The Age of Personalized Hardware Is Coming(geastack.com) |
The Age of Personalized Hardware Is Coming(geastack.com) |
I had Claude design an entire 4 layer rp2040 based PCB from scratch and PCBWay build it. It worked on the first go, other than some silkscreen overlapped, which doesn't hurt anything. That was before Fable.
Then I had it design a case for the new pcb to 3d print. Also worked the first go, but with minor cosmetic issues.
People have yet to even BEGIN to appreciate what these things can do with the right harness.
There's no reason to think a conventional LLM is the right approach, of course. But the LLM could delegate the task to a program it wrote, given the right prompting and the right feedback loop(s).
They are different constraints than the ones in SW development, but is there some fundamental difference that prevents replacing the human in the loop with AI in the loop?
Would I be on the right track if I guessed there's a DSL for designing PCBs that would help enforce functional correctness?
That’s probably enough for an LLM to check if you’ve mis-wired something, missed a part or chose the wrong resistor to set a regulator voltage. Plus good old DRC/ERC. If you pass all of those, there’s a good chance things will work unless your placement is really bad, but you could manually lay out and autoroute a lot of simple boards. Not to belittle the parent but a 4 layer board is actually simpler in some ways because you have a power plane which is one less net to worry about.
For analog work you can even run a SPICE simulation.
But that was a few months ago, getting high hope with fable and seeing killed before I could even try it for that project killed all my motivations.
Functional PCB "artwork" is actually not intended for visual consumption, it must meet constraints and advice scattered in documentation, prior "art" in the training corpus, etc.
It doesn't need to be surprising at all
A good static SVG cartoon can be viewed as a multispectral photographic video, with a lot of information thrown away and distorted or oversimplified. Different species will disagree about what information should have been kept vs thrown away. A bee would "complain" it can't see any of the plant's UV markers when presented with plant imagery. Many animals depend on motion or movement to identify predator or prey.
At a certain point culture and personal life experiences start conflicting even within the human subgroup, and we see people argue why this or that flamingo is better than the other.
I also believe most design related to a physical object have documentation justifying the choices.
Can you elaborate on this? What did you use to "design" it?
Im thinking of getting a 3d printer and would love to explore this intersection a bit more.
The reason why AR glasses are not a thing yes is because there isn't a big enough battery to allow them to function for more than a few minutes.
Glasses form factor have space for about .9watt hours now, and ~2 in 4 years time (assuming current trends) Assuming a 14 hour day, that means that your have 140mwhr to spend every hour. A not very bright light on your glasses is about 30mw, a decently bright one is 90mw. Processing imagery on that kind of budget requires custom silicon, and a huge bunch of optimisations.
For the most people, the risks outweigh the desire for tinkering. Personalization will grow right at the vendor offering, not in the hands of customer. People don't even have the time to cook their own recipes. People have their own chores to worry about. I'm talking about bulk of the customer base, not the geeks.
There's a whole continuum from "Buy an off the shelf unit" to "here's a barebones case you can slide ready made switches and caps into" to "mix and match custom PCBs and cases" to designing your own PCBs and/or cases. There are pretty clear pipelines and even some levels of tooling for "draw up the layout you want and get a bunch of files you can send to production houses"
Takeaway 1: A lot of this is grounded in economic realities. I did the full bespoke route (custom PCB, custom 3-D printed case) and figured it probably cost me about $500 all inclusive to get what I wanted, and that's honestly a lot of money for a keyboard.
AI won't solve any of the economic problems. They can't fix "the minimum PCB order is N units, so now you have a drawer full of spares you paid for". They can't make the expensive part you needed cheaper, especially if you're an individual buying quantity of 1/5/10 instead of an OEM buying reels and containers-full. They can't change the fact that a case for a large widget will be expensive to mill/3-D print/mould/etc.
Takeaway 2: Customers may have surprisingly limited imagination for bespoke gear. There are galleries (and even coffee-table books) full of exotic keyboards. But Micro Center is full of $50 interchangeable "tenkeyless with RGB lighting" boards; throwing on a random set of "custom" keycaps, and that's enough for a large part of the audience.
Will these customers want or benefit from more tools, or will it just give them rope to hang themselves on and give them an excuse to bail out of the purchase entirely? Even if you can provide them a gallery of vetted turnkey choices, there might be more choice paralysis than actual benefit.
Takeaway 3: Hardware is forever (relative to software). You have a lot of small firms and group-buy products that disappeared and now the owners can't get an exact replacement or repairs. Conversely, Unicomp can gut and rebuild a 1986 Model M with new innards in large part because they've been selling the same basic design since a 386DX/16 cost as much as a Toyota Tercel.
If your AI spawns a galaxy of 1-of-1 bespoke products, who services and supports them? That seems like it's only going to appeal to the enthusiast-hacker type who can keep them alive themselves, who is least likely to need AI help designing them. Design for disposability isn't a great look for anything but incredibly low-cost, limited-usecase items.
If you are keeping inside the LEGO level of complexity, almost. But just as with code, if the project is complex you need a real engineer herding the cats. That said, it certainly can extend the reach of an effort, and is a huge help doing board reviews and data sheet analysis, etc. but the real engineering decisions are very very hit and miss, just like with code.
I had an esp32-box-3 lying around from a lapsed "voice agent" project from a year or two ago. Had a baby. Baby moved to another room, sleep trained. Baby either: 1. wakes up a few times a night, babbles for a bit, goes back to sleep OR 2. baby wakes up and fusses for N (=10) minutes, at which point parents need to go in and settle (that's the sleep training routine we use).
In either case, we do NOT want to wake up every time the baby does. Baby can go back to sleep easily, we adults have a harder time. A few rounds with Claude and the esp32 is now our new baby monitor. It tracks cry/fuss duration and publishes an audio stream (via a web UI or direct with, say, VLC). The audio only comes through AFTER N minutes of fussing have elapsed. It also posts notifications (to ntfy) after 30s and N minutes. My log says baby often wakes up 1-2 times a night and resettles almost immediately. We only wake up if the audio comes through, after N (10) minutes.
Also during the day it's really handy to be notified when baby has woken up from her nap. Let's us be out of the house, or in a distant room, and still keep track of what's going on.
It's fun to keep improving and adding features to this. Never would have had the time/energy to get this done without a coding agent. I ordered a set of 10 more of the esp32-box-3s to give them out to my friends (well, some are for other projects... so much potential).
(EDIT: Yes, I know this isn't AI designing hardware, but even writing code for embedded off the shelf stuff feels like a huge new potential.)
So they put a web browser in the device?
All my experience tells me it can do it with or without a GPU meaning if you don’t have one it can easily write a software render for a UI
With hardware you get extra safety risks of fires and shocks, so let's see
By the way, this is not new either. In the 1990s Microsoft tried to convince people how great it is if their fridge gives data to MS so MS can order needed food etc..
Fast forward some decades - many don't want to yield any more data to private entities, no matter which alleged "benefit" this would bring.
Producing PCBs (say, 5 or 10 units) is pretty expensive, and components are also costly on such a small scale. Combining these two requires additional money or time. Beyond that, you need to consider that you probably won't get it right the first time, and every attempt multiplies the cost.
It'll be a deal-breaker for many people — the risk is just too high.
Have engineer sit there to design pcb is expensive
Do you have a different experience?
You're lucky if you're in a region where these open-hardware companies sell their wares, even though many of them will go under in the current market.
All the people that I know in tech and not in tech, don't do that, even if presented with the option they're too busy to also get into this endeavour that still requires a lot of expertise. I stopped reading there
It's not that I don't like your point of view. It's that I can't stand AI slop.
I tried building a health device few years ago and got completely lost in how to setup camera and touch display with a raspberry PI. Would imagine, with AI running in a command line, it would be much easier.
Battery life decreases by 25%
Adds telemetry to see why battery life is suffering
Battery lasts 10 minutes and the device gets hot to touch.
Thats going to be a brutal intro into embedded hardware
Also, I think many consumers wouldn't mind if the rims, frame arms, or frame generally (and optionally a VR mode visor) where covered photovoltaically. if the total area exceeds the area of both pupils (divided by efficiency) the environmental light could power an additive display. (with additive I mean for example pixel-wise LED's so dark doesn't consume power compared to subtractive displays which generate a uniform backlight and then block needless light).
Just on that point, if we are talking about a monochrome screen that is just displaying text, that is just an IO device for a mobile phone, I can see the argument but its not really the case.
Screens _eat_ power, waveguides are really not that efficient, so need mini projectors that are overdriven to produce enough light. Current off the shelf for VGA/25 degree is 95mw for the projector only, none of the other hardware. (https://www.jb-display.com/product_des/17.html)
that gives you 4000 nits to the eye. Which is about enough for normal day to day tasks in normal light. (assuming you've cracked waveguides, and thats an active area of research)
But, thats a poor experience, as its not "world locked" as in being able to attach things to real world objects. For that you need a whole new rendering system along with a SLAM stack to know where you are, a machine perception stack to work out what you are looking at. Bear in mind that 90mw for the projector is about 2/3rds of your entire power budget. bear in mind that even Bluetooth LE is quite expensive ranging from 12-40mw for streaming something like 1mb (although through LE thats a tough ask)
The other day, I took apart an old handheld vacuum cleaner to scavenge the internals. All there was inside was a small but powerful motor, and a couple of 18650 batteries with a charge controller and a switch.
Two small batteries, running a relatively powerful vacuum, which could do spot cleans of the whole house (corners and crevices and such) without needing a recharge.
I've been programming computers and tinkering with all sorts of hardware for more than 30 years. I first used FreeBSD in.....2001? and Linux not long after that. I've programmed OS code, I've grudgingly written VHDL, I've assembled a sound card for the Apple II I still have running - all this to say that I believe I'm in your tiny few.
And I'm so tired. Tired of having to debug all the things. Tired of having to pay attention to them. Tired of setting them up "just once" and then months later having to reverse engineer my own work because something failed.
So I don't. I leave nearly all my devices stock. I run Windows because I'm sick of debugging device driver issues. And I don't want personalized hardware with any electronics in it (bespoke wooden objects, those I love and make).
For me this has translated into a slightly different outcome: stock ubuntu with basically just zsh/powerlevel10k, I add component assembly when I buy a new desktop, I'm about to probably replace my home server with a NAS etc.
Lost me there. The rest is reasonable but other OSes (MacOS, Linux) also rarely have driver issues. Its not like Windows never needs fixing (these days more than Linux). These days you can buy computers with Linux pre-installed and compatible hardware.
My biggest problem with technology is with very stock stuff. Mobile apps that everyone pushes you to use that clutter up your phone and are often crap. Every time an app I use updates I wonder "what have they broken this time?".
I think so. There will be a short period, like now, where many will attempt to build their own products and the five good ideas out of five thousand will be incorporated into products built by those who know what they’re doing.
An absolutely technophobic friend asked me what agentic AI was today. I think Charlie Munger said “When even your barber is talking about it, sell and run”
The traditional baby monitor system had three states (ON and functional, ON and non-functional, OFF). It provides a constantly available and instantaneous test for "functional" - as long as you can hear sound from the other unit, it's almost guaranteed to currently be in a safe condition. Monitoring is constant, human-first, and all human. The system is relatively fail-safe.
Your design replaces this with a multi-state system and algorithms (adding ON in listening mode, ON in delay mode with output off, ON and partially-functional etc.). It removes any reliable method to prove an obviously "functional" state at any given time. Monitoring is non-constant (for the human), machine-first, and human as last resort (like corporate customer services). The system is not fail-safe despite being machine-first and at higher risk of error or malfunction due to complexity.
The risks include a few failed notifications or incorrect delay timing leading to early developmental trauma, such as fear of abandonment. It doesn't take much. We are still learning to deal with this human-human, before adding invisible unknowns.
If nothing else, what are babies likely to learn in a pre-verbal state where the days are spent sensing and observing their environment to develop the brain? It'll probably be how to game the baby monitor algorithm.
> Also during the day it's really handy to be notified when baby has woken up from her nap. Let's us be out of the house, or in a distant room, and still keep track of what's going on.
If you are going out of earshot of the baby by doing this, you are fully relying on the technology being functional. That would seem unnecessarily risky, and not nice for the baby (they can sense this stuff). The odds of a catastrophe are low in a singular environment, but still enough to worry. Almost inevitably it would end badly if this were scaled to a mass consumer product.
The idea is interesting from an engineering view, but from a human one it feels dystopian to insert a machine between parent and baby to this extent. It removes/replaces a layer of human-human connection. Where does engineering the natural human experience out of life end? Automated feeding? AI nannies and teachers? Then onto AI therapist?
My suggestion would be to do a lighter version of the features in a system which focuses on safety (always being able to monitor no matter what, for example on the ESP32, you could have the second core independent and direct-output the feed if the first core hangs or crashes). Feature-wise, rather than not alerting for 10 minutes, you could apply a DSP algorithm which reduces harsh frequencies for this period, whilst slowly increasing the output volume of the monitor.
I like the idea of alerts because it expands the base features without risk, and crucially makes the babies life better, as alerts could be sooner for important things that do need attention quickly. As an experiment, you could add an SD card and record snippets of cries which ended up in alerts, tag them with an ordered list of what you did to soothe or what was wrong, and see if AI can make anything of the data. Maybe certain cries can trigger alerts sooner, or cries over an extended period may indicate fever.
In my case, I have a separate heartbeat check process (running on another machine) which ensures that the monitor is correctly working (no silent fails). I also wired up a SECOND esp32-box-3 as a receiver, and it will complain loudly if the transmitter one isn't sending (much like most normal baby monitors do). The monitor (web ui) has an option of directly listening to the raw non-gated audio, which I use every now and then to confirm what's going on. I also record the streams sometimes for algorithmic improvement.
My aim is not to replace human connection with machines, it's to allow the parents a bit more sanity so as to be better carers for our children. So far we've gained in sleep, piece of mind, and reduced stress. The last one is important - listening to your baby yell for 10 minutes, even if you know they're just annoyed because they don't want to sleep, can be really draining (especially for mom). The notifications allow us to keep tabs on things without the direct audio line to the limbic system.
After a few rounds, I gave up on asking the LLM and just drew my SVG in BBEdit.
The altimeters are available OTC but i wanted to use my pi-zero and an IMU with a barometer. What I was designing was an i2c pyro board that could ignite up to 4 e-matches that sets off the blackpowder charges.
I've met the guy that made this altimeter, it's pretty amazing. The hobby is filled with engineers so some of the stuff they come up with is impressive. https://www.featherweightaltimeters.com/blue-raven-altimeter...
A human would have done a better job, and maybe used half the PCB... but it got it done and at the end of the day the board is 100% functional, and 100% machine designed.
Are you talking about fully assembled (pcb + components + solder mask + pnp + soldering + test) at 17EUR/unit? If so, that's incredible cheap for a finished product delivered to your door. If not, you're doing something wrong.
For example many are treating vibe coding like slot machines and the app monetisation gamble is unlikely to pay off.
I don't want to have to carefully pick hardware for compatibility. I want first class support for pen and touch. I value thin, light, and long battery over the ability to run a custom bootloader.
Here's a thread where I ranted with specific examples 3 months ago: https://news.ycombinator.com/item?id=47463982
Looking at your specific examples they are not a like for like comparison. Two are Ubuntu on Raspberry Pi issues (so on a device designed for experimentation, and not even running the usual distro). One is a bug in the GUI for True NAS. One is unhappiness with the default installed command line packages for Debian on WSL.
When I used Windows I had just as many irritants. It was very noticeable when I was using Linux at home and Windows at work (and is one reason I switched to Linux). it is definitely not true that Windows always works flawlessly for everyone in all circumstances. If you count every irritation you find with every Linux distro on a wide range of hardware it will add up to more than Windows on typical hardware. Have your tried Windows on your Pi? Have you replaced your TrueNas server with a Windows equivalent?
Those "tools" you throw out so cavalierly cost around $50,000 to $300,000 per annual seat license with all the optional packages needed for 40GHz signals (like Cadence Allegro + Sigrity X Speed / SystemSI). Even if AI could do it it'd be much cheaper ($3-8k) to get someone in China to use their license and knowledge to do it.
At some point the cost makes it irrelevant. If you drive a Bugatti Chiron, you don't care about the price of petrol. Humans or AI, there's not really a difference if the capital costs vastly outweigh the labor costs.
I’m curious if AI’s know enough to do some useful high frequency / data rate layouts without paying for $50,000+ simulation tools.
Then you can make it process existing good designs and make it propose a general design/routing methodology (and it's own scripts/tools for calculating/optimizing the "thumb" rules that would be used throughout) for initial placement/routing of key components based on all the collected data.
Then you may have enough context and tools to have it make use of them during design.