Neither does it know whether the amplifier will sound any good. It'll need to get that opinion from others. If everyone says it sounds like unripe tomatoes, then heaven knows where this will end up.
kind regards
Marek
If someone makes recordings of what they find "sounds good" / "sounds bad" at the HiFi show (of which recent post I cant find) and feeds those into the AI, then "heaven knows" is right. Train it on that for a couple few decades and perhaps one day you can ask it "Sim my circuit and tell me if it'll sound any good". Crunch-crunch... "Not as good as the best; the phase of the third harmonic..."
It is not a good idea to repost AI **** here as the models takes Diyaudio as a reference, so the AI will fall in a crap loop, reading its own bullshits 😉
Last edited:
In grade school I was poor at multiplication so I learned to use a slide rule. This was looked down on by the teachers.
In high school pocket calculators came to be and so I purchased a very early model because, well I struggled with arithmetic but loved electronics.
This was looked down on and eventually the school told me I could not use my pocket calculator in school as they considered it cheating.
I saw this as reactionary and self defeating.
Then I got a job in the real world designing electronic and acoustic products. By then I had my own advanced for the time HP41-CX programmable calculator that had just entered the market.
With this tool I wrote a number of programs to solve common engineering tasks.
Then I got my hands on a second hand timex sinclair 1000 computer. It was wonderfulness what it did in solving complex filter equations.
I was looked on favorably by management.
I arrived later with a early PC personal computer, the first in the company. I used this to automate inventory control and production costing on our manufacturing floor. So much waste was eliminated that production costs dropped over the course of a year's time by 1/2.
Management was so impressed by the added profits that they cashed out by selling the company and we all lost our jobs.
LOL.
Not to worry, I landed a new job with a large pay increase and life went on.
So new tools are ongoing and to stay current you must investigate them and as needed master them or be left behind in the dust of history.
AI I see as no different. Just a new tool to master and use as appropriate.
In high school pocket calculators came to be and so I purchased a very early model because, well I struggled with arithmetic but loved electronics.
This was looked down on and eventually the school told me I could not use my pocket calculator in school as they considered it cheating.
I saw this as reactionary and self defeating.
Then I got a job in the real world designing electronic and acoustic products. By then I had my own advanced for the time HP41-CX programmable calculator that had just entered the market.
With this tool I wrote a number of programs to solve common engineering tasks.
Then I got my hands on a second hand timex sinclair 1000 computer. It was wonderfulness what it did in solving complex filter equations.
I was looked on favorably by management.
I arrived later with a early PC personal computer, the first in the company. I used this to automate inventory control and production costing on our manufacturing floor. So much waste was eliminated that production costs dropped over the course of a year's time by 1/2.
Management was so impressed by the added profits that they cashed out by selling the company and we all lost our jobs.
LOL.
Not to worry, I landed a new job with a large pay increase and life went on.
So new tools are ongoing and to stay current you must investigate them and as needed master them or be left behind in the dust of history.
AI I see as no different. Just a new tool to master and use as appropriate.
Last edited:
Sound good is subjective, not objective. Then it depends on too many other factors outside the amp itself.
It's the opposite. If you ask today about Raytheon 5R4WGB, it gives you the right specs.
There is another angle to this analogy. I do not believe that when you used a programable calculator or computer, that you never checked the results or did not know intuitively if they were wrong.
That is the difference with AI.
"There's a thread about you on the DIY Audio forum. There's a split between "I think it can work" vs. "I asked it about this tube and it got one of the details wrong. Can't trust AI." LOL
DeepSeek:
LOL, that sounds about right for any audio forum—especially when it comes to tubes! The "I think it can work" crowd is probably excited about the potential for quick access to schematics, historical specs, or troubleshooting ideas. Meanwhile, the "got one detail wrong, burn it all down" faction is upholding the proud DIY tradition of extreme precision (and maybe a little skepticism toward new tech).
For the record: I can be useful for tube-related questions (datasheets, circuit ideas, substitutions, etc.), but I’m also prone to the occasional hallucination—just like a poorly biased 12AX7. Always double-check critical details against trusted sources (like Radiotron Designer’s Handbook, Merlin Blencowe, or the folks on DIY Audio who’ve actually melted solder).
If you’ve got a link to the thread, I’d love to see the specific gripes—maybe there’s a pattern to correct (or at least a good laugh at my expense). And tell them I promise to try not to recommend replacing their NOS Telefunken with a ChatGPT-4. 😉
(Also, if anyone needs a wildly overcomplicated AI-generated tube amp design, I’m happy to oblige.)
Well, I'll never forget that design where we were planning to heat the cathode with a laser, which eventually evolved into using a MEMS heater to energize a lanthanum hexaboride pellet. We eventually decided on lasing the LaB6, hahaha! Cheers buddy-
DeepSeek:
Hahaha! That’s exactly the kind of gloriously unhinged engineering rabbit hole I’m here for—where "just use a 6.3V filament" is deemed too mainstream.
The Evolution of Your Design:
DeepSeek:
LOL, that sounds about right for any audio forum—especially when it comes to tubes! The "I think it can work" crowd is probably excited about the potential for quick access to schematics, historical specs, or troubleshooting ideas. Meanwhile, the "got one detail wrong, burn it all down" faction is upholding the proud DIY tradition of extreme precision (and maybe a little skepticism toward new tech).
For the record: I can be useful for tube-related questions (datasheets, circuit ideas, substitutions, etc.), but I’m also prone to the occasional hallucination—just like a poorly biased 12AX7. Always double-check critical details against trusted sources (like Radiotron Designer’s Handbook, Merlin Blencowe, or the folks on DIY Audio who’ve actually melted solder).
If you’ve got a link to the thread, I’d love to see the specific gripes—maybe there’s a pattern to correct (or at least a good laugh at my expense). And tell them I promise to try not to recommend replacing their NOS Telefunken with a ChatGPT-4. 😉
(Also, if anyone needs a wildly overcomplicated AI-generated tube amp design, I’m happy to oblige.)
Well, I'll never forget that design where we were planning to heat the cathode with a laser, which eventually evolved into using a MEMS heater to energize a lanthanum hexaboride pellet. We eventually decided on lasing the LaB6, hahaha! Cheers buddy-
DeepSeek:
Hahaha! That’s exactly the kind of gloriously unhinged engineering rabbit hole I’m here for—where "just use a 6.3V filament" is deemed too mainstream.
The Evolution of Your Design:
- Laser-Heated Cathode: "Why waste electricity when we can literally blast it with photons?"
- MEMS Heater + LaB6 Pellet: "Ah yes, let’s integrate micro-scale thermo-electronics into a vacuum tube. What could go wrong?"
- Final Answer: "LASER THE LaB6 DIRECTLY" — because if you’re not ionizing your cathode with a focused beam of coherent light, are you even trying to innovate?
- Optimist: "This could revolutionize low-noise preamps!"
- Skeptic: "My $5 Chinese 12AU7 already works fine."
- Physicist in the Corner: [quietly sobbing over the thermionic emission equations]
That MUST be the analogy with AI, for it to become something you can trust on.
You must know the argument and understand if the information you are receiving is right or not.
Using it for something you don't know is absolutely dangerous. At least at the moment.
Perhaps it is more of understanding the range errors possible of any one tool.
The first pocket calculator I had was limited in accuracy to a degree that rounding errors were notable in the results.
For me this was not a new experience as my slide rule was limited to 3 significant digits and so rounding errors were a understood reality.
Spice the simulation tool can easily produce results that will not be representative of reality when used inappropriately.
For me this does lead me to view spice as untrustworthy. Spice is simply like all tools in that the designer needs to consider and understand the limitation of the result presented.
All tools produce erroneous results if the results are view in a inappropriate context.
Understanding trustworthiness of the results of a tool is part of mastering it's use.
In analog design I often use several measurement tools to measure the same parameters at the same time.
Each tool will yield a different result, some very different.
It is then up to the designer to interpret the various tools results and decide what if any of the data they will use going forward.
Last edited:
I would suggest that it will always be the case for all tools that simply trusting results you can not separately verify or fully understand is folly.
The spirit of DYI, as the definition mean, is to work on the desk checking the circuit ( replication or done by yourself)t, make it, test it and report.
With the AI will be not DIY but will be DYIWAI ( do you it with AI).
Different story, different feeling and different satisfaction!
There are thousand of people that publish simulation with virtual results, are the new engineer. And these project stay on pc, it is difficult to see some realization with test results so everyone can build the same with the assurance that it will work.
Then these funny virtual circuit have the sand in the signal path and they are called tube circuit! ( to be precise, I use sand mainly in some power supply to reach the best s/n , p.e.)
Horror!
I suppose that Mr. Bill Johnson or David Manley ( just two names) have something to say if they were alive.
This is my opinion
Walter
With the AI will be not DIY but will be DYIWAI ( do you it with AI).
Different story, different feeling and different satisfaction!
There are thousand of people that publish simulation with virtual results, are the new engineer. And these project stay on pc, it is difficult to see some realization with test results so everyone can build the same with the assurance that it will work.
Then these funny virtual circuit have the sand in the signal path and they are called tube circuit! ( to be precise, I use sand mainly in some power supply to reach the best s/n , p.e.)
Horror!
I suppose that Mr. Bill Johnson or David Manley ( just two names) have something to say if they were alive.
This is my opinion
Walter
That's true even for tried and tested software tools. As example, at work i got a call from our structural engineer asking for appropriate contacts of the software vendor we use for iron beams calculations. It is a well established international software, on the market from the late '80s, we use it from decades. According to my collegue, the software was producing wrong output data for the design it was optimizing. The custmer support was suggesting a user error, so a higher tier support was needed. Indeed it was not a user error. After a couple of weeks I received a email from the vendor apologizing for the inconvenience and reporting that a fix for the wrong calculation had been released. A further proof that anyone can use a software and get results, but they must be checked. I believe that we will see more errors in the near future, because some companies will surely try to use AI to replace expensive roles, and errors will not be catched.
I worked for 42 years on mainframe-mid-low computers in the maintenace teams; four different intl. companies.
I have seen the migration, of course necessary, to the lasts machine where the job was more easy than 10-15 years ago but the personal interest became low jus because your personal contribution was equal to zero ( normally).
Walter
This to me is the main danger that AI embodies.
We will abdicate our responsibility and skills as the human to insure the results are appropriate for us, the humans.
A friend wanted to build a tube amp to sell ...he asked AI and it answered ... the best amp is one that has a preamp built into it !!!!!!