Alex: Hello and welcome to a new episode of the Open Internet Pod live on X every Thursday, 12 pm UTC, and then again as a video on demand and a podcast every Sunday, 12 pm UTC on YouTube and your favorite podcast app. Today we wanted to have a look at a at a topic that might seem a bit risque because it kinda talks about bubbles and everybody likes the bubble while they're in it, but nobody wants to talk about it afterwards. So we thought we'd talk about it both after and before in a way, because we are a bit after the the crypto bubble that exploded with Terra Luna and we are in the midst of what is becoming, you know. More and more people, an AI bubble, right? And you have from people like Ed Zitron talking about it, you have even even investors like Brett Gershner from I forgot the name of his investment vehicle that you know, was Richard: Yeah. Alex: openly questioning Sam Altman during a podcast. I think it was like the All In Pod about Richard: Right. Alex: how much burn they have compared to to profits. And you know, Sam was very, very dismissive and said like, Hey, if you wanna sell your stock, I'm sure we can find someone to can take it off your hands. and it gotta shows, in my opinion, when you say that, first and foremost, it it was one of those admission of it kind of got past the filter because the answer was instant, right? So it was like a reaction. Richard: Mm. Alex: And I I don't think B Brett was the first person who asked that question. think it's it's becoming more and more obvious that you know the cost does not outweigh the benefits and we have some some new stats on hallucinations and and the percentage Richard: Hmm. Alex: of hallucinations and you know you you read about stuff that you know for example the the latest open AI model ninety two percent of the time would rather lie to you than tell you that it doesn't know the answer. Richard: Yeah, that's certainly, I mean, it's super interesting from my perspective because I mean, I use LLMs a huge amount for coding and they're just getting better and better. The way you do guardrails, they still need guardrails, but the way you do that has changed dramatically and what you need them for. But also the amount of time it comes back with something really quite good, it just nails the problem and it's a three line change where it should be and it works. and it defines it in seconds where you would have taken minutes, you know, is really great. And I think what that shows me is that there's a lot of doom saying, you know, I think you have this weird situation where you have a lot of people mindlessly boosting or not even mindlessly boosting, very tactically boosting to try and make it feel inevitable, et cetera, that we're going to have this AI revolution. And then you have people who smell the BS. pushing back against that. And what you end up with is no debate or actual thought on what's really going to happen or where the reality is. And I think the reality is super useful tools. Yeah, super Alex: Gonna get into that. Yeah. Richard: useful tools, but for certain tasks and not for others. So for things like software development, for a bunch of language processing tasks, et cetera, like machine translation, LLMs are clearly very good. Perfect, no, good. better than other options, almost certainly. But asking them for advice, et cetera, I think is something that's really in the interest, if they were risk averse, AI companies would say, don't do that. That's not what they're for. They hallucinate. They're not tools that can give you that kind of response. But they don't do Alex: If Richard: it. Alex: even the fact that it's called hallucination instead of lying, you know, I mean it l it lies. It literally lies to you, right? It's 'cause yeah. But they had to Richard: or just erroring, yeah, or just baffling, right? Alex: make up this they had to make up this other word altogether so they won't say that it's wrong or that it's it's making stuff up. No, it's hallucinating. Right? Richard: Yeah, yeah. Yeah, so. Alex: I I hallucinated a lot on my math quizzes when I was younger, you know. Richard: Yeah, it's super interesting to me because it's clearly about needing to justify these valuations and the approach and what you saying earlier, the model behind all of this of like super mass scaling very, very quickly and isn't realistically looking at where these things will be useful, et cetera, et cetera. And so. rather than having a model where, you know, lot of technologies are kind of introduced where the producers think carefully about who they should introduce it to first so that it gets the maximal, you know, positive reception from the right people who will then broaden its acceptance and stuff. And this is not what's happening. They're trying to just jam it into every sock closet and, you know, public park, wherever they can find a place to use AI, it's there now. and without much regard to what it's actually useful for and not useful for. Alex: Right, so so the the the approach for this for this episode before we we get too far into the weeds, Richard: Before I run off with something. Alex: yes, would be to to to like set up the theme, right? So so the first Richard: Mm-hmm. Alex: wave of crypto crashed under the expectations that they themselves created, right? Richard: Mm-hmm. Alex: A lot of this is very reminiscent of what is happening now in the AI sector. Right, because it's one thing if you talk about automation or you know, making low lower tasks automated through some sort of like intelligence that is derived from repetition and data, right? And it's another thing to say that we are achieving AGI, which is a whole different ballpark. I mean, the none of these models are capable of being self reflexive or taking a like a a map Richard: They're not intelligent, yeah. Alex: Yeah, matters of taste, they have literally no no no possibility to recreate that. So You know, that's the the kind of aspect. Like we know we've seen the AI bubble. We're seeing the AI bubble right now, we've seen the crypto bubble. Both of them have this overpromising, under delivering aspect to it. and and we want to talk a bit about that and how that reflects within projects and what the future can look like. But before we get into all that stuff, I thought we should do a bit of news. And I'll let you choose between Android hackers fulling biometric checks or Sony. remaining gamers they don't own their games. Which one do you want to start with? Richard: Let's start with the biometrics because I think this is something close to my heart because I've been yelling about it for years. yeah, mean, basically, folks built an app that can stand in for the photo, for the camera output during various KYC type liveness checks. So it can hop in and provide a pre-recorded session of view. in place of you actually holding up the camera and showing your face. So this is classically used for things like opening a bank account, definitely crypto accounts, but more and more other kinds of online services where they want to or they need to, by regulation, have fairly hefty identification of you. And what this just shows is that these technical solutions, quote unquote, just don't actually do that, right? there's always an end run around these kinds of checks and that I think legislators are often the patsies of salespeople at the kind of companies that say they can do this kind of thing because the legislators want to believe and when someone gives them the right story with their with a handout for a check, the legislators are likely to... to be drawn in because it's a nice story. It's what they want to hear. But time and again, we see this that it just doesn't work the kind of age verification stuff for kids to use social media in Australia is an absolute joke. And openly so on Reddit with kids talking about all the things they did to get around it. Alex: Yeah. Richard: And this is just another case of that same. same hubris and that same just willingness to believe in something that's obviously flawed and problematic. Alex: Yeah, what I think is even more interesting is the fact that now these tools are available Download for anyone with the with the phone. It's not just that you have to do some you know ex extreme level of hacking to get there, right? It's more like a it's hacking for all, right? It's you can get an Android phone, you can root it, and Richard: Mm. Alex: then you can run these apps on it, and then you're you're good to go, right? You you can do this stuff, and it's I don't know, it's it's it's one of those things where you end up thinking that you know, I already do this, but Every because I I I I I ended up having my identity stolen here in Sweden at at one point. you kind of have like have a shredder at home. Like you can't even throw anything in the trash because you don't know where it's gonna end up. And it's it's again not so much that these things happen and they're easier to get done, but it's also that you you're you're using systems that kinda incentivize hacking because the moment Richard: Yeah, absolutely. It's ridiculous because if you think about it, who is not going to use these hacking tools that we've just talked about to cheat? Well, it's the honest people, right? So criminals are going to find some way to either get a stooge to do it, or they're going to find software like this to hack their way through. So what you end up with is a database full of vulnerable, honest people. Alex: Yeah. Richard: This is the thing. So it becomes... You're actually putting all the people who do the right thing and, you know, use the, use the software appropriately according to what was intended to identify themselves. They're now all putting themselves at risk because they're providing video footage of their face and their ID documents online somewhere for someone to steal it and reuse, when applying for, whatever it is, an illicit bank account in that person's name. Alex: I think it's it's this harkens back to something you said in an earlier episode that stuck with me and you said something like, you know, well, we have all this infrastructure, you need someone to verify IDs, you know. Why don't we put these two things together? And whether you think about it or not, there is some public-private collusion happening there. It's obviously Richard: Mm-hmm. Alex: through these people, the you know, the the revolving door analogy, right? You you finish Richard: Hmm. Alex: your job at the local council and then you get picked up by some security firm as a consultant, you get paid a lot of money, and then you call your buddies at the city council and like, hey guys, why don't you use this app? And you Richard: Yeah. Alex: know, everybody, you know, you scratch my back, I scratch yours. And the people who you're supposed to serve, you know, they just get served instead. Richard: Mm. Alex: And and the the increased occurrence of such behavior also shows that we are no longer morally inclined to do the right thing. You know, it's you end up in a situation where more and more people are doing these and it kind of becomes the way to do things, and and it's bad because it's kind of like dis destroys the contract between the citizen and the state. Richard: Yeah, no, absolutely. Yeah. I think it's, it's again, it's this dream to keep selling themselves for various reasons, that the right place to put the preventative measures is like right up front, right? Alex: Mm. Richard: As if, as if, you like vet everyone coming into your country club, there'll never ever be a possibility of a crime inside the country club. And it's like, no, that's not how it works. The thing to do is to actually police. It's not like that. The thing to do is to actually police people's real actions. that problem is that takes hard work and it's difficult to handle with the PR. And it's what's absurd is that one failure of policing can get a huge amount of bad press, whereas Another huge hack on a database full of people's identity documents kind of gets shoved off as, well, the provider, we will find them. They were terrible. They didn't do their jobs, but it happens all the time. And it doesn't seem to stick to people who should be getting in trouble for like, why did you enforce a regime where people had to put their data in some random company's server? Alex: I think it's one of those things that Cory Doctor highlights. it's the small compromises. Like Richard: Hmm. Alex: a series of very small compromises that overtime, you know, the salami strategy, right? Like over time you or the the the he s thing that you talked about, right? With the Richard: Yeah. Well, it's also a little bit of a switch, right? So, yeah, you have to give your name to, I don't know, the one I got lost my identity on was a mobile provider, right? So, well, you know, criminals buy mobile phones. So you'll have to give your name. And then later on it's, will need your name and your address, maybe some other details. And then eventually we need your name and two forms of government issued ID. So when the mobile phone provider that I was with got hacked a couple of years ago in Australia, Optus, they had my driver's license, my passport, my address, my photo. They had everything they could possibly need. mean, the same information I have to use if I want to open a bank account or whatever, they had all of that. So the government had to rush out, reissue of driver's licenses and a bunch of other stuff to fix the problem. But no minister or kind of government authority got in trouble for that, where in fact they should have, because there was no need for me to provide all of that information to the mobile phone provider in the first place. And it's absolutely absurd. And at the same time, if you had to get a license, if I had to take all of that to the government to get a special number so that I could go off and get a mobile phone number, then people would actually object probably, because it would seem weird that you need a license to have a phone. And people would realize, actually, why do we even? want this regime where we have to know everybody who's got a phone. And Alex: Okay. Richard: it's a massive blame shifting operation if nothing else. Alex: Right, so let's move along then. Richard: Yeah. Alex: Sony reminds Gamer that they don't own their games. Sony is sending email communications to PlayStation users reminding them that they do not own their games purchased on the PlayStation store. Several people have reported receiving an unprompted email from the company containing a copy of the PlayStation terms of service that set out the rules regarding online services, digital purchases, and more. The email explicitly states that PlayStation users do not own their digital games and only buy a license to use the product when they make a purchase. And I think Richard: Hmm. Alex: this is very interesting because there's also a class action lawsuit happening in California s exactly about this thing where where you know the the wording of you know buy, right? What Richard: Mm-hmm. Alex: are you buying? It's a game and you're buying the game, but no, it's a game and you're buying the license. So Richard: I think the Sony one has wording. I think it was Sony where it's like they explicitly say it like deep, deep, deep in the document somewhere that like the use of the word by does not imply ownership. yeah. Alex: Right, right. But I'm saying like when you do interact with the store it says buy and you are buying a a game and Richard: Yeah, no, that's the thing. then 22 pages of legal document later, they tell you that when you buy, you don't own it. But that's what they say. They Alex: Right, exactly. Yeah. Richard: don't say, buy a license to use this for a week. They say, buy the game on their website. Alex: Right. Right. Exactly. And not only do they say that, but they also have stipulations in the terms of terms and conditions that say that you have thirty days to send them by mail, by snail mail, a complaint and saying that you do not agree to those terms of services and thus, you know, remove yourself from them. But who does that, right? So It's it's I think it's very Richard: That always reminds me of the start of the Hitchhiker's Guide where they're like, these, you know, the aliens that are about to pave over earth, tell the guy who's complaining, like, but these plans have been on show at the planning office at Alpha Centauri for two centuries. Alex: Yeah. Richard: It's a bit like that. It's like, well, thanks, Sony, for that. Alex: Yeah, and I'm just I just highlighted this as part of you know this change in how we do things and how we how do we perceive things, right? Because while we're transitioning from the physical world into this hybrid, physical, digital environment, we don't want to lose our rights in the process. And that's pretty much what's happening, right? Everybody's going like, but hey, this is you know, you agreed to this, you know, 20 years ago, right? In Alpha Centauri. Richard: Mm-hmm. Alex: No, nobody agreed to this. He just made it very difficult for people to understand what exactly you're doing. And then eventually, you know, they got to a situation where they wanted their games. They're like, Well, whoopsie, it's not yours, right? So the fact that these are happening more and more, and part of the Talking about technology has now become also talking about the political aspect of it, the ownership, Richard: Mm-hmm. Alex: the privacy, the surveillance, right? I think that's a very positive development because we are being a bit more conscious about how we use technology and what it actually does. Richard: Mm-hmm. Yeah, no, I agree. I think it's overall good to see the amount of extra discussion and awareness around these topics. It's probably a good thing to say. Alex: So now let's go into some stats on hallucinations and performance of LM models because this kind of ties up a bit into our Richard: Mm. Alex: discussion for today. So these are the latest statistics from Supermind. And the thing that stuck out to me the most, and I thought it would be interesting to talk about it, kind of ties up a little bit with what you were saying about the the robots helping you with coding. if you look here, you have the AA omni accuracy. Accuracy on hard knowledge questions across 42 topics. Higher is better. Right? and that's the the row here. And we see that, you know, this is the latest open AI model, it's fifty-nine point four percent accuracy, right? And the highest one is sixty-five percent, and I think this is the anthropic mythic, the one that got pulled off the you know, like it y you cannot use it outside of United States. But still, these are very, very low numbers for something that's supposed to, you know, change the world to have a gazillion dollar valuation. And then another another stat here, which is very interesting, and again, kind of like provider open AI. It's this AA omnihole stat, which shows the percentage of time where the where the LLM chooses to make up something instead of telling you that it doesn't know the answer. And for the latest OpenAI model is 92. 2.2%. But I mean, look, all of them are like the lowest one is like 30 35%, right? Or yeah, 35% clot opus 4.8. So I don't know what what what do we what do we where have we seen this before? You know, a a technology Richard: Hmm. Alex: that does something and it does it great, but the thing that it does. does it really work for the VCs behind it who wants to pump the valuation as much as possible and then you kinda end up you having to create all these little stories around it and then the stats came out and like whoopsie actually Richard: Well, I think you can kind of often see progressions, right? So I think you can actually even go before the crypto bubble through even maybe as far back as the original internet crash of 2000, right? Where people are staring at this new technology, nobody really knows what it can do. Things like YouTube don't exist yet, right? So people have a sense that it could be super useful. And then suddenly pets.com has like a billion, multi-billion dollar valuation in 2000. Right. And I think what happens is, you get this, the storm happening where people are rightfully very excited about the possibilities. Some people take the middle ground with the obvious looking products and get funded to try and build them. actually some really useful, interesting things get ignored on the sidelines because people don't. see, have the vision to see that they'll be really big later. And because there's no air left, that send has been taken, people start moving kind of outwards in concentric circles, try to promise more and more outlandish things because they want in on the pie, right? They want in on what's happening. And so you end up with, you know, internet for paper clips or whatever. And then, you know, in the kind of dot-com boom and in the crypto boom, you ended up with I mean, eventually it went to straight gambling, but on the way there you had chains that were going to do all sorts of things that were patently impossible to anybody who investigated the functioning of how they worked, et cetera. And those were the ones that came after the middle ground had been taken by some things that looked probably quite useful and sort of had a growing market, et cetera. And so you end up with this, you know, It's those little compromises, right? It's those little steps each time to something slightly more outrageous, slightly more far-fetched. And I think that's where we are with AI. It's like, there's a bunch of core things that you can demonstrate if this is what you were claiming it could do, right? Its ability to write a function for you in a given coding language that has the results that you need. And you could get a page of statistics like we just looked at, and it would be... very easily understood as, you know, after a certain point, it's going to be useful. Yeah. Spell checker. Not perfect, but useful. helps you. It's got some false positives, but it's, you know, it's good. But that's not enough. That's the problem is that's not enough. So they start saying, well people want chatbots and people want to use chatbots for all kinds of things. And, know, there's plenty of examples of people going to chatbots for therapy. And the companies having had legislators and other people tell them, look, you can't let people do certain things with these tools because it will harm them. And the companies trying to work out ways to get around those restrictions. So nominally putting them in, letting people not closing accounts when there are no people have problematic behavior and stuff like that. And I think this is the problem, right? They need every possible user and every... possible use case they can because they're trying to get every dollar they can because they've made such outlandish promises that you know they'll do anything to try and stave off failing. Alex: But you know, I'm looking at it from maybe from a bit more of a different angle. And I'm thinking to myself that why is there that we don't have any sort of progress in any other domain? Right? Like the latest thing that happened was this empic thing, right? but you know, I don't see the news talking about a new way of building houses that is cheaper and better. Maybe maybe it exists, but it doesn't get it doesn't get like that. No, I mean like Richard: You mean from AI or just in general? Alex: in tur in terms of like technologies and development and progress. Like you know we all the companies that are having these huge valuations are tech companies. Right, like we don't have huge valuations outside of the tech industry, right? There it's Nvidia, it's Google, it's open AI, it's Microsoft. You cannot become a trillion dollar company, you know, building houses or or you know, like doing I don't know, paper or you know, teaching people how to dance, right? So it's it's it's kind of like If this is the place where you get ultra mega hyper-rich, then that also attracts a certain behavior, a certain type Richard: Mm-hmm. Alex: of people, and then they meet together and they create their VCs and they have their, you know, like, how do we sell this? How do we push this? Right. And and that kind of like feels a lot like, you know, think Mike Novograd's right with Luna. How he was like pumping that, this is the one, this is the Richard: Absolutely. So was trying to think as you were saying that I was thinking what's a really great, amazing technique. Colored LEDs, the invention of the blue LED, which gave us the ability to have any color you want, because you've now got the base colors covered and you can mix and match. That's changed the world massively. A, we've got LCD screens everywhere, but B, household lighting uses a fraction of the electricity it used to. They last super long. They're amazing. They are genuinely one of the wonders of... modern world and we forget how bad the options were before them. But like you say, no trillion dollar companies, no massive overvaluations and stuff. And I think it's because there's a simple-ish kind of set of maths you can do quite quickly, which is like, well, people don't buy more than a certain number of LEDs, even if everybody goes crazy and replaces every light they can find with an LED lamp. It's a limited market, right? Whereas in tech, you have this ability to sell dreams, right? Beyond what's actually possible because you, can sell these disruptive stories that we're going to totally change how people work or think or do or whatever. And that that scales as fast as you can imagine, right? People, you know, will somehow need hundreds of times the amount of inference and compute that they use now on a daily basis, because this will change how everything works. and those stories can just get a lot further than we're going to replace every light globe in the world, which is just not a thing. Alex: I think that when when Steve Jobs did that kind of stuff, you know, they're talking about the reality Richard: Mm. Alex: distortion field. I think when he did it, it worked because the products were so well developed and it was like a very closed ecosystem. And he kinda like always delivered on what he said. Richard: Mm-hmm. Alex: And and then now you have this like newer generation that saw Steve Jobs do that, but they don't have anywhere near his worth work ethic or capacity Richard: Hmm. Alex: to just like pull all the best talent. And you know, they they talk the same talk, but the product cannot live Richard: Hmm. Alex: up to that expectation. You know, they create expectations, they're they're writing checks their bodies can't cash, as they would say in Top Gun. Richard: Well, I think it's also, and this is the scary thing, is that for a lot of companies, and I'd need to do the maths, and I haven't prepped on this one because it didn't occur to me till now, but I'm not sure that Apple's trajectory would be acceptable compared to the growth targets that OpenAI, for instance, have set themselves. So like, Nvidia are... projecting a 70 % increase in revenue, I think, for the next fiscal year, off like a projection of 350 billion for this year, which is just insane, right? And I'm not sure if Apple had had the growth trajectory that they even had with the iPhone, if that would be as aggressive or as impressive as what the AI companies are now claiming they're going to have and valuing themselves. against the claims that they're valuing themselves against. So yeah, think it's a combination of those things. Like, yes, the iPhone was a generationally impressive product. It totally changed the whole industry in several ways. But even that isn't as disruptive and as valuable as what people are claiming AI is. And I think that's... part of the issue is you have to make these more and more and more extreme claims on the upside to get people their money back, to get the VCs their money back, to be seen to be playing the game. And so you've got this insane race between Anthropic and OpenAI and others to sort of outdo each other in a way that isn't even vaguely realistic, if you ask me. Alex: Right. Just wanted to ask you real quick if you have your microphone right in front of you. Okay. Richard: I do, I also muted it because I was coughing and I think I might have changed how sensitive it was. So sorry audience if I'm sputtering. Alex: Yeah, no worries, no worries. Yeah, but I think I think that that you're you're on to something. but I would say that I would say that Steve Jobs, you know, going back to that, he was pretty pretty strict. with how he tackled these what do you call them like activist investors. I I think Carl Icon Richard: Mm-hmm. Alex: wanted him to do stock buyback Richard: Mm-hmm. Alex: to increase shareholder holder holder value and he just told him to like, you know, why don't you go sell your stock if you want to make a profit? Right? Richard: Hmm. Alex: You bought at a certain entry point, now you made I don't know how much money, go sell your stock. If you if you wanna make money with Apple, go sell your stock. Don't expect us to do anything in that regard, you know In other words, like we spend the money on the things that makes the company better, not on what you want because you wanna, you know, ten times your investment. And I think that was like a very healthy approach for the consumer because he had the product in mind, right? And when Tim Cook took over, he was much more open to these sort of things and then s you know Apple started doing share, you know, share buybacks, and you know, you get like the gazillion versions of iPhones, different SKU for every price point, which was his strength, you know, and then you get services and all that stuff. But you know, who's buying the the Google Vision Pro, whatever it's called, you know? Who's who's buying all this? I mean, I would say that the the the greatest thing that Apple's done recently is the series chip chips. Richard: Mm-hmm. Alex: which are really groundbreaking, you know, and like now, like five years later, people are still playing ketchup with them, you know, 'cause, you know, for for laptops and so on, they're in a class of their own. But the product in itself, I don't think they're they've done anything great since the iPhone. Richard: No, and there's gotta be a question of whether that's a reasonable expectation. I think that's one of the issues here is I think that reflects back on what we're saying about AI and bubbles, right? That there's this kind of expectation that you have to continue to set the market on fire and that you'll have another killer product. And I don't need as much circumspection about is there in the same way that people aren't prepared to say, the LLMs we have now are great for this subset of tasks. And you fu- Alex: Imagine how much money would disappear for the valuation if you'd say like, hey, these are great if you're coding and it helps you code six hours work in like twenty minutes, which is amazing, but it's not AGI. It's not gonna answer all your questions. Richard: Yeah, well, and that's the thing. think that the stories that get told to investors have to be around essentially, and it's funny because it just seems so sexy to finance types. it's around mass layoffs, right? we're to lose a whole lot of these like wages that hold us down in the economy, which is we're in the business, which is... strange because it will also obviously massively impact the economy, which is the place where the money comes from that these businesses use for their profits. But also is that actually realistic at all? Right. That's just that that becomes a necessary tale to tell. is it is it realistic? Alex: Yeah, because I'm cause I'm wondering, like, you know, is it is it because we can't believe that we can get funding for an idea unless it has like this, you know, moonshot, right? We always talk about the moonshot. But you know Ethereum is still like infrastructure bound, right? It's it's crippled by by its infrastructure. and y is it is it that you know the question is Is this whole VC model thing, right? We we created this thing, it worked, it gave us Silicon Valley, it gave us the dot-com bubble, it gave us, you know, Facebook, all this stuff. But you know, it's diminishing returns, like you said. But you cannot sell that. You cannot sell someone that has like a limited scope. We just want to do this much, but we want to do it good. That's the premium on that is probably negative because everybody's looking for that one in a gazillion bet that's gonna save all their bad bets. And what happens is that, in my opinion, you you you have a situation where the VC model kind of like incentivizes cons. And that's why Richard: Mm-hmm. Alex: you have so many losers. And out of all those losers, one actually makes it and saves everybody. Richard: I think, yes, I think you're right. I think there's a couple of ways that that starts to happen, right? One is not deliberate, but I think one of the ways that starts to happen is when you have the mentality of, need one of these to come off, but I don't need like, you my one in 10 or whatever it is, you know, makes me profitable. It changes your attitude to whether you need to genuinely vet everything. Right. And it starts to make you think about it in terms of like probabilities and the numbers game rather than like actually thinking about each, each instance. Is this something I want to be part of? Is this something that's realistic, et cetera, et cetera. So it's, it sets you up then mentally to start looking at it as like, well actually if I can, you know, if I have that mentality, then I can start thinking things like, well, what if I just managed to flip it before it crashes? Alex: Crypto. Yeah. Right. Richard: Right? Because I'm starting to think that way. Right? I'm just thinking of it as a numbers game and I've got to just like pull my money out before. So I stopped worrying. It's very easy to then stop worrying about whether you're actually creating a viable business and start worrying more about, there a way I can structure Alex: It's a gamble. Richard: it to make my gamble pay off? And so then you can also stop looking at things like, is this a real potential or... Is it just something where I can get my money out on the way up? Right? So are they really going to build factories that make widgets? Or can I just get my money out during the widget factory boom and not really worry about what happens after it? And that I think is one of the big issues with the model is you've got, you know, there are old school shops on Sandhill Road where they very diligent and they invest in. some of the things we're saying don't really get the attention, know, mid smaller things that have good scope and they think they'll make their money back, et cetera. But the brand name stuff, the stuff you hear about in the headlines has to play this more and more extreme game of either ponzonomics or gambling to get out there and get the attention. Alex: I've heard this one take where it said something to the I mean something like there is there is a desire within the AI world to make the bubble so big that if it crashes or when it crashes, it has to be bailed out by the state. And that would be a Richard: Right. Alex: way of getting the state to pay for all the infrastructure that is needed because they don't have the money to do it themselves. Richard: I mean, the too big to fail is a thing, right? And if it's Alex: Yeah. Richard: going to take out your economy, it's a thing. think there is also, particularly in the US model, a desire to set up the next bubble, right? The next. mega industry that comes out of this. So they want to keep being the center like they were with the original internet and then with the hyper scaling era. so they're wanting to make sure that the next one happens on US soil and they're prepared to like. go to extreme lengths to try and do that. And that's not the approach being taken elsewhere, namely in China, where they have a very different approach, where the open weight models have a much bigger share of things. And it looks like, to people watching it, that may actually be the superior way of building things. To my mind, the more open the better. So probably. Alex: NVIDIA is going that way as well. NVIDIA's leading towards open models as well, but not for the reasons that you might think. It's because OpenAI is doing deals with AMD. Richard: Right. Alex: And you know the way for NVIDIA to hit back would be to like, is that what you're doing? then I guess we don't have to have that deal with you guys, and Richard: Hmm. Alex: we can also support all these open models, right? Because they already have Hugging Face, which is the you know, the Richard: Mm-hmm. Alex: the Zure hub for all the open models that are available Richard: Yeah. Alex: to download. Richard: Yeah, I think that's, I think it's, it's very interesting because I think the, is a question of whether AI ends up in the cloud or local or combinations and how that actually plays out. And that's one of the absurd things is we've got this level of growth being pushed when people haven't answered that. And I think that's a place where it's incredibly like crypto, where you've got things like Bitcoin and Ethereum where in Bitcoin's case, nobody's answered the question of what happens when the subsidy for security runs out. It's just an accepted, like, too far in the future problem, but there's a known problem, like a really big known problem, aside from the 51 % attack. And people are just, were happy to pile in and just ignore that for now because it seems to be going all right. And I think this is the kind of thing that happens in these spaces where people are just so keen for it to work that they keep pushing. And I think this is true as well with AI. People are putting in this level of investment when they don't know really how it's going to play out, like not just what the winning model will be, because you can influence that, like how it even works for people. And I think that's really interesting, particularly as people starting to pay more attention to what's going on with their data in the cloud, et cetera. And I think we'll see a fairly big push for people not to want AI to be just like running their lives in public as far as the kind of, know, open to the company. Alex: I want you to have a look at this and tell me what you think of David Lynn's take on on AI. I thought it was quite funny and curious to hear your opinion. Yep. So I'm curious what you make of that. Richard: Well, think it's, I mean, we've seen, yeah, in crypto, you had the, the reinvestment Ponzi's, right? So people would make money on a, on one project, use that to fund the creation of another project, which they would then hype, dump on retail and recycle. That was huge, but it was limited in some ways by the needs to, you know, lure in fresh retail to, to. to extract from. What we're seeing with my understanding of the finances around the way Nvidia is operating is that we have a kind of advanced Ponzi because in a way, or the risk of an advanced Ponzi because you have companies buying, funding the purchase of their own product from their customers. So they're lending their customers the money to buy their own product. So the circular financing can lead to ungodly size bubbles because you can recycle that money then incredibly quickly. And when you're talking in corporate terms and when people are letting companies write billion dollar checks daily, that can happen at a speed. The growth of that bubble can happen at a speed that is hard to grasp. And it's not necessarily tied to too much because it can be tied to purchase orders for equipment, et cetera, that doesn't actually have to be made for the money to start moving in a circle. And so I think that's where David Lin's point is really trenchant is investors will want these companies to list at some point that will require some level of opening the books. And I think the way Doctora puts it, everybody loves AI when you're being sold hundred dollar bills for a dollar. And what we've seen so far is when they try and sell the $100 bills, even for $5, people reject it. What he's talking about there is it costs far more to generate the results from the AI, given the cost of the build out for the infrastructure and the energy costs than it does to actually, that the people will actually ever dream of paying for the result that they got. And I think that's really like even where I've been saying I'm getting good results with some coding tasks from LLMs, you know, at a realistic cost basis, would it be worth doing? I it's certainly at the levels that it's subsidized now, it certainly is, but you know. Well, I think there's a little bit of the gym subscription thing going on there where people, a few people go over their $30 plan a few times and so they kick up to the 200 and not a lot of people are using that all. But if you start doing agentic team workloads where you have agents spawning agents to do tasks, to build things, then yes, I think you end up with the ability to, yes. get thousands of dollars worth of energy burned on your behalf at the very least plus all the equipment costs for your $200 subscription, absolutely. Alex: Yeah, so so it this is this is I think it like ties up into the crypto world in some way because here you have the subsidies. hiding the actual cost, right? Because they want to get Richard: Mm-hmm. Alex: as many users as possible. But those are fake users because you know like the product is not that great that you would pay the full price once they stop Richard: Mm-hmm. Alex: the subsidies. Like if I have to to pay forty thousand dollars, I'd probably look for someone to hire then pay that every month, right? Because then I I I I skip all the all the all the levels of you know like having to sit there and like no that's wrong, that's wrong, that's made up, you're hallucinating again. And then I would probably use some like lower level LLMs and put them in the hands of the devs that I hire with that money. And then I would have like a augmented team of of engineers using a more basic version of LLM, but being more cost effective. Richard: Yeah, I think that's where we'll see like people really working stuff out. I mean, there's also all sorts of silliness going on right now, like using frontier models to sort through your email. Alex: Mm. Richard: When they're not doing a better job, particularly, and the speed is utterly irrelevant, because you could do it overnight or whatever. So I think companies will start being much more thoughtful about that. But guess what? That could be done by stuff like on-prem, and you could also possibly do it with people's workstations when they're... not around, right? Like you can use spare compute to do some of this stuff so you don't need these Alex: Right. Richard: huge investments. So I think there's a lot of these things that are un- unthought through and I think it's similar, it is similar to crypto in that way and that people didn't investigate the fundamentals. Like very few people got excited about and interested in the fundamentals of the things that they were saying were important because of the fundamentals that they represented. Alex: But at the same time, because I love it that you brought up fundamentals, don't they make it really, really hard for you to understand the fundamentals? Like, cause I was trying to look at a lot of projects that are fundamentals, and the deeper you go, the more convoluted it gets, the more, you know all these like terms come up, and you know, like for example, hallucination. It's lying. That's what it means. It's lying. But In order to like hide and obfuscate, you create all this fancy language, you make this long, long, long, long reports that nobody has the time to read. And in in one way it comes Richard: that was like, this was one of the skills in the crypto industry was people renaming stuff to get away from or around, like people serious questions. So how do you explain that? the system that you're building doesn't have a facility to do X. no, we've got nodes that do X, they're X-doer nodes. And you come up with a cool name for that and then people were very Alex: Mm. Yeah, yeah. Richard: happy. right, these are the first guys who have X-doer nodes without actually looking at, how do they do the X, right? And this is classic corporate obfuscation, but it was certainly an absolute blight on crypto and it kept people from digging into the fundamentals. And it's quite funny because often when talking about Soto I'd say to people, like, no, question is, who pays for this? And people Alex: Mm. Richard: wanted it in crypto terms. There's just the question of like, who is actually paying to do that? Or why are we paying for this other thing when it's not needed? And that's, know, people were kind of uncomfortable because they were what about tokenomics? And that was mostly just wanting to go back to the language that was only invented to obscure the reality and any understanding of what's actually happening. Alex: No, I I completely agree. I think that's that's going back to this idea that people just wanna believe and they don't really wanna work to get to where they wanna go. So Richard: Mm-hmm. Alex: if I mean i They were kinda like a match made in heaven. The the the the scammers and the con artists found exact all the lazy people that just wanna make a quick buck and they found each other and that's how you got the the bubble. But what's scary is that this AI bubble, I see the same behavior that I saw in the crypto bubble. But this is for real. I mean it's no longer I mean, compared to Luna collapsing like what is forty billion or eighty billion dollars, how much was it before the crash? I mean Anthropic, OpenAI, NVIDIA, these are like trillion dollar companies. Right? If they start behaving in the same way and you know the the rug gets pulled, I mean it it Richard: Yeah. Alex: it really makes me think that they do this on purpose because they wanna have that backstop of the state coming in and maybe implementing this like surveillance crypto AI, no, not mean surveillance AI state. Richard: Well, I think it's varied. think partially, like, it kind of doesn't matter to people. Like, they're just interested in, like, being part of the ride. The ride goes on long enough, they're happy that they were part of that, right? So there's no... Necessity to think that people are actually, you know, there for the long haul in these things. They're just there for, you know, to be this person right now. They're enjoying that. That's certainly something we saw in the crypto bubble was like people who had no, no intention of seeing their project through. they were just going to cash out at some point, either partially or do a full road pool and then just wander off. Right. There was a very common practice in the crypto bubble. And I think there's a lot of people in the space who just. somewhere between buying and genuinely believing in the hyper scale dream and kind of not caring. They're just performatively being the disruptive CEO. They're performatively being the next jobs or whatever it is in their head. And so they are enjoying that and they're not actually, you know, if you, if you question. things like the feasibility of the industry, then you just incur their wrath because it's popping their bubble, literally. Alex: Yeah, I think I think that's I don't know what to call it, but I think that it goes beyond trying to make money because it kinda like skirts the idea of morality, hard work. Richard: Mm-hmm. Alex: you know, all those things that are necessary elements of building a a high trust society and a high trust environment, right? And again, it's one thing to see it in the crypto world, another thing to see it in in the in, you know, in the let's say the real world, because the impact can be so much larger. And and I think that people Richard: Yeah. Alex: are starting to figure this out. And I wanted to share something with you and also with our viewers because our listeners won't be able to see this. But there's this guy, this Australian guy, countryman of yours, called Suthless, and he has a new is he? Okay. Richard: Although he's hiding in New Zealand. What are you doing over there, Struthless? Yeah. Alex: Right, so he has a new documentary called The Subtle Al Art of Normalizing Dystopia. And it's very interesting to see that, you know, a guy who's like a filmmaker has ended up going so deep. Obviously as a documentary filmmaker, you Richard: Well, it's Alex: do have to go deep. Richard: certainly interesting as well because the first thing I think that got his attention to me was like a wonky video on fonts, which Alex: Mm. Richard: I really enjoyed. But it's interesting, like this guy is not like a classic gray beard hacker internet freedom guy. He's an advertising and content and digital communications guy. yet Alex: Right. Richard: this has come to his attention. Alex: Right. Richard: So I'm very well camouflaged in apparently. Alex: Yeah. But no but but I I I think I think I think that you know is that also a bit on us to to make it simple for people to understand stuff, right? So because If you're going against these people who are, you know, hiding behind all these terminologies and hundreds and hundreds of pages of docs and so on, would it not Richard: Mm-hmm. Alex: mean that the antidote to that would be to have something that's very simple to understand and be open and transparent? Richard: I think so. And I think like the reason that someone like Cory Doctorow is a hero is he has the ability to do that when he speaks. He's thought deeply about these things and he has a very tight set. I mean, I've heard enough of him speaking out to hear the repetition. And that's also kind of heroic, being able to go out there and deliver it again and again and again to people who need to hear it. But he has a pretty strong and unique skill in like really boiling it down to know this is what's happening. Like, no, your printer ink is very expensive because the US... back ended its DMCA laws through trade treaties. And so now you can't, it's illegal for you to like mess with your own printer. Really simple explanation of like the ink is expensive because they use the legal trick to stop you from being able to put generic ink in it. And that's, I think the kind of thing we need to hope to, we need to aspire to across tech generally, but. you know, also when we talk about the internet, what we should be doing here, something we try and do at CIDO is to speak very plainly, right? To talk about things and, you know, consensus mechanisms and the the way blockchains work is complex and it's very hard. But I think the work to be done is to show people, you know, views into it, which, which helped them understand it, not to do the obfuscation. And that's, that's the clear difference is like a lot of the industry use these made up terms, use a lot of jargon obfuscation, cause they don't want you to dig in. want you by the time you've been introduced to your 32nd new piece of vocabulary, reading a white paper, you feel like, I'm just going to. to give up, right? I'm just not gonna look into this anymore. And I think it is on all of us to kind of try and actually speak plainly and clearly about the things we understand so that we get rid of that sense that people can't understand these things. Alex: Right, I t I I think that my favorite bit of explanation for Sato nodes is it pays the nodes that bring in traffic. And I feel like that's such a simple way of explaining it. Like look, if you work and you do your job and you bring in data to the network, the network pays you for it. If you don't do that, you don't get Richard: Right. Alex: paid. It's that simple. Richard: Yeah, it self-optimizes. How does it do that? Well, you get paid more if you interact directly with the user and less for doing backroom stuff. So everybody wants to do as little backroom stuff and as much serving users as they can. That's the result we want. It's as simple as that. It's as simple as explaining things that way, I think, most of the time. And that's something we need to be... pushing is that people can use those kind of analyses. It needs to be done properly. It's not like every rule of thumb holds every time or that can get, aphorisms will do, you need to look into stuff. But it's not something that the regular person can't do or understand. Alex: What I would say to the regular person would be something that maybe won't be that nice to hear, but in the short term, but in the long term, will prove very useful in my opinion. Is that this idea of trusting a third party, be it the state or some your favorite tech company or you know, your favorite distribution service or whatever, what have you. This era is coming to an end because you're no longer aligned from an incentive standpoint, right? There was a time where the corporations were incentivized to have customer loyalty, right? Because there was high competition and they had to develop a relationship with the customer in order Richard: Mm-hmm. Alex: to ensure that it can sell long term. Now, when they've made a deal with the state instead. They don't need to have that relationship with the with the customer anymore. And guess what? It's easier to have a relationship with the state because it's like one, two, three things that you have to keep a tab Richard: Yeah. Alex: on instead of having like hundreds or millions of people to interact with. Richard: It's one of the things I've been saying since I was in my early 20s, which is far too long ago now. But I actually think things like legislation for truth in advertising hurt people, right? Because if you have legislation that says advertisers have to be truthful. then there's a chance that they're bending the rules and trickery will actually work on people. Whereas if everybody knows, I can go on TV and say, you know, this milk extract will make you grow your hair back or whatever I like, you know. shaking this, you know, silver crystal will make you rich, then people won't believe me because they'll just be like, well, you can say anything on TV. Right. And so you actually have a degradation of people's ability to look after themselves because you're trying to protect them. And I think there are people who need protection, like children, for instance. You can't. expect them to be doing the judgment and everything. by and large, yes, I think people unfortunately need to do the work, right? You need to actually think of how can I find if I can't and don't have the time to do this analysis, where's an authority I feel like I can rely on? My friend. you know, Bernie who did it did a good analysis of this or someone, you know, someone on the internet who has skin in the game who will lose their reputation or, you know, actually think through how do I know this information is of value and, and is it, is that Alex: Right, but Richard: good enough for this situation? Alex: But you're expecting this sort of self agency from a global population that has been automized through an entire coveted lockdown, right? So it's like you're asking people who Richard: I think we've also... Alex: have lost a lot of social contact to become social again, which is hard. Richard: Yeah, also so much has been done over last 10, 15 years in the kind of hyperscale age to normalize the idea that it's the platform's responsibility, that it's someone else's responsibility. And it's like that example I've used a couple of times on the pod of like, you need to have social abilities, right? There's a guy being an absolute num-nuts at the bar in your local pub and... the public can either tell them to go away or the population do and you manage it as a group, right? If it's your friend's uncle and you know that his wife just died and he's being a bit of a jerk, you handle it differently to if he's some young guy who's trying to get a rise out of the crowd. You have a social set of ways of dealing with this and it doesn't always work perfectly, but you have those skills. I think what we're getting to is the point where you kind of, at every call, we kind of call the police. or something, Alex: Right. Richard: or it's the pub's job to make sure that everyone's separated. So we put in barriers or something between the people in the pub to kind of make sure that nothing can happen. KYC people adore that sort of thing. And I think we've lost that. Alex: Yeah, I think this idea of right, this idea of relying on on someone else or you know, the kindness of strangers, I don't think we live in that world anymore. And I don't think we ever did, but to some extent it made financial sense for these corporations to have a a proper and honest relationship with you. It doesn't make sense anymore, which is why we're seeing this shift. That's my opinion at least. Richard: Well, no, think it's the dynamics have changed, right? So we've seen a couple of things. We've seen them create systems that are for their own benefit, where they're the ones that are expected to manage and control and police everything. And that makes us dependent on them for that. But it also means that, they don't have to honor their side of the social contract with us either. They can just start to insurify their services because there's nowhere else we can go. Alex: Yeah, it's it's it's it's the topic that keeps on giving us Richard: Yeah Alex: reasons to meet every week and and have a chat. but yeah, thank you so much Richard for all your insights into this topic. I'm sure we could speak another hour on this and we will, but that will happen Richard: Yes, we have gone a long. Alex: will happen next week. So if you want another hour of me and Richard you you'll have to wait until next week or you can Catch the VOD shortened version of the spot on X on YouTube, sorry, and on your podcast apps every Sunday, twelve PM UTC. Thank you so much for this episode and catch you guys next time. Bye bye. Richard: Thanks everyone.