I have been seeing pictures like this all around the internet. Look carefully. The breadboard is messed up, the LED circuit is backwards (not to mention that the LED's arrows are pointing inwards not outwards), the voltage divider has odd markings around it (even though to its credit it did get the equation right), and there is a wire going right into the breadboard. AI is that it is just an algorithm that is good at finding patterns in the circuits that they see. In this case AI saw millions of breadboard circuits and smudged them all together into this odd circuit. It changed its mind several times like on the breadboard it mixed up resistors with jumpers.

 

It would be hard to list all of the things wrong with that "photo". Sad, but not surpising.

 

View attachment 372708

I have been seeing pictures like this all around the internet. Look carefully. The breadboard is messed up, the LED circuit is backwards (not to mention that the LED's arrows are pointing inwards not outwards), the voltage divider has odd markings around it (even though to its credit it did get the equation right), and there is a wire going right into the breadboard. AI is that it is just an algorithm that is good at finding patterns in the circuits that they see. In this case AI saw millions of breadboard circuits and smudged them all together into this odd circuit. It changed its mind several times like on the breadboard it mixed up resistors with jumpers.

Baby steps. When I was young, I drew the LED backwards and messed up all the wires.

Give it time.

 

Baby steps. When I was young, I drew the LED backwards and messed up all the wires.

Give it time.

But you (probably) didn't then go out and try to sell your work claiming that it was an ideal educational kit.

The problem is what AI is capable of -- that will improve with time. It's that the people using it are doing their basic due diligence to vet the results. If they don't care enough to do that, then why should anyone expect them to put any more effort into the products they are pushing?

 

But you (probably) didn't then go out and try to sell your work claiming that it was an ideal educational kit.

The problem is what AI is capable of -- that will improve with time. It's that the people using it are doing their basic due diligence to vet the results. If they don't care enough to do that, then why should anyone expect them to put any more effort into the products they are pushing?

I can't help that there will always exist those who'll buy the patent medicine.

But here is what I know: AI has pretty much conquered programming -- as I have just proven to myself this week to my amazement (I'll provide details, if you wish) -- and it has conquered much of math. It's doing pretty good with art (which I assume is just a process of merging existent ideas to create a new, but derivative, form).

Engineering may take longer. Why? Probably the fact that we use a lot of symbols (and not words) -- whose orientation and position matter -- with hidden (but taken for granted) parameters to represent complex ideas, and we do lots of rules-of-thumb rather than pure mathematical analysis. Is it fair to say engineering is a bit more experiential (as opposed to textbook) than pure math or science? I think so.

The low hanging fruit goes first. Engineering will get there.

 

Further: even after AI is 100% reliable in all facets of our lives, there will still be snake oil salesmen, and they'll probably use AI to push their wares: their agents trying to fool our agents.

 

There is a photo floating around the internet, where a young lady is holding a soldering iron from the heating element.
There are some wisps of smoke. Either from the flux or the poor woman’s fingertips.

 

AI will cure all diseases! It will kill you. But you won't be sick anymore.

 

There is a photo floating around the internet, where a young lady is holding a soldering iron from the heating element.
There are some wisps of smoke. Either from the flux or the poor woman’s fingertips.

That's a photo (and there are many copy cats) that's been around for decades. The original (or at least the first one I ran across) was a genuine ad for some tech school recruiting campaign.

 

'She' is a bot, so clearly doesn't feel pain. Holding the iron near the tip gives more precise control over tip position.

 

Yeah. I also observed the same problem when I tried to generate banners using Gemini or ChatGPT for my Arduino course for Kids. The mistakes that you pointed out still need an expert eye. The results that I got were more frustrating. Even a picture of an Arduino UNO connected to the laptop's USB port or an LED connected to a breadboard turned out to be absurd last year.

 

"AI" is an old term to describe a new concept. Therefore saying "AI is doing something well...." or "AI is not doing something well..." is meaningless since "AI" is just a concept. It's like saying "I tried driving a car and it was slow".

 

"AI" is an old term to describe a new concept.

Wouldn't the assembly line robots I used to repair and program, come under the A.I. Umbrella??

 

If you don't know what you're doing or how to do it, let AI do it...

 

Wouldn't the assembly line robots I used to repair and program, come under the A.I. Umbrella??

Depends. If they went through motions that were programmed into them, then no. The notion of "AI" generally has to incorporate some form of "adaptive learning" beyond simply repeating actions by rote.

 

and smudged them all together into this odd circuit

That's because that AI model is designed to be a general image generation model. What the model knows is constrained by hardware running that model. It's just like an average person drawing a tank. The tank they draw would be mechanically incorrect in many places, but it is unmistakably a tank. An AI model is just like that. An average person doesn't know all the mechanical details that make up a real tank. They just never spent the time, or their head is already full of more useful things. That's exactly the same as hardware constraints when it comes to AI models.

 

That's a photo (and there are many copy cats) that's been around for decades. The original (or at least the first one I ran across) was a genuine ad for some tech school recruiting campaign.

View attachment 372715

It should be obvious that those folks staging this photo did not understand the difference between a scope probe and a soldering iron. I have met people like that, some were in upper management positions.

 

AI is that it is just an algorithm that is good at finding patterns in the circuits that they see.

I've recently been experimenting with AI, just using the "AI mode" tab that appears in Google search results.

I was quite impressed by how good it is at "understanding" the purpose of a circuit diagram. I was working on a 3.3V rail to rail op-amp design to amplify the signal from an electret capsule to feed into the ADC on an ESP32.
I just took a screen shot of the circuit in KiCad and pasted it into AI. It gave a comprehensive and correct breakdown of what the circuit would do.

On the other hand, it's terrible at going the other way. E.g. it suggested a couple of minor changes using English. I asked it to show me a drawing of the proposed changes and it gave me the drawing in ASCII art, which was almost impossible to follow, and littered with mistakes.

I asked it to give me a graphical drawing. It still contained some very basic mistakes, such as wires going nowhere, components in the wrong place, random wire joining dots not placed on any wires, etc. I copied what it had drawn, marked up the mistakes, then posted my marked up version back into the AI. It was able to spot its own mistakes. I went through a few iterations of this process, asking it to correct its mistakes and show me the corrected drawing. Each time I did this it introduced new mistakes and wasn't converging on a mistake free drawing.

I asked it why it didn't check its own drawing for mistakes before posting them to me. I got a long explanation that I didn't fully understand, but the gist of it was "I can't see my own drawings. I give the input to <something> that then produces the drawing that you see, but I don't get to see it"

 

..... as I have just proven to myself this week to my amazement (I'll provide details, if you wish) ....

Please do, I'm very interested in your experiences.

 

I've recently been experimenting with AI, just using the "AI mode" tab that appears in Google search results.

i did the same test. my observation:
the drawing is generated using another tool and result is inserted into reply. and that tool sucks. if you don't ask it to draw something but rather tell what the connections are (like a netlist). results are much better. same thing if you paste screenshot of schematics. parsing that image is done by another tool (improving but not perfect) so it can miss things and will tell you what "wrong" was with it (imagines dot where no connection is on crossing lines etc). if you correct it by telling what you see, it scores way better.

so if using Google AI, try to treat it like you are playing chess with a blind chess master ;-)