Superhuman Hardware Design Is the Easy Part
a16z.news
a16z.news
For a lot of applications where signal integrity is not critical, high frequency switching is not required and power density is moderate, yeah. Anyone can do that, even an AI. But you don't even need to make a pcb for those cases, just throw the components in a protoboard and fire away.
When you start adding variables to the search space (emi interference, matching routing impedances, heat management, mechanical integrity, etc) the search space grows so big that any non-expert will become lost in it.
Will future cheap compute power be able to overcome this? I'm not betting on it.
Famous last words. https://en.wikipedia.org/wiki/Bitter_lesson
Also did you not see the quoted post at the end, they are already training ai on running material science labs so we are quickly moving away from the non-expert.
AI isn't brute-forcing, or it would have been completely unusable for code.
Btw. the expert had to learn this, its probably available in written form one way or the other on the internet and LLMs have read that too.
You should bet on it.
With code or math or simple text, or games, you can have swaths of documents easy to digest and readily available, you can run thousands of tests per second and you can check results right a way.
With a power converter, a ddr4 module, an fpga, you have to build the pcb, assemble it, test stuff little by little, fix thing as they break, increase power, etc.
You can take notes on stuff that has been changed , the reason for the change is usually a few words note if any. And then you find something weird and you see that there is absolutely nothing on internet about it and you have to find out on your own. At least that's my experience with hardware design.
For the important stuff that makes you lose a board or a month of work you make sure to document the reason and the solution, but it usually is in a internal document no one will (regretably) read.
If you then add mechanical stuff, thermal management, failure modes, etc. I just don't think there's enough compute power to bruteforce such a huge space of solutions, even if you don't take into account that you can't really teach the machine about hardware design apart from some open source stuff and development kits where documentation, part lists and gerbers for the pcb are available.
Also, Claude can already write a multiphysics simulator for you. Doing EM analyses on the fly during design is something that will make LLMs much better designers than humans.
Have you tried Astra?
I'm designing an analog ASIC. It will suggest e.g. transistor matching. You can also give it hints for layout.
You still need to validate with SPICE simulation (and obviously silicon).
I'm also looking into non-LLM scripts that use brute force aided by heuristics (e.g. lookup tables) to efficiently search the solution space.
Disclaimer for stockfish in particular: Even if it's mostly bruteforce nowadays, stockfish usually has an edge because it can bruteforce with "taste" and can manage the clock very well knowing when to ponder deeper a position.
> ... search space grows so big that any non-expert will become lost in it.
And I am saying that's a fallacious argument: chess has an enormous search space too, and that does not imply AI can't beat humans.
Also
> You would beat anyone if you could calculate a few thousands moves per second, even more so if you had a neural network that discards suboptimal moves early.
So like the AI can? Your argument's dont add up.
In chess, being able to check hundreds of lines per second makes you a superhuman. But in hardware design you get lost if you try to search the solution space for a circuit, schematic, component selection, layout, routing, stiffener and thermal solution.
There are a few like positioning and routing where the solution space can just be considered infinite as there are infinite copper geometries (even if you stick with straight line polygons) that can connect a number of infinitely differently positioned components.
Then, for the mechanical stiffener, again you have an infinite number of geometries that don't clash with your components, don't run over your copper traces and make the board resist the mechanical tests (which the AI could check with a lengthy FEM analysis per iteration).
After all that, or in parallel with the mechanical analysis you can run the thermal ones and tweak positions of the hot stuff and thickness of stiffener walls to get everything fine.
I just can't see an LLM do that with a board with a few hundred components, even if most are decoupling capacitors.
Reality turns out to be more interesting, it was an exhibit of Southern Pacific's first steam locomotive, the CP Huntington, built in 1863.
So while it's true to say that the 1915 worlds' fair exhibited Southern Pacific's first steam locomotive, it's also very misleading to suggest it was new technology of the time.
I would say there is between me and other people a gap of 1-4 years. But we have news, the internet, video and audio.
So what might be 1-4 yeard today, it might have just been 10-50 years at 1915.
A new technology for the avg person
The locomotive being exhibited in 1915 was a historic exhibition, exhibiting their first locomotive, which itself was actually from 1863.
It was a history piece. It was not demonstrating new technology!
It would be like saying that the first shuttle orbiter, Enterprise, was first exhibited in New York in 2012, inspiring crowds.
It's a pedantically true statement ( it was moved to NY in 2012 ), but it doesn't actually tell you about the state of technology in 2012.
There are so many processes involved in making a high-precision bolt that goes into an F1 racing car. Well, hardware is always limited by laws of physics, it's hard to accelerate. Maybe in the future, AI will do this too, through robotics. I am still waiting for the robotics ChatGPT moment.
Life of Bolt(2016): https://www.youtube.com/watch?v=iptAkpqjtMQ
Dan Gelbart recently appeared on a podcast where he shared valuable insights into precision engineering. It was really great!
Title: Solving Impossible Problems for Fun and Profit - Dan Gelbart
Artists, at least, can make the process be the product itself. But who cares what story is attached to a piece of hardware? Not on a meaningful scale, at least.
Drones should be handled the same way as weapons! Strict licenses and mandatory insurance!
https://www.lesswrong.com/posts/bGoo3NWzMAzsLQceJ/drone-wmds...
These things almost never happen because people just aren't that evil, at least most of the time. When it happens, it's an extraordinary event.
And yet tens of thousands of us die in the US one or two at a time due to guns.
Humans are odd little copy machines. For example when a big name suicide occurs there tends to be a pretty sharp rise in copycat suicides.
Things tend to add up not in big events, but the long tail sweeps up huge numbers.
As for antipersonnel quads, Ukraine currently uses 7-8 inch ones. They are not small, quite loud and not exceptionally maneuverable. They are also extremely fragile and can be shot down by hitting just about any part of the quad. Also the thing about being difficult to detect using radars is just not true. They have a fairly large radar signature relative to their size which is also fairly unique because the spinning propellers cause Doppler shift on the rotating props. Small autocannons are already being developed for this role and in the urban environments similar systems using nets are being considered. There are also some prototypes of special rifles or rifle attachments for counter drone use being tested in Ukraine for infantry use.
The two most realistic dangers come from concealing them in shipping containers and then deploying them deep behind enemy lines (I am fairly sure this is already illegal in most jurisdictions) and targeted assassinations, where defending against drone threat is not that different than defending against other assassinations threats (concealed bombs, gunmen, snipers etc.).
You know it's easy to get illegal guns right?
Building an autonomous drone swarm capable of nuclear-weapon-tier kill count is far easier than building a nuclear weapon. We already know how dangerous military drones are, and those are still mostly bottlenecked by relying on human-powered target identification. Building drones to indiscriminately kill unarmored civilians is an easier task.
We are literally seeing the large scale use of robotic weapons platforms AT SCALE.
Machine guns create a storm of ballistic objects, and have been doing so for over 100 years. Large scale murder isn't new. We just have systems that have cameras that make it easy to put the footage online now.
This is just as true for guns as drones. Both need fast reactions and fine control. A human cannot control a swarm of drones any better than they can control a swarm of guns, but you can add autonomy (sensors, actuators, compute) to either to get some of those capabilities back.
Conventional munitions will always be more economical. It's hard to beat a few grams of steel in price per kill. Drones are only relatively cheap when compared to rifles, but they're actually replacing bullets/bombs. Great for high value targets, less great for indiscriminate killing.
Where though? It's certainly easier in some parts of the worlds than others, and surely something in the environment makes this easier/harder, I wonder if we could identify what exactly?
Ammunition is a bit harder, but again, there is criminal markets that fill the needs.
> but again, there is criminal markets that fill the needs.
Again, those are bigger in some places in the world, and almost non-existing in most parts of the world, even though it's possible to obtain in the right circles.
If we stop assuming the US's "everyone should have a gun at home" mentality is how it goes everywhere, it's still not "easy to get illegal guns" in most parts of the world.
The pipe is metal. You don't print that part.
Guns are 500+ year old technology at this point, manufacture is possible anywhere, as the industrious artisans of Yemen and Northwest Pakistan have shown.
But this was all in countries where guns were and are outlawed, which does make it harder to acquire guns, since those countries don't suffer from the problem of everyone and their mother having a gun, school shootings and all the other jazz that stuff comes with.
It still isn't "easy to get illegal guns" in those countries, even though it's much easier in other countries that are a bit behind the rest of the world when it comes to safety.
And just because its not easy doesn't mean there needs to be no law. No law means no conviction. There is no 'common sense law' which allows the state to put you in prision because its obvious that using some weapon in an unsave way is illegal.
And yes sometimes society adds protection to normal people so that they don't do stupd things. Plenty of 'smart' people allow others to get killed due to neglect, stupidity etc.
Btw. i'm pretty sure that having easy and quick access to a weapon is also a danger in itself. Sometimes you just need a little bit time of cooling down or talking to someone.
A license requirement for a car makes sense - you're dealing with forces well beyond the norm for what a human body has. Small drones though? I just don't see it.
Your 'arguments' of paper planes and human bodies are just strawmanning.
You can't fly a paper airplane a few hundred meters atop easily.
Feel free not to see it, i want to have my air space free and I would also never stand under a drone show.
Because assholes ruin everything for everyone at scale.
> Scooters
Remember when scooters were rare, then some VC's dropped hundreds of thousands of them in major cities so drunk people could plow into the side of your car at 20MPH?
Drones themselves are one of the more wildly dangerous technologies we've really taken off with in the last few decades.
We don't make rules for the good people of the world. We make rules because there's always a part of the population that's willing to take a big'ole shit in your glass of milk and sell it to you as pure milk.
All these demands always seem like rich people wanting to make sure regular people can't do anything.
Maybe bureaucrats are fed up of regulating technology as primitive as basic 1960s quality guided missiles. Or whatever. But drones not considered straight up missiles without the bomb part is a bit of a mystery.
"Superhuman Hardware Design Is the Easy Part" but then says "Superhuman hardware engineers are coming soon". The first example is a tweet "Vibe hardware is here" that shows a blender animation of a hardware, that looks correct, but hasn't been built or confirmed in any way: https://x.com/eminimnim/status/2098072497182666987?s=20 - The article freely admits it can't yet do the design side. So - in theory the actual hardware design is the easy part, it's not even possible of doing more than a convincing (but very good and actually impressive on it's own!) 3d model.
I'd actually be interested in something that shows how they're trying to get LLMs to design hardware. I'm not interested in investor propaganda that tries to imply things are far further along than they actually are.