I wonder if there were some employees at the manufacturing plant confused and laughing at this thing
This is pretty funny
I wonder if there were some employees at the manufacturing plant confused and laughing at this thing
This is pretty funny
I guess that is what happens when you don't have a billion of open-source CAD projects to train your model on.
I hope the post is satire, because it's funny as hell.
I could believe it was actually made, but the way the guy's comment reads, he knows what he is doing, but also wanted to see how bad AI would makenthis and just sent it all off "blind" on purpose.
I tried using ai to create an openscad simple object. It's basic and open source with lots of available examples.
I got crap back. After 4 tries I ended up just hacking it's code to make it work.
Did the fab house call up like "uh.. are you sure?"
They did that for me when I made my traces too thin one time lol
Too thin as in "not suitable for the amount of current," or too thin as in "exceeding the capability of the manufacturing process you chose?" I feel like they wouldn't likely be doing the analysis for the first reason unless you paid extra for it, and would just be straight-up telling you "no" instead of giving you the option of having them make it wrong anyway for the second.
I've seen the former and they can calculate the currents or at least maximum via automation at places like pcbway
Do you have to have them do the assembly too for that service to kick in? I've ordered bare PCBs a couple of times and wasn't aware of it.
It is a pastiche of [thing], more than an actual [thing] itself.
This is exactly what "AI" does, this is precisely what it's for
I could easily see this being a YouTube video, iterating to the point of getting a working PCB. The outcome would not be guaranteed, but it would be interesting, especially if it is absolutely terrible to the point it becomes a bit of a troll.
Of course they're a Project Manager.
I would love to see the looks on the faces of the Chinese laborers making this stupid thing.
Skynet v1.0 doesnt make functioning machines. At least when a person asks...
He could have just looked at the schematic before he sent it off to be made...
Yeah I’m wondering what the purpose was for actually having this thing fabricated. What point was he trying to make.
It looks like he was trying to prove that using AI to make things is bad.
I get that, but you could have come to the same conclusion just by looking at the schematics. You don’t need to actually have the thing fabricated to make that point.
I get it as well, but some people need something a bit more solid to realise just how stupid something is. I haven't looked at circuits since high school, but if I see something that just straight up doesn't work, that's undeniable.
He could have received the same board by having an intern or high school electronics student design it with no oversight.
He also could have sent the result to another AI with the prompt “point out all the errors in this design and tell me if we should have it printed”.
This was just a dumb “hur hur, AI bad stunt”
An unsupervised high school electronics student could easily design a PCB to power a LED via USB power. It's an extremely easy task to accomplish, which is why it was picked.
Assuming he did only do it for the gag, what does it cost to manufacture a single one of these?
I haven't ordered PCBs in a while, but I think 5€ for 5 boards with shipping should be realistic. Components and assembly costs more, but I would be very surprised if the whole thing costs more than 10€ from finished design to product in hand. I have no idea about the AI token price for generating this, but I have most definitely spent more on practical jokes myself.
It depends on how many layers there are.
Update V2: "Well, my house burned down"
I can see at least one innovation there. Diodes make current go one direction. D1 ensures current goes neither direction.
Just wait another month bro I promise it will get better bro (every month for the past 4 years)
I like how ever capacitor expect c1 is useless. R3 isn't connected.
The design acts like it there a common ground instead of insulation.
also the trace patterns don't look like they're conductive
C2 also looks useless. Itnlooks like the trace goes in and out on the same side.
FYI, the actual circuit properly designed is stupidly simple:
That's it.
Now, assuming R3 and R4 are properly connect CC line resistors (though WHY THE FUCK are the two lines of R3 routed on the other side of the board!!?), the only two other things needed are R1 and D1, nothing else.
Instead, there are way too many extra components, most notably this thing on the middle, supposedly a microchip (judging by the "U" code, can't see the actual writting in the device), maybe a voltage regulator but what would be the point!?
Worse, all 3 legs of that U1 device are wired together. If we're really really lucky, they go nowhere. Otherwise at least one ends up connected to a Ground line (ultimatelly coming from USB) and the other to a power (most likely the 5V from USB) - in other words, it's a short circuit of the power from USB. Not, just not good, but actually a seriously bad "I've never touched electronics in my life" mistake: there is literally no topology where the 3 pins of a 3-pin component are wired together like that, since electrically that's the same as not having it there at all (so even if connected to something else than 5V + GND, at best that component would never do anything). This is like something you figure out in the first hour of learning Electronics.
This shit is not just a little bad, it's incredibly bad and probably a danger to connect to anything over USB.
This is kind of a neat novelty thing I guess but I would have made sure it was functional before printing it
Am I missing it? Or is there not even an LED to light up?
There is, D1 in lower right is an SMD LED
Does D1 have only one leg attached? I don't see any via hole that connects to the back side of the board.
Got it. Thought it was just a regular diode.
It may take away jobs from loyal workers, be less efficient, require an AI subscription, be environmentally devastating, but at least the end product also doesn't work at all.
This reminds me of the people who trained neural networks on stuff before ChatGPT and uploaded YouTube videos with titles like, "I FORCED an AI to read ALL of twilight, and THIS is what it wrote!" and then they laugh at the garbage that comes out of the model. Like... yeah, the model is not good at this task that it was not designed to do. Some of the text is funny, but in the same way people don't really emotionally respond to AI art because there was no human intent behind it, I don't respond to AI "humour". It's using a tool wrong and then laughing that the outcome is bad.
There's satirical comedy to be had here, but it needs to be grounded in what actual people are doing. Personally I haven't seen anyone seriously expect a language model to be able to assemble a functioning PCB, so I can't enjoy this as satire either.
Or could it be genuine curiosity, just seeing what happens? Always possible but such a predictable outcome doesn't tickle my curiosity either.
So, there are all the reasons I didn't find this interesting. Why did I reply it all? I dunno man.
This reminds me of the people who trained neural networks on stuff before ChatGPT
I did this with Buzzfeed clickbait headlines, not using neural networks, just simple Markov chains.
I read it as satire. Is it not?
Dude says in the post that he just wanted to see what happened. If idle experimentation makes one an imbecile, then I suspect your standards for stupidity are out of line with reality.
There's not much point in sending it to be manufactured when it would fail DRC and a quick visual inspection. Experimenting is fun but there was no point wasting the money to make it when it clearly isn't a working PCB.
It probably cost 5 dollars max including shipping for a funny joke object.
He had an ai slop that pcb design out and then had it shipped from China. Even if they’re perfectly fine wasting their own time, shipping isn’t free
So you're saying it generated the schematic and the pcb.. and then they didn't look at all?
Or that they did and didn't see what was wrong with it?
They absolutely deserve to be the butt of jokes. I bet they let the robot summarise books for them.
Hell, both Gemini and Claude have done well for me in throwing some code together for simple apps, but I still looked it over before running it. I try to do that for even human-made open source code, although at least in principle, if it's been up a while, others have at least tried it and given feedback.
I'm sure this is just a joke or a proof of the issue; you'd have to look at the PDF or whatever it gives and see that something is way off.
And for the record, the best code that I've gotten from an LLM has been the first few runs of an idea, one that was thoroughly explained in the prompt. If you put together a vague prompt and continue to add to it, it will get worse quickly, with the LLM even changing parts that were perfectly fine. Maybe turning the temperature down, if possible, will help with that randomness, but it's always better to keep sessions short and precise.
I'd argue this has incredible artistic value. And I think that was the point.
Wow, he's using those new invisible inductors!
Live stream plugging it in to your computer, whoever you are. And then in to your phone and then your router.
AI is, in the end, designed to subjugate us further. It's intention and utilization is innately corrupt. That said, it's important to understand how it works. This is badly enough done that it reflects more on the user than the tool.
top 84 comments