Bitcoin and AI: A match made in cyberspace

SPEECH TITLE: Bitcoin and ai: A match made in cyberspace
SPEAKER: Guy Swann
CONFERENCE: Bitblockboom 2023

Introduction

I am Guy Swann, the guy who has read more about Bitcoin than anybody else you know. And I have the honor today of kicking this thing off. You probably know the podcast Bitcoin Audible, but you may not know about another project I started, about six weeks ago now, called AI Unchained. I wanted to talk about the marriage between Bitcoin and AI. That’s why our talk is Bitcoin and AI: A Match Made in Cyberspace, a little corny subtitle, but I’m sticking with it.

These are the two most important technological developments happening in the world right now, and there are a ton of parallels between these two technologies that are really important, specifically for Bitcoiners to understand because the head start that we have on understanding the monetary shifts of the world, we can actually apply and add an at a multiplier if we properly understand the direction and the value of AI. I was actually chatting with Mark Moss last night, and it was funny because he used the exact same analogy that Svetski and I did on an episode of AI Unchained. Everybody remembers 2017, the height of the crypto retardation. Bitcoin was huge, Bitcoin was suddenly a big deal, people were making a lot of money on it. Blockchain. Everybody wanted to stick ‘a blockchain’ on everything. If you had some sort of an idea, didn’t matter what it was, stick ‘a blockchain’ on it, and it’s like innovation, it was just everything blockchain. I actually remember a company that was trading on the stock exchange, and they didn’t do anything to the company, they didn’t change the business plan, nothing changed, but they added blockchain to the name, and in a couple of weeks, the stock literally tripled. Bananas on the blockchain were literally a thing. I used that example because AI is basically filled with that right now. It’s got the crypto disease, and separating the signal from the noise is incredibly difficult because everybody’s just sticking AI into their thing. AI with email, AI notes, and God knows what else.

As soon as I had my first touchpoint with AI that really kind of blew my mind was Chat GPT and Stable Diffusion (the image diffusers). My first thought was, ‘Holy crap, is this going to be like a really terrible centralizing force against what Bitcoin is doing?’ Because I kept hearing that you have to have thousands and thousands of GPUs to do this, and Google has their new model, and it’s like a hundred trillion parameters, and it’s gonna be better, it’s got more neurons than the human brain. There’s all this hype about are we creating God and all of this stuff. I immediately was like, ‘I think it’s critically important that we understand this to make sure that we don’t go there, because if we do, all I could imagine was that AI was going to essentially be the worst surveillance nightmare that you could possibly come up with. If only a billion-dollar corporation can run these things, this is actually that valuable, then we’re all going to be plugging into it, and it’s going to have a contextual understanding of everything we do on our computers, everything we’re doing at all times, is going to be able to predict where we’re going to click before we click on it, and we’re not going to be able to get the productivity gains without plugging into those giant servers. Honestly, that kind of scared the shit out of me. I felt like I really needed to understand what it was.

Is AI a centralizing force?

In doing so, there’s this question, have we created God? Have we created a superintelligence? I want to say, no. In fact, with OpenAI and Chat GPT, they have an Open AI playground which will actually just show you what it’s doing essentially. It’s just a prediction machine for text, and it takes the context of the specific word, the group of words, and then the entire sentence, and it literally just has a probability for how likely the next word is. Like that little thing on your phone where it predicts what you’re about to type? It says like, ‘I went to the,’ and it said ‘store, bar, Bit Block Boom,’ in the fill in the blank, and you could select one. Well, it’s basically that on steroids. You’re just guessing, and the playground will actually show you the percentage of the likelihood, like 73% likelihood that this is the next word if you go play with it. But I’m out of free tokens, so I’m not paying for that. So we did not create God. It is a technology, and it is a form of intelligence, but again, it’s not some superintelligent being that’s going to eat us all.

With that, there is good news and there’s bad news. Unfortunately, there is actually a lot of bad news that we don’t have time to cover, but these are the main elements that are relevant to Bitcoin, and I want to hit a couple of them.

Another insanely deflationary force

First is that AI is an insanely deflationary force. It is going to really change how we do things, and replace an enormous amount of menial, tedious tasks and kind of administrative. The administrative economy is really going to get hit hard and everybody’s talking about how many jobs are you going to take. If you read any article, it’s literally anywhere between like five million jobs they’re going to be taken before 2030 to 300 million (I think Forbes article says 300 million jobs are going to be taken). It’s an exaggeration. You could have said the same thing about the internet, but anyway, the one interesting thing about this is that what it does is it actually takes a lot of the economic capital that we thought was not information technology and was not in the scope of the digital revolution and it kind of aggressively shoves it into that paradigm. So we’re going to get Moore’s Law in areas of the economy that we did not expect was going to be the case for potentially a really long time. In fact, if you have not listened or read the Price of Tomorrow, you should fix that. But if you haven’t read it in a while, you should read it again because in the context of the technological shift we’re seeing it’s just so much more potent with kind of the framing of AI.

The other problem is that if you happen to have 31 trillion dollars in debt, really, a massively deflationary force is a huge problem. So we’re going to have to get a lot of inflation. One of the things that inflation has to do is essentially outpace it, has to erase all of the natural price deflation that the growth would enable in the money supply, and because of that, there’s going to be a massive amount of money printing just to stay at baseline. In doing so, it means that more and more capital is going to be allocated towards the political and counterfeit class because the more they take our product productivity gains, the more all of this economic center basically sucks around them and their projects and everything that they want to do. That’s not exactly the best news.

Disruption as the norm

If we go back half a millennia, business models were a lot simpler, production structures were a lot closer to from production to product, especially in relative terms to what we have today. You can even use your grandparents as an example, they grew up in a world where you can just learn to do one thing really well, and then you had a career and you did that for 50 years and then you retired. That world does not exist anymore. It’s done. We are entering a phase where AI actually operates like a 3D printer because one of the things that a 3D printer does is it allows you to iterate, it gives you a production tool that isn’t locked to any certain object, usually would have to have like a giant machine and it has to have some very specific tooling or operation in order to print one gear and then you print like a million of those gears, whereas the operator can now just sit around and come up with a bunch of ideas and iterate on the same products a hundred times in a day. And if you’ve got a production line of additive manufacturing, you could completely change what your product is or what you’re offering to the market overnight. You just shut them all down, download a new design into the machine and then start printing again.

AI is kind of like that for task management and software and media generation. Stable Diffusion and these things are basically a form of media that generate media, and it’s software that generates software, and the operator is just trying to guide it as to what to create, you’re prompting it to give you back some sort of structure or idea. In that sense, there’s an interesting way to think about what AI is, it’s a little crude, but I think it works. It’s kind of like a really, really aggressive compression algorithm that doesn’t actually store, like an image diffuser is a great example is that it doesn’t actually store any of the images. If you search an image on Google of a cat, you’ll just get images of a cat. If you search the same thing again, you’ll find the same images of a cat. It’s just recalling something that’s saved, whereas when you train a model on thousand images of a cat, what it does is it stores the relationship between the pixels on the screen that give you the relative shape of a cat. So if you give it cat, it will literally generate an image of a cat that never existed but it will do it from all of the patterns that it’s stored that it learned from all of the images of the cats that you gave it. It’s like a hyper compression that none of the images actually exist there, but the patterns that the images represent are stored and can be regenerated.

What all this means in the context of disruption is that the norm will be a constant disruption of the economy and it will get faster and faster, it will continue to accelerate to the point that when we start into another phase of disruption in changing how we tool and operate, it actually will not have enough time to seat itself properly in the economy before the next disruption stage will already be started. Volatility is probably the first word that comes to mind.

Knowledge is power

Another element is ‘knowledge is power’. Companies are realizing that users are producing a new resource that they weren’t before. We were already being harvested for our metadata, our social graphs and all of this information. Now, we’re a great way to train their own company’s model. Zoom just changed their ‘terms of service’ that now they can use your likeness, your voice and your conversations to train their AI model. This is going to happen a lot. Because of that they’re going to start to silo a lot. They’re realizing that if they have anybody that has a community that has a user base that they can watch they think that they’re going to have an advantage over the other companies by locking in their user base so that they can watch you and they can use your feedback to train their AI and produce the better product. What this means is an aggressive increase in the amount of walls coming up on the internet and companies and platforms being siloed. Think of the problems of Apple and iPhone except spreading to a lot of different areas in the digital economy.

Prove that you are human

We have a lot of different ways of ‘proving that you were human’ online. Don’t mean shit anymore, they’re all useless. The number one method that we have to prove that we are human online is KYC, take a selfie with your ID, solve a catch, etc… If AI can regenerate your likeness instantly, it can regenerate your voice instantly. Now, actually, the newest models, you need five seconds. If you ever get a phone call from somebody you don’t know and you just say ‘hello, anybody there?’ That’s all they need to train a model and turn them all on your voice. And then they can turn around and call somebody that you know and be like, ‘hey Bob, what’s up?’ and it will sound exactly like you. You know the scene in Terminator? When the liquid metal Terminator kills his parents and calls them up on the phone, sounds like him. That’s the world we live in. That’s not science fiction anymore. Because of that, fraud is going to go through the roof. And when you have a product that you’re offering to somebody, a computational service that has immediate cost—in fact, Sam Altman said that every single time they run Chat GPT, it costs upwards of eight cents for every single time you ask a question to it. I want to talk about what that means. [I am not Morgan Freeman video running]—I saw that video as crazy. The fact that you can just do that live now is so nuts. What’s the good news? There’s actually quite a bit of a Bitcoin lining to this.

We have no moat, and neither does OpenAI.

Google Memo

There’s a leaked memo from a Google engineer that was sent around internally, and I almost did the entire talk on just that memo because it’s what got me starting AI Unchained. Essentially, open source has a really natural advantage. First, you do not need thousands of GPUs to run these things. Like I said, I can run some really fascinating stuff just on this machine, and this is not optimized for AI; the hardware isn’t even there yet as far as being geared toward it. All the hardware companies, are now making that happen. So the next iterations of products and chips are all going to be AI-optimized. But one of the really crazy things that open-source has is that companies building a model from scratch is insanely expensive. But if you’re working on an open-source model, you’ve already got a community that did all the work for you. There’s this natural advantage; there’s something called low-rank adjustments, and we’re seeing this same thing in LLMs, it’s not called exactly the same thing, but this is in image diffusers, and it’s basically micro-trainings on one specific thing. Let’s say, you want an image diffuser that’s a little bit better at generating images of people in a suit, and so you make a low-rank adjustment and you train it on 100 high-quality images of someone wearing a suit, and then you can actually retrain with the base model staying the same, you can actually retrain it to teach it to better make someone in a suit. And when you have a community doing thousands and thousands of these little micro-adjustments, you actually get extremely high-quality content because everybody’s very centered on getting the quality of what they’re trying to do. And you have this pool of knowledge that’s feeding back on itself, and nobody’s having to redo the work over and over again.

Another thing is that the establishment and the regulatory apparatus are crippling the shit out-of-their AI tools. They’re making them inoffensive and safe, which open source doesn’t care about, which is great because there’s all these sorts of use cases and restrictions in the establishment version of AI, the closed-source AI, that they simply cannot serve. They can’t meet. And then also, with the licensing problems is that OpenAI is going to restrict what you can actually do with their API and how you can use Chat GPT. But the free alternatives are actually not only matching but surpassing a lot of the closed-source, like the OpenAI models of Chat GPT. Falcon came out not too long ago and it immediately went to the top of the leaderboards of user feedback and the quality of answers. And it just keeps getting surpassed; the leaderboards are just active. Every few days, there’s another one that’s up at the top. Llama2, another open-source model, is now up at the top.

People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality.

Google memo

This is actually what we’re seeing. OpenAI is another, a leaked document or whatever that they might actually go out of business. They hold the record for the platform that has the fastest growth in users, fastest to a million ever. Ever. In June, they had 1.7 billion visitors to the Chat GPT service, and in July, they had 1.5. They’re actually losing because people are just realizing that they don’t have to pay a subscription and they can just run a free model and they can have it for their company. What this is actually showing is that because these middling models that have outsized quality and outsized of results is that we’re seeing a technology that actually greatly benefits the individual, the small and the medium service provider disproportionately than the giant corporation. And the overhead for the giant corporation trying to train its own model is something that the smaller guys don’t have to worry about.

The free version of Chat GPT, the smaller model, GPT-3, cost them $700,000 a day just to keep running. That doesn’t include GPT-4, and that doesn’t include their image diffuser. And we used to live in a world where API calls were largely costless. So, going back to the thing I said about, it costs eight cents every time somebody asks it a question, what you’re essentially doing is you’re being given an API access to burn through somebody’s GPU, just set it on fire. What that essentially means is that if you’re in a world of increasing fraud, what you’re actually doing is you’re providing an instantly delivered irreversible product, which wasn’t really the case if somebody’s just reading an article; there’s no real cost there. Somebody’s like, just watching a movie on your Netflix or whatever, and then you have to cancel their subscription in three days; that’s not that big of a deal. But if somebody can run up thousands and thousands of dollars in cost in just a matter of hours on your platform, and then you have to charge back, then you have to cancel, you can’t get that cost back; you can’t redeem that. That’s not free; it is a bearer asset delivery.

The future if orange

If corporations are crippling their own AI with licenses and trying to be inoffensive, then that means the open-source economy has a chance of producing higher quality alternatives with a much larger array of services than what the corporations can. If we see that the speed of economic change is going to get faster, and disruption is going to be the norm, this is actually a huge negative pressure on massive corporations. It actually contributes to the diseconomies of scale, and, again, it benefits the individual, the small, and the medium business far more because they are agile; they can change; they can retool quickly. Somebody who’s already invested two billion dollars in their product, in their design, or their production flow is going to have a hell of a lot harder time adjusting to constant disruption than somebody who’s small, agile, like the 3D printing versus the big machine sort of platform. And if none of the KYC measures can actually prove that someone is human, we’re going to see KYC everywhere. Everybody is going to start adding KYC because you just can’t have an unlimited number of AI bots imitating everybody as much as possible on everybody’s platform, just burning through capital and burning through the internet.

But on a long enough time span, KYC is dead; there’s nothing they can do about it. It’s gonna get worse before it gets better, but if you have a platform that actually runs on public and private keys and avoids the KYC, you have a huge starting advantage because that’s where things are headed. It’s going to be like the copyright fight against BitTorrent and file sharing. They tried to say, ‘Oh, the copyright’s going to get worse, and we’re going to have DRM on everything, and we’re not going to let the user control the content,’ and they failed; file sharing basically forced them to change their entire business model; they can’t restrict the music access; you’re going to get it everywhere that you want, and the longer they fought it, the more money they lost. And if you have a ton of fiat debt in a crazy price deflationary environment, the one thing you absolutely need is to be holding onto sound money; you need something that actually benefits from that price deflation rather than something that just turns the printing machine back on, and that the government has to essentially print to infinity to make sure that their obligations don’t get worse every year. Then if costs are instant and irreversible and the payment method is delayed and reversible, well, then if you’re building an open-source ecosystem with instantly delivered bearer asset settlement, where you send eight cents the exact same moment that you execute a cost of eight cents, your margins are going to be a lot better than the corporation that’s dealing with billions of dollars of fraud that gets worse every single year. Not only that, you’re going to be able to undercut them in price for any of the services that you provide. Essentially, you’ll have a better user experience. The beautiful thing about Nostr is that you’ll be in an environment where you can have an ecosystem where you can just Zap and get exactly what you wanted. And we’ll find ourselves in a place where people will wonder shouldn’t it have always worked exactly like this?

The path to Heaven is straight through the gates of hell

The problem here is that all of it gets a lot worse before it gets better. Unfortunately, that’s a truth that we have to deal with, but that’s a road that we can drive. That’s a path that we as Bitcoiners are ready to take because we already kind of knew that, and that’s a lot of the reason why we are here. But the thing that I love about Bitcoin is that essentially everybody here is a builder, it’s full of builders, and not only that, it’s full of people who actually have hope and actually realize that there’s a chance of fixing this, there’s no nihilism here; we recognize that there’s work to do, but there’s actually something better at the end of the road. This is why I started AI Unchained. If we properly understand these new tools and work together to utilize an open-source collaborative and self-hosted AI ecosystem, we will actually have a vastly better chance of getting the future that we are fighting for, so don’t sleep on it.

Thank you

Please, subscribe to AI Unchained. The point of the project is to unravel how important that is and how it’s tied to Bitcoin. Obviously, Bitcoin Audible. I’m still the Guy Swann on TwiXTeXr, and as long as it’s relevant, find me up there, and then obviously on Nostr. BitcoinAudible.com/AI is actually going to have a bunch of tools that I have been using recently. So if you want to kind of go down the rabbit hole, it’s actually going to be a good list of signal for things that would be really cool to check out.

Watch the original content: Click here
Also read: Tsunami of innovation

Disclaimer: Transcripts provided on bitlyrics.co represents solely the opinion of the speaker and is not by any means financial/legal advice or an opinion of the website. The content has been transcribed with maximum accuracy. Repetitions and fill words have been amended in order to enhance the reading experience. The full text may not be confirmed by the speaker. Please, refer back to the above-provided source of content for more certainty. If you are a speaker and wish to confirm/amend your speech please contact us.