> They don't sniff all the packets to find anything useful. Snowden showed that they intercept certain packets flowing between particular sources and destinations that might contain useful information
Their legal argument was that it was actually ok for them to grab and store all packets, and that it wasn't a search until they ran an actual search that matched against that stored traffic.
> In Snowden's leaks, one such application was email inbox backup transfers for big Internet companies that did not (at that time) encrypt their WAN traffic, from which they mined the sender and recipient to build a social graph, a program that the leaks said had already been shut down.
That was private fiber they had spliced into, not anything routed across public Internet. That's why the DCs weren't encrypting it to begin with.
> Their legal argument was that it was actually ok for them to grab and store all packets, and that it wasn't a search until they ran an actual search that matched against that stored traffic.
No, they never made that argument because they don't grab and store all packets. Instead, they stored the metadata extracted from these packets (the sender and receiver). This program has ended prior to Snowden's leaks according to his documents, but the telephone pen register collection was still ongoing.
> That was private fiber they had spliced into,
No, this was public Internet, at places like Room 641A. There was nothing in the leaks suggesting they had spliced private fiber.
> No, they never made that argument because they don't grab and store all packets. Instead, they stored the metadata extracted from these packets (the sender and receiver). This program has ended prior to Snowden's leaks according to his documents, but the telephone pen register collection was still ongoing.
The utah data center is their buffer. And they were storing large amounts of traffic because they had their diffie-helman hack.
> No, this was public Internet, at places like Room 641A. There was nothing in the leaks suggesting they had spliced private fiber.
I've talked to Google engineers about this. They spliced private fiber. Google was smart enough to encrypt anything routed over public internet, but thought that their private WAN was safe until the disclosures.
The Utah data center hadn't even been constructed at the time of Snowden's leaks, so obviously there was no evidence for that in the leaks.
> And they were storing large amounts of traffic because they had their diffie-helman hack.
Again, there was no evidence that they were doing this in the leaks. The leaks said that they did full take data collection only in a specific set of regions like Afghanistan.
> I've talked to Google engineers about this. They spliced private fiber.
Those engineers simply misread the leaks. What they did was insert their network analyzers in places like Room 641A, next door to an IX where subsea cables connect to private routers. These cables pass through that room first. No splicing required. If you look at the third slide in the PRISM/US-984XN deck, you can see that this happens only on international infrastructure like these subsea cables, not on purely domestic DC to DC links.
The engineers I've talked to did not misread the leaks and found non public information confirming it. The diffie-hellman stuff has explicit evidence in the leaks. Yes, the Utah data center is newer than the leaks. It was created because total traffic was expanding and they required more storage.
I will stop correcting you when you stop posting falsehoods. The conspiracy theories that people spread following the Snowden leaks that were contradicted by the leaks themselves helped lead to a the government we have today, so this is not some minor issue that I will let rest.
> The diffie-hellman stuff has explicit evidence in the leaks.
Nothing in the leaks says they were exploiting a Diffie-Helman weakness. That was hypothesized years after the leaks. I didn't address this because it was irrelevant compared to the larger lie that they "grab and store all packets" in the US instead of in some small set of regions with little Internet traffic and extreme national security interest as said in the leaks.
> It was created because total traffic was expanding and they required more storage.
Again, not because they were storing full take US data. That is not in the leaks.
It's also only 45 frames uncompressed at 4k. It's remarkably easy to hit that if your base assets are mostly raster rather than vector for a dynamic scene.
Obviously, they should try harder, and this is an explanation rather than an excuse, but the graphics assets are mostly why.
I mean, I usually check the weather by putting "weather" into a search engine. And this will show me little SVG icons that take up a tiny fraction of my low-resolution screen and probably involve like less than a kilobyte of path description. And I think this looks perfectly fine and I can hardly even fathom why anyone would want it to be fancier than that. I think that even animating them would make the display actively worse.
I think there's a chain of trust. The sensor signs raws, and the private cloud takes signed raws, does minimal processing so they're at least coherent, and re signs that output (maybe even including the original signed raw as well in the image file).
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
Indeed, LLM also stole that data as well, and encoded it based on frequency of occurrence and similarity with other previously compacted fowl related vector weights. Yet still has no idea what any bird actually looks like. bye =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
If anything, it's closer to the bin packing problem. The cost function isn't expressed in terms of lowest cost between nodes, but instead fitting all of the traces with the lowest board layer count while still meeting stuff like EMI crosstalk guidelines. Then there'll be some traces that are very constrained (think DRAM or SERDES links) that have to be length matched, but those are pretty much invariably point to point links that already are heavily automated.
Traces being, say 20% longer than they need to be for an optimal solution doesn't really change the performance of the board in the vast majority of cases since they'll be some of the lowest resistance components of the netlist anyway, as well as the fact that the PCB is normally oversized for the number of traces needed as it fulfills structural/mechanical needs as well as the netlist needs.
And on top of that, because the optimal solution probably is NP-hard, once again, the industry doesn't look for optimal, only good enough. If layout techs in Altium were regularly solving NP-hard problems, computer science would be a different place. The only goal I see is to meet that bar without hiring layout techs.
>Traces being, say 20% longer than they need to be doesn't really change the performance of the board
Impedance matching requires you know what you are doing, and Altium will only take you so far. For the price, Altium/Protel has always been a poor deal for what you get software wise.
Even free Qspice (a more modern Analog Devices LTSpice) is a far better option:
> Impedance matching requires you know what you are doing, and Altium will only take you so far. For the price, Altium/Protel has always been a poor deal for what you get software wise.
Pretty much every high end board is designed under Altium. QSpice is a different tool for a different purpose.
And like I said, those few cases were the length actually matters tend to be heavily assisted with automation as it is.
> Indeed, that is why your iPhone still weighs 3.7 kg. =3
I guarantee you that an iPhone board isn't an optimal solution, simply good enough.
>I guarantee you that an iPhone board isn't an optimal solution,
DFM almost certainly guarantees an iPhone is both form, and cost optimized. Jobs was very clear early on in the product development that the EE work had to be miniaturized due to the energy density possible setting volumetric minimums on the battery technology at that time.
> simply good enough.
A common philosophy for people that make cost optimized low-end products. Some people love their 3.7kg phone, as it comes with a stylish shoulder strap. =3
Their legal argument was that it was actually ok for them to grab and store all packets, and that it wasn't a search until they ran an actual search that matched against that stored traffic.
> In Snowden's leaks, one such application was email inbox backup transfers for big Internet companies that did not (at that time) encrypt their WAN traffic, from which they mined the sender and recipient to build a social graph, a program that the leaks said had already been shut down.
That was private fiber they had spliced into, not anything routed across public Internet. That's why the DCs weren't encrypting it to begin with.
reply