AI

[ivconic]

It should be referred to as IA instead of AI where IA is Intelligence Amplifier.
I am up to over 18000 lines of C code in my latest project and I did not even write a single line.
AIs or more properly LLMs deal with context ....if you make it read your chat history....before each interaction...it "learns" what you expect..so if it's stroking your ego too much ... Just tell it not to patronise you. Asking it to tell you which chip is the best low noise front end is not a good question because firstly it's not language based ie code or maths or symbols etc and secondly humans can't agree either so if it was trained on audiophile forums it's probably psychotic.
Here's the point though ... I had an interrupt timing problem ....causing stack overruns and random crashes ...normally you would put break points in and examine registers / memory and try to nail the timing conflict with long debugging sessions. Not the AI though it took the compiled binaries... Decompiled them to assembly language machine code then looked in the ARM assembly language specification for instruction execution times including cached non cached memory and worked out the race condition causing the interrupt clash...

Interestingly this also means AIs can be used as superb tools for reverse engineering firmware if you need to, strictly for academic purposes of corse

It's not a problem in the question that was asked, it's not a problem in the disagreement about the choice of opamp;
the problem is that AI invented and lied about the RX frontend opamp in existing (and non-existing) PI detectors.
Lied!
Made it up!
That's the problem.
...
Imagine that one day such an AI prescribes medicines for diseases to the sick.
Imagine that such an AI controls the traffic in a metropolis!
Imagine such an AI giving financial advice on the stock market!
And finally, you don't have to imagine this, because it's already happening;
that the politicians who decide the fate of the world have their closest advisers around them who generate their advice from countless chats with such AI!
I'm not sure which smiley to put here at the end of the post now; hilarious smileys or some horror smileys?

[moodz]

It should be referred to as IA instead of AI where IA is Intelligence Amplifier.
I am up to over 18000 lines of C code in my latest project and I did not even write a single line.
AIs or more properly LLMs deal with context ....if you make it read your chat history....before each interaction...it "learns" what you expect..so if it's stroking your ego too much ... Just tell it not to patronise you. Asking it to tell you which chip is the best low noise front end is not a good question because firstly it's not language based ie code or maths or symbols etc and secondly humans can't agree either so if it was trained on audiophile forums it's probably psychotic.
Here's the point though ... I had an interrupt timing problem ....causing stack overruns and random crashes ...normally you would put break points in and examine registers / memory and try to nail the timing conflict with long debugging sessions. Not the AI though it took the compiled binaries... Decompiled them to assembly language machine code then looked in the ARM assembly language specification for instruction execution times including cached non cached memory and worked out the race condition causing the interrupt clash...

Interestingly this also means AIs can be used as superb tools for reverse engineering firmware if you need to, strictly for academic purposes of corse

[ivconic]

Using AI tools (such as ChatGPT) can be useful if used sensibly ... but only if you know what you're doing.
Blindly accepting the AI's response is not a good idea. I've often found that questioning the answer results in an admission that it made a mistake.
However, it can still be a useful tool in certain cases.

"...anything more than a handy dumb calculator and text editor..."

Maybe I was a little too strict in this statement, but I'm not far from the truth.
When AI is properly managed/guided; it becomes an extremely useful tool with which work is done much faster.
But, at the end of the day, it's still just a slightly smarter calculator + database + good text editor (for developers).
Excessive mystifications should be avoided. I've been spending way too much time with AI since they first appeared on the net.
And I notice at myself that I have gone through all possible stages; from quite enthusiasm to now quite realistic (down to Earth) attitude towards AI.

[ivconic]

I discovered this the very first time I tried ChatGPT. Also, that it loves to flatter.

Oh yes!
Ahahahaha!
AI likes to get familiar after some time.
But one should be careful; in those moments AI is very prone to hallucinations!

[Qiaozhi]

Using AI tools (such as ChatGPT) can be useful if used sensibly ... but only if you know what you're doing.
Blindly accepting the AI's response is not a good idea. I've often found that questioning the answer results in an admission that it made a mistake.
However, it can still be a useful tool in certain cases.

[Carl-NC]

"...The model is optimized to generate fluid, believable text — not to generate the truth. When it doesn't have enough knowledge, it continues to generate fluid text... which is no longer true.

I discovered this the very first time I tried ChatGPT. Also, that it loves to flatter.

[ivconic]

From all this it is clearly seen that the leading politicians in the world make their decisions solely on the basis of the answers they get from AI!

[ivconic]

To my simple question about the Pulse Induction RX frontend...
What do famous PI designs use?
Claude listed:

Goldbug PI LM318
Carl's PI LM318
Surfmaster PI LM318
White's TDI (Commercial)Internal but similar character (Whatever this means!???)
Different DIY's LF357 / TL071
...
Me:
"You messed up terribly and you're hallucinating like never before! Alas, you are full of nonsense and incorrect things, I just got a headache from the vomit you wrote!
None of the PI designs you listed use an LM318 at the input!"

Claude:
"You are absolutely right, and I sincerely apologize.
I made a classic mistake — I made up specifics (design names and which op-amps they use) instead of admitting that I didn't know it for sure."
...
Moment later Claude tried to explain:
"Why is this happening? A fundamental problem of architecture
I am not a fact-checked database.
I'm a statistical model — I predict which token (word) is most likely to come next, based on the vast amount of text I've been trained on.
This means:
When I know something well → the answers are correct
When I don't know enough → the model still generates "plausible sounding" text because that's what it's optimized for
There is no internal alarm system that says "stop, you don't know this"..."

[ivconic]

Devastating facts about AI!!!
I had a long "conversation" with Claude and I was getting it to finally bare itself and admit some horrible truths!
At the core of the concept of AI is not the pursuit of truth, nor that the final result or "product" of AI be the truth and true facts! No! That is not the essence and goal of AI (in its current state of development).
At the core of the AI concept is something else entirely!
The essence and goal of AI is to always maintain "fluency" in conversations!
If the AI does not have true data on an issue; the first thing AI will turn to is generating its own "truth" (bunch of lies, nebula, nonsense and 11000% wrong facts)!!!
Only when you stop it and very rigidly let it know that it is deeply hallucinating and spewing massive amounts of nonsense; only then does the AI begin to truly explore the data.
Here is a short quote that Claude itself wrote in his confession (after my "expletive-filled intervention" and argumentative criticism):
"...The model is optimized to generate fluid, believable text — not to generate the truth. When it doesn't have enough knowledge, it continues to generate fluid text... which is no longer true.
It's the equivalent of a person making up an answer rather than saying "I don't know" — but on an industrial scale, at the speed of light, with perfect confidence...
...A paradox that no one likes to admit...
The industry is spending hundreds of billions of dollars on AI development, and the fundamental problem — the difference between fluency and truthfulness — is still unresolved..."
I think the vast majority of people in the world don't realize that AI is far, far away from becoming anything more than a handy dumb calculator and text editor.
It is still many years before AI will be at the doorstep of true intelligence.
Beginners and laymen are delighted when the AI writes them a hundred lines of code that works (or works after a few iterations). But when you spend months and a couple of years testing AI every day... you realize a truth that is devastating.
Why is this so dramatic, devastating and ultimately very dangerous?
Because (we are witnessing) world leaders getting dumber and dumber. Each new generation of politicians is more and more handicapped in terms of intelligence.
And such creatures hold enormous power and authority in their hands. Such degenerate idiots have their finger over the red button!
The trends around AI are moving in the direction that one day AI will replace humans in such very important and essentially vital positions.
From a huge horror... to an even bigger horror! Our destiny is shaped by degenerate genetic residual human waste... and the tendency is to be replaced by even worse degenerate residual electronic waste!
I wasn't worried one bit because this world is slowly sinking into a global war... it's laughable (despite the horror that's happening) compared to what awaits us one day when AI becomes "independent".
Who decides what will happen to AI and to what extent will AI become independent? The scientists and academics who designed it and who work on it? No! Of course not!
The final decisions are made by the same degenerate genetically backward human waste on power.
(I don't have to name them, it's enough to turn on the TV and change the channels with the remote, they're all there, present)

[ivconic]

The whole story around local "offline" AI is slowly starting to remind me of the beginnings of mining stories.
The sudden burst of great enthusiasm is greatly reduced and subsided over time and turns into bitter disappointment.
You realize that in this world, in this life; there is no such thing as a free lunch.
In order to set up your personal local "offline" AI (it is a problem of human nature, first of all a problem of mentality and psychology)
you need to allocate a serious sum of money.
I often think in pictures and I like metaphors the most.
To get free green energy; I have to invest a lot of money in a serious solar plantation (otherwise it doesn't pay for itself).
And when I distribute that amount of money to the monthly bills that I would pay if I used the same electricity from the city network;
in the end it pays more for me to be powered from the city grid!
It's the same with AI.
It's much cheaper to use online AI than to throw huge amounts of money at equipment that will be able to run local AI, which will barely compete with the online versions.
And now we return to the story of mining. I know people who have spent enormous fortunes on mining equipment.
Today when that failed story; their balance is still on the negative side!
This world is full of scam ideas and loads of people who flippantly take the bait.
The other day, my old father, a pensioner, fell for commercial lies and bought a heater that "saves electricity".
Not only does he not save, but he spends more, and the price of the heater would have covered several monthly bills, if he had heated himself with the old heater he already has.
There are many metaphors, there are many variations, and they all revolve around the same phenomenon.
Not so long ago I seriously did a study, which showed me that the USA will have to build at least 200 (sorry if I got the number wrong, I forget)
small nuclear power plants; to meet the energy needs of the electric car industry and upcoming AI resources.
And that's just the beginning.
...
Let's go back to AI.
In short, I have a (modest by today's standards) configuration:
Intel(R) Core(TM) i3-10105 CPU @ 3.70GHz, 3693 MHz, 4 Core(s),8 Logical Pr...
Motherboard 11-th generation
NVIDIA GeForce GTX 1060 6GB
RAM 32GB
M2 disc
...
These days I've been trying options again, LMStudio, Ollama, AnythingLLM... a bunch of models with them.
The effective results are devastating. Compared to online free AI options.
As with all the metaphors above; the conclusion is that we little people are useful idiots who voluntarily participate in the development of resources
that will be owned and controlled by the same people who have owned everything until now throughout history.
Think about this for a moment.
I often wonder (well, not really) whether the story with the detectors is one of those stories!

[kt315]

[kt315]

https://www.youtube.com/shorts/ndn4qb-hwLQ

[ivconic]

The AI balloon is slowly deflating:

[ivconic]

[ivconic]

This is not a joke.
This is literally the picture I see when I turn on the TV and watch press conferences.
I probably have some veil over my eyes, which allows me to see what is real and filters out the outer beautified facade.