Transcript
This is an AI transcription and may contain errors
Frank: You are here with me, Frank Prendergast, and my co-host Justin Collery, to argue about all of the latest AI news as usual. But actually, before we get into arguing about AI news, Justin,
Justin: Yes.
Do AI projects justify taking private land?
Frank: Have I told you about the film that I watched last weekend?
Justin: You can. I love films.
Frank: Well, have you heard of a film called Little Pink House?
Justin: No.
Frank: It is a film starring Catherine Keener who, I dunno if you know her, but she is fantastic. She’s phenomenal in everything she’s in, and this is no exception. She’s brilliant. So Catherine Keener’s character buys a house in the town that she grew up in, and it’s a bit of a fixer-upper, and she does it all herself. Puts a huge amount of effort into it, gets it just the way she wants it. It’s absolutely stunning. But unbeknownst to her, the city is trying to attract this big development, this big Pfizer development into the city. And it’ll be brilliant, and it’ll be jobs, and they’ll be collecting loads of taxes, and all the usual…
Justin: Sounds great.
Frank: But Pfizer need a whole load of land, so they decide they’re just gonna take everyone’s houses.
Justin: Yeah, but that was a film, Frank. It’s not real.
Frank: It was a film.
Justin: Dreams and reality.
Frank: Based on a true story.
Justin: Yeah, the character may have been real, but Pfizer just taking their land probably wasn’t real. There’s laws, there’s checks and balances, property rights, all that sort of stuff. America’s big on that sort of stuff. They’re good when it comes to property.
Frank: I won’t spoil, I won’t do any spoilers for the film. But I will tell you why I’m telling you about this film on a show called The AI Argument. I came across an article in The Guardian.
Justin: Mm-hmm.
Frank: And it was kind of about the same thing, but not for Pfizer developments, but for getting power to AI data centres. So they’re building, as we know, as we’ve talked about on the show, these massive data centres. One of the stories I read was actually specifically about Stargate, the OpenAI development that has the Trump administration backing and all that malarkey. And they need power. They need to power the data centre. And so they need to run new power lines. And so they need to go through people’s land. So they want to buy the land, basically. And a lot of the farmers are saying, no thank you. No thank you very much.
Justin: This isn’t a new problem, Frank, surely. I mean, power lines have had to go through people’s land for the last hundred years, and people haven’t given out about it. And the attitude generally is, look, it’s a power line. It’s not taking up all your land. And this is a public good. We’re giving power to the masses.
Frank: Well, this is it.
Is AI just clickbait in an eminent domain debate?
Frank: They can take the land, they can take private land for public use, which, public use, it just feels a little bit different when it’s running power lines to a data centre as opposed to running power lines to power the next town over, which desperately needs power. I mean, there was one case, right, there was this guy, I think this guy was in Virginia. I did some research and it’s happening in all kinds of places. It’s happening in Wisconsin, in Kentucky, where my wife Marcy’s from, in Pennsylvania, in Virginia, in Texas. And I think this story was in Virginia, where this guy, it was a famous artist actually, and he had cultivated the land himself into this beautiful prairie, and that’s where he gets his inspiration from. And he does these landscapes of the land he lives on, and now they want to run these huge Statue of Liberty-sized power lines through his land. And he’s like, no, this is not just my land, not just my life. This is my livelihood as well. And in that case, there’s another route that they could go, but it just costs them more.
Justin: So, hold on a second, right? I’m gonna give you the other side of the argument. And I have sympathy with the guy, right? I have sympathy with the guy, but it is a public good, right? So let’s say we were building a massive data centre, actually because people don’t like building things, so we build it in Cork, right? So nobody cares what goes on in Cork. We have to get some huge power lines there from the… It’s the ESB, right? It’s the power company that’s putting in the power lines, right? And that’s the public good, right? The fact that it’s going into an AI data centre is just clickbait, right? It’s just, oh, we’re gonna blame AI for this bad thing that’s happening. The power company, that’s just another customer that creates jobs, that needs a utility. And they have a mandate not to waste our money because they tend to be… In America they’re probably not public, or they’re not private, they’re not like state-owned companies, but they’d be public benefit companies, right? And so you can’t go around, you know, if they were to reroute their power lines every time somebody objected, there’d be power lines everywhere. And it would cost a fortune. So there’s laws in place that have been around for a hundred years that says, no, this is a utility and if we need to run sewage pipes or if we need to run water pipes or electricity lines, we’re not taking over your land. We’re putting a metal structure there. The land around the metal structure is still yours, and we’re delivering public good, which is the electricity. It’s not AI. The public good part is the electricity.
Frank: Yeah, but, I mean, in this case it is specifically running electricity to a private company’s development, and I do think that that’s a key difference.
Justin: No, but it’s always… It doesn’t matter, right? If it was a Pfizer factory, they’d do the same thing. If it was a factory that made baby milk for disadvantaged children, they would still run the power line through your man’s field, right? But people wouldn’t give out about it, and more importantly, it wouldn’t get any headlines because it wasn’t related to AI. I would say this happens hundreds of times every week.
Frank: It probably happens, I don’t know if it happens hundreds of times every week, but it’s the scale of the data centre developments that is causing the issue, I think, because they’re all private developments. And what is it that they’re talking about, like needing a hundred per cent more data centre development? So it’s the scale of this issue. It’s the scale of the land being bought up for this particular reason, and to power AI, which a lot of the people who are being asked to give up their land are like, well, we don’t want AI. Leave us alone.
Justin: Frank, I’m sure, right, I know that you keep a close eye on the economic signals and economic data, and specifically the economic data in America, and it will not have escaped your attention that all of the economic growth over the last quarter was from building AI data centres. There was zero growth in America last quarter, except for building data centres. So not only is it a public good, it’s also doing the entire economy good.
What’s your price in the AI land grab?
Frank: That’s an interesting take, but I’ll ask you this, right. If it was your house,
Justin: Yes.
Frank: If it was a letter pinned to your door saying, look, we’re taking this, we gotta run a power line through here. Thanks. How would you feel then?
Justin: Well, look, this comes down to then, what’s your number, right? Are you gonna offer me some money in compensation for this power line that’s gonna go across my…
Frank: Yeah, you would get money. That’s a good question. I think that depends. I’ve seen stories on either end of the spectrum. There was one guy who was fighting the utility company because he felt that they were offering him under market value. And then there are other cases where people are being offered life-changing amounts of money, huge amounts over market value. So yeah, it’s happening at both ends of the spectrum. So you’re saying if the price is right, you’re okay with it?
Justin: I totally, yeah. Take my land. Ooh, he’s gone. Take my land already, is what I’d say if people are offering me land, and if they’re not offering me, like, 60 million quid, well then I’m just, I’m not gonna take their money. They can go and do it. My question, Frank… Yeah, the tech demons are against us. So you don’t know what my answer is to that, and I’m not gonna tell you. You’re just gonna have to watch. What I was gonna say is, let me throw a question back at you, right? So let’s say you have a choice, right? The data centre’s gonna be built there anyway. They’ve gone through, right? You can’t just build a data centre without some sort of regulatory approval, unless of course it’s a Stargate one and the administration in the US say that it’s a fundamentally important piece of infrastructure and you can bypass those laws. But other than that, they’ve got their planning information, they put the data centre there. Would you prefer a wire going through your land or a nuclear power station beside the data centre?
Frank: Oh, that’s, thank you. That’s such a wonderful question, is…
Justin: No, you gotta pick one or the other. This is the sort of game kids play, like what would you…
Frank: I actually don’t know because my opinion on nuclear power is based on very outdated emotional attachments that I have from when I was a kid, and when we’d all just heard about Chernobyl and then they were building Sellafield across the water from us. I kind of feel like my knowledge on nuclear needs to be updated before I could answer that.
Justin: I’d like, I’m gonna take a step back from this argument.
Will AI’s power hunger drive chip efficiency?
Justin: And there’s an interesting side effect, right? Because this argument is real and people do give out about it and people do get upset about it, right? And I think there’s a positive side effect to the argument. And that is there’s now an economic push to make chips less power-intensive. And this now creates a societal push to make chips less power-intensive as well because obviously if the chip uses less power, you need less power lines. And so I still think you’re gonna end up with photonic chips, right? So chips that work on light instead of electricity. And we know from the research that they’re about a hundred times more efficient than a regular chip. So that seems like a good deal. We had the brains in Petri dishes last week, not so keen on that one. And then there are other companies, I can’t remember what the name of it was, but there was another company and they were making what are called ASICs. So those are chips that are designed for a specific purpose. And again, they had multiples of less power consumption. On the other side, I did notice Jensen, whatever his name is from Nvidia, this week was at a conference and he was talking about the demand for inference. So the demand for tokens was going parabolic and it was exponential growth and stuff. And I agree with him. So, you know, if you think that the power lines and the data centres and so on are bad today, I would put it forward that in 12 months from now, if we don’t have gains either in the software and the hardware, you’re gonna have a lot more demand for power centres, and that’s gonna make this argument get worse and worse. I wouldn’t like a power line going across my land either. I do have sympathy for them, but if I was being offered 60 million quid, yeah. In fact, if I was offered 60 quid, I didn’t invite them.
Frank: Yeah, there was one woman who was offered 33 million for a 650-acre farm in Kentucky. And I have no idea what that means, but I asked ChatGPT and it said it’d be about maybe 5 or 7 million at normal market value.
Justin: They’re very fat power lines, aren’t they?
Frank: What, what do you mean, very fat power lines?
Justin: Fat, very fat power lines.
Frank: Oh, fat. I mean, the things that I was reading about that they’re constructing are huge. We’re talking…
Justin: Does she have a very narrow farm? Like 650 acres seems like an awful lot of land to run a power line over.
Frank: Oh, I see. Yeah, I’d have to go back and read up on that story. There may be some crossover between some of the power lines being constructed and also some land being actually required for data centres themselves.
Justin: And that would make a lot more sense.
Frank: And there’s middlemen making a fortune right now, local developers spotting the opportunity, buying up land, and then selling it on to tech companies when they move in.
Justin: Listen to me.
Can AI spot land deals before developers do?
Justin: That’s gonna be an interesting… So you might have noticed a week or two ago, Andrej Karpathy, the ex-head of AI for Tesla and OpenAI and all these guys, now he’s a teacher and a lecturer, he likes to teach things, and he released an open-source project called Auto Researcher. And so I was playing with it last weekend. That was my weekend project. And Auto Researcher, it’s like a loop, it’s like an agent loop that everybody knows and loves, but you can use it to train models or train anything you want, right? So if you have a goal and you can sort of quantify the goal. So I want to maximise money, let’s say, which is obviously the one I went for. It will just run in a loop and it’ll write the program, or train a model for you and optimise for that. And it’ll just keep on iterating, iterating, and iterating. Anyway, I’m getting to the point. Slowly. I went to Claude last weekend and I said, Claude, look, this is just a game, right? We’re not really doing this for real, but if we wanted to come up with a way to make money out of this, can you come up with a hundred ideas and then go through a thing and rank them and whatever? Go through… I gave it the process and I said, I want you to give me the top five. And so I’d been playing around with football matches, data from football matches, so it said, actually your data from football matches is probably one of the best ones. Second one that it gave me was land in the UK for exactly what you described. So its suggestion was that you would get all the maps from the UK, you would look at the laws in all the maps, and then you’d look at history. It had a whole process, and the suggestion was that you would identify parcels of land which were not yet developed, but were ready for development and classified as agricultural. And then you would buy them, get planning permission, and flip them and sell them. And you could optimise a model to do that.
Frank: Interesting.
Did Claude get Justin his travel refund?
Frank: That actually reminds me though, we forgot. I forgot to check in on your ongoing drama, your other ongoing drama with Claude. Have you been reimbursed for the travel that Claude told you you would be covered under your insurance agreement?
Justin: Well now, Frank, are you aware of the term weaponised incompetence? I am not gonna mention any company names here, right? But I booked those aeroplane tickets through a booking agency, right through a, I booked the hotel and the flights all together. And so I lodged the complaint. I had the transcript, I kept everything. I got a response during the week. It says, thank you for your request for a refund from the hotel that you stayed in. We’ve got in contact with them and they’ve refused your refund. I’m going, I didn’t ask for a refund from the hotel. I asked for a refund from the flight, the airline that was supposed to fly me. I got another email today to say that the flight people are very slow to get back to us, so no is the short answer. But I do have a job.
Frank: Watching this space.
Justin: They’re gonna get an email tomorrow morning about nine o’clock from Claude. Oh no, Claude said they’d give it another two weeks, but it’s checking my emails constantly.
Frank: Okay, so Claude is on it.
Justin: Let me just explain for those who don’t know what that is, right? Because this is so within cowork in Claude, if you’ve got the desktop app now, within Claude you can connect it to your email or whatever else you want to connect it to, and you can run scheduled tasks and tell it what to do. And you can now create projects. So this is a project, right? Get the money back is a project, and you can say your goal is to get the money back from the holiday. And you can say, right, set up a scheduled task every week to check my emails to see do we get any response? And it sort of, we set it up together, me and Claude, and it said, fine, we’re gonna give them a month and then after a month we’re gonna escalate and send them mail and stuff. So I actually spend zero mental energy knowing that Claude is in the background, looking after it for me.
Frank: Excellent. Excellent. Well, let us, yeah, keep us posted on that.
Is Microsoft about to sue OpenAI?
Justin: Let’s talk a little bit, let’s talk a bit about OpenAI.
Frank: Yeah, you’ll have to fill me in on what’s going at OpenAI because I’ve been obsessed with AI taking people’s land.
Justin: I see you’re very upset about that. That’s okay. So OpenAI’s had a very interesting week, right? So everybody’s been paying a lot of attention to Anthropic and how Anthropic are being sued by the Pentagon, and so on, and that’s working its way through the courts right now. So this week it was announced, it was rumoured that Microsoft were about to sue OpenAI. Now, Microsoft, if you remember, are one of the biggest investors in OpenAI. They have a deal that gives them the right to all of the IP from OpenAI, and they have a right, I think there’s some clause that says if they achieve AGI by 2028, they get all of the IP for everything up to that point, right? Or whatever. So OpenAI signed a deal recently to get investment from Amazon and part of that deal was that OpenAI models would now be available on Amazon as well as Azure, and that’s why Microsoft are suing them because they say you can’t do that. We have an exclusive deal. Your models are served by us and us alone, or yourselves, but not by Amazon. And OpenAI have come up with some clever legal wording that says that the inference isn’t being done by Amazon. The model is hosted on Amazon, but somebody else is, and they have some sort of clever legal wording. They’re not gonna win that one, right.
Frank: They seem to be very good at coming up with these clever little get-out-of-jail-free-card, or attempts to get-out-of-jail-free-card, clauses anyway.
Is OpenAI finally killing its side quests?
Justin: I think, right, because there’s a lot of other things going on with OpenAI this week as well. So another thing that came out is that the washing has been, or the things are being, the clothes have been washed in public. Now it’s come out that OpenAI are gonna stop all these side quests. They’re gonna focus on enterprise adoption. And clearly what they’ve done, like the Johnny Ive thing. Is the Johnny Ive thing ever gonna become reality? How much did they spend on that? 6 billion. 6 billion.
Frank: I can’t remember. Yeah. So, and are they really gonna drop it now?
Justin: It seems like they might, right? That’s what the rumours are saying. All of these other side quests, they’re literally gonna do what Anthropic do, they’re gonna try. And the other rumours this week is that they’re going to unify their app into one app. So if you have the Anthropic Claude desktop thing, you’ve got one single interface, one pane of glass, and you’ve got three options across the top, chat, cowork, or code. And what the rumours are is that OpenAI are gonna do exactly the same thing. They’re gonna have one pane of glass, it’s gonna have a browser in it. Who’s using their browser? I mean, what was the point of them releasing the browser? Who knows? But they’re gonna try and integrate that along with Codex and so on. What I think is going on is that the commercial realities are starting to come into play now, and they need to make money. That’s why they did the deal with Amazon. That’s why they’re trying to unify everything, right? They’re having their arse handed to them on a plate. I’d say one more thing, by the way, that happened this week that was interesting in terms of, you know, chickens coming home to roost and so on. Mustafa Suleyman, who wrote that great book that we both read and whatever, well respected. But I remember he became the head of Copilot within Microsoft. He’s lost that job this week. He’s been fired.
Frank: I missed that completely.
Justin: Yeah, he’s been fired. He’s now the head of some other thing, but it’s clearly that he’s been pushed aside. And two interesting things to note on that. One is the adoption of Copilot appears to be terrible from Microsoft. So I’ve seen numbers that say that the paid take-up is about 3% of their user base. So absolutely terrible. And the other thing is, I saw that even though he was one of the founders of DeepMind, I did read when he became the head of the AI division within Microsoft, I did read rumblings at the time that he was not the right person for that job. So it’s interesting that that also seems to have happened this week.
Frank: It is weird to go back to OpenAI and their abandonment of side quests, as they’re calling it now. And I dunno where that quote came from, but I did spot that, but I was like, okay, but didn’t they declare a code red just a matter of months ago or weeks ago? I’ve lost track of AI time at this point. But, you know, they already declared a code red and said they were gonna focus in on the core product. And there was really no sign of them genuinely doing that because they went off and they hired the guy from, and I can’t even remember how many side quests we’ve had since they said we’re declaring a code red. So I’m kind of looking at this going, okay, but are you really gonna do it this time?
Justin: Let me remind you of one of my predictions from earlier on in the year, and I’ll tell you why I think that’s a good question. A great question, in fact. So one of my predictions is that Sam Altman will not be the CEO of OpenAI by December 31st, right? That’s my bold prediction. And let me tell you why. Look what happened between OpenAI and Anthropic with Open Law. OpenAI bid against Meta in order to get, and I don’t know how much, it was never really revealed how much money they paid, but they paid a reasonable amount of money and they have him on board. I don’t even know what’s happening with Open Law, but at some point they’re going to somehow integrate it into the ecosystem, right? I think it’s been spun off as a separate sort of non-profit-type entity, right? Anthropic didn’t bid for it, but they have a great coding model. They just went and coded it.
Frank: Yeah.
Justin: It costs them way less than what OpenAI paid. They own the IP now. So if you have great products that are AI and help you to code, why are you spending all this money getting some guy to come on board and then, like, why didn’t you just get one of your, like, those sort of questions, like why didn’t you just write it? Why didn’t you just copy it? It’s open source. You don’t even have to write it yourself. Surely you have the technology. So why did we spend all this money? And I think those questions, the side quests and the Johnny Ive and the other acquisitions that they were doing, at some point you would’ve thought the investors would start to say, look, our job is to make money, not to spend money. Why aren’t you using the technology to make money?
Frank: Yeah, and it’s interesting as well, I think because we’ve talked before about how OpenAI appeared to be playing at the biggest scale they possibly could, and basically kind of trying to take on Google and become the new Google for the AI age. And now it seems like, yeah, because of the pressures exactly as you said, they’re kind of going, oh, you know what? We should really be focusing on making sure that Anthropic don’t eat our lunch.
Justin: Yes, and Anthropic, by the way, also this week I’ve seen a lot of metrics around enterprise adoption of AI tools in the last short period, three or four months, and Anthropic is up at, I think, 70% or thereabouts. It must be less, maybe 50%, right? But certainly it was by far and away above everybody else, and OpenAI were down at 23 or 24%. So, you know, and that’s where the, and if you, let me just link that back to Jensen’s comment about the demand for tokens. The greatest demand for tokens is not gonna be from people doing stuff for free. You don’t get paid for those ones on their iPhone. The greatest demand for tokens is gonna be from companies who automate the processes within their organisations. So, I mean, I think that realisation is dawning on OpenAI now.
Frank: And then meanwhile, because Google are Google, they are doing all kinds of interesting side quests.
Justin: Yes, because they make a hundred billion quid a year. They can.
Frank: Yeah, yeah, exactly. Exactly. Yeah. So it’ll be interesting. So to go back to the thing that kicked this off where you mentioned that Microsoft were suing OpenAI, I guess the big question is like what scale of a suit would that be, and can OpenAI come through the other side of it if they lose?
Justin: So I would say that the, see, when you put the question like that, right, I don’t know. They could sue them for an amount of money that’s not gonna be consequential. It just stops the deal from happening. Now that’s an interesting thing because Microsoft, in a way, are suing themselves, right? The less money that OpenAI make, the less money that Microsoft make. So that seems kind of a weird setup.
Frank: Well, if you want to be really Machiavellian about it, you could look at the musings that we made on this programme previously where we talked about how, well, if OpenAI don’t make it, Microsoft would probably hoover them up for peanuts.
Justin: Well, and then if we were to look at the Copilot experience, will they do a great job? I mean, if that happened, you might say, well, Google and Anthropic are winning that race. They seem to be the best company at commercialising the technology.
What did ‘Sam Altman’ say in The Onion?
Justin: In unrelated news, by the way, Sam Altman was interviewed this week. Very revealing interview.
Frank: Oh yeah. Who interviewed him?
Justin: The Onion.
Frank: The…
Justin: The Onion. They are, and I think, because I know that you love Sam Altman, so I’m gonna ask the same questions and I’m gonna put you into the shoes of Sam Altman. You can try and guess the answers. How about that, right?
Frank: Sure. Okay.
Justin: Okay. So Sam Altman was asked, what purpose will humans serve once AI dominates every aspect of our lives? Sam Altman’s response, according to The Onion.
Frank: Oh, I, I would imagine something like humans, there’ll only be one. It’ll just be me. It’ll be singular human.
Justin: Good. Good answer. “Humans have always been and will always be important for watching ads.” That’s a good answer. Okay. You like this one? Sam Altman, according to The Onion, was asked, “What informs your personal sense of morality?”
Frank: What’s moral? Explain morality to me.
Justin: The Onion says his response was, “Previous things I’ve gotten away with.” Okay, I’ll give you one more. Sam Altman, apparently, according to The Onion, was asked, “Why did you decide to devote your life to AI?” This is the most meaningful one for me. I mean, the answer was deep.
Frank: I mean, I would imagine he saw how much money there was to be made.
Justin: Oh, that’s a good answer. But no, the real answer according to The Onion was, “I just saw so much suffering in the world that needed to be automated.”
Frank: That’s fantastic.
Did AI help Rosie the dog fight cancer?
Justin: We have, you were brought to attention to be a guy in Australia, was it?
Frank: Do you know, I’ve actually completely forgotten where he was. I assume you’re talking about Rosie the dog.
Justin: Rosie the dog. Yes.
Frank: Yeah. So this guy, I think he works in AI. To be fair, he’s probably not your average user, but he used AI to develop basically a personalised vaccine for his dog, Rosie, who had cancer. Basically this form of cancer, they said, look, it’s not really treatable. I think they had tried various treatments like chemotherapy and radiotherapy. It prolonged her life a little bit, but it didn’t shrink the tumours at all. And they said, look, Rosie probably has months to live. And this guy, instead of accepting that, used AI to do a load of research on what kind of possible routes there might be to explore. He then managed to convince a lab, based on his research, to analyse Rosie’s DNA and the DNA of the tumour. Then he went on to do all kinds of other complicated stuff and, you know, AlphaFold was involved. And so, you know, I’m kind of pointing out as well, this wasn’t just going to ChatGPT and saying, “Hey, could you cure my dog?” But it was enabling somebody to do massive amounts of research to enable them to go to experts and say, “Look, I’ve done this research, I’ve done the groundwork, I think we can do this.” And then the experts were looking at that work. Initially they were kind of like, “Who is this lunatic?” But then they looked at the work and they were like, “Hmm, there is something here. Let’s help this guy.” So absolutely fascinating story. And I think the main thing to take away from it is the things that we can do now, because we can do the research, we can do the groundwork, and then we can get human experts involved.
Justin: It is a brilliant example of how AI augments your capabilities.
Frank: A hundred per cent. Yeah, a hundred per cent. I did read, well, just because, you know, it wouldn’t be a Frank story if it didn’t come with this cautionary tale at the end of it. But I did read one article from somebody in the medical field and they were just saying, you know, we are gonna have to be careful now because this guy did it the right way, involving the experts. He made sure that he got all of the right permission, got in touch with people who were authorised to actually give the dog the vaccine that they had developed. So he did everything to the letter of the law properly. And of course the danger is not everyone might be quite so letter-of-the-law as this guy. And that’s something we’re gonna have to be careful of.
Justin: Yeah, I want the vaccine too. Gimme the vaccine. I’ll take the…
Frank: And you don’t, you don’t, you don’t care about the proper channels. You’re just gonna set Claude off on the task of…
Justin: Damn right. Yeah, yeah. I’ll take the vaccine too. Frank, it’s been a pleasure as always. Have a fantastic weekend and a great week.
Frank: Chat to you next week. Cheers, Justin.
Justin: Take it easy.
