It's not really that magical. As TFA points out, RFIC design, way beyond normal RF engineering, is close to black magic that relies a lot on the knowledge and experience of the designer, assisted by what would have been supercomputer-level-a-few-decades-ago modelling and design tools. What AI can do is a breadth-first exploration of all possible outcomes and then pick the best-performing one rather than the human-level "this seems like a good path to go down, let's explore it further".
the biggest question for me is how robust are these designs.
in the journal articles they did show measurements of real devices which agreed fine with predictions, but i didn't find them addressing it explicitly in the text. also, some systems they presented contained subblocks that were conventionally designed that could be carrying some of the weight.
or maybe i'm just sour that they're coming for my job? or maybe that's what they want us to think?
i think what wins in practice is simple ideas that can work in spite of all manufacturing and environment variations, and model limitations -- think stuff like feedback and symmetry. and what they show here is the opposite of that.
i've done blind optimization of circuit parameters some times only to end up realizing some pretty simple such ideas that i'd missed (like "you need symmetry here" or "you just need more bandwidth here") and made complete sense when you thought about them. so i wonder if we can't tweak a few pixels in their structures and reveal something simpler.
Since you beat me to it, I'll add something that relates relates you were saying on "realizing some pretty simple... ideas".
I think a big plus of computer aided design like this is "innovization"[1]. Somewhat awkward term. But, a system like this leading one to deeper understanding of a particular process is the general idea. It's a fun feeling in practice.
> but i didn't find them addressing it explicitly in the text
Yes, this is exactly what bothers me about this article and about a few similar articles published in the past, that they do not contain any evidence that their claims about the usefulness of AI in design are true.
In TFA it says that the role of AI is replacing the electromagnetic simulator in the optimization process, by guessing the behavior of the structure, which is many orders of magnitude faster than a simulation.
This sounds plausible, but in order to believe this I would want to see the differences between AI guesses and real measurements, in the case of structures with geometries that are very different from those used in the training of the AI.
Also I would want to see exactly with which simulators they have compared the speed of the AI model.
There are various simulation approaches for electromagnetic fields and electronic circuits, that can trade-off accuracy for speed, so I am not convinced that AI inference takes necessarily much less time than some faster low-accuracy methods of simulation, which would still be more accurate and more reliable than AI guesses.
> the biggest question for me is how robust are these designs.
Maybe it doesn't matter?
I mean, of course it matters. But most of this sort of design space is effectively NP-complete, where the creation starts with a blank schematic page and has an impossibly large search space, but where the checking of the design is much simpler.
> also, obligatory mention: "genetic antennas"
Exactly. How does this work? When confronted with the question, of course, everybody gets all excited about the constrained randomness of the GA, but if you think about it, what really makes it work is that there is a comparatively cheap test for fitness for purpose.
> How generalizable are these methods? Can they consistently deliver truly high performance? Can we get to a place where AI produces designs that maximize every conceivable trade-off, holistically optimizing every parameter to its most ideal physical state? .... AI can hallucinate a design that creates bad circuits that don’t work. This means verification methods need to remain under human oversight.
And they are essentially correct. We need better validation and verification methods, both software and hardware to keep in check the mistakes of automated random processes.
I am confused, every day I read on HN that AI's can just interpolate the data they have seen in training, and that they are structurally incapable of coming up with something new, creative and not in the training distribution.
> I read on HN that AI's can just interpolate the data they have seen in training
No. That can be said about LLMs, but not about all forms of AI. The technique used is not a LLM.
Sadly we've bastardized the term AI that, if it ever meant anything, it's meaningless now. The currently most voted thread in this post discuses the topic.
This is wrong - the training data is necessary but insufficient. There are a lot of other parts of the architectures used that add a lot of value - otherwise Markov chains would be all you need. There are layers upon layers with non linear activation functions, learned residuals, etc. They still absolutely must interpolate but the space they interpolate through is much more complex than the training data, and they can definitely create things not in their training data. What they can not do is wander outside their non linear parameter space’s convex hull. But this is a really permissive constraint on what they can do “creatively.” People generally under estimate the advantage the architectures confer on that constraint. This is why there was a step function change in expressive power as the architectures (attention, self attention, transformers, diffusions, others) evolved given the same training data. Generally though I challenge you to define “creative” in a way that is precise enough to measure and isn’t self referential or refer to concepts ill defined.
The key tho is can they solve problems not easily solved before with prior techniques. Further can they identify problems not readily presented. Then identify novel solutions. Etc. The answer is emphatically yes they can. These features don’t have to literally exist in their training data, but the supporting highly convoluted network of associations of all their training data does have to in some complex space allow for it to produce these answers. It’s not the same as they’re stochastic parrots at all.
Are they creative? No, because they don’t have awareness. My personal imprecise definition of creative requires both self and awareness as well as free will. There is no driving awareness in all AI architectures, it all derives from extrinsic impetus. Creativity is derived, IMO, from a layer of our minds that is not readily assessed or measured and is only indirectly expressed through language, art, and music. Hence it is not directly trainable and therefore a learning model can’t learn it by reinforcement. It can learn the proxies, but the proxies are not, as we all deeply know, the same as our experienced awareness. We are not our words, our art, our music. We try hard to bridge it, but it’s impossible and you and I know this to be true from experience. In fact we can not even examine our own awareness because it’s not directly observable or possible for us to directly reason about. This is core to a lot of philosophy, especially mid and far eastern philosophy of the mind, the self, the five aggregates of Buddhism, etc. Psychology points at it, and modern psychology avoids it because it’s practically difficult for outcome oriented treatments.
GA’s optimize only combinatorial problems though — where you have discrete set of choices (~genes) for each variable, and therefore do not have a gradient
The other side is Cognitive Radio [1] which also evolve the OTA protocols for cooperative diversity from IEEE 802.22 onwards. Now I can see AI, via a local SLM/NPU plus agentic GNURadio loops for new radio use cases. This is going to be much more wide spread in the upcoming 3GPP 6G releases in 2030.
"Humans couldn't even imagine" seems like overselling it, but I'm sure that machine learning algorithms can brute force their way to chip designs no one has tried before and that some of those might be useful to us. That seems like a pretty reasonable thing for a computer to do.
It's marketing bullshit. For one, it's like proving a negative; you can't prove to me that humans couldn't have imagined it. Second, humans have already imagined quite a lot of crazy stuff...
I was going to post this as well, its delightful to see that other people enjoy it since it was really mind-blowing when I read it.
It's interesting since I saw another comment near yours that raised the question of robustness of the lab-grown design, which I thought was kind of the most fascinating part of the damninteresting article was the revelation that the evolved programs were inseparable from the single physical FPGA used in the training. Since this RFIC training model employs a simulator, do you suppose that the quirks of the physical hardware on which the simulator runs are sufficiently isolated from the training such that a pair of designs would behave similarly when the simulator was run on distinct hardware? And I guess the even more obvious question is whether a design evolved on a simulator would have any hope of behaving as expected in physical hardware?
My hunch about the latter is no, although it still seems like an interesting study, and I often find myself thinking that really understanding what was going on with the FPGAs might be a prerequisite for really understanding how to master reinforcement learning.
Anyway I'm glad you posted this and if you have any other favorites related to this domain send them my way!
>I thought was kind of the most fascinating part of the damninteresting article was the revelation that the evolved programs were inseparable from the single physical FPGA used in the training.
100% agree with this!!! It gave me this weird feeling the first time i read it, like the onset of some alien intelligence. xD
I definitely think a simulator is the way to go, but I'm guessing tools like that could find problems and edge cases in the simulator that nobody thought to test for.
I'm just glad they still have the article up. I bet I've shared it 50 times over the past 20 years lol.
In case anyone feels déjà vu, Popular Mechanics wrote about this professor's lab in Jan 2025, with almost the same title: "AI Designed Computer Chips That the Human Mind Can't Understand".
I feel a bit of unease when I read this title, not because of the threat of AI, but because the prevailing aphorism that "RF is black magic" is a slap in the face to the millions of physicists and RF engineers who DO understand every bit of this. It's a fun harmless anti-intellectual saw that I don't believe is harmless at all. We need more RF engineers and telling people it's all "black magic" and "wizardry" (and worst of all, saying "even RF engineers don't understand RF") makes it seem like it's not worth studying.
> telling people it's all "black magic" and "wizardry" (and worst of all, saying "even RF engineers don't understand RF") makes it seem like it's not worth studying.
I think the opposite is true. It being advertised as difficult to understand is one of the reasons I personally decided to study RF Engineering. The prospect of learning something so challenging pulled me in. The Smith Chart helped.
Chips? I've tried to task Opus, Gemini and Codex with a simple PCB. All of them placed holes correctly but can't understand that the traces should not cross physically.
I’m a bit frustrated. AI can do a looot of things; but I think as we continue to muddy the waters between LLMs and more traditional machine learning like Monte Carlo, Genetic Algoriths, Expert Systems and other Statistics magic tricks, we’re too aggressively conflating established and morally neutral activities in ML with the concerns that people have about LLMs and Stable Diffusion.
I have been practicing saying ML for traditional machine learning and LLMs for LLMs for just this reason. Trying not to say AI anymore. Too ambiguous. Sometimes I'm talking about game AI even, I'll try to use shorthand for whatever algorithm I think the AI is using (often I'll talk about its flowchart, though not always sure it's literally using that under the hood).
I wish I could wave a magic wand and just make the word "AI" go away. It has no actual meaning. It could mean anything from your opponent in Mario Kart to Stable Diffusion.
Recently I heard some people conflate procedural generation and generative AI and I had to explain why there isn't some legal or ethical issue with what breaks down to essentially scattering some points.
It's really getting annoying having to have these conversations.
"AI" is a term cursed by cool sci-fi implications. It makes it a kick ass marketing term because most people are going to have some familiarity with sci-fi AI and "X media predicted Y technology" is a pretty widespread belief for a lot of values of X (star trek, Hitchhikers Guide to the Galaxy, Arthur C. Clarke) and Y (internet, cell phones, VR). If you want to tell someone we're making big strides in something, linking it into some popsci understanding of sci-fi being the great predictor of human achievement is low effort and high impact for quite a few people.
People aren't trying to communicate accurately if their first priority is getting you excited about the thing!
Just as more successful machine learning fields distanced themselves from the term during the AI winter, I suppose we will (and perhaps are?) be seeing them adopt it again, now that we are in an "AI summer".
AI is not a real thing or a natural kind but a perspective. Whether something qualifies as "AI" or not cannot be decided by the objective features of the thing. Ergo, it can be defined at the author's pleasure.
> conflating established and morally neutral activities in ML
LLMs are no more or less morally neutral than other ML techniques.
Yep, can't wait until everything is free and costs nothing to generate content. Free hosting and electricity will be super sweet too. Won't need admins or even the Internet. Everything I want will just be free for me because I don't think anything has value.
> That’s not even to speak of all the movie plots that would have been ruined.
I clicked on all the links. Pretty much all of those movies could still work with wired technology. Even the one called cellular, in which a woman is trapped in an attic with a broken landline phone and manages to connect wires and dial a random number.
Yes I'm nitpicking. I guess I'm glad we have Wi-Fi and all, but don't try to sell me on it as a crucial plot device
One great application of AI design is patent poisoning. Use AI to churn out masses of variant designs, make them publicly visible on a web site, and if future patents come out use any collisions to invalidate them or at least restrict their scope (generalization of a patent is limited by prior art.)
The methods outlined in this article aren't new. Scientists were using "genetic algorithms" to design antennas that weren't understood by anyone, but worked well, decades ago.
I wonder if our common expectation that true theories somehow had to be beautiful and elegant is going to survive the coming century. What if "real" nature phenomenon were actually best described by horrible mess of impossible equations, that only machines could actually manipulate and reason about ?
One takeaway from the article is that they had to get rid of the tried and tested fundamental building blocks of chip design to generate this advancement. I wonder if the same applies for mundane coding. Are the incredible innovations in AI coding actually hampered by rust and python? Should we let AI tools just code in the lowest level possible?
63 comments
[ 4.2 ms ] story [ 77.0 ms ] threadin the journal articles they did show measurements of real devices which agreed fine with predictions, but i didn't find them addressing it explicitly in the text. also, some systems they presented contained subblocks that were conventionally designed that could be carrying some of the weight.
or maybe i'm just sour that they're coming for my job? or maybe that's what they want us to think?
i think what wins in practice is simple ideas that can work in spite of all manufacturing and environment variations, and model limitations -- think stuff like feedback and symmetry. and what they show here is the opposite of that. i've done blind optimization of circuit parameters some times only to end up realizing some pretty simple such ideas that i'd missed (like "you need symmetry here" or "you just need more bandwidth here") and made complete sense when you thought about them. so i wonder if we can't tweak a few pixels in their structures and reveal something simpler.
also, obligatory mention: "genetic antennas"
Since you beat me to it, I'll add something that relates relates you were saying on "realizing some pretty simple... ideas".
I think a big plus of computer aided design like this is "innovization"[1]. Somewhat awkward term. But, a system like this leading one to deeper understanding of a particular process is the general idea. It's a fun feeling in practice.
[1]: https://dl.acm.org/doi/10.1145/1143997.1144266
Yes, this is exactly what bothers me about this article and about a few similar articles published in the past, that they do not contain any evidence that their claims about the usefulness of AI in design are true.
In TFA it says that the role of AI is replacing the electromagnetic simulator in the optimization process, by guessing the behavior of the structure, which is many orders of magnitude faster than a simulation.
This sounds plausible, but in order to believe this I would want to see the differences between AI guesses and real measurements, in the case of structures with geometries that are very different from those used in the training of the AI.
Also I would want to see exactly with which simulators they have compared the speed of the AI model.
There are various simulation approaches for electromagnetic fields and electronic circuits, that can trade-off accuracy for speed, so I am not convinced that AI inference takes necessarily much less time than some faster low-accuracy methods of simulation, which would still be more accurate and more reliable than AI guesses.
Maybe it doesn't matter?
I mean, of course it matters. But most of this sort of design space is effectively NP-complete, where the creation starts with a blank schematic page and has an impossibly large search space, but where the checking of the design is much simpler.
> also, obligatory mention: "genetic antennas"
Exactly. How does this work? When confronted with the question, of course, everybody gets all excited about the constrained randomness of the GA, but if you think about it, what really makes it work is that there is a comparatively cheap test for fitness for purpose.
> How generalizable are these methods? Can they consistently deliver truly high performance? Can we get to a place where AI produces designs that maximize every conceivable trade-off, holistically optimizing every parameter to its most ideal physical state? .... AI can hallucinate a design that creates bad circuits that don’t work. This means verification methods need to remain under human oversight.
And they are essentially correct. We need better validation and verification methods, both software and hardware to keep in check the mistakes of automated random processes.
No. That can be said about LLMs, but not about all forms of AI. The technique used is not a LLM.
Sadly we've bastardized the term AI that, if it ever meant anything, it's meaningless now. The currently most voted thread in this post discuses the topic.
The key tho is can they solve problems not easily solved before with prior techniques. Further can they identify problems not readily presented. Then identify novel solutions. Etc. The answer is emphatically yes they can. These features don’t have to literally exist in their training data, but the supporting highly convoluted network of associations of all their training data does have to in some complex space allow for it to produce these answers. It’s not the same as they’re stochastic parrots at all.
Are they creative? No, because they don’t have awareness. My personal imprecise definition of creative requires both self and awareness as well as free will. There is no driving awareness in all AI architectures, it all derives from extrinsic impetus. Creativity is derived, IMO, from a layer of our minds that is not readily assessed or measured and is only indirectly expressed through language, art, and music. Hence it is not directly trainable and therefore a learning model can’t learn it by reinforcement. It can learn the proxies, but the proxies are not, as we all deeply know, the same as our experienced awareness. We are not our words, our art, our music. We try hard to bridge it, but it’s impossible and you and I know this to be true from experience. In fact we can not even examine our own awareness because it’s not directly observable or possible for us to directly reason about. This is core to a lot of philosophy, especially mid and far eastern philosophy of the mind, the self, the five aggregates of Buddhism, etc. Psychology points at it, and modern psychology avoids it because it’s practically difficult for outcome oriented treatments.
https://en.wikipedia.org/wiki/Evolved_antenna
Starting with: https://sci-hub.ru/storage/moscow/4324/11d145b2c2c3ab320f70b...
[1] https://en.wikipedia.org/wiki/Cognitive_radio
The problem isn’t the design: its manufacturing restraints.
This is nothing new or impressive.
One of my favorite little morsels of internet goodness.
It's interesting since I saw another comment near yours that raised the question of robustness of the lab-grown design, which I thought was kind of the most fascinating part of the damninteresting article was the revelation that the evolved programs were inseparable from the single physical FPGA used in the training. Since this RFIC training model employs a simulator, do you suppose that the quirks of the physical hardware on which the simulator runs are sufficiently isolated from the training such that a pair of designs would behave similarly when the simulator was run on distinct hardware? And I guess the even more obvious question is whether a design evolved on a simulator would have any hope of behaving as expected in physical hardware?
My hunch about the latter is no, although it still seems like an interesting study, and I often find myself thinking that really understanding what was going on with the FPGAs might be a prerequisite for really understanding how to master reinforcement learning.
Anyway I'm glad you posted this and if you have any other favorites related to this domain send them my way!
>I thought was kind of the most fascinating part of the damninteresting article was the revelation that the evolved programs were inseparable from the single physical FPGA used in the training.
100% agree with this!!! It gave me this weird feeling the first time i read it, like the onset of some alien intelligence. xD
I definitely think a simulator is the way to go, but I'm guessing tools like that could find problems and edge cases in the simulator that nobody thought to test for.
I'm just glad they still have the article up. I bet I've shared it 50 times over the past 20 years lol.
I feel a bit of unease when I read this title, not because of the threat of AI, but because the prevailing aphorism that "RF is black magic" is a slap in the face to the millions of physicists and RF engineers who DO understand every bit of this. It's a fun harmless anti-intellectual saw that I don't believe is harmless at all. We need more RF engineers and telling people it's all "black magic" and "wizardry" (and worst of all, saying "even RF engineers don't understand RF") makes it seem like it's not worth studying.
I think the opposite is true. It being advertised as difficult to understand is one of the reasons I personally decided to study RF Engineering. The prospect of learning something so challenging pulled me in. The Smith Chart helped.
Though I also imagine that that is the point.
It's really getting annoying having to have these conversations.
People aren't trying to communicate accurately if their first priority is getting you excited about the thing!
AI is not a real thing or a natural kind but a perspective. Whether something qualifies as "AI" or not cannot be decided by the objective features of the thing. Ergo, it can be defined at the author's pleasure.
> conflating established and morally neutral activities in ML
LLMs are no more or less morally neutral than other ML techniques.
I clicked on all the links. Pretty much all of those movies could still work with wired technology. Even the one called cellular, in which a woman is trapped in an attic with a broken landline phone and manages to connect wires and dial a random number.
Yes I'm nitpicking. I guess I'm glad we have Wi-Fi and all, but don't try to sell me on it as a crucial plot device
That said we’ve had some success internally having Claude do parameter sweeps
That would be really sad..