LCEnzo
1yr ago
Firstly, afaik China does not have EUV machines, but I'm not sure. What they surely lack is support from ASML. The machines are somewhat custom built for each customer, and companies don't just buy the things. They also buy maintenance, support services, etc.
Secondly for the same reason that if Putin had the design docs for 5th gen fighters and an intact model, Russia would be building 5th gen fighters any time soon. It's bloody hard, and requires a lot of rare expertise and tacit knowledge in multiple domains. It's not impossible, China has after all built things like planes. But it's really hard. And, as is, companies do not have state support for such a venture. The West won't stand still.
For a historical example, let's take Japan. Their government in the mid 70s organized it's 5 big players in the so called VLSI project, and granted them state support. They also had both a large domestic and export market for it's chips. They became a leader, at least for DRAM, for a time. Unlike China, where Xi is no mood to support anything related to computers, and I'm unaware of any big companies which might be able to pull it off on their own. Maybe the industry self organizes, and deals with this on their own. I would bet that they don't, at least for 4 years.
LCEnzo
1yr ago
In theory, they have cracked 14nm. In practice, using DUV means they will, and already do, have very poor yield. Without access to ASMLs EUV machines, they might eek out a symbolic advance in some lab, but will not advance in any way that matters (ie, producing chips at scale, economically, which is what matters). ASML makes the most advanced lithography machines, and they are deeply dependent on a global supply chain, including firms from America, Germany, etc. AFAIK SMIC has followed Intel in doing 14nm with DUV, and is also having problems with it. Intel eventually started using EUV, though it's now lagging behind TSMC, and even having some CPUs fabbed by TSMC because they lag behind in their process. SMIC now doesn't have the option to buy EUV machines to even try transitioning to them. ASML is something of a system integrator like Boeing and Airbus. Replicating it means replicating all their suppliers, including leaders in optics, robotics, etc. How long will it take? And when they manage, how far will have the West advanced? China will not only be competing with the USA, but all the nations which in some form or another contribute to making the chip industry function. The USA, EU, Japan, UK, SK, India, Israel, and Taiwan, together have almost 2x population, are far richer, and have a good head start.
LCEnzo
1yr ago
The fact that they're made up of humans doesn't seem to be all that relevant, because the corporation itself is not human despite humans being the "material" from which it is made.
The problem with corpos being made up of humans is similar to trying to make ever better computers without changing transistor size. You can optimize the layout, cooling, etc, but you'll forever be bound by the size. Corpo capabilities and architecture are chained by their components. They would be a lot more dangerous if they could produce better humans at scale (compare the performance of Jane Street vs retail investors, or special forces vs green Army grunts), or produce a new part to do mental and social work (AI).
LCEnzo
1yr ago
not all birds fly, and anyway birds have slop similar to those of reptiles.
Nor do species lose adaptations the moment the original reason they got them disappears, but I concede the point.
larger bodies have about same number of components as smaller ones [...] And the only parts which larger bodies have more complex are gastrointestinal tract and lungs
'Components' is an artificial category. The number of cells which sense are what matters (since the collected data needs to be processed, otherwise the cell is worse than useless). If you need to sense with the same precision on 1 mm^2 of skin and on 5 mm^2, you'd need more neurons for transport and processing for the larger patch of skin. Not necessarily 5 times more, you can compress the data or whatever, but you definitely need more than for the 1 mm^2 patch.
AFAIK, two things come together to give birds small but powerful brains, and better mass/compute scaling:
-
Cell size can vary between species, and birds have pressure to miniaturize to reduce weight.
-
Smaller bodies have less stuff (skin, muscles, etc) that need a nervous connection and a part of the brain to process data from and/or issue commands to. See Encephalization quotient. Eg. Women have the same IQ as men, despite having, on average, 90% of the brain size.
EDIT: This post by Scott might be of interest.
LCEnzo
1yr ago
[...] point of the royal family after the end of monarchy
What end? The monarchy is still in place.
What is the actual difference between her passing and like a Kardashian passing?
Part of it is that her lack of controversies and the sheer length of her reign bred a pleasent familiarity, and another is that she's a symbol and common cultural touchstone for British (and more broadly and to a lesser extent Commonwealth) citizens. She has sentimental value for people.
LCEnzo
1yr ago
Peter Watts - Blindsight
Great Sci-Fi book touching on consciousness, truly alien aliens, firs contact, mental disorders, etc. Pretty good, tho the writing could be better.
The Martian
Got a movie, haven't watched it. It's near future, about a guy trying to survive alone on Mars after his colleagues though him dead. Has a solid grounding in real science.
Iron Kingdom: The Rise and Downfall of Prussia, 1600-1947
I'm curious about governments which were/are outliers, and thus wanted to learn more about Prussia which united Germany, and was famed for the quality of it's armies. It was readable for a layman, though some jumps across dates were a little jarring for me. I enjoyed, though couldn't read more than 2 chapters a day.
Na Drini ćuprija/The Bridge on the Drina
The one book I really enjoyed reading in Highschool. It was written by Ivo Andrić, a Nobel prize winner, diplomat, and a lover of history. This book depicts a small town on a river that divides modern day Bosnia and Serbia. Though most of the events in the book are fictional, it presents the reader with a rich and colorful picture of life in the town around the bridge (and through it the events in the surrounding lands) from the bridges construction in 16th century, to it's partial destruction in World War 1.
AI has been noted as a priority in the most recent 5 year plan, but that is talk.
Chinese tech companies have been corralled and restrained, starting with Jack Ma and the Ant group. You might say that they aren't AI companies. This would be mostly right. AI hasn't had great success in being monetized in it of itself. Companies like Google and Meta can pay for research and large amounts of compute which need not be profitable by the way of their other businesses. Even OpenAI, is getting investment from Microsoft (as a former employee, I've talked to an intern working on fine tuning GPT3). The real estate bubble popping and the zero covid policy put more strain on tech companies in the midst of their humbling.
Let's look on the three things needed for AI R&D (besides money): data, compute, and human talent/capital.
China as a society gathers a ton of data. The state and companies collect a lot. The question is how much of that is available to researchers. Despite the Chinese advantage in data collection, most of human genetics research uses the UK biobank dateset. Researchers in firms like Alibaba and Tencent are bound to have some good stuff.
China has received a ban on buying advanced compute. It's domestic industry has been hard hit, and received no state support to keep them propped up. YMCA didn't get it's billion dollar deal with Apple to supply memory. On the design front, China has some companies. Their electronics industry in general is good and that includes design. I'm aware of Moore Threads designing GPUs, but haven't looked into other compute hardware (there was a x86 CPU a couple of years back, but that was in partnership with AMD, so it might be dead now).
China has a lot of good engineers, being an AI researcher and engineer is one of the most sought after careers due to the high compensation. That said, a lot of them, especially the very talented, head over to the USA.
As for money, tech companies are hurting right now, and I don't know the state of funding in academia.
Yes, AI needs compute. Where you're wrong is that the support for domestic chips ain't here. The companies are bleeding money and can't wait forever for state help. They're conducting layoffs as we write.
More options
Context Copy link