.
right? It's text ('knowledge') generated by humans. So, LLMs are no
stupider than their input. (I am side-stepping the delicate issue of
humans who ask these questions of AIs, rather than researching their
questions the long-hand method.)
Dear occam,
My point was that when the "A.I." /knows/ two different points of a
factiod [1] it still cannot /connect the dots/ to make an overall ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
On 2/10/2026 2:29 a.m., Johann 'Myrkraverk' Oskarsson wrote:
On 10/1/2026 2:27 PM, occam wrote:
My point was that when the "A.I." /knows/ two different points of
a factiod [1] it still cannot /connect the dots/ to make an overall
ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
So this will be the next benchmark to watch for: when AI can perform
like a HIK-I-A -- only way, way more insufferable? I begin to sense
what "could destroy us all" means.
On 02/10/26 12:20, Ross Clark wrote:
On 2/10/2026 2:29 a.m., Johann 'Myrkraverk' Oskarsson wrote:
On 10/1/2026 2:27 PM, occam wrote:
My point was that when the "A.I." /knows/ two different points of
a factiod [1] it still cannot /connect the dots/ to make an overall
ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
So this will be the next benchmark to watch for: when AI can perform
like a HIK-I-A -- only way, way more insufferable? I begin to sense
what "could destroy us all" means.
Arrogance seems to be a key feature of AI responses. "I know the exact
reason for your problem. Here is the single unique reason for the
failure." It then goes on to give an explanation that turns out to be incorrect. Whey you point out its error, it responds "Yes, you are
right. This is because ...", and it comes up with another incorrect answer.
On 02/10/2026 05:04, Peter Moylan wrote:
On 02/10/26 12:20, Ross Clark wrote:
On 2/10/2026 2:29 a.m., Johann 'Myrkraverk' Oskarsson wrote:
On 10/1/2026 2:27 PM, occam wrote:
My point was that when the "A.I." /knows/ two different points of
a factiod [1] it still cannot /connect the dots/ to make an overall
ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
So this will be the next benchmark to watch for: when AI can perform
like a HIK-I-A -- only way, way more insufferable? I begin to sense
what "could destroy us all" means.
Arrogance seems to be a key feature of AI responses. "I know the exact
reason for your problem. Here is the single unique reason for the
failure." It then goes on to give an explanation that turns out to be
incorrect. Whey you point out its error, it responds "Yes, you are
right. This is because ...", and it comes up with another incorrect answer. >>
Peter, does that [admitting its error] sound like arrogance? A human
would pretend he had been misinterpreted or misquoted (PDT comes to
mind). An AI is more like an autistic savant. It takes no offence, but
does not condescend either. It just plows on.
[Has anyone had the experience of a human having the last word with an
AI? (No.) That is because it regards that every statement that is typed
as something it needs to react to. ]
On 10/2/2026 4:13 PM, occam wrote:
[Has anyone had the experience of a human having the last word with an
AI? (No.) That is because it regards that every statement that is typed
as something it needs to react to. ]
Isn't that how Sam Altman codes them?
Or instructs his human minions to
code the A.I? So can we infer that Sam Altman is an insufferable know- it-all, and believes that is normal human behaviour [sic]?
On 02/10/2026 05:04, Peter Moylan wrote:
On 02/10/26 12:20, Ross Clark wrote:
On 2/10/2026 2:29 a.m., Johann 'Myrkraverk' Oskarsson wrote:
On 10/1/2026 2:27 PM, occam wrote:
My point was that when the "A.I." /knows/ two different points
of a factiod [1] it still cannot /connect the dots/ to make an
overall ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
So this will be the next benchmark to watch for: when AI can
perform like a HIK-I-A -- only way, way more insufferable? I
begin to sense what "could destroy us all" means.
Arrogance seems to be a key feature of AI responses. "I know the
exact reason for your problem. Here is the single unique reason for
the failure." It then goes on to give an explanation that turns out
to be incorrect. Whey you point out its error, it responds "Yes,
you are right. This is because ...", and it comes up with another
incorrect answer.
Peter, does that [admitting its error] sound like arrogance?
A human would pretend he had been misinterpreted or misquoted (PDT
comes to mind). An AI is more like an autistic savant. It takes no
offence, but does not condescend either. It just plows on.
[Has anyone had the experience of a human having the last word with--
an AI? (No.) That is because it regards that every statement that is
typed as something it needs to react to. ]
On 02/10/26 18:13, occam wrote:
On 02/10/2026 05:04, Peter Moylan wrote:
On 02/10/26 12:20, Ross Clark wrote:
On 2/10/2026 2:29 a.m., Johann 'Myrkraverk' Oskarsson wrote:
On 10/1/2026 2:27 PM, occam wrote:
My point was that when the "A.I." /knows/ two different points
of a factiod [1] it still cannot /connect the dots/ to make an
overall ass- essment to answer a larger question.
Unlike a human insufferable know-it-all.
So this will be the next benchmark to watch for: when AI can
perform like a HIK-I-A -- only way, way more insufferable? I
begin to sense what "could destroy us all" means.
Arrogance seems to be a key feature of AI responses. "I know the
exact reason for your problem. Here is the single unique reason for
the failure." It then goes on to give an explanation that turns out
to be incorrect. Whey you point out its error, it responds "Yes,
you are right. This is because ...", and it comes up with another
incorrect answer.
Peter, does that [admitting its error] sound like arrogance?
I didn't quote the full response. In fact it doesn't admit its error. Sometimes, as here, it praises its interlocutor, but
it doesn't go as
far as saying it was wrong. In some cases it contradicts something it
earlier said was absolutely true, but in such cases it avoids mentioning
its earlier claim.
A human would pretend he had been misinterpreted or misquoted (PDT
comes to mind). An AI is more like an autistic savant. It takes no
offence, but does not condescend either. It just plows on.
Yes, I agree.
[Has anyone had the experience of a human having the last word with
an AI? (No.) That is because it regards that every statement that is
typed as something it needs to react to. ]
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted: >>Anyway, a human insufferable know-it-all could have corrected my ques- >>tion, and followed up with an /achtually/ stating that we are merely >>talking about four years, from fourteen hundred and ninety seven to fif- >>teen hundred and one [1].
OTOH, more than once I got answers where the model apparently
tried to guess my intend, and I had to write in the next prompt,
|Please answer my question literally.
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted:
Yet, I am not impressed. It looks like some mathematicians put con-
jectures into a machine, and got randomized gibberish out, that just
happens to be a valid proof. How is that impressive?
How is a valid proof randomized gibberish?
If the machine could trivially solve simple conjectures, it would have
solved Collatz' Conjecture years ago. Yet nobody has written about that
to my knowledge.
The Collatz conjecture is by no means simple to prove or disprove.
So how many centuries do you think it will take for a machine to gene-
rate a valid proof of Collatz' Conjecture? Can a human do it in a num-
ber of years instead? And what about the power bill for the datacentre
in the mean time? Who pays that?
I don't know.
On 10/1/2026 5:27 PM, Stefan Ram wrote:
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted: >>> Anyway, a human insufferable know-it-all could have corrected my ques-
tion, and followed up with an /achtually/ stating that we are merely
talking about four years, from fourteen hundred and ninety seven to fif- >>> teen hundred and one [1].
OTOH, more than once I got answers where the model apparently
tried to guess my intend, and I had to write in the next prompt,
|Please answer my question literally.
Dear Stefan,
Usually, when that happens to me, it seems to want to make up the quest-
ion. As in, answer something completely different. Sometimes, when I
ask a search engine for a definition of a word, the /ai assist/ will de-
fine a completely different word.
So I have trained myself never to read the /ai assist/ summary when ask-
ing a dictionary, and always click merriam-webster, oxford, cambridge,
or some other well known human-edited dictionary.
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted: >>We are also not told -- at least I have never seen such texts -- how
long it took the L.L.M. to /solve/ these puzzles, nor how many false
starts it took the human team behind the machine, before they got some- >>thing they could use to gaslight regular people with in the news.
Here's one example:
Claude was prompted:
|The problem is to compute the Six-particle (hexagon)
|amplitude in planar N=4 SYM at nine loops.
. It did the computation during several days for about USD 1500 using
two different approaches and making sure the results agree. So far
humans were only able to do eight loops, which already took years.
So this is the first time this nine loop calculation was ever done.
Human experts later confirmed Claude's results to be correct.
[email protected] (Stefan Ram) writes:
Johann 'Myrkraverk' Oskarsson <[email protected]d> wrote or quoted: >>> We are also not told -- at least I have never seen such texts -- how
long it took the L.L.M. to /solve/ these puzzles, nor how many false
starts it took the human team behind the machine, before they got some-
thing they could use to gaslight regular people with in the news.
Here's one example:
Claude was prompted:
|The problem is to compute the Six-particle (hexagon)
|amplitude in planar N=4 SYM at nine loops.
. It did the computation during several days for about USD 1500 using
two different approaches and making sure the results agree. So far
humans were only able to do eight loops, which already took years.
So this is the first time this nine loop calculation was ever done.
Human experts later confirmed Claude's results to be correct.
The thing here is: in the thing about Iceland, it simply cannot reason over periods in time, since, well, it does not "recognise" the facts humans can see: periods of time, connected to countries being tied together, and reason from there.
It only can, if additional coding is applied which allow for these data to become facts (not factoids ;) ).
In math, I suppose, the "only thing" (allright, that's an understatement)
one has to do is connect / combine outcomes, up to the point that the
given proposition yields. Or not. But I assume, in the latter case, it will continue forever, or until all combinations are exhausted.
| Sysop: | DaiTengu |
|---|---|
| Location: | Appleton, WI |
| Users: | 1,142 |
| Nodes: | 10 (0 / 10) |
| Uptime: | 93:07:26 |
| Calls: | 14,704 |
| Calls today: | 2 |
| Files: | 186,497 |
| D/L today: |
2,349 files (779M bytes) |
| Messages: | 2,551,657 |