We can find local optima, but there's no real way to prove a given design is the best of all possible designs. We can only find designs which work and meet the given criteria.
One can prove that a design is correct, but it requires the same class of compute as an LLM to properly simulate. Circuit boards are hellishly complicated once you start simulating EM and RF responses.
As for how a good engineer does this, it's the same way pilots can fly helicopters: intuition. It's a skill built up from lots of reading the rules and theory, and lots of practice. One gets a feel for how the various fields and energies are moving.
Which, when you think about it, is just about how neural nets learn, isn't that neat?
I think the real answer here is that machine learning is absolutely capable of producing satisfactorily correct circuit boards. In principle. But I think the problem space is far too large for today's ML techniques and verification/iteration is just way too expensive. Maybe in another few years, I just don't think we have enough compute for it yet.
Only decades of standards compliance, workmanship training, and experience. Human beings are very good at spotting patterns in noisy jungles, and making rational design tradeoffs.
Auto-routers have been around far longer than even smartphones. Somewhat functional in the trivial problem domains, and always useless where design choices mattered.
Computationally what’s the difference between a human and an LLM besides scale and speed? Trained humans are still using heuristics and shortcuts. Just because these are subconscious and only possessed by talented, trained professionals doesn’t make the skill computationally special.
LLM do not think because they are not real "AI", but it does copy the linear patterns people exhibit if statistically salient within the granularity of the higher dimensional vector search space proximity. Watermarking does skew the compaction slightly, but not far from resulting output patterns. =3
I do think that slightly smarter LLMs and a good (although not perfect) autorouting algorithm could solve 99% of hobbyist's projects and simple industry boards, due to the repetition of those patterns in real life.
The LLM would learn the heiristics (example: data lines first, power lines later, etc.) and would request the autorouter to do that routing, then take the image output and request a different part (depth-first). If later no routing is posible with that configuration, after some retries it could try another way.
It wouldn't solve complex boards, and engineers are always needed for short comings (and even if not, for research), but I wouldn't say this is something different than poetry, music or pixel art, LLMs can imitate although most of the times in a uncanny way.
Yes, the optimal solution won't be achievable, it will burn lots of compute and we need better simulations so that less errors are made (and will be made).
Still, if LLMs are capable of writing working code, art and solving math problems, they qre definitely capable of doing suboptimal routing (with some algorithmic aid)
That is the leap in logic, as an LLM doesn't think/create/feel or understand. It simply copies billions of patterns, and finds the closest fit.
> writing working code,
You mean a compacted collage of stolen codified work from real people. It will require a continuous parasitic relationship of user data to remain coherent. Given the firms have already stolen everything, it is unlikely to improve much as weights are refined.
> art
Again, a statistical output of salient feature clusters mimicking a humans output is not a creative process. For example, a glass-blower only sees the glowing art-form in its true state before it is annealed into its final form. Or a painter undergoes a creative process to find meaning, and abandons the work when there is nothing left to add.
Emotional projection onto an algorithmic output is simply recognizing the millions of peoples aesthetic choices stolen, and compacted with some degree of lost granularity. It is peoples art forms, but no creative thought or intent behind the results.
> solving math problems,
LLM are very good at context search, but require persistent data streams to mine for weighting relevant vector proximity. Again, everything that could be stolen, has already been scraped.
Only user input data remains in a sea of slop, and the process is degenerative given most human chat-users appear to be losing 17% cognitive function. As people discover the intelligence campaign against users is not in their interest.. paying companies to rob you and your friends makes less sense.
Being a sentient turnip, I am probably unaware of such things. ymmv =3
Finding a consistent pattern by conflating most papers, resolving a similar isomorphic vector search proximity result from an entire field of research.
LLM are very good at brute force context search spaces, and people see patterns in nonsense even when it is nebulous. Similar to the Newton's Apple story people are fond of telling each other. =3
> Somewhat functional in the trivial problem domains, and always useless where design choices mattered.
This is laughable when you consider that semiconductor design is the least "trivial" problem domain, and it happens entirely in RTL. There's basically zero human intervention when it comes to layout: even floorplanning is being automated nowadays.
Fab standard cell libraries are already the practical optimization of their current capabilities, and even FPGA try to minimize routing proximity with various levels of success. However, there are again named problems that constrain what naive people can get away with....
Procedural generation or parametric design is not the same, as the behavior and constraints were rationally engineered by people.
LLM would simply look at the outputs during training, abstract it as looking similar to a potato-chip factory, and generate a nonsense answer some fool assumes is appropriate.
Humans are wired to see meaningful patterns in chaotic systems even when they aren't really there. Neuromorphic computing may create real "AI" someday, but it almost certainly won't be from LLM cults. =3
Simple: we aren't.
We can find local optima, but there's no real way to prove a given design is the best of all possible designs. We can only find designs which work and meet the given criteria.
One can prove that a design is correct, but it requires the same class of compute as an LLM to properly simulate. Circuit boards are hellishly complicated once you start simulating EM and RF responses.
As for how a good engineer does this, it's the same way pilots can fly helicopters: intuition. It's a skill built up from lots of reading the rules and theory, and lots of practice. One gets a feel for how the various fields and energies are moving.
Which, when you think about it, is just about how neural nets learn, isn't that neat?
I think the real answer here is that machine learning is absolutely capable of producing satisfactorily correct circuit boards. In principle. But I think the problem space is far too large for today's ML techniques and verification/iteration is just way too expensive. Maybe in another few years, I just don't think we have enough compute for it yet.
Only decades of standards compliance, workmanship training, and experience. Human beings are very good at spotting patterns in noisy jungles, and making rational design tradeoffs.
Auto-routers have been around far longer than even smartphones. Somewhat functional in the trivial problem domains, and always useless where design choices mattered.
Best of luck solving a named problem. =3
Computationally what’s the difference between a human and an LLM besides scale and speed? Trained humans are still using heuristics and shortcuts. Just because these are subconscious and only possessed by talented, trained professionals doesn’t make the skill computationally special.
LLM do not think because they are not real "AI", but it does copy the linear patterns people exhibit if statistically salient within the granularity of the higher dimensional vector search space proximity. Watermarking does skew the compaction slightly, but not far from resulting output patterns. =3
https://en.wikipedia.org/wiki/The_Subservient_Chicken
I do think that slightly smarter LLMs and a good (although not perfect) autorouting algorithm could solve 99% of hobbyist's projects and simple industry boards, due to the repetition of those patterns in real life.
The LLM would learn the heiristics (example: data lines first, power lines later, etc.) and would request the autorouter to do that routing, then take the image output and request a different part (depth-first). If later no routing is posible with that configuration, after some retries it could try another way.
It wouldn't solve complex boards, and engineers are always needed for short comings (and even if not, for research), but I wouldn't say this is something different than poetry, music or pixel art, LLMs can imitate although most of the times in a uncanny way.
Every trace on a PCB is a traveling salesman problem with ballooning complexity.
https://en.wikipedia.org/wiki/Travelling_salesman_problem
While physics informed models do exist, they are still going to burn a lot of compute to generate failure modes people didn't know were possible. =3
https://www.youtube.com/watch?v=T4Upf_B9RLQ
Yes, the optimal solution won't be achievable, it will burn lots of compute and we need better simulations so that less errors are made (and will be made).
Still, if LLMs are capable of writing working code, art and solving math problems, they qre definitely capable of doing suboptimal routing (with some algorithmic aid)
>if LLMs are capable of
That is the leap in logic, as an LLM doesn't think/create/feel or understand. It simply copies billions of patterns, and finds the closest fit.
> writing working code,
You mean a compacted collage of stolen codified work from real people. It will require a continuous parasitic relationship of user data to remain coherent. Given the firms have already stolen everything, it is unlikely to improve much as weights are refined.
> art
Again, a statistical output of salient feature clusters mimicking a humans output is not a creative process. For example, a glass-blower only sees the glowing art-form in its true state before it is annealed into its final form. Or a painter undergoes a creative process to find meaning, and abandons the work when there is nothing left to add.
Emotional projection onto an algorithmic output is simply recognizing the millions of peoples aesthetic choices stolen, and compacted with some degree of lost granularity. It is peoples art forms, but no creative thought or intent behind the results.
> solving math problems,
LLM are very good at context search, but require persistent data streams to mine for weighting relevant vector proximity. Again, everything that could be stolen, has already been scraped.
Only user input data remains in a sea of slop, and the process is degenerative given most human chat-users appear to be losing 17% cognitive function. As people discover the intelligence campaign against users is not in their interest.. paying companies to rob you and your friends makes less sense.
Being a sentient turnip, I am probably unaware of such things. ymmv =3
> That is a leap in logic
It is not, it doesn't matter if a brain's neurons or a prediction algorithm managed to get the solution if it has done so in a (sort of) reliable way.
https://en.wikipedia.org/wiki/Duck_test
Errors are made by both machines and humans, so that's not a differenting factor.
What is clear is that the human element will be more valued on art as time goes on.
What were the chicken's thoughts on the Jacobian conjecture?
Finding a consistent pattern by conflating most papers, resolving a similar isomorphic vector search proximity result from an entire field of research.
LLM are very good at brute force context search spaces, and people see patterns in nonsense even when it is nebulous. Similar to the Newton's Apple story people are fond of telling each other. =3
https://en.wikipedia.org/wiki/Pareidolia
Explain how this is different from what humans do.
We all just muddle through life, predicting the next token. That's it. That's all there is. And it's enough.
>Explain how this is different from what humans do.
The same reason every human understands how Einstein brushed his teeth in the morning. Best regards =3
That logic is so goofy, I don't even think the Greeks had a name for it.
> Only decades of standards compliance, workmanship training, and experience
If that could solve NP-hard problems, computer science would be a very different field.
> Somewhat functional in the trivial problem domains, and always useless where design choices mattered.
This is laughable when you consider that semiconductor design is the least "trivial" problem domain, and it happens entirely in RTL. There's basically zero human intervention when it comes to layout: even floorplanning is being automated nowadays.
Fab standard cell libraries are already the practical optimization of their current capabilities, and even FPGA try to minimize routing proximity with various levels of success. However, there are again named problems that constrain what naive people can get away with....
https://en.wikipedia.org/wiki/Clock_domain_crossing
> floorplanning is being automated nowadays.
Procedural generation or parametric design is not the same, as the behavior and constraints were rationally engineered by people.
LLM would simply look at the outputs during training, abstract it as looking similar to a potato-chip factory, and generate a nonsense answer some fool assumes is appropriate.
Humans are wired to see meaningful patterns in chaotic systems even when they aren't really there. Neuromorphic computing may create real "AI" someday, but it almost certainly won't be from LLM cults. =3
https://en.wikipedia.org/wiki/Pareidolia