• Please note that Audiokarma will be offline briefly for updates early on Monday morning, September 28th.

AI to the rescue

How is chatGPT at providing transistor substitutes? I'm afraid to use my email to access the AI
No idea. I'm not that familiar with transistor part numbers, but here's what it said for 7199 tube substitutes:

What are the best substitutes for the 7199 tube?

Answer:
The 7199 tube is a vacuum tube that was commonly used in audio amplifier circuits. However, the 7199 tube is now obsolete and difficult to find, which has led many enthusiasts to look for suitable substitutes.

There are several substitutes that can be used in place of the 7199 tube, including:
  1. 6GH8A: This tube is a direct substitute for the 7199 and is electrically equivalent. It has the same pinout and similar characteristics, making it an ideal replacement.

  2. 6U8A: This tube is similar to the 6GH8A but has slightly different electrical characteristics. It can be used as a substitute for the 7199 in some circuits, but it may require adjustments to biasing and gain.

  3. E180F: This tube is a high-gain, low-noise pentode that can be used as a substitute for the 7199 in some applications. It has a different pinout than the 7199, so wiring modifications may be necessary.

  4. 6BL8: This tube is a triode/pentode combination that can be used as a substitute for the 7199 in some circuits. It has a similar pinout to the 7199, making it a convenient replacement.
It's important to note that while these tubes can be used as substitutes for the 7199, they may not sound exactly the same. Different tubes have their own unique sonic character, so it's worth experimenting with different substitutes to find the one that works best for your particular application.
 
I thought there were no direct drop ins for the 7199. I did replace mine with 6GH8's but had to rewire the sockets.

I do have a diagram that shows the rewire from 7199 to 6GH8.
 
I thought there were no direct drop ins for the 7199. I did replace mine with 6GH8's but had to rewire the sockets.
Ha ha yeah the AI is still fumbling on some things, but improving. Of course it only knows what it reads on the internet so still garbage in garbage out.

I’m guessing the next time it crawls Audiokarma it will see your response and slightly deemphasize equating these as pin compatible substitutes. If enough people start saying they are not pin compatible it will eventually rise into the AI’s conscience and it will then have “learned”.
 
Just got an "invitation" to register for Google Bard. Curious to see if it will "think" twice before answering and will use true information sources instead of super fast scouring of the first google results page
 
Ha ha yeah the AI is still fumbling on some things, but improving. Of course it only knows what it reads on the internet so still garbage in garbage out.

I’m guessing the next time it crawls Audiokarma it will see your response and slightly deemphasize equating these as pin compatible substitutes. If enough people start saying they are not pin compatible it will eventually rise into the AI’s conscience and it will then have “learned”.
Creepy! The Beast is coming!
 
Can the "beast" download a Pdf scan of say an old magazine or tube data sheet where the text is an image, then absorb or interpret the information?
Does everything need to be typed out or can it read?
 
Can the "beast" download a Pdf scan of say an old magazine or tube data sheet where the text is an image, then absorb or interpret the information?
Does everything need to be typed out or can it read?
Most of the old pdf magazines at worldradiohistory are searchable, so I'd assume yes.
 
Not audio related, but I was asked to supply abstracts/short summaries of (sometimes very long) texts on my website. One of my friends suggested running it through ChatGPT and getting it to summarise the texts. It actually did reasonably well, just needed some tidying up and minor rewrites of a few passages.

My wife says that they are starting to investigate the use of AI to make first drafts of things like press releases. AI is also showing great potential in medical fields as well, particulary in areas that require a lot of data sifting. It's only going to get better.
 
It seems that AI at present is just an aggregator of information out there. Surely much depends on where ‘there’ is and how it assesses the accuracy of its information for its ‘deep learning’?

For our hobby it would be great if we could ask something like: ‘I have a xxx amplifier and it is showing yyy faults. What is the cause of these faults and how can I fix it?’ Would it access the relevant schematic and database of common faults and point to possible causes and solutions? How would it assess the relative likelihood of these causes and the efficacy of the solutions?

Or is it just ‘engaging user-centred e-services’? (Thanks bullshit generator for that one.)
 
this is probably more than you wanted to know, but ChatGPT and similar AI systems are Large Language Models (LLMs). They basically have digested the entire contents of the internet and have billions of nodes that an AI system uses to make decisions. But AIs can also state conclusions to hypothetical problems that there is no "written" answer for on the internet.

For example, ChatGPT is pretty good at summarizing information, so if you feed it RDH4 for example as a text stream it would be able to summarize it probably quite nicely and give you sort of an automated help system to answer questions. That's one use for this kind of technology anyway.

So in general ChatGPT is just another AI system, but what makes it interesting and the reason it's got all the buzz right now is because it more or less knows in summary form all of what's on the internet, whether that be good or bad...you decide.
 
It seems that AI at present is just an aggregator of information out there. Surely much depends on where ‘there’ is and how it assesses the accuracy of its information for its ‘deep learning’?

For our hobby it would be great if we could ask something like: ‘I have a xxx amplifier and it is showing yyy faults. What is the cause of these faults and how can I fix it?’ Would it access the relevant schematic and database of common faults and point to possible causes and solutions? How would it assess the relative likelihood of these causes and the efficacy of the solutions?

Or is it just ‘engaging user-centred e-services’? (Thanks bullshit generator for that one.)
For that, you need human intelligence.
A tech can remember how they fixed things in order to do a better, more complete job next time, and do it faster. I fix cars all day but I do not document in the computer what fixed the car. Therefore the computer has virtually no data when it comes time to diagnose things.
It can only provide generic information and dont count on it ever knowing much about symptoms and problems with regard to tube amplifiers, because nobody is reporting what actually fixed the amp.
"Did you try changing the electrolytic capacitors"
The chat bot can only repeat what it reads on the internet.
Ask yourself, is that something we need more of?
Repetition of hogwash that seeps from the internet seems to be all anyone knows anymore, and it's free, everywhere...
Finding a seasoned expert in a particular topic is getting hard, and in some cases impossible.
When you find such an expert, be sure and pay them well.
 
Chat AIs as we’ve discussed in this thread needed something big to digest to prove out their algorithmic capability. This has been proven certainly for general knowledge, for which the internet fed as input works okay to put the AI engine through its paces to see if it can digest that much data. What better place to test than feeding it all of the internet. Understand that there are billions of decision nodes built up in ChatGPT that just a few years ago would have been impossible.

But you should not place a high level of trust in large AI models that feed as input the “entire internet” because we all know that half or more of the internet is garbage anyway.

So the real breakthrough here is that such systems can be used for specific knowledge summarization. For example if we wanted to build a Chatbot for AudioKarma to help users understand design and build concepts for tube audio, we would not use the general model that was trained from all the internet. Rather we would stand up our own instance and train it with known good and trusted information such as college texts, RDH4, Blencow, Jones, RCA tube manuals, etc.—in other words known trusted sources. We would then find that this kind of AI system built from the ChatGPT base would deliver highly accurate and trusted answers to inquirers’ questions.

That being said, If you wanted it to answer “why does my amp buzz or hum”, well in the same way a technician might have a hard time answering that without seeing and testing the actual amp, an AI system would have an equally hard time. In some cases there is no substitute for a human who can put the amp on the bench and run it through its paces.

Of course this makes sense, we all know that to answer generic questions about issues always requires the amp on your bench before you can offer specific diagnosis.
 
Last edited:
Back
Top Bottom