Bill Gates has been on a speaking tour. He's talking about potentially a billion people dead. He's talking about this being as dangerous as nuclear weapons. How do you think about that in that context?
First of all, leading with fear is not the right model for engaging anyone in what you're trying to do. I'll start on the cyber side. These models are very cyber capable.
You're listening to the Big Tech Show with Adrien Weckler, sponsored by Deel. Deel lets you hire, manage, pay, and equip anyone, anywhere. Visit deel.com/bigtech.
Hello and welcome to the Big Tech Show with me, Adrien Weckler. Now, I had the chance to sit down with the most powerful woman in Irish technology, Sarah Friar, who is from Tyrone and is the chief financial officer of OpenAI. We talked about doom. We talked about the economics of AI, whether it holds up. We talked about jobs in Ireland. There's 2.5 million people now in Ireland using ChatGPT and a whole lot more. This is how the interview went. So Ireland, jobs announced here. Yeah.
A lot of jobs, actually. First, before we get to that, usage in Ireland, do we have any update on that? There's a million here in Ireland. That figure, I think, is over a year old.
That figure is very old. So today we see about 2.5 million people, so more than double in a year, using it every single week, and that's about half the population of internet users here in Ireland. So we're super pleased. And it doesn't surprise me. Ireland's always been foot forward on technology. People are innately quite curious. The good thing is we see it used throughout Ireland as well. It's not just a Dublin phenomenon. We see farmers using it. We see small businesses using it. We had an event last night here in Dublin just for women, Ladies Who Launch, actually owned and operated businesses, and you know, we had examples of people doing healthcare-oriented things. People doing actually one of my favorites, Linda.ai, two wonderful Irish women. Their dad's a dentist, and so they named it actually after his assistant, Linda, because they kept saying, "We need to clone Linda." So they've created now an AI-driven dental, I don't even call it application, I guess, that helps with scheduling. So for example, they will say there's 600 people in this practice. They should be coming back to get their teeth checked every six months, but they don't. No receptionist can call 600 people, you know, and do that. And yet, Linda's doing it. So they saw something like 250 people said, "Oh, yeah, I need to do that." And so for a dentist, you see the schedule filling in. That's good for business.
And so, in terms of Ireland, we think of ourselves as a tech country. How do you think we're doing? Are we holding our own as a base to develop AI in European terms?
You are. We just put out an economic research report that's going to look at by country usage, and Ireland's right up there. Now, at the very top, it's interesting, you've seen countries like Latvia, Estonia, the Balkans. Nordic countries tend to be very foot forward, but Ireland's right up there in terms of both usage and I think willingness to embrace. You mentioned jobs a little bit earlier. We're investing here because we see a population that is highly educated, works really hard, and is tech curious. So today we have almost 200 people just here in our OpenAI office today. In July we announced another 250, and that was to continue to build out areas like go-to-market and so on. But the announcement we're making this week around the AI summit is another 200. And what's important to me is in that 200, we're really trying to drive more towards technical-oriented jobs. So as AI is progressing, you're starting to see this new job category emerge called FDEs, forward-deployed engineers. What are they? Those are folks who come inside a company and say, "Okay, tell me what's important to you." Like, what do, if you're an airline, are you trying to fix, for example, how you make money or how you save money, right? Are you trying to upsell your customers to maybe do more with you, or are you trying to make sure the airline's operations are more efficient? If you're a supermarket, maybe one of the things you want to do is increase the basket size of who's buying from you. That FDE will come in, they come in educated about your particular vertical, they come in educated about your business, but then they know how to go about getting AI deployed, not in a superficial way, but actually getting it into the guts of your operations. So those FDEs we think are going to be trained up through the university system here in Ireland. We're starting with University of Limerick, but we love all the universities.
That's the Collison University. They have a software program.
That's why, because there's that kind of genesis already. But I'm kind of interested in how do we do this across the nation. There's also those FDEs, we're going to focus them in two areas, which is chip design or chip building. Semiconductor chips, not the fried things. And cyber. Sorry, I couldn't help myself.
Well, that would be an interesting diversification.
Well, it would be, and it would go down quite well here in Ireland. But back to chips, chips. We already have quite a pretty vibrant semiconductor industry here. I think people are always pleasantly surprised when they see the scale. So we know we can get the expertise there. And that's, we just developed our own first-party chip at OpenAI that's going to help compress the cost to serve our customers. So that's an area of focus. And then cyber, we should have a big conversation about cyber, but cyber is an area that, again, Ireland has expertise on. So that's why we want to go after this first 200. It's just the beginning. It's just to get a number out to push against. I think we'll be back in a year with a bigger number and a bigger number.
I'm going to come back to you on cyber in a moment. But for all of this infrastructure, data centers, you have a lot of data center requirements in Europe. You've got Stargate in a couple of countries. Ireland, would you ever consider a data center here?
We absolutely, we'll put data centers around. We want to be close to our customers because more and more when you start serving AI and particularly now when you're starting to move to the frontier in places like voice, images, video, latency really starts to matter a lot. Our goal, remember, as a research lab, we are very broad-based. It's our core competitive advantage, I would say, even over other kind of, you know, deemed frontier labs. We're building the big large language models, but we're also frontier in images. Right? If you're in ChatGPT, you should get it to create an image for you. It's super fun. If you haven't done the 80s meme, do it. We're very good on voice. Like, I feel like our most recent voice model now is, it's like you and I having a conversation, right? You can interrupt me, you can steer me slightly. Easy back and forth. So you need to have those data centers next to the people using the product. What would we like about Ireland? I mean, number one, it's a place that has a lot of renewable energy in particular. I think you're the second largest in the world with wind farms, if I recall. You know, in addition, you tend to be quite tech-forward and quite tech-welcoming. And then I think the third thing is anytime we're building data centers, we lead with the community. I think you can get a lot of talk up here from, you know, the politicians, but the most important thing in any data center build is how do we engage the community? And the thing I learned from being the CEO of Nextdoor is never tell a community what they need. Ask them what they think they would like. So I'll give you a good example. Savannah, Georgia, back in the US, we just put in a, we just bought a big tract of land that's powered. We're going to start building data centers there. We asked them, we put $80 million into the community. We're giving them $1,000 back every single year right now in taxes. But we asked them what would they need. The first thing they asked for, free school lunches. And we said we can do that. And since we just funded the lunch program for kids that have food insecurity, we're seeing the uptake on the school lunches go up something like 30 to 50%.
So is that the kind of approach you might consider then for Ireland? You identify an area, you might consider a data center there, you talk to the community, similar incentives you'd consider?
Absolutely. I think in most places where we'll put down data center footprints, we'll do community investment. Right? We've done the same in Ohio. We have another big data center coming up there. One of the first things we did was give every kid, every teenager, I guess they don't like to be called kids anymore, access to our technology for free. So we gave a free Codex license to every student in the state of Ohio.
So, but and maybe free lunches, maybe $1,000. Yeah, whatever the...
Again, it goes back to, it's, you're trying to balance that line of it's truly investment in the community. Because we're going to get out of it people who come work in those data centers, real jobs. So things like electricians, HVAC, like, you know, true skills, trade skills. We'll get people that will run those over the long run, and we want the community to feel part of it.
Just need to ask about risks and security. I've seen you talk a lot about that recently. You yourself mentioned Astra scored 100% on a recent cyber vulnerability test. We've also seen stuff like, and this goes to the high-level stuff, like in Yemen, we've seen weapons being built with another model's code. Bill Gates has been on a speaking tour. He's talking about potentially a billion people dead. He's talking about this being as dangerous as nuclear weapons. How do you think about that in that context?
I think, look, first of all, leading with fear is not the right model for engaging anyone in what you're trying to do. I'll start on the cyber side. These models are very cyber capable. You're right on CyberBench, which is one of the benchmarks, we are at 100%. Chinese models are not that far behind, maybe three to six months behind. In fact, the GLM 5.3 model was the one that I looked at and was like, wow, this is very cyber capable. So if the offense has the tools, we need to make sure that the defenders have the tools. And so what you'll see coming in short order from OpenAI is not to sell you problems or fear, but to sell you solutions. So we actually learned, we effectively experimented on ourselves. We put Astra, that model, against our infrastructure. We looked for the vulnerabilities. We used our own Codex tools, which is our coding tools, to be able to patch at machine speed. And so we're creating a much more resilient, hardened environment for OpenAI. But that created a playbook that we can now talk to customers about, and we'll talk to customers around the world on that front.
So the idea there is that as maybe the models exploit or look for exploits or vulnerabilities, that machine learning can fix in real time.
Fix in real time. Yes. Because the machine speed is the thing. The minute that vulnerability is found, the likelihood that others have found it is also high. And so we can't afford to then take months or years to patch. Right? If you talk to most CISOs, they'll have a list of their, you know, their highest severity vulnerabilities, and they have a long list. And if you're on a board, ever you show up and they show you all of this, right? We do not have the benefit of time here. And so that's actually, it is very top of mind for me. On the one hand, it can feel like talking your own book, Sarah, right? Because you want to go sell this. But more importantly, I just want to make sure that our core mission-critical infrastructure is safe and secure. Right? I started my career in InfoSec, and one of the first things I experienced was the first set of real internet viruses that happened. It's funny Bill Gates would talk about it because, by the way, it's because Microsoft had a lot of holey code at the time. Yeah. And what did it actually mean? Like, get away from the techy speak. It meant that in real life, if you called 999, an ambulance didn't show up, or you couldn't take cash out of an ATM or cash point, you couldn't get on an airplane because the booking system went down. That's when real people feel it. I think last night, overnight in France, there was a ransomware attack against a supermarket where they turned up the temperature on freezers. So they effectively spoiled all of that food. That's very costly if you're in the supermarket business. So these are kind of, when you take tech into like how it feels to real people. So that's kind of number one on your point about, you know, how will we talk about it in the world. I also think if you lead with like, you know, a more negative point of view, you're not bringing people into where the world is going with AI. So to me, the best analogy is electricity, right? You were walking into people's houses, and originally they tried to sell electricity. There was a lot of fear, like you could get electrocuted.
