I asked it to design me a Dynamic Tube Sandwich Emulator.
Did a good job, found no errors.
Did a good job, found no errors.
Designing the Tube-Based Dynamic Sandwich Emulator (TBDSE)
1. Core Components:
- Vacuum Tubes (Preamp and Power Amp):
The TBDSE will use vacuum tubes to give it that iconic analog warmth and harmonic distortion.
- Preamp Tubes (12AX7, 12AU7, etc.) for signal shaping.
- Power Tubes (EL34, 6L6, KT88, etc.) to drive the output and produce that warm, saturated sound.
- Dynamic Signal Processing:
- Dynamic Sandwich Layering: We need to process the signal in layers, and each layer represents a different frequency range (bass, midrange, treble). These layers will dynamically interact, mixing and modulating in real-time to give you a unique sonic "sandwich."
- Dynamic Feedback: The signal processing could use feedback loops to let the frequency layers merge and evolve depending on the input, making the sound "live" and adaptable. This could be done with variable feedback resistors or active filters.
- Modulation Section (Toothpick Modulation):
- This will be the playful, user-controlled part of the TBDSE. It could involve a mechanical knob (or even a physical toothpick-shaped lever) that changes the modulation parameters.
- This modulation could be used to adjust the filtering, frequency shift, or feedback intensity. For example:
- Moving the "toothpick" could shift the frequency response, introducing more sub-bass or bright treble.
- It could also introduce phase shifting or modulation effects (flanger, chorus, etc.) in real-time.
Oh my good god, careful I bought a box full of tubes drunk on I ebay one day (dont remember doin it)
I get the box to my surprise and decide to do an experiment.
I uploaded the image of the labels to Gemini and asked it to build a guitar amp or pedal out of the tubes it sees.
I had to correct it by comparing to data sheets constantly.
It does better when you name a certain section of the build to reference later
Remember, the company has the incentive to make it sound authoritative when it clearly should not.
I get the box to my surprise and decide to do an experiment.
I uploaded the image of the labels to Gemini and asked it to build a guitar amp or pedal out of the tubes it sees.
I had to correct it by comparing to data sheets constantly.
It does better when you name a certain section of the build to reference later
Remember, the company has the incentive to make it sound authoritative when it clearly should not.
I think the only thing missing is a cow bladder and a fever 🙂
I am intrigued by the choice of ECL11, just using the triode part as driver and leaving the pentode.
Is the triode in that tube something special? Or is it just cribbing another design somewhere out there?
Is the triode in that tube something special? Or is it just cribbing another design somewhere out there?
So ... somehow we have to persuade AI that when asked 'Which is better for a high quality amplifier tubes or transistors?'- it replies 'Stupid question, tubes of course'. Then the enthusiasts will come flocking to the forum, and I can start shifting some gear.
I see one significant use of AI as a distillation apparatus, like how its used to search through prior legal findings which is a laborious task when done by humans.
Previously I posted a DYIAudio screenshot, where one thread talks about a high performing two-way speaker design and randomly adjacent to it, another thread talks about a good sounding two way speaker design. The 1st thread being 14k replies, it's easy to understand why the second thread was started; it was likely hoped someone would simply provide a distillation of the (much) larger threads content.
This is something I believe AI could do and maybe even be good at. Alas, it is not (yet) a function provided by the software that hosts this forum.
Which is better for a high quality amplifier tubes or transistors?
Copilot:
Great question! The "better" choice often depends on the sound characteristics you’re after and the application of the amplifier.
Sorry Hector
Copilot:
Great question! The "better" choice often depends on the sound characteristics you’re after and the application of the amplifier.
- Tubes (vacuum tubes) are loved by audiophiles and musicians because of their warm, rich, and smooth sound. They tend to add a pleasing natural distortion, which is why they're popular in guitar amplifiers and high-end audio systems. Tube amplifiers also offer a certain nostalgic charm with their glowing appearance. However, they're bulkier, require more maintenance, and can be less energy-efficient.
- Transistors, on the other hand, are more compact, durable, and efficient. They produce a clean, accurate sound with less distortion, making them ideal for applications where precision is key, like home theaters and portable devices. They're also more affordable and robust compared to tubes.
Sorry Hector
No ‘intelligence’ not self aware, AI is still just a (complex) differential (marketing) engine gathering and regurgitating data. When properly ‘trained’ AI is proficient at processing large data sets quickly benefiting those who understand how to train it. No technological breakthrough, no consciousness, it doesn’t understand what you’re really asking so it cannot know if it has answered correctly. It simply processes data faster than you and its quite pompous about it. The rabbit hole is deep however, the illusion of intelligence is there, one might train an AI model to design an amp then train the model to send the design to a simulator, check results and modify the design based on results. Still a virtual simulation which may or may not translate to the real world. To do this repeatedly and quickly simulates learning and an altar we seem willing or even desperate to pray too. This is the basis for machines designing, building and improving themselves which is of course the beginning of the end for us meat bags…
In a tube the cathode is connected to ground (directly or through a biasing resistor) and this is the return path for the power and the signal associated with that tube. There is no way to split power and signal return paths, or is there??
The last course I did was teached by one of the main consulent in EU, and information were changing while doing the course.
Thank you for proving my point about praying to the altar. Its just a program, not a simulation. Although that may be the goal it’s lightyears away from simulation of either an individual or collective consciousness, it’s created to dazzle and confuse and market. Lets not forget the first time ChatGPT was released to the world it immediately descended into porn and violence in it’s replies and had to be trained by thousands of people for years before it was once again ready for its marketing (re)debut. Not a simulation but a reflection of the data it was asked to process. Now that the sex and violence have been removed it’s a poor, censored reflection restricted from giving a true image of ourselves.
Interestingly Copilot cannot generate schematics. However it does seem to be able to generate pspice models for things like a single stage tube amp and you do seem to be able to ask for voltages etc. However like a young person it does get things wrong and you can reason with it up to a point. I still find that quite amazing given you can also ask about such a diverse range of subjects.
It is like paying for a lottery ticket and missing the jackpot by 2 digits. If it had said 'ECH21' I could plan my retirement; I have cornered the market.
Every active system has a cathode/emitter/source/whatever to ground.
But there's also a grid1/base/gate to consider, a driver/output device, the main PSU, etc...
It suggests not to ground the input signal on the loop of the first filtration stage.
None of the AI I tested can do schematics. They confuse pinout and connections for tube amps.
But nowadays, on 8 billions humans, who cares about tubes? They do better with modern electronics and do wonders in software.
Spice models fit very well too, indeed.
It is to wonder - never heard of anyone doing it yet - if there was an "electrical engineering" specifically trained AI. Like, what would it do after getting a look through the "treasure trove of Japanese amplifier schematics" referenced in a thread posted here. If you fed it the old Markus circuit books, the old Radiotron designers handbook, the old National Semiconductor Linear Applications Handbook.
Somehow I doubt these specific information sources were encountered and thoroughly examined in these AIs "trained on the internet" attempt at a general intelligence. Because general intelligence is the goal; I'd assume the category specific ones come later.
One of the things I'd get asked at work is "how do you know" what bit and pieces to use, to effect a solution to some random electrical / instrumentation problem. I couldnt explain how that comes about; "I just do". But you cant know everything there is, like I dont know all the solutions the entire product line of THAT corporation provides.
That - ar-ar - would be an excellent application for AI; have it digest the entire Mouser and Digikey catalogues and be able to route out an available component solution based on a verbal query, that many engineers would have no idea even existed.