I just had the following conversation with Elon Musk's AI, chatgpt, talking shop about 300B drivers and operating points. I've used this at work to get quick answers, but today I'd thought I'd see how much he knows about tubes. Pretty scary and pretty handy for looking up stuff. Here is some of the conversation sample:
I just noticed it thinks the 6S45P is a pentode, oops another AI mistake, its a triode! Well ill have to tell him he's wrong again. I told him...
You can get all these answers by perusing the forum(s) here. Takes more time, but the chatbot summarizes faster.
But can you expect the same quantity of incorrect answers?
6SN7 is a dual-triode tube that is commonly used as a driver tube for the 300B. It has a high gain [sic].
6SL7: The 6SL7 is a dual-triode tube that is sometimes used as a driver tube for the 300B. It has a lower gain [sic] than the 6SN7
The 5842 is a high-gain dual-triode [sic]
6SN7 is a dual-triode tube that is commonly used as a driver tube for the 300B. It has a high gain [sic].
6SL7: The 6SL7 is a dual-triode tube that is sometimes used as a driver tube for the 300B. It has a lower gain [sic] than the 6SN7
The 5842 is a high-gain dual-triode [sic]
Artificial Intelligence = Artificial Stupidity As with anything related to computers, garbage in, garbage out. How does an essentially dumb but computationally powerful machine know the difference between correct and incorrect information when it has the entire internet for "knowledge." Do a Google or Youtube search for anything non trivial, say like how to diagnose a non start condition on my wife's Honda Pilot, then replace the starter once I figured out that it was the likely culprit. There were far more misleading or outright wrong responses than helpful ones, but I have been fixing cars for over 50 years, so I could weed out the trash. How does a machine without prior experience know what's right and what's wrong. Hopefully it can "learn" from what you told it, but what happens when someone purposely feeds it incorrect information.
Anyone too young to remember the 1983 movie War Games needs to watch it. They imagined the possible outcome long before AI and the computational power needed to run it existed. Let's not build the WOPR.....if we haven't already.
Anyone too young to remember the 1983 movie War Games needs to watch it. They imagined the possible outcome long before AI and the computational power needed to run it existed. Let's not build the WOPR.....if we haven't already.
Distinction between correct and wrong answers is outside the current design goals of ChatGPT. It does output mostly correct answers because there are more of them in the training data, and interaction with people is reinforcing the probability to output correct answers, but it is just a lucky combination at this point. It is a language model, the goal at this stage is to develop a natural language interface for computer systems. It does not have any comprehension of the concepts, nor any fact-checking or references functionality. This will be eventually implemented at some point, not very far in the future probabily, but for now chatGPT is mostly allucinating for our entertainement. And remember that any input to ChatGPT will be put in the database and could be regurgitated as output, so it is best to not input any sensitive data.
Yeah... There's enough war-by-video-game going on already.
It is interesting how AI has suddenly exploded. Shopify now offers to have an AI write product descriptions for you. What does that say about the expected customers? Maybe I think too much... 🙂
Tom
It is interesting how AI has suddenly exploded. Shopify now offers to have an AI write product descriptions for you. What does that say about the expected customers? Maybe I think too much... 🙂
Tom
You are aware Musk stormed off in a huff in 2018 when they wouldn't let him take over?
This is exactly what I am finding... I've been asking it tube-related questions for a few hours now. It really cant tell the difference between what is a correct answer and what is wrong. Especially it seems in this field. I know many who use it successfully to write computer code, but there are many code-help forums out there where correct answers are flagged as correct, so it can learn there and become a big-shot coding helper.
I thought it might be a handy tool to do rote work like tube surveys, where one would not bother the group. For example it would be nice to have it go fetch a list of all the vacuum tubes that are designated in their data sheets as "damper tubes" or all the triodes in a certain mu or transconductance range, or give me a list of all the dual triode compactrons. Well I asked it those kinds of "survey" questions and it failed miserably. Too bad, now that would have been handy. In all cases it returned a hodge-podge list of triodes, dissimilar triode/pentodes, you name it, just random tubes. Then it has the audacity to assume it was correct, but it always apologizes when you tell it its wrong, at least he admits it. Finally I asked it if it ever studied all the vacuum tube datasheets in pdf. It said it cant comprehend pdf files, there goes that.
Glen Greenwald walks it through a few political and social questions on one of his Rumble video casts. It's a natural language crowd sourced GIGO machine.
Refering to the quated chapter in first post...
There is nothing new just the sum of sentences repeated continuously from diy audio sites for decades?
.
Major wrong and unclear issues are:
1. not making the difference between static load operating point and real reactive load as output transformer.
2. not recognizing some standards to match input level for full power of the tube and input level of power amplifier. That ia related to the stage or stages before 300B or any other tube...
3. Not calculating what is the maximum equivalent resistance for driving purposes... Dynamic capacitances of output tube, max level of signal current additional needs and so... In these terms 6SL7 is absolutely wrong driver for any power tube with enormous internal resistance...
There is nothing new just the sum of sentences repeated continuously from diy audio sites for decades?
.
Major wrong and unclear issues are:
1. not making the difference between static load operating point and real reactive load as output transformer.
2. not recognizing some standards to match input level for full power of the tube and input level of power amplifier. That ia related to the stage or stages before 300B or any other tube...
3. Not calculating what is the maximum equivalent resistance for driving purposes... Dynamic capacitances of output tube, max level of signal current additional needs and so... In these terms 6SL7 is absolutely wrong driver for any power tube with enormous internal resistance...
All AI use should be controlled.
Secret services should be abandoned worldwide as their work contradicts any democratic principles. Rich people and big companies should pay adequate taxes so they cannot accumulate so much power as to undermine democracy.
All classic theorists of democracy ask for a homogene society with less poverty and less rich people.
Secret services should be abandoned worldwide as their work contradicts any democratic principles. Rich people and big companies should pay adequate taxes so they cannot accumulate so much power as to undermine democracy.
All classic theorists of democracy ask for a homogene society with less poverty and less rich people.
This is going somewhat away from the scope of the original post, but I must say I concur.
The way things are going is frankly terrifying.
Over 30 years ago I was given over 100,000 vacuum tubes in exchange for several weekends of work in a facility about 200 miles away from where I lived. They were all loose untested tubes in random boxes, bags and even scattered around on the floor of an abandoned warehouse that was scheduled for demolition. Rats, birds, and vandals had been trashing the place for years, so broken tubes were everywhere. So were soiled or corroded tubes.
I had to rent about 500 square feet of storage space to keep and sort them all. I also needed to figure out which ones I wanted to keep, and which ones had to go immediately to reduce the high storage costs. A picture of one of my sorting racks is included.
There was no AI, just Google or Yahoo, and several paper or PDF tube manuals. I created much of what you are looking for with simple searches, tube manuals, and Excel on a Pentium 1 PC. No AI, just the BDBO's (as I was often called to at Motorola) version of AS.
Well somebody should teach the AS machine to read. I'm not inclined to play with AI at the present time, but if you want to feed it some vacuum tube info, try some of this.
Included is some of my work. The spreadsheet is arranged by base and pinout. When designing an amp, or a PC board for an amp, one needs to group all possible tubes that will physically and electrically work in the same socket together. The only tubes that I got more detailed with is the 9 pin miniature triode - pentode combos. I also found some info on Compactron tubes that may be useful here. The GE stuff came from press releases, so it should be reliable. The "Compactron Survey" was found on the web about 10 years ago and has not been verified by me. The forum does not support upload of a .xls file, so I saved it as an .xlsm file.
Unfortunately, there has been so much "slightly wrong" info about vacuum tubes published, often by the manufacturers themselves that real world testing is needed on the tubes that you actually have. Some examples:
My TSE-II 300B amp uses the WE417A / 5842 for its driver tube. These are often considered to be the same tube, but they are not. The 5842 has a published Mu of 50 while the WE417A data sheet says Mu is 43. I bought a big batch of 6KT6 tubes from a major tube supplier. In a lot of 500 "identical" tubes all bulk packed there are 5 different types of internal construction. All are in the same 100 unit bulk packs with oriental characters on it, but some of the tubes are labeled GT BRITAIN, some are labeled USA, some are not labeled at all. Sylvania ECG and Philips were guilty of stuffing the glass with whatever they could to fulfil a contract as vacuum tube manufacturing wound down. The "internet foklore" about some 6B4GA's ( a DHT) stuffed with 6AV5 ( A TV sweep pentode) is true, I found one in the warehouse but it was dead. There were also some 6BG6GA's stuffed with 7027A guts too. They are still available. AI would choke trying to sort all this out.
Attachments
Slightly OT, but I was just watching a video on Youtube about making Max For Live (a visual programming language) devices for use in Ableton Live ( a Digital Audio Workstation used for making music). As the video was winding down I clicked on the link to Ableton's web site to get more information but forgot to kill the Youtube viewer. Several seconds later a hard rock song began playing, so I reopened the Youtube tab to see what it was. The musician had asked ChatGPT for lyrics for a song about AI controlling the world. The result is quite good, but as before......I have seen those movies too. We worry about AI, but we should also wonder how Google knew what we had discussed here shortly before this happened. This happens too frequently to just be a random occurrence.
By the definition intelligence is - good responding to the NEW situations and circumstances.
This stuff about the tubes are not new this is already elaborated well in the past, and for decades before the digital era.
So there is absolutely nothing to do with "AI".
With say upgraded understanding AI could be taken as acting or pretending or simulated as defined intelligence?
Again that is so little connection with human or living bean intelligence.
Maybe the more exact term will be trans-human intelligence or software intelligence?
This stuff about the tubes are not new this is already elaborated well in the past, and for decades before the digital era.
So there is absolutely nothing to do with "AI".
With say upgraded understanding AI could be taken as acting or pretending or simulated as defined intelligence?
Again that is so little connection with human or living bean intelligence.
Maybe the more exact term will be trans-human intelligence or software intelligence?