ChatGPT designed a gain cell - your comments?

LOL, so I got this apparently stupid idea to only do what ChatGPT suggested, I was not allowed (my own rules) to modify or add anything that ChatGPT did not suggest or recommended. The reason for saying apparently stupid is that I think it took about 3 weeks, on and off. The below description does of cause not contain all the tedious work it was to get it to understand what I wanted, sort out its many errors (talking about shielding where it suggested a garden fence because it shields for the wind). 🙂

Here is what I asked for and why it may look as it does.
1) Sound quality is most important.
2) Analog all the way. (just to avoid the usual digital volume control)
3) Possibility for adjusting the volume without ever using a pot.
4) Sound degradation is not allowed due to mechanical wear.
5) Everything has to work at the same voltage (It ended being +/-18VDC).

I started out asking for an op-amp, and it started by suggesting the OPA627, that I accepted but still did I ask about if it was the best, just to see how it reacted, it ended up in ChatGPT suggested a discreet solution and yes that is of cause a better solution, but I did simply not dare to begin letting ChatGPT guide me through a design of a discrete op-amp. (lol, my sanity is too valuable for that.) So I asked if there were some ready build op-amps, and it ended suggesting Sonic Labs 990Enh-Ticha that I have never heard off but hay it has great marketing.

Now to a way to adjust the volume and I asked it for a solution, again did several digital solutions come up, but I reminded the AI about the 100% analog solution. Then did it suggest using a FET, a combination of LED and LDR, or a gain cell solution. As you properly can read from the title did, I chose the gain cell to see how it would turn out.

To adjust gain did the AI suggest using a pot, but I reminded it about the possibility for adjusting the volume without any pot, and it got the AI to suggest a resistor of unknown value and a VCR in the feedback, but it was fairly limited what it could find and finally did it suggest a THT THAT2181.

Lastly after the AI had talked about using a pot for the 879th time did I ask how it would implement a pot, thinking of my rules about sound quality and wear, it had two suggestions, one was to get the best most expensive pot I could get or placing a buffer between the pot and the VCR, so I tried.

The result do you see below. It looks very strange, but this is how ChatGPT would have me build it.
Any comment? 🙂
 

Attachments

  • gain cell.png
    gain cell.png
    11.9 KB · Views: 392
Your analog-all-the-way ruled out tons of great hybrid solutions of course, such as PGA2311, which are all analog signal path, but digital control, even a stepped attenuator is in this class really!

But why didn't it choose the application circuit from the THAT2181 datasheet? I guess it didn't what to bother you with having to use a distortion meter to set up your circuit properly trimmed? Is it wasn't trained on datasheets? Maybe not. And it missed out on a heater/thermistor gain-control technique despite knowing about LED/LDR, which suggests no analytical thought processes, just cargo-cult copying - it Googled its homework.
 
I have found out it is in fact not able to read pdf files, and is not able to see pictures of any diagram, so it may simply what it could find?
I could also not get any impedance at all, so I am not sure what value it should have given. 🙂
 
It's interesting to hear about your experience with ChatGPT while discussing op-amps. ChatGPT's suggestions of the OPA627 and a discrete solution demonstrate its ability to provide different options based on the input provided.

Considering the complexity involved in designing a discrete op-amp, it's understandable that you chose not to pursue that path. Designing discrete op-amps requires expertise and thorough understanding of electronic circuits. It's important to prioritize your comfort level and expertise when embarking on such projects.

Asking about ready-built op-amps is a practical approach, and the suggestion of the Sonic Labs 990Enh-Ticha is intriguing. While you may not be familiar with this specific model, it's worth researching and exploring its features, specifications, and user reviews to determine its suitability for your needs. Evaluating the reputation and credibility of a product, beyond marketing claims, is always a good practice.
 
Thanks for your interests! 🙂
To avoiding that, I unwillingly did "force" the AI to select better solutions, change components I may like better or correct for obvious errors, have I not read anything about each component, I had in other words no idea if it would even work at all.

After going through the solution, is there some things I have changed.
1) The +/-18VDC was min for some components. That made me change it to +/-19VDC.
2) The buffer had a max of +/-15VDC, so it got changed to LT1462ACN8
3) The THAT2181 Sym got added a 20k pot to GND

After reading about the components, must I say that ChatGPT did fairly okay on the sound quality, it may even have gone a bit overboard. The Sonic Labs 990Enh-Ticha op-amp looks like it is may be at the level of the V6 Vivid or Classic, or may even surpass them. The VCR looks like it may be really great or the best THT. Unfortunately have I not been able to find the input and output impedance for the 990Enh-Ticha, does anybody know about that?

By the way, do any of you know why ChatGPT chose THT? Can it be it looked at the Sonic Labs 990Enh-Ticha op-amp and saw it was a THT and kept it that way?

My conclusion is that you in fact can get the ChatGPT to design a circuit for you, but with a speed of one week per component, do I not dare to think how long it would take to build a complete amplifier!! 🙂
It also had some strange errors like ignoring the "sym" on the VCR and selecting an 15V buffer on a 18V circuit, it did also not see it as a problem running a component right at it's border, it saw min 18V so let's use 18V! 🙂

So if a beginner need advice on how to connect a chip, how to select a chip, is the AI great at it, if you just do as it says and read the datasheet yourself.

Would I do it again? Heck NO, not as the AI are right now! 🙂
 
You are all going the "rabbit hole" ... how unfortunate !
I'm glad the EU has the foresight to attempt to regulate it.
Here , in the Fascist USA , it will fully destroy us. 3 years ... at the most. Exponential.
I'm not a luddite , but cause and effect is quite predictable.
Lazy humans , design your own amp ! F AI.
OS
 
  • Like
Reactions: aditya and rayma
ostripper, I remember when the personal computer came out at everybody was extremely concerned about people loosing jobs, that the computer would take over, but there were also those who predicted that nobody would use papers anymore and lots more.

You here exactly the same now about AI that i fact should be called TSS (Text Search Scripts), all are concerned about the exact same things and is debating the same. As the "AI" is right now are they nothing more that extremely fast search scrips that can repeat what it has been trained on, it has no creativity, no thoughts, no ability to give new knowledge, it can only combine old knowledge and maybe combine info across fields to highlight not previous discovered patterns.

No need to be concerned, what is the most concerning is bad programming. 🙂
 
ostripper, I remember when the personal computer came out at everybody was extremely concerned about people loosing jobs, that the computer would take over, but there were also those who predicted that nobody would use papers anymore and lots more.

You here exactly the same now about AI that i fact should be called TSS (Text Search Scripts), all are concerned about the exact same things and is debating the same. As the "AI" is right now are they nothing more that extremely fast search scrips that can repeat what it has been trained on, it has no creativity, no thoughts, no ability to give new knowledge, it can only combine old knowledge and maybe combine info across fields to highlight not previous discovered patterns.

No need to be concerned, what is the most concerning is bad programming. 🙂
But this truly different. this is NOT a "luddite" response. The very creators of this tech warn as I do. why ?
They know lazy minds will fall prey to the glitter and convenience of it. Just look at the present web , it's a piece of $hit.
corrupted with spam , porno and disinformation . 90% of server web traffic is ADS and porno.
AI will be "horny" , if it scrapes the majority of traffic. And , it will try to sell you something.
OS
 
  • Like
Reactions: fab and rayma
Humans invented the knife, it has helped our civilization to survive and get so big as it is right now, but it has also costed properly millions of lives throughout time, both by accidents and murder, I do still not think the knife should never have been invented. The computer has changed things for nearly everybody to the better, and yes you can find porn and things that are way, way worse. Is that a reason for not inventing the computer, I don't think so. And the same will go for the AI. You'll get virtual girl and boyfriends, you will get new kinds of porn, and you'll maybe get a tool to get away with murder 100% of the time, but it will also help us to reach new highs and achieve things we right now are not smart enough to think of.

Unfortunately, is what we call AI, as far from AI as a cockroach is from being Einstein. 🙂
 
I fall prey to LTSpice and I have not done calculations with pen and paper for years.

That’s how dark ages happen. When the tech disappears, people don’t know how to do anything anymore.

But pandora’s box is open. You can join the crowd or be trampled by it.

As a software engineer, I probably have 3-5 years to learn something new. That’s for sure.
 
The true "end game" of AI application ... It will be able to code and create new organic molecules and cures. BUT ,
what is most important to our Fascist masters ... is - that it will be able to sift through all the data that has been already amassed
from 1990-2022. They are 'retentive" - western US has 10 sq. mile server farms.
I suppose "if you nothing to hide" WTF ? But AI will surely fast-track personal info. Not to our benefit ?
In perspective , technology itself complicated our lives. Very few impersonal relationships , don't speak to neighbors.
I have not seen a time in my 59 years that we are more alienated and divided ?
Not a "rant" , but a stark truth....
OS
 
  • Like
Reactions: rayma
Do you still know how to carry burning wood in a pouch, how to make a coat out of a wolf skin, or how to perform bloodletting?
Not all old knowledge will ever be useful again and if you no longer are able to use LtSpice is it most likely replaced by something better, or we do no longer have electronic to build computers of, no matter why would the manual calculation not be necessary, just as transporting burning wood is no longer necessary to fast and reliable make a fire.

But this debate is exactly as it was when the computer got out, all this has been debated before than also, i.e. when the calculator came, the sliding ruler, the abacus before that and so on. New is scary if you look on the downsides.
 
AI already has a subheading on google news. Speculators are willing to dump billions into a technology unproven and
uncontrolled. Quantum computing is a converging (co)- technology that will coalesce into a true AI that will make us extinct.
To not see this is consistent with our species.
The EU ,at least has the mindset to vaguely realize this. Private corps. can not be trusted !!!
 
As more and more ingenious people from all wake of life use it - the smarter it gets.

And these smart people loose the capacity to think and design. Etc just my 2c’s

That’s one possible scenario.

Another possibility is that such tools become good enough to teach with infinite patience and care. Hence, people who could not acquire the knowledge and skills to be useful contributor in their field suddenly can and human-driven progress continues or accelerates.
 
  • Like
Reactions: aditya