cross-posted from: https://lemdro.id/post/38940130
Taking into account that such a circuit would only need one resistor and one LED (apart from the connector), this shows that AI is still of no real use.
Exactly. People act like LLMs are sentient or intelligent. It’s just a “next word” predictor, no cognition or understanding if reality.
It’s a USB-C connector, so it would need slightly more that a single resistor to trigger power delivery. If I recall correctly, you can get away with two resistors formong a voltage divider on a specific pin to trigger a basic 5v supply though, so it’d be three resistors.
One 5.1k pull-down resistor on each of the two CC pins.
OK, conceded. Still quite different from the LLMs attempt.
Oh the LLM’s attempt is nonsense on all sorts of levels. The board is too big, conponents only have one terminal connected, the socket is on backwards, there’s random holes in the board, and I initially wondered if it was trying to use a power delivery IC, which would be massive overkill, but it looks more like a transistor, and it appears to have connected all if the terminals together. Oh, and unless d1 is a tiny LED, it hasn’t actually included the very LED it’s supposed to light.
Gaving the LLM desugn the board as an experiment is fine (result: fail), but sending it to be manufactured without even checking it was astonishingly wasteful. It’s just more e-waste. The more I think about it, the more cross I get.
That’s the first physical manifestation AI slop I’ve ever seen. I hope OP frames it and hangs it on the wall. It’s a historic moment and having that remind you every day will be worth it.
If you don’t proofread the text, check the diagram and think about the code, you’re taking a huge risk, and this is the result. If you outsource your thinking to a feeble AI like that, you deserve a humiliating price like this.
Journeyman of electronics here. That is not true.
Yes, you can in theory just use an LED, two resistors and build a voltage devider. You could as well just use a screw driver and look how deep you can get in an USB C port. In theory nothing should go bad, in practice, that is not how the USB C standard wants it.
Using something like an LM317 is the correct way.
Heck, technically the resistor isn’t even needed with an led that can handle the power from a USB brick
Note to others: Don’t do this unless you’ve verified that your particular LED already has the resistor built into it, which is how those work.
Otherwise, hooking a bare LED up to DC voltage tends to eventually convert it to a smoke emitting diode, especially since LEDs have an inverse temperature/resistance relationship. The hotter it gets the lower its resistance becomes so the more current it draws so the hotter it gets so the lower its resistance becomes so the more current it… pop.
At least for a little while!
/s
As an electrical engineer I’m reasonably confident my job is a very long way from danger. Analog electronics never made their way to internet forums at scale, and for humans the only real reference you need is an old book called the art of electronics. AI needs a ton of training data to be good, and outside of code and potentially law, I just don’t see what other fields have sufficient training data.
Maybe I’m a fool, but I’m a lawyer and I feel very safe. Everything I’ve seen so far is very bad, and in ways that don’t make me think it can get better. “next word prediction” is just not the right avenue to do legal analysis haha. It can do a fair approximation sometimes, but I think that’s honestly more dangerous because I’ve seen LLMs say things that look plausible but definitely don’t work. When I’ve seen it, I’ve been able to jump in and stop catastrophic mistakes…but how many people are just going for it without talking to a lawyer? Gonna make a lot of work for lawyers to fix all that lol.
same here. AI doing machine design? a company demo’d just AI drawing creation for us a while ago, and it could barely do that. how the fuck is an LLM going to fit together a few hundred custom made parts in a way that can be assembled, maintained, function properly, be manufactured without excessive cost, etc
I have yet to see AI perform a task effectively once you look deeper than the surface.
AI lowers the barrier to entry for people to do things. that doesn’t mean it’s going to do them better than professionals.
Yes! AI has been used to help with minimizing circuit designs and layouts for decades. It weirdly upsets me that AI has become synonymous with specifically LLMs, when the reality is that there are many different types of AI that has been an active field for years. Idk, maybe I’m just old and grumpy because I did my master’s thesis on AI before the rise of LLMs and a lot of modern ML implementations.
Yeah, I am not here to argue with “sometimes good enough is good enough.” Like…if you’ve got a raspberry pi making your garden spray water to the beat of “flight of the bumblebee” or some shit, and an LLM helped you do that without knowing anything about code…that’s cool I guess. But when people actually need stuff to work and are willing to pay for it, I just don’t see this tool ever getting to that level.
Also true when troubleshooting technical problems.
If you’re a complete noob and you have super simple tech problems, an LLM might occasionally help you out. If you have complex problems, an LLM will confidently send you on a wild goose chase. If you’re a noob, you won’t know until you’ve already spent a lot of time banging your head against the wall. ~(Don’t ask me how I know.)~ It takes experience to notice that the output is completely useless or even harmful.
I mean, maybe the LLM could at least file documents on time and at least attempt to do what it, on record, promises it will because that seems to be the biggest problem with all the lawyers here in Belgium 😂
My girlfriend’s case was fucked because the lawyer, with around 7 reminders and my girlfriend pre-writing the whole document, still didn’t file the document on time and also not “officially signed”, thus shifting the entire burden of proof from her former employer to her. Then lying about it for a year. Now we know why she pushed for a very very bad settlement so hard. (Like, the company could burn your house down and murder your children and you still would be sued into bankruptcy if you said anything bad against them, with about 10% of the legally required compensation, type bad)
Then a different law firm hired to clean up her mess is doing their best in that case, but also got hired to fight the bullshit invoices from the other lawyer’s “work” (she billed about 5x the hours she possibly could have done). And just never filed against the other lawyer so now that is very very late after 3 reminders (they literally have to send one email) and we might have to pay thousands of euros out of pocket.
Sorry, end rant, but at this point using an LLM would have been more useful than either of the law firms. Plus we had to compile all of the documents and pretty much write the entire thing (and the lawyers just slap their names on it) anyway 😂
That sucks man, I’m sorry to hear about that. Sounds like everyone’s malpractice insurer should be paying out big time (missing a deadline like that is practically always negligence).
Another reason to prefer a lawyer though…nobody to sue when the LLM misses a deadline! That said, I don’t doubt an LLM can be helpful to a self-represented party if the court has really good documentation. The LLM could probably explain clearly how to get documents filed and the like. Of course…if the court has really good documentation, a human being could also just read it and get it done.
This weekend i decided to adapt an old analog phone i thrifted, so that i could use it as a microphone on stage. Last time i soldered a resistor was a long ass time ago during the 20th century so i had to rely on Claude to help me every step along the way, including designing the overall circuit, telling me the exact components i needed to buy and how to assemble them.
It was a lot of fun and we did a piss poor job of it but it worked ! And we already have ideas for improving it in V2. Granted it was a very simple design but it felt like a barrier was finally coming down. I’ve been having this kind of ideas for ages but i never really committed to any of them, and now i know that i can knock one out of the park in a few hours of focus, it’s great.
I don’t think it’s threatening your job as a professional’s experience is tough to replace, but expect to see a lot of vibe-soldered side projects in the coming years. Probably a bunch of house fires too hehe…
ME here. I couldn’t care less about it…outside of the jobs that will be lost (erroneously or not).
The art of electronics and the student book with the exercises if you can find it.
Behold, future engineering of everything.
Bout to have some bridges and building collapse and cars going right off the road at random.
Take me back to 1999 please.
When people put “e.g.” instead of “i.e.” 🤌🤌
E.g. = eggsample
My big brain at work.
An eggsample is a valid measurement of how trying the times are.
He’s a senior product manager who probably makes several times more than you, how dare you imply he is stupid.
I know right
What’s the point of this LinkedIn post?
look, everyone, I’m completely stupid and my brain is not working, this is my name and contact info, please hire me
Did he want to accomplish something like this???
The LLMs are good on python and react because they’re trained on billions of files stolen from GitHub, not because they’re actually intelligent
They did not stole that much for PCB designs, so the copy machines will not work as well
And by the way, you send to production something without any kind of supervision???
Pretty sure he’s just trying to prove that his job Is still not able to be taken over by LLMs.
I think that’s what we all are doing here.
Looks like it’s just a fun experiment 🤷♂️
So fun, but there was no need to let some company to actually assemble and ship the e-waste from another continent. Anyone with basic electric knowledge would have known that it wouldn’t have worked.
He’s a product manager, his actions are like this? They pay him to ship code directly from Claude? Why aren’t firing him and just put Claude on the server?
I would think this is a personal hobby project just to test what it’s capable of, either to see what’s possible or for giggles. But to be sure you’ll have to ask the actual person.
tbh its kinda funny, shipping could be a bit costly but negligible if you’re making other boards too and ordered this vibeshit together with it
The copy machines will learn eventually with massive PCB SaaS like Altium and AutoCAD Eagle who’ll probably be working out, if they haven’t already, how to reword their EULAs so they can use PCB designs from industrial companies to train LLMs and then sell that back to their customers as a way of cutting down on design staff.
I have heard whispers in the wind that software like Autocad will transition to web based apllications, they say “oh you wont need a big beefy PC with good graphics card and lots of ram, just a simple workstation” and all i hear is “do all your designs on our servers and in return for saving money on hardware we will retain the software at all times, train AI on your IP and onsell your designs/concepts/engineering as we see fit.”
Does anyone else spot the irony that we’ve been using AI (simulated annealing) to help us design circuit board layouts for decades, only it isn’t an LLM so I guess it doesn’t count?
I guess people don’t think of SA as artificial intelligence. Does it even rely on neural networks or other traditional machine learning tools?
It doesn’t rely on Neural Networks, but it does rely on gradient descent/hill climbing, which is basically what a neural network does under the hood.
Annealing works, and therefore isn’t AI.
The problem is that you forgot to add “and don’t make any mistakes” to your prompt.
It’s like modern art but for EEs.
Thankfully, still OK to use ChatGPT for healthcare decisions. The human body and it nervous, endocrine, cardiovascular, musculoskeletal, respiratory and other systems aren’t as complex as turning on an LED… /s
I bet the tech that printed this was like wtf
omg , ofc its on linkedIn !
“Senior Product Manager, AI”
Not sure if he did this to show how stupid it was on purpose, but if that wasn’t his purpose, then this was really stupid.
First of all, you can’t boast an “open-source toolchain” if you’re using fucking Claude to operate it.
Second of all, did he have it assembled and shipped to him? He could have just had the PCB printed and then soldered it together himself.
Third of all, why would you do anything “blind” with AI? Its only real use is expanding your own capabilities, so forfeiting your own agency and giving it full control is idiotic.
Fourth of all, “stay tuned,” like he’s planning on trying this again…
How do LLMs have a ‘real’ use-case? Only actual beneficial application I am aware of is in large dataset processing for research purposes.
You literally just answered your own question.
Besides, you really had to torture my point in order to present it as somehow a defense or a promotion of LLMs, when clearly it was a critique.
What are people thinking when they post something like this? Do they think theyre conducting valuable experiments
Ikr. Btw what was your last contribution
I can’t tell if you think I’m criticizing the op or the guy in the screenshot.
But anyway my last contribution was sharing a video of a thai guy cooking in a wok while singing a linkin park song. Which i think outweighs sharing a busted ass AI made circuit board.
LLM is not logic! Ahhhh
