Microphone schematics by chat GPT

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
I found it exceedingly dumb. I asked it a series of questions about audio transformers during which it recommended a book on audio transformers by Bill Whitlock. it even gave the ISBN number. I looked it up but the ISBN it had given me was for a completely different book - not even about transformers. So I told it this. it apologised and gave me a different ISBN. Long story short I did this about 10 times and each time it gave yet another wrong ISBN number. II tried once more and also asked it if it was absolutely sure this was the correct ISBN number and and it confirmed that it was but it still gave the ISBN of an unrelated book.

Not in the least impressive.

Cheers

Ian

Annoying isn't it?

The problem isthat ISBN codes get recycled once the original book is out of print.

My personal pet-peeve with internet info is that sometimes the date isn't correct, or completely missing. And that's a big hurdle for AI.
 
Here is a repost from another forum.

In this instance I find the results highly intelligent.

I just asked Chat GPT 4...

Does the following sentence make sense?

"Then perhaps when people see that man was clever enough to not only arrange the elements in a manner that creates AI, (Algorithmic Intelligence) but was even more clever by getting his creation to arrange itself."

And Chat GPT 4's response was...

The sentence makes sense grammatically, but it could be rephrased for better clarity and precision. Here's a possible revision:

"Perhaps people will be impressed that not only did humans arrange the elements in a way that created AI (Algorithmic Intelligence), but they were even more ingenious in designing their creation to be able to organize itself."

This revision clarifies that the focus is on the intelligence behind creating AI and the design of self-organization, rather than on a specific man or person. It also avoids the ambiguity of "getting his creation to arrange itself," which could be interpreted as the man creating AI, or as the man creating a system in which AI could emerge and self-organize.
 
I somewhat agree. But I wouldn't say it's exceedingly dumb.

You are absolutely correct, it is not smart enough to be that dumb! See this:

1682076934440.png

So it took all the words and values I gave it and came up with some wishy-washy and technically totally incorrect BS, that is actually just using what was in my question and inaccurately.

So I directed it to the patent and got this:

1682077267398.png

The values have no relevance to the circuit in the patent and are just randomly pulled out of thin air (I'd say it's arse - but AI's don't have one).

Again, everything presented is utterly bogus and shows zero comprehension of the actual question, zero comprehension of the patent and the circuit, a total inability to find the example values from the patent and present these (even if they are wrong for my question).

So here is what ChatGPT had to work with:

US5450034A - Reflected plate amplifier - Google Patents

It contains this:

"By way of demonstration, and not to limit the present invention in any way, the values of components shown in FIG. 4 may be as follows: V101 is one half of a 12AT7 dual triode, C101 is 0.1μF, R101 is 100KΩ, R102 is 620Ω, R103 is 7.5KΩ, R104 is 620Ω 2 watts, and Q101 is a 2SD631 pnp power transistor with small heat sink."

And this:

1682077947603.png

"The operation of the circuit of FIG. 4 may be described as follows.

An input signal is applied to the control grid of tube V101 through capacitor C101. The control grid is biased to zero D.C. volts by the grid leak resistor R101. Cathode bias for the tube V101 is created by current flowing through R102 in the manner of a conventional self biased tube amplifier. The collector current of the tube V101 flows to the positive 15 volt source through the forward biased base-emitter junction of the PNP transistor Q101. The plate current of V101 is responsive to the input signal according to the tube's transconductance factor.

The base current of Q101 is obviously equal to the plate current of V101. The collector current of Q101 is dependent on the base current, and will be many times greater than the base current due to the transistor current gain. The collector current flows through R104 and produces a voltage drop across R104 which is proportional to the collector current. The collector voltage is defined as the amplifier's output voltage. The output voltage at the collector of Q101 swings in-phase with the grid voltage of V101.

This is because of the fact that the collector current increases in proportion to the plate current which increases in proportion to the grid voltage. As the collector current increases, so does the collector voltage because the collector current is a reflection of the plate current. This is in contrast with a conventional plate load circuit where the plate signal is in inverse phase with the grid signal. The output impedance of the circuit of FIG. 4 is mainly determined by the value the load resistor R104, which in this case is 620 Ω. This is a much lower output impedance than provided by conventional tube stages, thus meeting another objective of the present invention.

Negative feedback to the cathode of tube V101 is used to stabilize the operating point and gain of the RPA of FIG. 4. This is provided by coupling a portion of the output signal to the cathode through the branch of R103 and R102. Because of the negative feedback, the voltage gain of the RPA of FIG. 4 is calculated to a reasonable accuracy as R103 divided by R102. The D.C. operating point at the collector of Q101 is a function of the tube's threshold bias, the transistor's beta, and the feedback ratio. For example only, the desired collector operating point might be zero D.C. volts. The proper D.C. voltage bias at the cathode of V101 can be established by adjusting the value of R103 until the zero volt operating point was reached on the collector of Q101. The circuit values given by example for R102 and R103 reflect the experimental results of this adjustment for a specific tube and transistor which were used."

If that is "AI" we are safe, no need to stand at street corners with placards reading "The end is nigh"....

1682077536585.png

Now here is roughly what I would expected from "Average Intelligence":

Theory, Practice and Sexlife of the "Reflected Plate Amplifier" Hybrid, low Voltage Tube circuit

That whole things is so bogus, no substance, all just remixing and rephrasing what you gave it and random added bits.

Thor
 
Last edited:
You are absolutely correct, it is not smart enough to be that dumb! See this:

View attachment 108155

So it took all the words and values I gave it and came up with some wishy-washy and technically totally incorrect BS, that is actually just using what was in my question and inaccurately.

So I directed it to the patent and got this:

View attachment 108156

The values have no relevance to the circuit in the patent and are just randomly pulled out of thin air (I'd say it's arse - but AI's don't have one).

Again, everything presented is utterly bogus and shows zero comprehension of the actual question, zero comprehension of the patent and the circuit, a total inability to find the example values from the patent and present these (even if they are wrong for my question).

So here is what ChatGPT had to work with:

US5450034A - Reflected plate amplifier - Google Patents

It contains this:

"By way of demonstration, and not to limit the present invention in any way, the values of components shown in FIG. 4 may be as follows: V101 is one half of a 12AT7 dual triode, C101 is 0.1μF, R101 is 100KΩ, R102 is 620Ω, R103 is 7.5KΩ, R104 is 620Ω 2 watts, and Q101 is a 2SD631 pnp power transistor with small heat sink."

And this:

View attachment 108158

"The operation of the circuit of FIG. 4 may be described as follows.

An input signal is applied to the control grid of tube V101 through capacitor C101. The control grid is biased to zero D.C. volts by the grid leak resistor R101. Cathode bias for the tube V101 is created by current flowing through R102 in the manner of a conventional self biased tube amplifier. The collector current of the tube V101 flows to the positive 15 volt source through the forward biased base-emitter junction of the PNP transistor Q101. The plate current of V101 is responsive to the input signal according to the tube's transconductance factor.

The base current of Q101 is obviously equal to the plate current of V101. The collector current of Q101 is dependent on the base current, and will be many times greater than the base current due to the transistor current gain. The collector current flows through R104 and produces a voltage drop across R104 which is proportional to the collector current. The collector voltage is defined as the amplifier's output voltage. The output voltage at the collector of Q101 swings in-phase with the grid voltage of V101.

This is because of the fact that the collector current increases in proportion to the plate current which increases in proportion to the grid voltage. As the collector current increases, so does the collector voltage because the collector current is a reflection of the plate current. This is in contrast with a conventional plate load circuit where the plate signal is in inverse phase with the grid signal. The output impedance of the circuit of FIG. 4 is mainly determined by the value the load resistor R104, which in this case is 620 Ω. This is a much lower output impedance than provided by conventional tube stages, thus meeting another objective of the present invention.

Negative feedback to the cathode of tube V101 is used to stabilize the operating point and gain of the RPA of FIG. 4. This is provided by coupling a portion of the output signal to the cathode through the branch of R103 and R102. Because of the negative feedback, the voltage gain of the RPA of FIG. 4 is calculated to a reasonable accuracy as R103 divided by R102. The D.C. operating point at the collector of Q101 is a function of the tube's threshold bias, the transistor's beta, and the feedback ratio. For example only, the desired collector operating point might be zero D.C. volts. The proper D.C. voltage bias at the cathode of V101 can be established by adjusting the value of R103 until the zero volt operating point was reached on the collector of Q101. The circuit values given by example for R102 and R103 reflect the experimental results of this adjustment for a specific tube and transistor which were used."

If that is "AI" we are safe, no need to stand at street corners with placards reading "The end is nigh"....

View attachment 108157

Now here is roughly what I would expected from "Average Intelligence":

Theory, Practice and Sexlife of the "Reflected Plate Amplifier" Hybrid, low Voltage Tube circuit

That whole things is so bogus, no substance, all just remixing and rephrasing what you gave it and random added bits.

Thor

Your statement "If that is "AI" we are safe, no need to stand at street corners with placards reading "The end is nigh"....', is naive. There are other applications of AI, such as deep fakes, which already significantly threaten humans. Seeing isn't believing anymore, neither is hearing.
 
@Thor,

You wrote: "The values have no relevance to the circuit in the patent and are just randomly pulled out of thin air (I'd say it's arse - but AI's don't have one)."

interesting philosophical point: There is a well known saying amongst organic intelligences that opinions are like arseholes - everyone has one. The corollary is that anything that does not have an arsehole cannot therefore have an opinion and as we all know, anything without an opinion is not intelligent.

Not to mention that fact that all current AIs are simple inference engines.

Cheers

Ian
 
Your statement "If that is "AI" we are safe, no need to stand at street corners with placards reading "The end is nigh"....', is naive.

Hardly. I completely understand what a real AI is and all the potential issues, moral, philosophical and in terms of the survival of the human species.

I am strongly opposed to attempts to make real AI

There are other applications of AI, such as deep fakes, which already significantly threaten humans. Seeing isn't believing anymore, neither is hearing.

Deep fake is not as such AI (again). So far the deep fakes I encountered only fool fools. This can of course be improved.

Technology always poses challenges. And it has the potential to disrupt.

But seeing what ChatGPT actually does today, with a patent and a clear and simple task, it utterly failed. It still completely failed on the second attempt. I could go through multiple more iterations of the same.

Give the patent and task to a tenth grader without significant electronics background and average reading comprehension. They should be able to restate in different words the patents operation explanation and copy the section with values from the patent.

No real understanding of electronics needed.

Thor
 
The fact that we are actually debating wether an "AI"is interpreting/analyzing circuits correctly is scary enough....

Does it need to be scary? It’s happening anyway so better to work out how to take advantage of it than let it get ahead of us. The absolute worst thing we could do is stay set in our ways and let things get to a point where AI has the upper hand.

Look at the music industry and streaming, rather than adapting to this new thing the industry was stuck in its ways and now streaming services like Spotify are screwing everyone over.

So right now it can help you analyse circuits, so use it to design circuits faster. Eventually it will be good enough to design circuits from a brief, so make sure you’re coming up with good enough briefs. Idk what comes after that but it doesn’t all have to be doom and gloom. Technology’s there to make our lives easier.
 
Last edited:
The fact that we are actually debating wether an "AI"is interpreting/analyzing circuits correctly is scary enough....

We are not debating AI doing that. Because demonstrably it cannot do so.

It pretends to do that, but obviously fails.

I gave it a task that should have been trivial.

Take a patent that describes the function of a circuit.

Restate this function/operation in plain language.

Give example values for the circuit that are different from those on the patent, but can be found on line (I had posted them).

ChatGPT failed at each task completely.

Now what I asked it to do SHOULD have been an ideal test for it.

Thor
 
Hardly. I completely understand what a real AI is and all the potential issues, moral, philosophical and in terms of the survival of the human species.

I am strongly opposed to attempts to make real AI



Deep fake is not as such AI (again). So far the deep fakes I encountered only fool fools. This can of course be improved.

Technology always poses challenges. And it has the potential to disrupt.

But seeing what ChatGPT actually does today, with a patent and a clear and simple task, it utterly failed. It still completely failed on the second attempt. I could go through multiple more iterations of the same.

Give the patent and task to a tenth grader without significant electronics background and average reading comprehension. They should be able to restate in different words the patents operation explanation and copy the section with values from the patent.

No real understanding of electronics needed.

Thor
The junk circuit design output of ChatGPT could cause the same or worse outcomes that the flawed electrolytic capacitor specification did when it was used to make faulty capacitors that were used in the manufacture of electronic devices, the capacitor plague: Capacitor plague - Wikipedia.

Deep fake output by ChatGPT or other AI programs can be extremely harmful just like junk and fake news, conspiracy theories, racism, are. Deep fakes undermine trust.
 
So right now it can help you analyse circuits, so use it to design circuits faster.

Nope, it CANNOT do that. Yet.

Eventually it will be good enough to design circuits from a brief, so make sure you’re coming up with good enough briefs.

It will only copy circuits from datasheet's prepared by humans. And it will copy their mistakes.

Thor
 
The junk circuit design output of ChatGPT could cause the same or worse outcomes that the flawed electrolytic capacitor specification did when it was used to make faulty capacitors that were used in the manufacture of electronic devices, the capacitor plague:

No. Because what it output was completely unrelated to any circuit that could remotely work.

The Capacitor plague (I remember it well, I had a Dell guy on site swapping out motherboards for maybe 1,000 PC's) was so serious because the resulting products worked, but suffered rapid and reliable early failure.

Deep fake output by ChatGPT or other AI programs can be extremely harmful just like junk and fake news, conspiracy theories, racism, are. Deep fakes undermine trust.

Deep fakes are made by humans using certain tools. The problems are not the tools.

This is a parallel argument to the current debate of "banning" AR-15 rifles because some mass shootings (arguably high profile ones) used them.

We can look to china for an indication of the results. China completely banned private guns, not even airsoft toys are allowed. They then ruthlessly implemented their gun ban. Before that many full auto Chinese AK's were in private hands.

And ever since there have been zero school shootings or mass shootings.

What China has instead are massive numbers of knife attacks in public spaces and at kindergartens and schools invariably with multiple victims dead and other violent mass killings including fertiliser bombs and vehicles.

Fake News and what'snots are created by humans for their uses. Deep fakes are actually surprisingly harmless, next to what humans do without them.

Just take that lil New Jersey b!tch Sarah Bils aka Donetsk Devushka. No deep fakes, just an absolutely basic con with a fake identify. Before the internet she would even have showed her face and gone on TV talkshows. Now what she did caused massive damage to all sorts of things. And all to bilk a few pee-pull out of pocket change and I guess feel important.

Thor
 
Yes Thor you and most people on this forum know it failed the task you asked of it, my wife nor 90% of the people out there would not. Its more troubling to me that it does not know uts limitations, it tries to bullshit its way through....almost child like...
 
Back
Top