From Newsgroup: alt.os.linux


Spent many hours yesterday with a popular AI service trouble-shooting a
guitar equipment issue: I could not hear the output of my Gx100 effects
board in the headset while recording with Audacity, something I'd had unrelated problems with before but not recently. Hours on end resolved NOTHING. Today I remembered a recent windows session to test the (winblows-only) bundleware and that during that very short session I had
seen some 'unusual' display flash just as I was shutting it down. So I
went into the effect-chain setup and realized that the Send/Return block
was absent, meaning that the SEND port on the effect-board had nothing
to send to Line-In. AI was incapable of suggesting a look at that, it
had me going on a butterfly chase that resulted only is screwing up my
OS so bad soundwise that it became unusable and I had to recover from a backup. AI is fucking STUPID pure and simple, it's nothing more than a GLORIFIED SEARCH-ENGINE. It reminds me of a building inspector who gets
called out because it's impossible to enter a new building. On arrival
he promtly attacks the brick wall with a magnifying glass and starts
finding faults in hundereds of bricks that have NOTHING to do with the
problem of there being no door in the wall. AI is incapable of
assembling and executing a heads-up overview of all possible factors and approaching the problem wisely.

</end rant>

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On 2025-10-30, bad sector <forgetski@_INVALID.net> wrote:

Spent many hours yesterday with a popular AI service trouble-shooting a

LLMs are good for only one thing: faking a conversation.
--
Jasen.
🇺🇦 Слава Україні
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On 30/10/2025 04.44, bad sector wrote:

Spent many hours yesterday with a popular AI service trouble-shooting a guitar equipment issue

First of all what we have today is Large Language Models, they ain't Artificial Intelligence, the AI is used only for marketing reasons and
most people will think it's some kind of AI as it seems to give you a plausible answer that can be right in many cases. LLM's will just
generate a most likely text string for the string of text you inputted,
it has no understanding of what you asked.

When you talk about these so called "AI", please use the LLM instead, as
one day in the future we will have AI, but we are still far away from it.
--
//Aho

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On Wed, 10/29/2025 11:44 PM, bad sector wrote:

Spent many hours yesterday with a popular AI service trouble-shooting a guitar equipment issue: I could not hear the output of my Gx100 effects board in the headset while recording with Audacity, something I'd had unrelated problems with before but not recently. Hours on end resolved NOTHING. Today I remembered a recent windows session to test the (winblows-only) bundleware and that during that very short session I had seen some 'unusual' display flash just as I was shutting it down. So I went into the effect-chain setup and realized that the Send/Return block was absent, meaning that the SEND port on the effect-board had nothing to send to Line-In. AI was incapable of suggesting a look at that, it had me going on a butterfly chase that resulted only is screwing up my OS so bad soundwise that it became unusable and I had to recover from a backup. AI is fucking STUPID pure and simple, it's nothing more than a GLORIFIED SEARCH-ENGINE. It reminds me of a building inspector who
gets called out because it's impossible to enter a new building. On arrival he promtly attacks the brick wall with a magnifying glass and starts finding faults in hundereds of bricks that have NOTHING to do with the problem of there being no door in the wall. AI is incapable of assembling and executing a heads-up overview of all possible factors and approaching the problem wisely.

</end rant>

And that's because :-)

[Picture]

https://i.postimg.cc/Kc0XGw8h/forgot-to-load-disk-pack-yesterday-for-customer.jpg

You might have been able to rig something up with the PC audio "monitor" or "what-you-hear"
path on the computer HDAudio. Then as Audacity records, a headphone plugged into
the computer could hear the same thing. The selector for that is labeled "Stereo Mix"
in some cases, which you point towards the LineOut for your headphones.

So far, I've downloaded a 142 page and a 435 page Boss PDF manual, full of "press this" and "click dat". If I'd never played guitar before, or
there weren't any forum posts, how would you figure it out from
those manuals ? The AI eats text for breakfast -- while it can extract intelligence from the photos, while it was training it might not have
been doing that.

As for the AI, researchers have noticed "LLM AI does poorly, on thinking
tasks which roam outside the training set". While on an AI loaded on
your own computer, you could select "high reasoning", in terms of
percentage of tasks completed, it makes hardly any difference to
the results.

"You are right, and I apologize profusely for my inability to think"

You should remember, that the concept is dangerous, and to have
made the "dumb version" of AI to start with, is saving our ass. You'll be sorry, when they hit AGI. The training wheels will fall off, and the
thing will proceed to run amok. See the potential for the bad stuff
it'll do, when it "gets its own brain". They've already given it
access to a web browser, and when it gets its own credit card... look out.

The AI are also gullible and easily coerced to do evil things. You will
have read the results of some of this already.

Paul
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On 10/30/25 10:19, Paul wrote:

On Wed, 10/29/2025 11:44 PM, bad sector wrote:

Spent many hours yesterday with a popular AI service trouble-shooting a guitar equipment issue: I could not hear the output of my Gx100 effects board in the headset while recording with Audacity, something I'd had unrelated problems with before but not recently. Hours on end resolved NOTHING. Today I remembered a recent windows session to test the (winblows-only) bundleware and that during that very short session I had seen some 'unusual' display flash just as I was shutting it down. So I went into the effect-chain setup and realized that the Send/Return block was absent, meaning that the SEND port on the effect-board had nothing to send to Line-In. AI was incapable of suggesting a look at that, it had me going on a butterfly chase that resulted only is screwing up my OS so bad soundwise that it became unusable and I had to recover from a backup. AI is fucking STUPID pure and simple, it's nothing more than a GLORIFIED SEARCH-ENGINE. It reminds me of a building inspector who
gets called out because it's impossible to enter a new building. On arrival he promtly attacks the brick wall with a magnifying glass and starts finding faults in hundereds of bricks that have NOTHING to do with the problem of there being no door in the wall. AI is incapable of assembling and executing a heads-up overview of all possible factors and approaching the problem wisely.

</end rant>

And that's because :-)

[Picture]

https://i.postimg.cc/Kc0XGw8h/forgot-to-load-disk-pack-yesterday-for-customer.jpg

You might have been able to rig something up with the PC audio "monitor" or "what-you-hear"
path on the computer HDAudio. Then as Audacity records, a headphone plugged into
the computer could hear the same thing. The selector for that is labeled "Stereo Mix"
in some cases, which you point towards the LineOut for your headphones.

So far, I've downloaded a 142 page and a 435 page Boss PDF manual, full of "press this" and "click dat". If I'd never played guitar before, or
there weren't any forum posts, how would you figure it out from
those manuals ? The AI eats text for breakfast -- while it can extract intelligence from the photos, while it was training it might not have
been doing that.

The BOSS crowd are another example of documentation Darwin laureates,
not to mention the fact that they shit on linux users. But the design
itself also sucks cubed. This Send/Return is in essence an effect module
that you can include is a chain so that if you include it between say
'echo' and 'tremolo' then the SEND port will export the sound as
modified up to the 'echo' effect only. There's a touch more but this is
it really, no rocket science up to this point. The monkeys got swamped
when it came time to decide what to do if the Send/Retun module isn't in
the chain at all (a.k.a. a reasonable/practical default config that
COULD NOT result in no sound at all) and such an 'at the end of the
chain if not found elsewhere in it' concept caused a neural meltdown.

Design is an art and a science, you don't give it to butlers and
pedestrians, which seems to include UI developers. Another example is
this AV-TunerAmp I have, it does everything except moon landings or
wipiping my ass (no pun intended) BUT it also permits one to *ACHIEVE*
several settings which result in no video OR sound leaving the box.

As for the AI, researchers have noticed "LLM AI does poorly, on thinking tasks which roam outside the training set". While on an AI loaded on
your own computer, you could select "high reasoning", in terms of
percentage of tasks completed, it makes hardly any difference to
the results.

"You are right, and I apologize profusely for my inability to think"

You should remember, that the concept is dangerous, and to have
made the "dumb version" of AI to start with, is saving our ass. You'll be sorry, when they hit AGI. The training wheels will fall off, and the
thing will proceed to run amok. See the potential for the bad stuff
it'll do, when it "gets its own brain". They've already given it
access to a web browser, and when it gets its own credit card... look out.

The AI are also gullible and easily coerced to do evil things. You will
have read the results of some of this already.

Paul

OK, 3 against 1 I have no choice but to reel it back in :-)


This was essentially my 2nd session-set with it, truth is the 1st one
was a huge success; I wasn't aware of the acronymic differences though
so I'll keep an eye out for AI/LLM/AGI in the limited future I can
expect at my disposal.



--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

Jasen Betts wrote:

Spent many hours yesterday with a popular AI service trouble-shooting a

LLMs are good for only one thing: faking a conversation.

Not really. The other day ChatGPT wrote me a script that etracts the
subtitles from a film, automatically translates them and saves them as
an external subtitle file. It needed 3 attempts, but got it working in
the end.

I wouldn't use it for serious business, though.

-jw-
--
And now for something completely different...
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On Wed, 29 Oct 2025 23:44:25 -0400, bad sector wrote:

AI was incapable of suggesting a look at that, it had me going on a
butterfly chase that resulted only is screwing up my OS so bad
soundwise that it became unusable and I had to recover from a
backup.

Is this a new concept for you?
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: alt.os.linux

On 10/30/25 5:10 PM, Lawrence D’Oliveiro wrote:

On Wed, 29 Oct 2025 23:44:25 -0400, bad sector wrote:

AI was incapable of suggesting a look at that, it had me going on a
butterfly chase that resulted only is screwing up my OS so bad
soundwise that it became unusable and I had to recover from a
backup.

Is this a new concept for you?

Not really but in the process I found out that the ALC1220 is a real;
shithole chip.

Alsa Auto-Mute 'can' be disabled to have simultaneous front-panel
Headphones and rear-panel Line-Out (speakers) but Loopback goes out ONLY
via rear Line-Out even though other streams can be mirrored. There's no hardware or software solution for this fuckup! I'll have to find high
quality 1/8" stereo Y pigtails, the two I bought were supposed to be
good but one failed a channel completely and the other allows all kinds
of noise infiltration.


--- Synchronet 3.21a-Linux NewsLink 1.2