Episode 278 - Proof of Stake, Map Powers, and Google answer OpenAI
Max talks about Ethereum's move to proof of stake and what tradeoffs it brings to the crypto markets. He also discusses Google's plan to upgrade maps at their IO conference, and their new product Bard in relation to chatGPT.
Links
Ethereum.org: The Beacon Chain
Unchained: Why Proof of Stake won’t be used for intergalactic first contact: A Bitcoin Astronomy essay
The Defiant: Ethereum Researchers Remain Mystified After Blockchain Briefly Fails To Finalize
Google: Immersive view coming soon to Maps — plus more updates
Twitter: Amjad Masad: “Bard can read your Replit or Github projects in less than a second and make suggestions🤯“
Related Episode
Episode 5 - Christian Lundkvist on Ethereum, Bitcoin, and Smart Contracts
Transcript
Max Sklar: You're listening to the Local Maximum episode 278.
Narration: Time to expand your perspective. Welcome to the Local Maximum. Now here's your host, Max Sklar.
Max: Welcome everyone, welcome! You have reached another Local Maximum. Good to see you. We have so much news coverage to do or just updates in terms of all the things we've been talking about on the Local Maximum. Plus, I've got a lot of things on my mind that I need to do some commentary on which we're not going to do today but hopefully, we'll get Aaron on next week.
Today on the docket. First of all, we're going to talk about some of the latest updates on Ethereum and what that means for the cryptocurrency field going forward. I know in crypto, we had a rough year in 2022. I think that, well if history is any guide, we might be heading into a crypto spring. And some of you are saying, “Hell yeah, crypto spring! I’m so excited.” And some of you are probably saying, “Oh no, not this again.” But look, a lot has happened since way back in episode five, we talked about Ethereum. And a lot has happened over the last five years so I just want to talk about what I see as where we stand today in terms of proof of work and proof of stake.
Then I'm going to talk about some of the new Google developments when it comes to AI. Google has answered Chat GPT in a few important ways. We're going to mention Google Bard, and what Google Maps is promising, such a fan of all mapping software. And then finally, we're gonna end with our segment probability distribution of the week.
Okay, so first of all, let's review a little bit on what the purpose of Ethereum is. In episode five, all the way back five years ago, I started talking about this. I interviewed Christian Reitweissner, who was one of the early developers on Ethereum. We talked about smart contracts and we talked about the idea of a blockchain, which is what Bitcoin is built on, which is what Ethereum is built on.
You could think of a blockchain as kind of a giant, decentralized database in the sky. I know, maybe some of you who want a more technical definition, that's not a very technical definition, but it's a good mental model. It really is. And so why is it called the blockchain? Well, oftentimes, these are databases that get updated over time. And then each block contains the updates. Hey, how has the database changed in the last 10 minutes? That's Bitcoin. Ethereum, I don't know how many minutes it is. But each block tells you how the database has changed recently. And so it gets built block by block, change by change, each block has a lot of changes.
So Bitcoin is kind of easier to wrap your head around, where the database is just kind of one giant ledger as to who owns what, and then each block just becomes just kind of a list of transactions. Who has transacted with whom? And of course, those transactions have to be backed up by transaction signatures. So you know the original owner had given permission. And then you have to kind of line it up with the previous end of the blockchain to know that the previous owner had the funds to spend. That's why, oftentimes, a very confusing term in Bitcoin is UTXO which is like an unspent transaction something. But basically, it's just what's left in a particular dress that hasn't been spent yet. That's the UTXO. So that's sort of the idea of a blockchain.
And Ethereum is also a ledger as to who owns what in terms of Ethereum. But it's a more general system that kind of can run code and can focus on smart contracts that pay people based on certain things happening. Bitcoin, certainly smart contracts can be built on top of Bitcoin as well. But Ethereum is sort of made where you can do that on the base layer. And so Ethereum still needs a currency to make it all work, Ether. Eth- E, T, H, to regulate the system and to pay for executing smart contracts.
So Bitcoin and Ethereum both have development teams. The development teams cannot force their decentralized node to accept all their changes. Although, in most cases when the development team comes out with an update and it's agreed upon, most of the nodes throughout the decentralized universe that are securing these things, will agree to add the new software because it's kind of broad consensus that these are good changes. So you hear consensus a lot, both in terms of the changes of the software, in terms of the consensus of what the next block is.
And Ethereum just had a new release. Used to be called the beacon chain, and some people say it's the proof of work versus proof of stake release. But now Ethereum runs on an algorithm called proof of stake and that term is sort of meant to play off of the old consensus mechanism that Bitcoin and Ethereum used to use and of course, the Ethereum offshoot, Ethereum Classic also uses proof of work.
Now, Ethereum has successfully moved over to proof of stake. There wasn't a big backlash from Ethereum miners. Basically, people who were mining Ethereum, some of them decided, “Okay, you know what? We're going to keep mining Ethereum. And so we're going to create an Ethereum Classic two where we're still on proof of work”. But it turns out that just was not very popular. It was not popular enough to create an Ethereum. It was not popular enough to fork the system really.
Ethereum Classic, which was a system fork back in, I don't know, 2017? That occurred because, or maybe it was 2016? because there was a giant hack. And so there they asked for a mulligan and the Ethereum Classic people said no mulligan. If there's a hack in your smart contract, too bad, so be it. And so, Ethereum Classic at the time actually had enough developers and enough miners to say, “Hey, we have an offshoot currency.” And in fact, Ethereum Classic still exists today.
But that's not what we're talking about. We're talking about Ethereum and how they've moved to a proof of stake system. And so this is sort of interesting. Now, that there are all these arguments over what's better? Proof of stake versus proof of work? What's the difference?
I don't think I can explain everything today. But I'll give it a shot. And I don't think there's an easy answer in terms of quote-unquote, what's better, but it's all about how the blockchain gets updated. How do these people determine what the next block is? And when I say people determined, it's really servers set up by people all over the world. How do we determine in Bitcoin? Okay, I agree that this is the next block. This is the next set of transactions that are true. We kind of all agree that a transaction is not a valid transaction unless it has a valid signature and all that, but which ones have made it into the block? We have to kind of agree on what the block is so that we know what we're allowed to do on the next block.
And so, proof of work requires processing power. Because in order to create the next block, we all agree. It's part of the protocol, we as Bitcoin miners, all agree. I mean, I'm not a Bitcoin miner. But assuming I was, we all agree that the winner of the next block is the one who, first of all, produces a valid block. Here's a list of valid transactions that I know that are out there. And by the way, if you win the block, you get the transaction fees as well. So that's another incentive to include all the transactions that are out there on it.
And I also created the answer to a hash problem. Hash problem, how do I think about it? Well, basically, I take some hash that I have to know from the previous block. I take all the information from this block, kind of package it up into data, jumble it up in some well-defined way. And then I add a seed to it, jumble it up again. And when I jumble it up again the second time, I need to have a certain number of leading zeros.
And because it's cryptographic cash, okay, let's not get into all the details of what that is. But the property is that there's no way to kind of figure out which input is going to give you all those zeros so you have to guess. And so how do you find-? Okay, I'm thinking of a number between one and a million, but every time you guess the number, I can tell you if it's right or wrong, but I can't tell you whether it’s higher or lower. The only thing you can do is guess the number.
How long is that going to take? Well, you could think on average, in the worst case, it's gonna take a million tries. Maybe, if you think on average, probably not exactly, but on average, let's say it takes 500,000 guesses. So you know that the person who got the right answer tried this thing 500,000 times.
Of course, the numbers, when it comes to the blockchain system, are astronomical. And it's not like everyone tries it all those times. It's just that the system as a whole is trying all these random numbers and then the winner is… You're more likely to win if you put more compute power into it. And you're less likely to win if you put less compute power to it. But either one could win.
And so the amount of, the unlikelihood of finding the answer kind of gives us an idea of how much work was put into the system. And so you have an incentive to mine because if you mine, then you can find the next block. And if you find the next block, then you get all the transaction rewards and you get the mining rewards so great. And, of course, it takes a lot of energy to do this, because you're doing that so-called work.
Ethereum was using that, they moved to staking. And finally, as they were developing it, you could stake your Ethereum but you couldn't remove it. So that is kind of a disincentive to stake. There were some people saying online like “Oh no, once they allow people to remove their staked Ethereum, then people are going to remove it and they're going to sell it. The price is going to tank. But the problem is, once you allow people to remove their staked Ethereum from time to time because you've completed the system, then there's actually more incentive to stake. Because people who didn't want to stake before are like, “Okay, maybe now I'll stake because I don't know what the development timeline is.” The people who are staking before were probably like, “I don't even know if they're gonna get this done.”
So now, there's an incentive to do it. So you need to put a certain amount down to become a validator. And now, your chance of winning is not based on how much compute power you throw at it. It's based on how many validators you have. So in theory, you have 32 ETH. That's currently almost $60,000. So it's quite a bit of money to put down. You become a base-level validator. There are some services where you can use less, but of course, let's not get into that right now.
So once you're a validator, you need to spend some amount of energy still to do the validation. You have to keep your servers on, but you can't just throw more compute power at more work at the job and make more. Your reward is based off of how much you stake.
So you might think, well, which one's more expensive? I would argue that from the miners' perspective in Bitcoin or the validators' perspective in Ethereum, they're both very expensive. For the miner, you actually have to invest in all that compute power and you have to have quite a bit before you make anything significant.
In the Ethereum sense, you don't have to invest quite as much in terms of compute power. You have to invest some, but it's not crazy. In fact, it's actually quite low, but what you do have to invest in is all of this Ethereum to put on the system. So if it's $60,000, it could probably be higher one day. And then let's say you're borrowing that, that's kind of your opportunity cost. So your cost is going to be the opportunity cost of the Ethereum, that you're staking away and whatever risks are associated with that. There are some rules where if you try to cheat, you can lose your Ethereum. But that could worry people that, okay, I better do this properly.
Many of the arguments in the proof of stake, proof of work debate, I am skeptical of. But there are many more arguments that make a lot of sense. And so I thought it would be interesting to kind of ask you, the audience, what you think. Do you have a preference in terms of proof of stake or proof of work? And how do you think that's gonna play out?
One argument for proof of stake that people make is they say it's more decentralized because mining will now take place not just in areas with the cost of low energy. Because if you're a Bitcoin miner, you really only want to mine in places where there's low energy, whereas staking could happen everywhere. So maybe it's more geographically decentralized.
On the other hand, every argument has a flip argument here. On the other hand, people say well, mining can be done by anyone who can find the compute power. Whereas for staking, you need to have Ethereum. So it's going to be in the hands of the early adopters and the people who bought Ethereum early. On the other hand, again, if you want to mine, you could buy the Ethereum, you can borrow it. So it's sort of unclear how those arguments play out.
It reminds me of a lot of negotiating where for every argument, there's a flip argument. I think I heard this once where there was a small company negotiating with a large company. And the large company said, “Well you shouldn't be paid as much in this deal because you're small, and we're big. And so this small amount of money matters to you a lot. And so we need the bigger part because that's what matters to us. Because we're big.” And then the flip side of that argument is, well, “If we're the small company, you should give us more. Because if you're big, and you still get a good deal, but just not as good, no one will notice.” So I'm trying to think of where I heard that before. I think it's a theme in negotiation, we can turn it on a person in terms of the argument.
So here's another one, for example, in terms of proof of stake. An argument against the proof of stake. This one is one that I kind of agree with. I think it kind of works. As an argument, it doesn't mean proof of stake is bad but I think there's more complexity in regulating the system. I think that a proof of work system is easier to self-regulate because cheating the system becomes more difficult. It's obvious the correct chain is going to be the one with the most work put into it. And you could see from the mathematics, as I explained before, how much work has been put into a blockchain. Whereas for proof of stake, you could still see the blockchain, see how long it is, but you can't really see how much work has been put into it.
So there has to be a more complex system of governance. You have to make sure there aren't people who are going to go back days, weeks, months, and then redo the whole chain for kind of cheaply as compared to a proof of work system. And then somehow manipulate the system so that this new chain gets accepted and they undid a bunch of transactions. You don't want that.
So there are-. Sorry, I'm having this cough lately, folks. I was sick a few weeks ago. And now I'm feeling fine, I'm feeling great. But it's just this cough is residual, and they won't go away. So no, I'm not earning residuals on the cough. It is just residual.
Alright, back to the proof of stake or proof of work. Yes, there's more that can go wrong with proof of stake. I think you need a more complex governance mechanism. On the other hand, what if that's solvable? What if you can solve the complex mechanism? So then maybe that's not a big deal. So one argument that Bitcoiners make against proof of stake is that there's more attack vectors when the coin is weak. But as we've seen with weak proof of work systems, there are also a lot of attack vectors, 51% attacks, etc. For the largest chain like Bitcoin, you're not going to have a proof of work attack. It's just too expensive.
So more arguments for proof of work is that the resulting blockchain actually proves that all the work has been done. Which is a very interesting property, very different. I have an interesting article, right here. It's called, it’s on Unchained, it says Why Proof Of Stake Won't Be Used For Intergalactic First Contact: A Bitcoin Astronomy Essay. I'm not gonna go into it entirely but the point that the person is making is that we can show an alien civilization the Bitcoin Blockchain then they would know that a lot of compute work has gone into it and then that's proof that that has happened. Assuming that this alien civilization has not found a way to break encryption, which we don't think it's possible.
So great. Maybe there's something really special about proof of work in the long run. But does this make Bitcoin valuable in the short run? And by the short run here, I mean like 10 years? I am not so sure.
So what is in it for proof of stake? Well, the proof of stake people say, “Okay, it might not be cheaper for the miners, for the validators to stake as opposed to the miners in Bitcoin. But it uses a lot less energy and it is a lot less expensive than proof of work.” And so what does that mean?
So one of the things that Bitcoin has prided itself on is that it has a very low issuance rate, in fact, has a maximum amount of coins at 21 million. There could never be more than 21 million Bitcoins. In Ethereum, they were a little bit more wishy-washy on that. And I would sort of be concerned that the issuance in Ethereum is maybe easier to change than Bitcoin. You might feel that, well, people wouldn't want to change the issuance. But it seems like it's easier to change in the long run over the course of 20, 30 years.
But as of now, Ethereum has a transaction fee that is mandatory in order to regulate the system. That's part of their new kind of governance mechanism. And so everybody has to pay this mandatory fee and the fee doesn't go to the validators like it goes to the miners. In Bitcoin, the fee actually is burned. In other words, that Ethereum is destroyed forever.
And so that means every block, and of course, there's an additional fee you can include to make your transaction kind of get ahead of the pack. And so that means that every block, Ethereum is destroying part of the money supply. It's a tiny, tiny, tiny percentage of the money supply that's destroyed. But still, it seems to kind of exceed the amount of issuance that's being created. And so that means it's kind of a deflationary currency. And so could that mean a much more restricted supply for Ethereum in the long run that could make it more valuable? And so that value, does that go into the token? Does that go into the Ethereum token?
Let's see, Ethereum is 1800 right now. Bitcoin is 26. Does it increase in value in relation to Bitcoin because of this property? Is this part of the proof of stake, the dividend? I'm not sure. It remains to be seen. But that's one of the things that could play out in the next Bitcoin, crypto cycle. We'll see.
Now, as could be expected with these types of things, there are some big problems. According to the Defiant, and this has been all over the news, last week. Ethereum briefly stopped finalizing blocks, raising concerns across the web tree community, despite transactions continuing to be processed normally. Quote, I'm not sure that any of us fully understand why, Ben Edgington of the Ethereum foundation said. It's still under analysis exactly what the root cause of the issue was, and why the chain recovered. It's the first major incident suffered by the beacon chain, Ethereum’s proof of stake consensus layer that merged with the main net execution layer last September, and serves as a cautionary reminder of the experimental nature of blockchain technology.
So that is what you get for complexity of governance. Now, this problem doesn't mean people couldn't trade, people couldn't transact and stuff. But it seems to be a little blip, a little problem. And it's not really comforting when the developers don't really know what's going on. So hopefully, they'll figure it out. But this is definitely under one of the downsides of proof of stake, I believe.
So ultimately, in my view right now, and I would like some experts who are listening to weigh in, is that the trade-off here is you're giving up simplicity and some security for potentially huge monetary gains and that it's cheaper to maintain the system. So what's the right trade-off? I can see people valuing both. and that's why I think proof of work and proof of stake chains currently have value. I don't want to speculate which one is going to do better in the next bull run, assuming there is a next bull run, which I think there will be, so we'll have to see.
Questions about the crypto community or for the crypto community. I have noticed some really interesting projects out there in crypto. There's Zcash, Filecoin, maybe even Solana, maybe. Although I'm not so sure if that's good tech. Maybe Cosmos, although that's done a little better than what I'm about to say. But certainly, Zcash and Filecoin, which were very compelling when I read about them, have not done well in the marketplace. Even as some of these so-called meme coins take off, although those usually crash and burn at some point.
But why are these coins, particularly Zee cash with such good tech, why has it not been doing well in the marketplace over the course of many years? Are the markets not efficient? Are there some issues with these coins that the markets know about that individuals don't know about? Seems unlikely? Maybe people just don't value a lot of this privacy tech or this sort of decentralized file, database. Or even library LBC, maybe people don't value that as much? Or maybe these coins will have their day in the future. Maybe you'll see five a bunch more years where they're low and then all of a sudden, you'll see like a 1,000x recovery? Although, I don't know, do you put a little bit in case that happens or just forget it? Very hard to say, what do you guys think? Send it to our locals maximum.locals.com or localmaxradio@gmail.com.
So Google has been under fire recently, including from this very podcast, Local Maximum, about how they're going to answer Chat GPT. They can't push a compelling product. Well, it looks like they're trying because in Google IO they made some impressive presentations over the last few weeks. When was Google IO Google? Recently it was I believe, it was on May 10. Alright, so it was last week.
And so the first one that was really interesting to me was Google Maps. We have not seen a lot of innovation on Google Maps over the last 10 years. I was amazed by street view in Google Earth when that came out in 2010 and I'm still amazed by it. But it's been there for a long time, like a decade, over a decade. And improvements have been slow and steady, but nothing spectacular. Like when I was at Foursquare, early on in 2011, 2012, we talked about what would these maps look like in 10 years. I would have thought they would be a lot more incredible than they actually turned out. I would have thought there would have been something like live view everywhere. Although that, if you think about it, could be very dystopian.
But Google is very excited about this new immersive view, which will use AI to visualize your full route. And it won't just be a group of pictures. They've done a very good job of stitching the pictures together, all taken at different times. So right now, you kind of have a group of pictures that are all taken at times kind of stitched together or a video of kind of a badly rendered terrain. But I think this new immersive view is going to kind of take all the data that they have, and kind of put it together in an AI system and sort of create an immersive view. Whereas this is what your route is going to look like, this is what any place is going to look like. And so this will be rolled out slowly over the next few months. Very exciting thought. Let's see if it lives up to the promises.
Google Bard. Google also released Bard as an answer to chat GPT. I tried it a bit. I see no reason to switch right now. I'm kind of happy with Chat GPT 4 and the interface. But what do you think? Have you tried Bard? Do you see a difference? One advantage Bard has over Chat GPT is that it can read the internet and it can have access to more up-to-date information. Chat GPT is trained in information up to two years ago
As one Twitter user, and I'll post this on the website, localmaxradio.com/278, Amjad Masad said on Twitter, “Bard can read your Replit or GitHub projects in less than a second and make suggestions.” So I tried it with my project, New Map Data AI. It gave some suggestions and praise. It told me that my code was a little bit complex and needed to be cleaned up. It encouraged me to keep going, but it was still fairly generic. Like, “Oh, add more documentation, your code could be refactored.’ It's almost like a horoscope type thing where you're like, ‘Oh my god, it's talking about my projects.” But it can almost apply to any project.
So it looks nice. It summarized what I was doing kind of well, but it also really just kind of gave me the first two sentences of my readme. So I don't know, it's a compelling use case but it's not that impressive yet. But If you give it a try, let me know. I, for one, am happy with Chat GPT for most of my use cases. So we'll see.
All right now, and hopefully, I could press the right button on to our segment.
Narrator: And now, the probability distribution of the week.
Max: Yes, ladies and gentlemen, the probability distribution of the week. Today, we're going to talk about the Laplace distribution. Remember last time, we talked about the multivariate Gaussian distribution. That is just like a normal curve, a bump curve, that you know and love. It has one mean, and either one standard deviation or the covariance matrix when you take it into multiple dimensions. But you could just think of it as like a mean, and a dispersion. Great.
So the Laplace distribution is similar to that and it matches exactly the exponential distribution. Wow! So the exponential distribution, if you remember, is on positive numbers. And it's something that if the most probability density is zero, then that probability density kind of decreases exponentially as you move away from zero. If you have a high exponent, and then it's always negative, then it decreases fast. If you have a low exponent, then it decreases slowly. So high variance, low variance.
But now, we kind of take this distribution that exists only on positive numbers and now we have it on the full number line. And so basically, you have an exponential decay of your probability as you move away from zero in any direction. So it could be on a number line, or it could be on a plane, or it could be in a space, it could be in any dimensional space. So like the normal distribution, it's symmetrical around zero. So in one dimension, it goes up, it comes down. Just like that bump curve, it goes up and comes down. In two dimensions, you kind of draw a contour map, it's kind of like a hill. Still a contour map of concentric circles around the origin, where the origin has the highest probability. Again, a lot like the Gaussian, a lot like normal distribution. Also, like the normal distribution, it can be squashed and stretched using a covariance matrix. So if you don't like contour map being circles, if the data has some kind of a correlation there, it can be ovals.
So if it's a lot like the normal distribution, what's the point? Well, unlike the normal distribution, first of all, things get really weird at the mean, which let's just say is zero at the origin. It’s this weird point. It's kind of pointy. It's continuous, but it's not differential, in the words of calculus. So whereas the normal distribution kind of looks like this nice, neat bell curve and that's how data is always distributed. The Laplace distribution, it kind of looks like someone pitched a tent, with the pole at the origin. And so it kind of has this weird pointy shape. If you're into functions that are smooth, and you're into functions that are analytic, Laplace distribution is not for you.
Also, one of the nice things about the normal distribution is that slices of it are also normal distribution. If I just take a random line through the plane, and then kind of slice the plane there, and then kind of look at the one-dimensional normal distribution that's along that line, it’s also normal. Normal is great like that. Slices of the Laplace, you get some weird non-standard thing. So that is quite strange.
But it does have some interesting properties. It has an association with something called lasso regularization. Like the normal distribution, the maximum likelihood point is exactly zero or exactly the mean. However, if you use it as a Bayesian prior to another process the maximum likelihood point will sometimes stay at exactly zero. In other words, if you use the Laplace distribution in larger problems, it will often create a situation where even if you have some noisy data. If you have some noisy data, for example, in a normal distribution, then you're going to get something even if it's close to zero. In a Laplace distribution, it might stay at exactly zero. So this means that it helps us choose features that are unimportant to throw away. So it's a good way to reduce model size, not gonna get into the whole mechanics of how that works. But I think my next part is going to kind of explain a little bit more.
If you have a bunch of data from a normal distribution, and you want to estimate the mean, then you take the average. So that is the mean. The mean is the mean. But for a Laplace distribution, it actually causes you to take the median. And so when we're talking about the normal distribution, we kind of use this least squares technique where everything that is off the mean, we kind of want to look at the square of the distance between it and the mean and we sort of want to minimize the sizes of those squares. You've probably seen this in high school or whatever. But when it comes to the Laplace distribution, you're actually trying to minimize the total absolute distance from the mean. Which in some ways, makes more intuitive sense. So if you love medians, and you hate means, then the Laplace distribution is for you. It's the normal distribution, where we're switching to median. So I think that's a good way to think about it.
Again, this segment is not about trying to show all the mathematical equations and how it works but just try to wrap your head around what all these things are and what they're used for. And so the median is very interesting because the median is not affected by wild pitches in the data. So for example, if I have a data set where I have one 0, and I have ten negative numbers and ten positive numbers, then the median is zero. The mean? The mean could be anywhere. So there's some kind of stability of it that helps with some problems and that a lot of people like. So again, Laplace distribution is to median as normal Gaussian distribution is to mean. I think that's a good analogy.
All right. Hope you enjoyed that today. I'm really hoping to get, had a very busy couple of weeks, but I'm hoping to get back into the swing of things here. I'm hoping to have Aaron on soon. I have so many things I want to talk about as a solo show. So have a great week, everyone.
Narrator: That's the show. To support the Local Maximum, sign up for exclusive content and our online community at maximum.locals.com. The Local Maximum is available wherever podcasts are found. If you want to keep up, remember to subscribe on your podcast app. Also, check out the website with show notes and additional materials at localmaxradio.com. If you want to contact me, the host, send an email to localmaxradio@gmail.com. Have a great week.
