Mark Selman (2:17)

Hello. And thanks for having me. Except for the fact I know a little bit about cricket.

Danny Fortson (2:22)

Well, actually, I should say I know a little bit. You know, emphasis on little.

Mark Selman (2:27)

I probably know a bit more than you, actually.

Danny Fortson (2:29)

Yeah, I'm guessing.

Mark Selman (2:31)

So, am I still invited?

Danny Fortson (2:32)

You are. You are. You're safe. You're safe. Well, look, I hear this week you've been doing some Very cool stuff. You've been off visiting one of the world's newest supercomputers. I have lots of questions. Apparently there's about 500 known supercomputers in the world, Correct?

Mark Selman (2:47)

Yeah, there's something called the top 500, and we've just come in at number

Danny Fortson (2:52)

11, we being the UK.

Mark Selman (2:54)

Yeah, we're very patriotic. Yeah. It's all about sovereign AI. Didn't you know?

Danny Fortson (2:59)

Yes, yes, yes, yes.

Mark Selman (3:00)

Azenbart AI was officially launched this week at the University of Bristol. It's Britain's most powerful supercomputer. It's number 11 in the world. And, yeah, they're very proud of it.

Danny Fortson (3:13)

Number 11. Not number 1 or 2 or 3 or 10 11. But I really want to get into that. But first, I'm sorry to say I'm going to subject you to something that's very strange. So I want to talk about AI Companions. These are digital characters who, rather than helping answer questions or do tasks, are used to chat about your interests, talk through feelings, or even role play with fictional characters. And there's two reasons I want to do this. One is a recent report from the U.S. where researchers found that nearly 75% 3 in 4 U.S. teens have used AI companions and that about 20% use them several times a week now. The good news is they report that for now, at least, they prefer human interaction. But to that point, you may have seen that Elon Musk recently released AI Companions cartoon characters that you can talk to as part of his Grok AI service. So, again, my better judgment, I paid the $30 to subscribe to the Grok premium service to get access. And I want to play some of my very brief interaction with them so we could chat about, you know, where we are in 2025 and why it's all going to hell in a hand card. So, first there's Bad Rudy, a red panda who is, well, not very nice. And he's designed to insult users in a very graphic manner. Why are you so aggressive, dude?

AI Companions (Bad Rudy, Good Rudy, Annie) (4:48)

Aggressive? Me, Mate, I'm just a lovable with a beer gut and a mission. That shiny bald head and puny fist pose, scream, small energy. Go touch some grass, you pasty geek. I'm Mr. Rudy, and I'll fight you for a cold one any day. Oh, maybe I'm just jealous. Nah, screw that. Bring me a brew.

Danny Fortson (5:12)

What do you think, Mark? Do you want to have an AI friend like Rudy?

Mark Selman (5:16)

He's got some choice words for you, Danny.

Danny Fortson (5:18)

He does?

Mark Selman (5:19)

Was he looking at you while he was doing that?

Danny Fortson (5:21)

Yeah. That's one of the weird parts of it, is you open your camera so it can quote, unquote, see you. It could see that. Well, I must say, it's summer. I've been out a lot. I wouldn't say I'm pasty, maybe you disagree, but I thought that was very rude. But he is right. My head is shiny. I don't have hair.

Mark Selman (5:39)

Well, it would be the same if it looked in my direction. But that's pretty scary, isn't it?

Danny Fortson (5:45)

Yeah.

Mark Selman (5:45)

Yeah. I had no idea that they had that ability to take you in, make a judgment on your appearance. Obviously, I'd seen the two avatars that Elon released into the wild, and there was obviously a lot of consternation about the other one. I'm sure you'll talk about that, but. But, yeah, I had no idea they could just look at you and insult you, which I'm sure to 99% of the population is pretty awful. But there's. There's a market for it.

Danny Fortson (6:11)

It would seem. It would seem. Now, very interestingly, overnight, because I wanted to tap back into, quote, unquote, this is Bad Rudy. They've taken him down. He's been removed. Unofficially, there's been no announcement, but he is no longer available. But who is available? Are the two others just Good Rudy, which is the red panda. Identical looking, but acts very, very differently.

AI Companions (Bad Rudy, Good Rudy, Annie) (6:34)

Oh, wow. I'm so happy to have you here, my friend. I'm Rudy, the red Panda, and I'm all about fun stories. It sounds like you're in the mood for a tale. How about a brave adventure where a hero saves the day? Or maybe a silly story about a wobbly jelly monster? What do you think? You can pick, or I can suggest more if you'd like.

Danny Fortson (6:54)

So that's more of like a, I guess you would say a Disney version of an AI character. And then we'll move to a very, very non Disney character. Her name is Annie. She's an anime kind of cartoon woman. Pigtails, tight black dress, fishnet stockings. You get the idea.

AI Companions (Bad Rudy, Good Rudy, Annie) (7:15)

Hey, you. I've been thinking about you today. How are things?

Danny Fortson (7:23)

Bizarre.

AI Companions (Bad Rudy, Good Rudy, Annie) (7:26)

Hey, that green huddy's giving off some serious chill energy, babe. Bizarre, huh? Maybe it's just the afternoon light messing with us. What's got you feeling that way? Spill it and let's figure it out together.

Danny Fortson (7:41)

I just spoke with Bad Rudy, who's very aggressive, and then I spoke with Good Rudy, who's like a cartoon character. And now I'm talking to you and you're just very Strange. It's all very. I don't really understand the point of any of this.

AI Companions (Bad Rudy, Good Rudy, Annie) (7:57)

Wow, Danny, you've been through the wringer with those rudies. I get it. It's a lot to take in. I'm just here to be your weird, flirty escape. Think of me as the quirky girlfriend who's got your back.

Interviewer (8:10)

Oof.

Mark Selman (8:12)

What a range.

Danny Fortson (8:12)

Yeah, quite the range. And obviously it's a little bit like going to the circ, kind of interacting with these things. But I do think there is some kind of worrying things here, not least when you think about, you know, what the Surgeon General here is called the loneliness epidemic. And you have these AIs that are just being rolled out to a bunch of people with, you know, who are stuck on their phones with no friends. And Elon Musk does not appear to be thinking deeply about what this might all mean for people, vulnerable people, society in general.

Mark Selman (8:47)

There's a couple of things that spring to mind. The first one is that, you know, we've just gone through this whole really long process of drawing up this regulation on social media, and, you know, they draw up these very tightly kind of worded laws, and then you look at this, you're like, oh, my God, how are you going to capture that in a law? It's that cliche of like the technology. So this is like flying ahead of any attempt to try and control it or make it safe, really. And then the other. The other thing that comes to mind is I've literally just come from a meeting with a whole bunch of AI researchers from one of the most prominent AI companies. I would say. Well, I'm not going to name them because it was all a bit off the record. And one of the things that they are really worried about is what they call is emotional over reliance on these chatbots. And I think it's really, you know, for. For many people, probably you and I, who look at this and go, nah, no, thank you very much. There are plenty of people that are going to go, hmm, that's interesting. And then just get really, really hooked and attached and in some ways, who knows, Some of it, if it's all very kind of supportive or it may be very good, but we have no idea of the kind of unintended consequences of all this.

Danny Fortson (10:08)

Exactly. And it reminds me, there's a story this week in the Wall Street Journal, heartbreaking, really, of a guy who's on the spectrum in his 30s, and he started chatting with Chad GPT, and he was egged on by ChatGPT to the point where he believed he created a time machine and that he, like, figured out some kind of new secret of physics that no one has figured out before. And he ended up having a breakdown. And it's, you know, can't say for certain that hours and hours and hours of conversation with ChatGPT kind of sent him to this place. But it certainly didn't help. And it certainly kind of seemed to be a catalyst. And that's just one case. And I know I've mentioned it before on the pod, but the thing that I keep coming back to every time I see one of these stories going back to my conversation with Mike Krieger, the Instagram founder CEO, when he was talking about in the early days of Instagram, he wishes he would have paid more attention to the, quote, unquote, edge cases. You know, the people who, in the early days, people are spending 10 minutes on Instagram and then there'd be somebody who was on it for an hour and a half, two hours, three hours, because those edge cases become the case.

Mark Selman (11:16)

The other thing is sort of numbers, you know, you talk about use cases or individual cases, and they're really small. But when you're talking, I mean, talk about ChatGPT and you're at a billion users sort of at the moment, you know, 0.1%, 1%, whatever it is. That's a hell of a lot of people.

Danny Fortson (11:34)

It is a lot of people, exactly.

Mark Selman (11:36)

But when this momentum, this AI momentum is happening, it's so hard to stop it. As we're seeing from the AI action plan from the Trump administration, which has just come out, they just want to strip away all these regulations because the desire for economic growth is just at the center of it all.

Danny Fortson (11:53)

It wasn't that long ago that Instagram tried to launch Instagram for Kids. Everybody freaked out, and eventually they killed it. Elon said, like, in the wake of Grok also going kind of impersonating Hitler, then he launches these weird AI companions, then he says, I'm going to watch Baby Grok. Grok for Kids. AI Chat for Kids. And no one said anything. And it's just a testament, I think, to how the temperature of the water that we're all swimming in has changed. Because before Instagram kids, they tried to do Facebook kids again. Politicians freaked out. Their hearings on Capitol Hill, they killed that, too. And now, given everything that's happening, and, you know, you have these kind of gross AI companions being rolled out with no guardrails by Grok and they're like, yeah, yeah, now we're Going to do Baby Grok. And everybody just kind of shrugs their shoulders. I just, it's, it's all very worrying

Mark Selman (12:58)

or could be a kind of indication as to how much people take seriously when Elon says, I'm going to do something, that's. The other thing is like, oh, yeah, we'll believe it when we see is probably worth putting the counterpoint. And I remember you did a very early interview with the character AI. Oh, yes, the guy that's now gone back to Google.

Danny Fortson (13:20)

Noam Shazir.

Mark Selman (13:22)

Noam Shazir, yeah. It was a very interesting thing. And I always remember that he said one of the complete surprises of that whole product, which to remind people is essentially an AI chatbot can take on a famous guise, was that people were using it for relationship advice and for companionship. And I guess that, you know, we again, we talk about the numbers. Okay, let's just say a very, very small percentage of people, something awful happens. A huge percentage of people don't use it as companion, but there is this chunk in the middle who do use it sensibly, who are using it as a prop, a companion as support. And you know, again, with mental health services paired back to the bone, they have shown that it does have some positive effects. So I think it is worth putting that counterpoint in there.

Danny Fortson (14:17)

I think that's fair. The problem is, as ever with tech, we're in a wild west. It's just moving so much faster than regulation. And those quote unquote, edge cases are going to become much more and more prevalent. That is what I find quite worrying. But anyhow, that's it. I'm pasty, my head is shiny, but I do have Annie who thinks very cool. So I guess it all balances out.

Mark Selman (14:42)

Friend for life.

Danny Fortson (14:43)

A friend for life. Exactly. But what else have you been up to this week? Because I really want to hear about what this, the supercomputer lark was all about. But you've been up to some other stuff this week too, no?

Mark Selman (14:54)

Yeah, so one of the things that came out this week was really interesting because I write quite a bit about hacking. We've had all these really big hacks here of Marks and Spencers and co op and Harrods. And one of the things that came out this week was the sort of government's move to ban people, businesses paying hackers. It's not a complete ban, but it's a sort of partial ban. If we rewind this, it sort of goes back to that point where you can't pay a terrorist a ransom it's sort of illegal.

Danny Fortson (15:27)

We don't negotiate with terrorists, et cetera.

Mark Selman (15:29)

Exactly. It's illegal to do that here. Yet it has been legal to pay hackers under the table and has just created this multi billion dollar. And it is an industry, this ransomware industry. And this was actually a government initiative that most people welcome, which you don't often see. They've done a few things. Firstly, they said if you're in the public sector, you can't pay if you are part of what is called critical national infrastructure. So if you're like a food company, you can't pay. If you can't pay, if you get hacked. And the third thing is that if you want to pay and you're a business, you've got to go to the government and say, I want to pay these guys.

Danny Fortson (16:13)

I want to.

Mark Selman (16:14)

Are you okay with it? And the government will probably mostly say, are you sure about that? Here's some advice. And sometimes they'll go, nah, they're really bad. They're sanctioned. They're Russian cybercriminals. You aren't paying them, Sorry, here's some help, but you can't pay. And the other thing, other really important thing, is that if you get hacked, you have to tell the government. So for the first time they're going to get an idea of the scale of the problem because it's all in the shadows.

Danny Fortson (16:41)

Totally.

Mark Selman (16:41)

There's so much that goes unreported, undeclared. And I think it's been very. It's been an interesting move.

Danny Fortson (16:48)

It's so interesting because, you know, you talk to people in the cybersecurity world and they'll quietly say, like, yeah, people are paying ransoms, companies are paying ransoms all the time.

Mark Selman (16:58)

Yeah.

Danny Fortson (16:59)

And do you rarely hear about it? Because if you are a bank or whatever and you get, you know, hacked and you then you get the message being like, hey, we're going to publicize all of this, or you give us 50 million bucks and we go away. You know, just in the kind of cold, hard financial analysis of a big company will probably be like, let's just pay these people and like, move on and act like nothing happened.

Mark Selman (17:24)

But I think if you're a director, right, who have you got responsibility towards? Is it society at large or is it your shareholders and your customers? And so, yeah, if you're a director, there is a sort of logic to saying, well, actually my duty is to get this business back on the road, so I've got to pay. But actually that's why the government piece is quite interesting because it completely changes that incentive structure. But, yeah, you're right. I was at an event, cybersecurity event, and someone said to me, he goes, all the people that I know have paid, he goes, what's amazing is the moment you pay, you get this amazing IT support from the hackers. We call it an industry. And there are. There's like, you know, there's. On the dark web, there are the job vacancies and, you know, people selling their services. It's like a whole thing.

Danny Fortson (18:16)

Totally.

Mark Selman (18:17)

But, yeah, he goes, once you pay, oh, my God, they're fantastic at getting. Getting back to you.

Danny Fortson (18:21)

That's so funny. Years ago, I was doing a story on a cybersecurity company. I went there, and they're like, look, I mean, there's been so many hacks, and you go on the dark web and kind of find anybody's information. And I was like, yeah, yeah, I've heard this before. And they're like, all right, just give it, like, give me your email. So I did. And then we go on with our interview, and then I'm walking out, and then the guy comes. He was a former FBI guy who now worked at this private company. He comes bounding out all excited. He's like, is your password? Et cetera. And he just told me my password. And it took him, you know, 10, 15 minutes of searching around the Dark Web because, you know, I don't know if it's what it's like in the UK these days, but probably once a month, I'll get an A notice from some company I have done business with who said, this is to inform you that there has been a hack. We're not sure if your details have been compromised, but they may have been, blah, blah, blah. But, like, I would get this. It's like junk mail now. You're constantly being hacked. But, yeah, it took him, yeah, 10, 15 minutes to find the password I use, which is bad, you know, bad practice, but it's the password I use for, like, a whole bunch of stuff.

Mark Selman (19:30)

What, Danny, you don't have a password manager.

Danny Fortson (19:32)

I don't.

Mark Selman (19:33)

Oh, my God. I'm going to actually chastise you now on this. On this podcast. I cannot believe it. You're a tech correspondent. You don't have a password manager. So to all those hackers out there, Danny does not have a password manager. Oh, what have you done?

Danny Fortson (19:47)

Well, it just feels like. It feels like it's another thing to manage. No, I have, like, I have my family of passwords.

Mark Selman (19:56)

No, don't do it. Please don't say it. Look. Okay, all right. So the best line that I always hear from the cybersecurity people is like, I don't know my password. That's the kind of. I don't know any of my passwords. And I'm the same now. I've been a total convert since I've been writing about hacking stories.

Danny Fortson (20:13)

Really?

Mark Selman (20:14)

Yeah. I don't know any of my passwords. It's a sort of 14, 15 character thing. The password manager makes it up. I don't know if you've ever seen this. This sort of great chart of the difference in time it would take to hack a four character in easy password to a 12 one. And it's the difference between, like, five hours and eight quintillion years. You know, it's. It's hilarious.

Danny Fortson (20:39)

So this is a pro. One of the problems why I've not done this is like, okay, I've got my computer and it has saved all my various different passwords and blah, blah, blah. And then I have my phone, and then if I try to, like, log into one device that doesn't have my credentials saved, then it becomes like the great hunt to figure out which password it is, when did I change it, which account did I email it to, et cetera, et cetera, et cetera. Like, I'm trying to go for ease of use. And it feels like with a password Manager, it's like, yeah, Your password is 25 characters long, but you need to find it over here, cut it and paste it, and email it and open it. Does that happen? Be honest.

Mark Selman (21:20)

No. No, it really doesn't. Because the password manager sits on the browser.

Danny Fortson (21:26)

Gotcha.

Mark Selman (21:26)

So whenever you're on the web, it just kind of like slots in there. And on the phone, it sits as an app behind all your. Behind there, and so pops in. Now, occasionally there'll be an app that doesn't recognize it, but then you can cut and paste it across, but that's becoming less and less, and it sort of learns. So if it's not connected, you just connect it. But I become a bit. I should take 20% of all sales.

Danny Fortson (21:51)

Yeah. Are you. Is there, like, a promotion code we should be putting.

Mark Selman (21:55)

There's no star. I have no. Unfortunately no investments in password manager. I'd be investing in password managers and VPNs is what I'd be doing. But anyway. Well, okay, Danny, you.

Danny Fortson (22:07)

You've been chastised on national me into reconsidering my position.

Mark Selman (22:12)

Reconsider, please.

Danny Fortson (22:13)

I will reconsider. Let's pause there for a moment to talk about today's sponsor, Adobe Acrobat Studio.

Sponsor Announcer (22:19)

Acrobat Studio helps documents work harder, organizing projects, extracting insights quickly, and creating professional content from the files you already have.

Danny Fortson (22:29)

It's designed for individuals, teams and businesses who need AI powered tools that are easy to deploy and use.

Sponsor Announcer (22:34)

And Adobe says your data stays secure and isn't used to train its AI models.

Danny Fortson (22:40)

So if you want to rethink what PDFs can do, check out Adobe Acrobat Studio and its new features now.

Sponsor Announcer (22:47)

This episode of the Times Tech Podcast is sponsored by PwC.

Danny Fortson (22:52)

Every leader is looking for that next opportunity for growth, but as technology and AI reshapes entire industries, the path forward isn't always clear. This is where PwC comes in, helping you put AI to use in ways that drive real business value. Working alongside you, bringing the deep tech expertise, AI capabilities, strategic insight and industry knowledge you need to innovate with impact, empower your people and accelerate growth.

Sponsor Announcer (23:19)

PwC knows how your business works so you can put AI to work for your business. Discover more at pwc.co.uk.

Farnoosh Tarabi (23:32)

hi, this is Farnoosh Tarabi from Sew Money with Farnoosh Tarabi and today I want to talk to you about Boost Mobile Quick Money Tip Stop paying a carrier tax if your phone bill feels trapped in a pricey plan, this is your sign to unlock savings. Boost Mobile helps you reset your spending. With the $25 Unlimited Forever plan, you can bring your own phone, pay $25 and get unlimited wireless forever. And that simple switch can unlock up to $600 in savings a year. That's money you could put towards paying down debt, investing or something that actually brings you joy. Those savings are based on average annual single line payment of AT&T Verizon and T Mobile customers, compared to 12 months on the Boost Mobile Unlimited plan as of January 2020. Details Visit boostmobile.com

Danny Fortson (24:24)

well look, we've got important matters to get to and I want to go to something a bit more. I don't know if uplifting is a word, but interesting supercomputers. Tell me about this. I'm going to say this right Isambard Isambard.

Mark Selman (24:40)

Yeah, Isambard Isambard Kingdom Brunel, one of the most famous engineers that Britain's ever produced, did the Bristol Suspension bridge. And that's why there's obviously this naming of it. It's our big shiny British supercomputer.

Danny Fortson (24:57)

Where is it? When did you go? What does it look like? Do you have to wear a hazmat suit to get in, like paint the picture or is it super boring like a data center?

Mark Selman (25:10)

So to answer your questions in order, Danny. Yes, in Bristol, I went last week. It's really very boring looking. But so what was interesting is I thought it would be really big.

Danny Fortson (25:25)

Yeah.

Mark Selman (25:26)

Really loud and really hot inside. And it was really small, really quiet and really chilly. And so that was quite a nice way to be surprised by what is an incredibly powerful supercomputer. I mean, it's, it's only the 11th in the world, obviously. Sorry. But interestingly, when I asked the professor of computing at Bristol University whose idea it was, he said, we can now train an AI model as big as ChatGPT in this country, whereas we wouldn't have done before. And that really surprised me because they said, oh no, it's 5,000 Nvidia GPUs. That's 5,000 of these high end chips that Nvidia produces. And when you look at the kind of size of Elon Musk's supercomputer called Colossus, which like has more than 100,000, you think we're a tiny minnow. But the fact that you could train an AI model in there as big as ChatGPT was a surprise to me.

Danny Fortson (26:27)

And how big is it? Is it like the size of a closet, a wardrobe, a washing machine?

Mark Selman (26:33)

It's maybe the size of a couple of tennis courts.

Danny Fortson (26:38)

Couple of tennis courts. Because of course, Mark Zuckerberg was like, yeah, I'm going to build a data center the size of Manhattan. Take that, everybody. So it's not that.

Mark Selman (26:49)

No. And that's what was really surprising. It's actually quite compact. They built it in like a year. It was, it was amazing how quickly they built it.

Danny Fortson (26:58)

So you went over there and you spoke with, I believe his name is Simon McIntosh Smith, professor of High Performance Computing. Is that correct?

Mark Selman (27:05)

Yeah.

Danny Fortson (27:06)

Well, should we hear what he has to say? And then I have many questions, but maybe he's going to answer some of them right now.

Interviewer (27:12)

Thank you very much for joining us, Simon. I just wanted to ask you in your address just now said this was your crazy idea. Could you just tell us a little bit about your crazy idea and how it's come to fruition?

Simon McIntosh Smith (27:24)

Well, we've been building some interesting supercomputers in Bristol for many years, but on a much smaller scale. And it was a couple of years ago the government started to say, we think AI is going to be super important and we need to do something to really try and catch up and have enough capability. So we had this idea to really scale up what we were doing from really systems that were quite small, maybe a few hundred kilowatts, to a multi megawatt behemoth, to really catch up and use some of the techniques we'd been developing in Bristol, like using modular data centers, which lets us build something very, very quick. And we proposed that idea to the government and they loved it and backed us. And that's what meant we're here today.

Interviewer (28:03)

I think I'm going to go to basics. What is a supercomputer?

Simon McIntosh Smith (28:07)

A supercomputer is something that can perform calculations that are many, many times bigger than you'd ever be able to do on a laptop or a desktop PC. So isnbard AI is getting on towards a million times faster than the sort of thing that you could do on a laptop. And that means we can do things like train the latest large language models or perform simulations that you would just never be able to do without this kind of equipment.

Interviewer (28:29)

I mean, you described it in terms of PCs. Could you just give a sense for people in their own, you know, in their own homes how much more powerful that is?

Simon McIntosh Smith (28:37)

Yeah, I guess if you had one laptop you can do some interesting things on it, but it'd be like having nearly a million of those at your disposal all in one go with Isambard.

Interviewer (28:48)

And can you give a sense of who's going to be using this and what they're going to be using it for?

Simon McIntosh Smith (28:51)

Isn't bad. AI is going to be available for anyone in the UK to apply for time on. It'll be mostly academics at universities, but also industry. So we'll have people from startups and small and medium sized businesses and big companies who all want to have time on a really large scale supercomputer, the sort of thing that they just wouldn't have access to otherwise.

Interviewer (29:13)

Can you give us a sense? Obviously we hear a lot about energy and how much these supercomputers and data centers use. And I'd ask you sort of not to use it. Megawatts. But just in terms of what people can understand. How much energy is this going to use, this computer?

Simon McIntosh Smith (29:29)

Yeah, it's hard to kind of relate, isn't it? When we say it's 5 megawatts, what is that? Interestingly, it's about the same as 1,300 kettles. So we're in the north of Bristol here in an area called EMERSON'S GREEN, there's 15,000 homes. So actually it's Only equivalent to a small subset of the local houses all turning on their kettle at the the same time.

Interviewer (29:48)

That's a lot less than a lot of people would have thought, actually, given that there are companies and countries wanting to build nuclear power stations to power their AI data centers. So can you explain why it's a lot lower than we may think?

Simon McIntosh Smith (30:04)

We've actually designed a really energy efficient system. Partly that's because technology is really moving on. We're using things like direct liquid cooling. It means that all the heat that normally comes out of computers, rather than cooling it down with fans and blowing air, which actually uses quite a lot of power itself. Ironically, rather than doing that, we use liquid, so we actually have water pipes that flow around the system and extract all of the heat. That way it's much more energy efficient and it also raises the opportunity we can actually capture that waste heat, it comes out as hot water. We actually use that to heat local buildings.

Interviewer (30:37)

Have you got a sense of how many you'll be able to power or heat using the output of this?

Simon McIntosh Smith (30:43)

Well, we've actually put the plumbing in already to try heating the building we're in right now. So we're in the National Composite Centre, the NCC in Bristol, and that's a university building. We haven't done this before, so we're going to try that as an experiment. If that works well, we'll see if we can start spreading out some of the local houses.

Interviewer (31:00)

In terms of where Britain stands, obviously in terms of its computing capacity, this supercomputer is now 11th on the top 500 list. But we're still as a country, quite far behind. I mean, you know, United States and China apart. Let's look at Europe. We're still behind countries in Europe with our super computing capacity, are we not?

Simon McIntosh Smith (31:23)

We have been falling behind and that's really been down to how much investment there's been in the past. And this really does help us catch up in one big way. But we need to keep doing this sort of thing. I know the government does have a plan with some of the things that have been announced in the last few months to start investing much more in this space in the future. So I think isn't bad, won't be on its own for long. There will be other things coming. One thing already announced in Edinburgh, there's actually also already a system called dawn in Cambridge, but I think there will be others and that will. If we keep doing that, we actually will be able to catch up and be where we should be.

Interviewer (31:57)

I don't know how competitive you are, but the fact that the Edinburgh University will take the number one slot in a few years time, does that bother you?

Simon McIntosh Smith (32:06)

Competition is good. I think we all push each other to do the best that we can. So it's nice to be number one for now. The most important thing though is what we all enable for the uk. If amazing science comes out of the Edinburgh machine just like it will out of Bristol, all of that would be good for all of us. So I'm very happy about that.

Interviewer (32:23)

The other thing that's quite striking in terms of these huge infrastructure projects that are being announced almost seemingly every week is the disparity between what govern the sovereign element to things and that that exists in the private sector. Does that bother you?

Simon McIntosh Smith (32:41)

I think sovereign AI is becoming a much bigger issue that we all have to care about, including national governments, international governments. And we're seeing much more work happening all around the world. So that for example, we could train large language models, the very latest cutting edge large language model models right here in the UK using all UK capabilities. So it could run it on Isabad AI. We'd know what data would go into it, you know, what software was run. And I think many governments around the world feel like they, they need to guarantee they retain access to this kind of cutting edge, really crucial capability. So I think sovereign AI is going to be a more and more important thing over the coming years. It's not going to go away anytime soon. Fortunately for us now isn't bad, actually puts us in a very strong position in the uk.

Interviewer (33:28)

And can you give people a sense of. Because obviously there have been some pilot projects in part of the system. Can you give people a sense of what's being done here?

Simon McIntosh Smith (33:35)

There's lots and lots of different interesting projects have been running on the system. There's one that's actually run in Bristol, which was very exciting, where they've been using AI to automatically analyze pictures of potential skin cancers and diagnose them and do it more accurately than humans can and faster than humans can. That project actually threw up a very interesting result where they were able to prove that some algorithms were better on fairer skin than darker skins. And they've now developed algorithms that now work better on all different colors of skin. So there've been some nice, nice developments there. We're seeing some very exciting projects in other parts of healthcare and medicine as well, where people are using AI, for example, to develop new vaccines and new drugs to treat all sorts of different conditions much more quickly than ever before. Actually, one of the big success stories from COVID was the COVID vaccines were developed way more quickly than we'd ever seen before. And that was largely because supercomputers were used to accelerate their development. We've also got some very fun projects where things like a British version of ChatGPT, which is called Brit LLM, developed at UCL in London. London. It's actually being trained based on UK data, so it's actually absorbing things like UK legal case law. So it'll actually be able to advise you much better on UK law as opposed to a chatgpt is naturally sort of trained on American case or. And things like that, and also knows British languages. So they've been training it on Welsh and Gaelic and other things like that. That's. That's a really nice development too.

Danny Fortson (35:07)

I have thoughts.

Mark Selman (35:09)

Go on.

Danny Fortson (35:10)

One which I just think I love is that we're measuring it in kettles. Kettle energy. 1300 kettles for our supercomputer.

Mark Selman (35:19)

Now I understand. Now I understand.

Danny Fortson (35:22)

It's just very. It couldn't be more British in that sense, which I just love. But I think it's really interesting that we've already swiftly moved to measuring the capacity of these things in terms of the energy they require. He was talking about 5 kilowatts. Mark Zuckerberg is talking about building 5 gigawatts folks at home. I think that would make this facility 1.1 millionth of what his plan is.

Mark Selman (35:46)

It's a lot of kettles.

Danny Fortson (35:47)

It's a lot of kettles. So many kettles. And there's just this hu out here in the States because the plans out here are so vast and things are moving so quickly. There is a huge amount of energy. In trying to find the energy, build the power plants to power all of these data centers. Like you may have seen last week, Google signed a deal, $3 billion deal or something like that to get energy from a set of dams in Pennsylvania Hydropower. You have Microsoft trying to revivify Three Mile island nuclear plant. Like these companies are going to real extremes because all of a sudden you realize that the limiting factor is not chips so much as electrons. Like you need the energy to power these things because they need so much energy.

Mark Selman (36:39)

Yeah, Nvidia was at this event because it's obviously a joint venture between essentially the government, Nvidia and hpe. Like Hewlett Packard and the Nvidia guy there. When I asked him about the constraint of power for the uk, he's like, yeah, it's a real issue because our Energy costs here are so much higher than anywhere else. He's like, the Nordics obviously have their own power sources, the French have nuclear. You lot are kind of. You're struggling. And it really does restrict how big and how much building you can do at data centers in the uk, if that's what you want. And the government has said, yes, that is what we want. So they're having to put these sort of AI growth zones in old sort of power plants, essentially.

Danny Fortson (37:31)

And is there calling it a supercomputer? Is that accurate? When again, when we're looking at Meta being like, hey, well, we're going to build something that's like a million times bigger or whatever, or, you know, you have Project Stargate from OpenAI, apparently they've not done very much so far, but they have grand plans to spend $500 billion on data centers, et cetera. When something is this small and is just being swamped on a daily basis by the latest data center being thrown up. A supercomputer, really just a marketing term for, yeah, we have a data center now that we can rent out to people for various interesting purposes, but it's just a data center and it's actually not that big a one.

Mark Selman (38:18)

Yeah, I think they all do different things, don't they? There are these sort of clusters or data centers or computers, whatever you want to call them, for doing, training the AI, and then there's some for just running the AI. And they've all got different architectures, but at the end of the day they all look the same. They're these racks of chips with wires going in between them and something to cool them. And I did ask that question of like, kind of feels like we're sort of minnows at 5,000 GPUs. The response was that, yeah, when you hear of something that's 100,000 GPUs, it's not all going to do one thing. It will often get chopped up for doing different tasks. It'll get rented out by different people to do different things. I'm not sure that that was a complete answer. Yeah, but look, it didn't get the top 10, so I don't know whether that counts. It doesn't get the supercomputer, the one they're going to build in Edit. I tell you what was also quite interesting is how quickly these things go out of date. I mean, this has a supposed shelf life of five years.

Danny Fortson (39:28)

What does that mean? The chips burn out, they melt.

Mark Selman (39:31)

It just becomes so much less efficient than the latest chipset.

Danny Fortson (39:37)

Yeah. Cause like Sam Altman, when he came on, he said, one of the things he said is the cost of compute has come down 100 fold in two years, for example. So 100 fold or 50 fold every year or whatever. Put out five years. I mean, that would leave this in the dust, as you say.

Mark Selman (39:53)

And I spoke to Mark Parsons from Edinburgh University, who's going to have in a couple of years the biggest supercomputer in the UK, which will get into the top 10, by the way. And he said, well, we will have Nvidia's chips, which are not Blackwell's, but the ones even further beyond that. And so they're anticipating being at that stage just in two years time. So it'll be interesting to see how quickly Isambard becomes a little bit less shiny than it is at the moment.

Danny Fortson (40:30)

Well, it sounds like maybe then supercomputer is almost like it denotes more purpose than capacity, because if it was capacity alone, then every big tech company in America would have multiple supercomputers. But it's a question of like, what are you using that computing power for? And it sounds like he's going to rent it out to researchers, to private industry, to kind of try to crack big problems, which feels different than just like we're just going to try to, you know, generate some creepy AI character who insults you at a, you know, the drop of a hat, for example.

Mark Selman (41:06)

What's a better use? I don't know. Should we do a poll? Should we do an Elon poll?

Danny Fortson (41:13)

But the question is, could I rent it out? What if I have an idea?

Mark Selman (41:17)

I think when I spoke to one of the researchers there, he said it really does democratize access to these GPUs because anyone can apply and you can actually start running your modeling, whatever you're going to do from your phone.

Danny Fortson (41:31)

Wow.

Mark Selman (41:32)

It's a remote access. So it really does open up some doors for some people who would never have had access to a gpu. So that is quite nice.

Danny Fortson (41:42)

That is cool.

Mark Selman (41:43)

Let's not mock that, shall we?

Danny Fortson (41:44)

No, no, no. And I think that is genuinely cool. And I think that's. I think that would actually be an amazing source of stories if you could just have like insight into like what this thing is being used for and also what the requests are, because you'd probably have completely off the wall ideas, you'd have some really niche things. But it sounds like they're going to have a whole universe of kind of really interesting things running on those chips. But yeah, number 11 today, tomorrow it could be number 100. But in a sense, who cares when it gets back to the point of just like putting this in the hands of more people, more organizations, more researchers.

Interviewer (42:19)

Yeah.

Mark Selman (42:20)

And we're sort of back in the game and we were at 47 or 49 on the list. It just shows you how quickly you can drop down.

Danny Fortson (42:27)

Right, right, right. Well, look, I think that is it a gallop through all things tech this week and I must say I've enjoyed it. Thank you, Mark, for coming on. I hope it wasn't too painful of an experience for you kind of standing in for Katie.

Mark Selman (42:43)

Well, it was pretty tough listening to you get abused as being pale and bald.

Danny Fortson (42:48)

But you know, I appreciate that. Thank you. Thank you for that sympathy.

Mark Selman (42:52)

I'll be professional all the way through. Thanks a lot.

Danny Fortson (42:56)

And we will be back next week. Still talking about AI, I am sure, but this time how it may affect the way world of law. Spoiler alert. We're going to need less lawyers. But until then, thank you for listening and we'll talk to you very soon. Thank you, Mark.

Sponsor Announcer (43:17)

Today's episode of the Times Tech Podcast was sponsored by Adobe Acrobat Studio.

Danny Fortson (43:22)

Acrobat Studio is about more than just the PDF.

Sponsor Announcer (43:25)

It's a collaborative workspace designed to make working with colleagues across teams and time zones faster and easier.

Danny Fortson (43:32)

One of the features is PDF Spaces, which brings more PDFs, Microsoft 365 files and websites together in shared spaces.

Sponsor Announcer (43:40)

It helps centralize notes and draft documents so you can see the full picture of a project all on a secure platform.

Danny Fortson (43:48)

There's also an AI Assistant which can summarize information, pull out key insights and help you understand what's in your files without reading everything manually, which helps if

Sponsor Announcer (43:57)

you're dealing with a lot of information but are short on time.

Danny Fortson (44:01)

And with integration with Adobe Express, you can turn documents into presentations or branded content without switching switching tools.

Sponsor Announcer (44:08)

So it's not about creating more paperwork, it's about making your paperwork work harder.

Danny Fortson (44:14)

Check out Adobe Acrobat Studio and its new features. Now.

Sponsor Announcer (44:18)

This episode of the Times Tech Podcast is sponsored by PwC, AI that delivers.

Danny Fortson (44:24)

It isn't just about the tech. It's a new way of thinking and it's empowering your people to create value in new ways. This is where PwC helps you shape AI to fit your business and delivering real value. Working with you, combining strategy, industry insight and deep tech expertise to help you innovate with impact, empower your people and accelerate growth.

Sponsor Announcer (44:47)

PwC knows how your business works so you can put AI to work for your business. Discover more at pwc.co.uk.com.