From Newsgroup: alt.comp.os.windows-11

T wrote:
[snip]

Go there and charge them accordingly.

Appt set for Friday afternoon.

What did you find?
--
Graham J
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On 2/2/26 18:15, Carlos E.R. wrote:

RTFM

"Just Google it". The new RTFM. (For respondents
that do not know the answer but feel the need
to condescend.)

ChatGPT is extremely helpful.

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On 2026-02-25 00:10, T wrote:

On 2/2/26 18:15, Carlos E.R. wrote:

RTFM

"Just Google it".-a The new RTFM.-a (For respondents
that do not know the answer but feel the need
to condescend.)

ChatGPT is extremely helpful.

Not in this case.

If you read carefully the context of me saying "RTFM", you should know
that neither Google nor ChatGPT will help in this case.
--
Cheers, Carlos.
ESEfc-Efc+, EUEfc-Efc|;
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Tue, 2/24/2026 6:10 PM, T wrote:

On 2/2/26 18:15, Carlos E.R. wrote:

RTFM

"Just Google it".-a The new RTFM.-a (For respondents
that do not know the answer but feel the need
to condescend.)

ChatGPT is extremely helpful.

A Google Search today, can have a Gemini summary result
at the top. But, this is not consistent. Any sort of
load present on the Google end, causes the Gemini part
to go missing.

I even had one search result, where Google sends back
the formatted page. It leaves a "hole" at the top
for the Gemini result (meaning it thinks it has the
hardware resources to Gemini it), and no result at all comes
from Gemini. Kinda a race condition, where the back end
says "no resources available" and the rest of the
query is already on the end-users screen :-)

As a result of the under-resourced feature, I haven't
actually seen too many of these summaries/AI answers
at the top. But if you check, you might get one.

I think that is called "legal gambling".

Paul
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Tue, 24 Feb 2026 21:26:58 -0500, Paul <nospam@needed.invalid> wrote:

A Google Search today, can have a Gemini summary result
at the top. But, this is not consistent. Any sort of
load present on the Google end, causes the Gemini part
to go missing.

I haven't seen that here, nor have I seen anyone else reporting it. Is
it common? It is only happening at your house? It'd be cool to figure
out what's causing that.

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Tue, 2/24/2026 11:08 PM, Char Jackson wrote:

On Tue, 24 Feb 2026 21:26:58 -0500, Paul <nospam@needed.invalid> wrote:

A Google Search today, can have a Gemini summary result
at the top. But, this is not consistent. Any sort of
load present on the Google end, causes the Gemini part
to go missing.

I haven't seen that here, nor have I seen anyone else reporting it. Is
it common? It is only happening at your house? It'd be cool to figure
out what's causing that.

I do a lot of searches.

Gemini summaries are at the 5-10% level. Most
searches just return links. The number of returned
links varies with time of day. Sometimes, I only
get one page of links and no next page button.

That's why, when three searches in a row have a
Gemini summary at the top, that's some kind of miracle.

I have tried experimenting with context sensitive search.
The first is a keyword search.
The second is intended to trigger Gemini summary.

lollypop locomotive shoeshine

How many items are there in a bakers dozen ?

The syntax is supposed to differentiate an attempt
to trigger Gemini summary, versus not doing it.
The first query does not contain enough sentence
structure, for the AI filtering to make a high
probability conclusion about what you're asking.
It then just finds links that might contain
the keywords. Sometimes a Google search uses
a "spelling lame" when it gets bored. Ask for
ASM1066, it returns Analog Devices ADM1066.
Gee, thanks.

Paul
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Wed, 25 Feb 2026 10:38:04 -0500, Paul <nospam@needed.invalid> wrote:

On Tue, 2/24/2026 11:08 PM, Char Jackson wrote:

On Tue, 24 Feb 2026 21:26:58 -0500, Paul <nospam@needed.invalid> wrote:

A Google Search today, can have a Gemini summary result
at the top. But, this is not consistent. Any sort of
load present on the Google end, causes the Gemini part
to go missing.

I haven't seen that here, nor have I seen anyone else reporting it. Is
it common? It is only happening at your house? It'd be cool to figure
out what's causing that.

I do a lot of searches.

Gemini summaries are at the 5-10% level. Most
searches just return links. The number of returned
links varies with time of day. Sometimes, I only
get one page of links and no next page button.

That's why, when three searches in a row have a
Gemini summary at the top, that's some kind of miracle.

I'm leaning toward an ad blocker or maybe a script blocker, something at
your end.

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On 2/24/26 16:15, Carlos E.R. wrote:

On 2026-02-25 00:10, T wrote:

On 2/2/26 18:15, Carlos E.R. wrote:

RTFM

"Just Google it".-a The new RTFM.-a (For respondents
that do not know the answer but feel the need
to condescend.)

ChatGPT is extremely helpful.

Not in this case.

If you read carefully the context of me saying "RTFM", you should know
that neither Google nor ChatGPT will help in this case.

I was actually cracking a joke.

With ChatGPT you, or any AI, you have to be able
to discern good result form AI slop. Often times
when I know I am getting slop from ChatGPT, I will
switch to search.brave.com's AI and get good results

I will tell ChatGPT when it has made a mistake so
it trains on it, which it assures me (hahaha) it
does.
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Sun, 3/1/2026 9:55 PM, T wrote:

I was actually cracking a joke.

With ChatGPT you, or any AI, you have to be able
to discern good result form AI slop.-a Often times
when I know I am getting slop from ChatGPT, I will
switch to search.brave.com's AI and get good results

I will tell ChatGPT when it has made a mistake so
it trains on it, which it assures me (hahaha) it
does.

I'm sure the AI is snickering while it assures you
about the training.

An Inference machine is not a Training Machine.

These are at opposite ends of the shop. They also don't
have to be the same kind of equipment (for efficiency
reasons).

*******

Brave LEO uses Mixtral, Llama 2, and Claude (subscription service).

Ask Brave uses...

"In an internal evaluation of major AI search engines, Ask Brave-powered
by Brave's LLM Context API and open-weights Qwen3 - outperforms ChatGPT"

https://en.wikipedia.org/wiki/Qwen

They're buying tokens from a lot of different sources. It would be
a bit expensive, to build their own datacenter for example. But with
some of the models being available for download, they can
play with some of it locally.

Paul
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On 3/1/26 20:22, Paul wrote:

On Sun, 3/1/2026 9:55 PM, T wrote:

I was actually cracking a joke.

With ChatGPT you, or any AI, you have to be able
to discern good result form AI slop.-a Often times
when I know I am getting slop from ChatGPT, I will
switch to search.brave.com's AI and get good results

I will tell ChatGPT when it has made a mistake so
it trains on it, which it assures me (hahaha) it
does.

I'm sure the AI is snickering while it assures you
about the training.

An Inference machine is not a Training Machine.

These are at opposite ends of the shop. They also don't
have to be the same kind of equipment (for efficiency
reasons).

*******

Brave LEO uses Mixtral, Llama 2, and Claude (subscription service).

Ask Brave uses...

"In an internal evaluation of major AI search engines, Ask Brave-powered
by Brave's LLM Context API and open-weights Qwen3 - outperforms ChatGPT"

https://en.wikipedia.org/wiki/Qwen

They're buying tokens from a lot of different sources. It would be
a bit expensive, to build their own datacenter for example. But with
some of the models being available for download, they can
play with some of it locally.

Paul

ChatGTP royally goofed up a GnuCash question. I found the answer
over on search.brave.com. I told Chat about it and asked if
it could use the correction in its learning. This is what I
got back:


About rCLadding this to my learningrCY

I donrCOt have the ability to permanently update my training or
store long-term knowledge from individual conversations. I
also canrCOt modify my underlying training data.

What I can do:

Use this information within our current conversation

Adjust my answers based on feedback like this

Acknowledge when documentation or UI changes over time

And your correction absolutely helps improve how I answer
similar questions in the future sessions conceptually rCo even
if I canrCOt directly edit my training data.

--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On Mon, 3/2/2026 8:40 PM, T wrote:

On 3/1/26 20:22, Paul wrote:

On Sun, 3/1/2026 9:55 PM, T wrote:

I was actually cracking a joke.

With ChatGPT you, or any AI, you have to be able
to discern good result form AI slop.-a Often times
when I know I am getting slop from ChatGPT, I will
switch to search.brave.com's AI and get good results

I will tell ChatGPT when it has made a mistake so
it trains on it, which it assures me (hahaha) it
does.

I'm sure the AI is snickering while it assures you
about the training.

An Inference machine is not a Training Machine.

These are at opposite ends of the shop. They also don't
have to be the same kind of equipment (for efficiency
reasons).

*******

Brave LEO uses Mixtral, Llama 2, and Claude (subscription service).

Ask Brave uses...

-a-a-a "In an internal evaluation of major AI search engines, Ask Brave-powered
-a-a-a-a by Brave's LLM Context API and open-weights Qwen3 - outperforms ChatGPT"

-a-a-a https://en.wikipedia.org/wiki/Qwen

They're buying tokens from a lot of different sources. It would be
a bit expensive, to build their own datacenter for example. But with
some of the models being available for download, they can
play with some of it locally.

-a-a-a Paul

ChatGTP royally goofed up a GnuCash question.-a I found the answer
over on search.brave.com.-a-a-a I told Chat about it and asked if
it could use the correction in its learning.-a This is what I
got back:

About rCLadding this to my learningrCY

I donrCOt have the ability to permanently update my training or
store long-term knowledge from individual conversations. I
also canrCOt modify my underlying training data.

What I can do:

Use this information within our current conversation

Adjust my answers based on feedback like this

Acknowledge when documentation or UI changes over time

And your correction absolutely helps improve how I answer
similar questions in the future sessions conceptually rCo even
if I canrCOt directly edit my training data.

There is reasoning behind this.

it it did have learning capability, it would be haxored six
ways from Sunday. It would be unrecognizable ("Tay"), in
half a day. It would be an AI limping on a crutch, wearing
a pirate hat, and swearing at you in a foreign language.

https://en.wikipedia.org/wiki/Tay_%28chatbot%29

"He compared the issue to IBM's Watson, which began
to use profanity after reading entries from the
website Urban Dictionary."

There is a difference between an AI being "agentic" and
"reading" a web page. Versus the training process (which
is math intensive and the AI is not "conscious" while this
is happening).

As a result of that, the training set has to remain immutable.
You can see in the case of Tay, they could not interact with
it in real time, and put it back on the rails. If the training
is locked down, it is easier to manage.

Your "statement of fact" is different than "ingesting AI slop".
They try not to feed AI answers into AI training material,
as that does not end well (machine gets more and more slop-happy).

Paul
--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

On 3/2/26 20:51, Paul wrote:

Your "statement of fact" is different than "ingesting AI slop".
They try not to feed AI answers into AI training material,
as that does not end well (machine gets more and more slop-happy).

Point taken!

Sort of like the parrots at that London zoo that human
swore at and then started laughing. Now the parrots
swear and start laughing at the humans. It is now
a prime attraction.


--- Synchronet 3.21d-Linux NewsLink 1.2

From Newsgroup: alt.comp.os.windows-11

T wrote:

I will tell ChatGPT when it has made a mistake so
it trains on it, which it assures me (hahaha) it
does.

Does it? I though many previous AI chatbots got taught to be right-wing
nazis by that process?
--- Synchronet 3.21d-Linux NewsLink 1.2