"Google’s Quantum Chip Is a Death Sentence for Bitcoin and Nvidia" - Chamath Palihapitiya

요약 보기

구글이 양자칩 'Willow'를 발표하며 표준 벤치마크를 5분 내 계산(최고 슈퍼컴퓨터 10^25년 소요 대비)했다고 소개. 논리 큐빗 오류율이 3x3→5x5→7x7로 갈수록 감소함을 시연해 기술적 전환점으로 평가. 다만 실사용까지 오류율 10^-6 목표 대비 아직 수 자릿수 차이가 있고, 암호 해독(쇼어 알고리즘)은 수년 내 위협이 될 수 있어 포스트퀀텀 암호 전환이 필요하다는 논의. 패널들은 상용화까지 10년 가량 걸릴 것으로 보고 양자컴퓨팅 주식 투자는 아직 하지 않는다고 언급.

다듬은 글

※ 자동 자막 원문(다듬기 전)

01. Quantum chip is super impressive freeberg and I we're talking about that on the group chat on Monday Google announced its latest Quantum chip it's called Willow here's the chip if you haven't seen it it's uh beautiful it was fabricated in Google's new chip plant in Santa Barbara they started this project back in 2012 their Quantum Computing project and the headline basically is Willow performed a standard Benchmark computation under five minutes that would have taken today's fastest Supercut 10 septian years or 10 to the 25th power which is billions of times older than the universe if you don't know what Quantum Computing is freeberg will expand on it but basically computers are binary you've heard this before one and zeros quantums use cubits you know those are 01 or both at the same time and uh Google got a a 5% pop they're up 13% in the last 5 days probably on the other news that Gemini 2.0 is out as well which is unbelievable I've been playing with it what do you think freedberg of this big announcement Google's announcement is a paper published in nature that follows a pre-print they actually put out in August so this news has been out for a little bit there's obviously a press cycle this week around it to kind of make a big thing about it but it is a very kind of important milestone in the evolution of quantum Computing so do you want me to kind of talk about Quantum Computing again I think we've talked about this in the past like I mean maybe a brief primer for people but like what what does this mean practically I think what people want to know is when did these things actually have an impact in the way say nvidia's gpus have had yeah the big breakthrough here is that the whole basis of a quantum computer is called a cubid or Quantum bit it's radically different than a bit a binary digit which we use in traditional digital Computing which is a one or a zero a Quantum bit you can kind of think about it as a wave function it's sort of a Quantum state of a of a molecule and if we can contain that Quantum State and get it to interact with other molecules based on their Quantum State you can start to gather information as an output that can be the result of what we would call Quantum computation and that sounds complicated but what it really means is that instead of doing kind of binary computation where we're adding numbers together or doing kind of other traditional arithmetic there really interesting functions you can do with cubits cubits can for example be entangled so two of these molecules can actually relate to one another at a distance they can also interfere with each other so canceling out the wave function and then when you read it out you get a a result that is basically a very very complex problem that is solved through this Quantum interpretation it's really hard to kind of highlight how different this is from traditional Computing so Quantum Computing creates entirely new opportunities for algorithms that can do really incredible things that really don't even make sense on a traditional computer they're not possible to kind of resolve on a traditional computer and sorry let me just State one thing the quantum bit needs to hold its state for a period of time in order for a computation to be done and so the Big Challenge in Quantum Computing is how do you build a quantum computer that has multiple cubits that hold their state for a long enough period of time that they don't make enough errors that you can actually do a computation with them so what Google was able to demonstrate here is they created these call it logical cubits so they put several cubits together and by putting several cubits together they were able to kind of have an algorithm that sits on top of it that figures out hey this this group of physical cubits is now one logical Cubit they balance the results of each one of them so each one of them has some error and as they put more of these together what they were able to demonstrate for the first time ever is that the error went down so when they did a 3X3 Cubit structure the error was higher than when they went to 5x5 and then they went to 7 by seven and the error rate kep going down and down and down so this is an important Milestone because now it means that they have the technical architecture to build a chip or a computer using multiple cubits that can all kind of interact with each other with a low enough fault tolerance or low enough error rate that they can start to do these Quantum calculations this is is a a big area of opportunity one of the very interesting areas that a lot of people are talking about is in cryptography so there's an algorithm by a professor who was at MIT for many years named Shore it's called Shore's algorithm and in 1994 1995 I think around that time he basically came up with this idea that you could use a quantum computer to factor numbers almost instantly and all modern encryption standards so all of the RSA standards everything that Bitcoins blockchain is built on all of our browsers all server technology all computer security technology is built on algorithms that are based on number factorization so if you can factor a very large number a number that's 256 digits long theoretically you could break a code and it's really impossible to do that with traditional computers at the scale that we operate our encryption standards at today but a quantum computer can do it in seconds or minutes and that's based on Shore's algorithm and if you want there's some great YouTube videos that describe sh's algorithm and how it works but it's like mind-blowing when you look at it it's like this really like non-intuitive but simple set of steps that when you put them together on a quantum computer it's like this thing can instantly figure out all the factors and then you can break a code one of the things that this highlights that in a couple of years theoretically if Google continues on this track and now they build a large scale cubid computer they theoretically would be in a position to to start to run some of these Quantum algorithms like shortes algorithm and so we're now kind of spitting distance or a couple year it's not really clear is it 3 years 5 years s years but a couple years away from having computers that theoretically could crack all encryption standards and there are a set of encryption standards that are called postquantum encryption and all of computing and all software is going to need to move to postquantum encryption in the next couple years so there's like this big kind of push now to like how do we do that how do we accelerate it I saw Sundar post it I saw it in my feed I ended up missing my next meeting because I had to figure out how long will it take for us to crack the encryp standards that we use for Bitcoin Nick here's the answer CU I was so tilted by this idea so if you think of Willow as essentially like one stable logical Cubit equivalent in a chip we need about 4,000 to break RSA 2048 and we need about 8,000 to break sha 256 which is the underlying encryption framework for Bitcoin so I think you're right I think we're in the sort of like the end game two to five year shock clock no I mean I think what'll have to happen is some of these chains will need to obviously reimplement something at a pretty foundational level the weird thing as free Brook says is like the willow chips error correction gets better the more of these things you start to use together now there are some really big problems in inside these chips like logical interconnects are very complicated if you put two chips on a board like the ctoc communication is comp this all this stuff that we haven't figured out how to do but this is a big deal and I was I was really like my God what's going on here other projects at Google are finally Landing you have and you have this now I mean project Lon might be gone but you know I think those projects we're going to see a couple of them change the world yeah just to give you a sense on the numbers like Google's Target for FSE Tolerance on a Quantum chip to make it logically useful is 1 * 10 -6 right now this Willow chip is kind of running at 99.7% so it's still a few orders of magnitude away they have a long way to go in getting the fall tolerance low enough to actually build uh logical Gates using cubits that can resolve kind of computational output and so there's still there's still a build cycle ahead and that pathway is a little bit unclear but what they've shown is this almost feels like the shockly transistor moment it's like you know here's this transistor now everyone's like you have a lossy transistor and then you'll figure out P Junctions you'll figure out all of these ways of just like getting the error correction down but by the way this reinforces what you said before which is it's hard for an outsider like us to comprehend what's really happening inside of Google because the business they've built was able to fund this I mean and and I I went down a rabbit hole because I'd never heard of who this guy was that that runs his helmet neevan he's in Santa they have a whole team in Santa Barbara they've been running for like 10 years he has his own law nevan's law and then I went down the rabbit hole of that but what's amazing is so valuable for Humanity Google had the money to fund him for the last 12 years exactly and the greatest money printing machine of all time is paying dividends yeah but isn't it great to know that Google takes these resources from search and sure maybe there's waste and or maybe they could have done better with the black George Washington or maybe they could have done better with YouTube but the other side is they've been able to like incubate and germinate these brilliant people that can toil away and create these important step function advances for Humanity it's really awesome Deep Mind is on that list as well Keith what are your thoughts Deep Mind yeah yeah so I I think first of all I think there's a long time before this becomes a commercial product or application of any sort so you know it's great that they're taking money but think about it like almost like Stanford takes money or the US government funds basic research in some ways this is at least a decade out kind of thing um there's another I mean this area Way Beyond my expertise but I've been talking to a lot of smart people because I do do Financial Services Innovation and obviously encryption is pretty critical whether it's Bitcoin or other places and there's a couple concerns one is it's not even clear that you can verify that this is true um by the way like standard Computing to explain sort of the magnitude of difference standard Computing would take 10 to the 25th years uh to verify that what uh Google analysis accurate so there's a chance that it's not even true for the thing for the for the sampling test they ran yeah to 2050 number of years that's the big I think hole in the whole RCS Benchmark that they use that it only is it's a it's a framework that only a quantum computer could theoretically even so how you know the answer is correct is I guess the question that long to solve and then to be practical like so assuming you solve all this and it's accurate and blah blah you make it fast all second then there is the postquantum Computing encryption which you know a lot of people a lot of things a lot of important things have switched over to so you have historical Communications that were encrypted under an old Paradigm that would be vulnerable and you know every year that goes by like the embarrassment level or the threatening level of old historical Communications will probably has some Decay function or some halflife so if it takes another 10 years Communications that were drafted 20 years ago yeah there'll be some embarrassing things and blah blah blah blah but the more time it takes the more safe you know sort of private Communications and exchanges will be so I think that's positive third is there's a question of order magnitude here you mentioned you need like three orders of magnitude sort of improvement is each step function you know incrementally easier and faster or is each step function you know 10 years and I don't know that anybody knows the answer to that that's right that's right yeah a lot more work here to to to be done you're not buying any Quantum Computing stock yet uh not yet we have looked at KB you know talk about like technology forward you know VCS over the years I've sat through partner presentations and we've never really pulled the trigger um there's other reasons like including like even if you have Quantum Computing you have to rewrite software on top of it uh from a different it's a completely different you know world that who know nothing Maps nothing maps at all nothing so you're starting from scratch so you have an application layer which might be actually an interesting business opportunity yeah how are these things actually going to be coded and how are developers is going to interact with them if at all Maybe by that time it'll just be AI run how do you build a compiler who the hell knows how do you build the language properly these are all complicated yeah Who's Gonna Write the basic who's gonna write like the Microsoft basic the interesting thing is there's a lot of work that's been done in the space like thinking about Quantum Computing and Quantum algorithms is like an entire Branch people do spend a lot of time thinking about this working on this and there ways you can kind of simulate and test and start to build out models for how you could utilize quantum computers but obviously we just don't have you know industrial scale systems at this point there was one Interstellar Marvel Easter egg in their announcement that I wanted to get your thoughts on freeberg Google said that this massive jump in performance quote lends Credence to the notion that Quantum computation occurs in many parallel universes in line with the idea that we live in a Multiverse so is that somebody in PR is high AF or reads too much science fiction dud you know you know the crazy us so the crazy thing about quantum physics is such a mind have you guys taken quantum mechanics like have not I have yeah I remember the summer I took it like the quantum the first quantum mechanics class and I was like glad it was a summer course because you really have to like think pretty deeply about what you've learn in quantum mechanics there's just nothing about it that's intuitive like the way we kind of think about the world is not the way the quantum World operates in the case of a cubit as soon as you measure the Cubit it collapses to a value if you try and measure it if you try and look at it it goes to zero or one the probability by which it goes to zero or one is to find by you know the the quantum state right at the moment You observe it it's just such a mind so effectively this thing is existing in a super position in multiple States at the same time until you try to observe it and that's the case of quantum mechanics so what's kind of happening I think in that language J is nothing novel was kind of discovered or represented that's just quantum mechanics it's a mind and you could go watch hours of YouTube videos if you want to like get taken down the Mind Rabbit Hole of quantum mechanics and realize one thing I've always found fascinating about this discipline is that looking at a cubit changes it like it understands it's for any particle there's a slit experiment and if you try and observe a light as a particle versus a wave it actually changes what happens what the the outcome is of it being a particle or a wave same with electrons the thing about quantum mechanics is the observation of a particle changes what happens serious question question here it comes here it comes puckle in when you look at the the quantum can you get a better idea of the scale of Uranus let's move on no let's move on I saw jamat was doing a joke he was queuing up a joke at the same time no I was I was thinking about Schrodinger's cat that's like a another ridic ridiculous thought experiment that when you the same concept it's the same Quantum the quantum state of the cat is it's both in the box and not in the Box you don't know whether it's in the box or not until you open the box until you open the box and then you open the box and there's a x probability it's in the Box X probability it's not in the box but when when you don't see it it's both it's both and if you're a cat lady you have three of those boxes and I'm a dog person well we got more engagement from Keith on that science Corner than we have from did so he's more he's more open to science I need to I need to Def they both said this is stupid and I'll never touch it except Keith was Kinder and more articulate in getting there did right s would have just been snoring yeah he would have done this he would he would have done his move where he goes stupid and then like he would have done his next topic there's an advantage of Monday partner meetings is I get to watch the science fiction stuff you know every week even if I don't really understand it but like a decade of watching science fiction you pick up some tricks do do you play chess with Peter teal while that discussion is going on play chess actually so I don't play Chess at all ever Okay the reason why is if I do something I want to be really proficient at it I don't have the time got it got it also you have a life you have a life yeah I like to do