DeepSeek and you shall find? With Professor Neil Lawrence
33:35 Share

G'day America, it's Tony and Ryan from the Tony and Ryan Pine podcast from Down Under. This episode is sponsored by Boost Mobile, the newest 5G network in the country. These guys are no longer the prepaid wireless company you might remember. They've invested billions into building their own 5G towers across America, transforming the carrier into America's fourth major network alongside the other big dogs who

Yep, they're challenging the competitors by working harder and smarter like this amazing new network they've literally built. They have a blazing fast 5G and plans for all the latest devices. Visit your nearest Boost Mobile store or find them online at boostmobile.com. This episode of the Times Tech Podcast is sponsored by Vanta.

Let's talk about something that might be keeping you up at night. Cybersecurity. According to Vanta's latest State of Trust report, it's the number one concern for UK businesses. And that's where Vanta comes in. Whether you're a startup, growing fast or already established, Vanta can help you get ISO 27001 certified and more without the headaches.

And Vanta allows your company to centralize security workflows, complete questionnaires up to five times faster and proactively manage vendor risk to help your team not only get compliant, but stay compliant. So stop stressing over cybersecurity and start focusing on growing your business in 2025. Check out Vanta and let them handle the tough stuff. Head to vanta.com forward slash the times tech to learn more.

Because when it comes to your business, it's not just about keeping the lights on. It's about keeping everything secure.

Welcome back to the Times Tech Podcast with me, Danny Fortson, in the valley of very shaken Silicon Valley. And me, Katie Prescott, in the city, the city of London, where I'm feeling quite smug, Danny, actually, because not one, but two of my predictions in our New Year pod have come off. Talking about the AI advances in China and also changes in chips. And it's only the end of January.

As we record this pod, the sense of untouchability, if that's a word, that seemed to imbue the tech giants, the magnificent seven, driving the US stock market ever upwards, it seems to be unraveling. Yeah, so it was just over two years ago when ChatGPT was launched by OpenAI. And in that time, the company has grown to be worth $157 billion and rising. And within

I think it was easier when they were with a fang. Yeah.

Yeah. And it was only last week on the pod where we were discussing seeing the heads of many of these companies paraded right next to the president, President Trump. And, you know, competing promises were made to build up to $500 billion worth of infrastructure to build these AI data centers to kind of really build the plumbing of this new AI age so that we all arrive at artificial general intelligence, this kind of holy grail, this idea that

These machines will be better than the best human at any and every cognitive task. Only last week we were saying it was going to be the Gilded Age, the medieval court of the Broligarchy. And it might still be, but then...

Back then, a Sputnik moment as described by the tech venture capitalist Mark Andreessen. That was when the Russians pulled the rug from under the Americans and set in motion the space race, launching the first satellite. This time, the Chinese are coming. That's right. Indeed, on the face of it, a smallish company called

called Hongzhu DeepSeek Artificial Intelligence Basic Technology Research Company. Someone needs to work on the name. It's like chat GPT. Anyway, that's another story. So the company is headed by Liang Wenfeng, who's a hedge fund founder. And the Chinese media says his company comprises fewer than 140 people, most of whom are what the internet has proudly declared 100%.

homegrown talent from elite Chinese universities. - That's right, so that's DeepSeek. And on January 20th, they released DeepSeek R1. - And of course there've been many different version of these large language models, different AI bots to help with statistics, accounts and design,

But the key thing about this one was... What's the key thing, Katie? It worked. Oh. It was cheap. People used it, shared it. And in a blink of the eye, it was topping the Apple App Store charts. And... And... Crash, bang, wallop. Right. So just in one day, shares in NVIDIA, one of the largest, most valuable companies on the planet, fell 17%.

wiping a mere $593 billion from its market value. And for those keeping track at home, that is four open AIs worth of value vaporized in a matter of hours.

So and shares in another semiconductor company, Broadcom fell 17% as well. The Philadelphia Semiconductor Index was down 9.2%, the biggest drop since COVID. ChatGPT maker or backer, I should say, Microsoft down 2%, Alphabet down 4.2%, all because...

of this little new Chinese chatbot. And it's got a name. This crash, it's been called the DeepSeek dip. But was it a dip and a chance to buy some Nvidia shares more cheaply than before or a sign of something more fundamental about the whole business model around AI? I should say, though, I was just looking at some of the

graphs earlier, some of the share prices don't cry a river for NVIDIA shareholders. It's up over 15% in the past six months, over 100% in the past year. And Broadcom has been up 38% in the past six months. Yeah, and that's after the quote unquote dip, right? And there's a stat that's been circulating around Silicon Valley that something like 80%

of NVIDIA employees are millionaires just because what has happened with the stock price over the last two, three years. So I think they're okay. I think we don't need to worry about them. But we're going to use this whole episode to try and get to the bottom of all this. And there are two things to think of. One is what is DeepSeek able to do and how?

And alongside this is the suggestion that deep seeks creative engineering means it needs far less computing power. So we're going to be joined in just a moment by a remarkable guest to discuss just that. But first let's,

Let's discuss what is it? What is DeepSeek? So it started, as said, by Wen Feng, the hedge fund boss back in 2023, when he pivoted his business, his fund called High Flyer. And this will ring some bells to try and achieve artificial general intelligence. And

And its latest model on the 20th of January is just, it's really getting everyone very excited. I mean, I've just started playing with it over the last few days and it's a bit like open AI and it's a reasoning model so you can see it thinking. It sort of,

It shows it's working out, to use a kind of school child term, as it's coming up with its final answer. It seems to have the capabilities of some of the chatbots already on the market, but it's just been developed for a fraction of the cost. Yeah, and benchmark tests on the model show it may already be better than ChatGPT's 01 model.

And critically, it's open source, which means the source code, the weights that make it work are made freely available to anyone, for anyone to view, modify, distribute, build on top of. It's interesting though. I mean, as a user, just mucking around with it, I appreciate the capabilities and the benchmarks might be excellent.

But there are some things that it won't tell you. So I asked what the relationship was between the US and China, and it started producing an answer. It came up on the screen. I screen grabbed it. I said, what's up to Overtube? But then it finally spat out, it kind of erased what it had written on the screen and said, sorry, I'm not sure how to approach this type of question yet. Let's chat about math, coding, and logic problems instead. Safe. Safe stuff. Let's get back to the safe stuff. Let's go on with it.

We don't talk about sex or politics. Yeah, exactly. On this chatbot. I didn't try the former, but yeah, it was very, you know, kind of stick to the, stick to the maths and the coding and the logic.

Right. But Danny, what's really interesting is, and it's amazing timing, you've been spending a bit of time at OpenAI this week doing some interviews. Yeah. What is it like inside the AI mothership? What was the reaction to DeepSeek? So it depends on who you talk to, right? So I spent some time with some OpenAI executives talking about what they're up to, and it just happened to be on the same day as like the world was freaking out about DeepSeek. So it was good timing, just kind of by happenstance.

And the broad kind of takeaway was like, look, we're not surprised by any of this. They were putting a very brave face on it and saying, you know, if you go back to two years ago, look at ChatGPT 4. That was released a few months after the initial kind of big bang moment of GPT 3.5, ChatGPT, the kind of the initial launch of

The cost for a model that is as smart as that one two years ago, which today looks dumb in comparison, is a thousand times cheaper than it was then to develop. The point is this stuff is getting these models are getting so cheap so quickly. And I think open AI, I think they're a little...

Miffed is perhaps the word that like, look, we are spending hundreds of millions of dollars, billions of dollars to really be at the leading edge, to constantly push the envelope. And then you have somebody like Deep Seek who basically comes in, looks at everything they're doing and effectively reverse engineers a lot of the breakthroughs that they have made and done it better, faster, cheaper.

And then all of a sudden you have a competitor that is, you know, something like costs a 30th of the cost to run and maybe a 10th of the cost to train.

But they're also like, look, you know, this is kind of the cost of doing business when you're out in the lead like this. Others copy you, figure out ways to do what you're doing better, faster, cheaper. And that's exactly what's happened here. Did they talk to you at all about how they felt it being a Chinese company? And that just feels like a very, very sensitive time in Sino-US relations right now. It did. And no is the short answer. I think, you know,

Open AI, my sense of the place is it's very, very academic, or at least the people I talked to. I talked to a lot of people in research and whatnot. And they're very laser focused on how do we make the best, most interesting models. They just came out with this thing, Operator, which is like...

freeing the chat bot from their little search window to go out and do things on the internet for you, go, you know, book me a flight, go, you know, do some research for me, come back, give me some answers, whatever it may be. Um, those are all the, you know, so they're constantly thinking, you know, just like sprinting ahead and trying to kind of push the envelope. Um,

And so they didn't really talk about the fact that it's Chinese. I think they're really looking at the technical paper that DeepSeek put out and being like, well, if they can do that, obviously we can do that. And if we can kind of, we too can cut our running costs by, you know, 30X, what does that mean for us as a business? You know, there's bigger questions about, well, who's going to pay for OpenAI when you have this effectively free alternative, right?

that's an unanswered question. But they're, you know, they were putting a very brave face on it all. You talked about Silicon Valley being shaken at the top of the program. Does it still feel shaking to you? We're a few days into this. I know there's a moment where everyone kind of went, oh, still reeling? It's both. Like if you talk to startups, they are over the moon.

And it gets back to like one of my pet theories around AI will become like electricity. It's just going to be everywhere, ambient, very cheap. And it's going to power a whole new layer of stuff we can't imagine right now.

And this is just like a big leap forward toward that future. I was reading an interview with a startup investor and he's like, look, now if you have a Stripe account, you know, Stripe, the payment processor, which are super cheap. I mean, if you have a Stripe account, a deep seek account and, you know, three guys, you have a pretty powerful, potentially pretty powerful company. And so that kind of little tech is very excited. Big tech is,

is, depending on who you talk to, is either very excited or very freaked out. Well, I think we might have reached the end of our computing power on that one. Yes. Let's double click on that.

We're going to double click on this whole AI race and really get underneath the covers of it. You know what I mean? We're going to peel back the layers of the onion. I'm glad you said that before our guest is on the line. I'd be embarrassed. Embarrassed for such a notable individual to hear you using these things. I'm totally comfortable with double clicking. Our guest today is joining us from Cambridge University here in the UK, which is why I'm sure they don't use those sorts of phrases at Cambridge. Oh, clearly not. Clearly not.

His name is Professor Neil Lawrence. Are you ready for this? He's the inaugural DeepMind professor of machine learning at Cambridge. He's been working on machine learning models for over 20 years. He recently went back into academia after being director of machine learning at Amazon,

And he's a senior AI fellow at the Alan Turing Institute. And he also never pulls his punches, as you know, because I know you sat down with him at the recent Times Tech Summit. And that's why, as well as all of his many notable achievements, we've asked him to join us to unpick all of this. It'd be great.

Hi Neil, thanks for joining us. Thanks for having me Katie. Good to see you again. Before we start our conversation I want to play this clip of you speaking a year ago from what I imagine was very snowy Davos. I've never been, I watch my more important colleagues go year after year. I hear it's great fun. So let's listen to that.

We are in the 1925 of the jet era. I believe the probability that large compute is going to be-- is required to generate such high capability is approximately zero, right? And at the moment that someone works out how to build these things with--

with a much more reasonable compute and much less data, as Jan was talking about, is going to be like the dreadnought moment. If you don't know about the dreadnought moment, it's when the British Navy that had a policy of maintaining a navy twice the size of everyone else's built one battleship called the dreadnought that could sink any battleship afloat. And thereby, they had a navy size of one. And that set off a naval--

battle to between Germany that leads to the First World War, Battle of Jutland, because all of a sudden your advantage is gone. And there will be a dreadnought moment with these models. I always heard academics like having their own words quoted back to them. Well, it's embarrassing because I remember at the time I needed to say that that led to an arms race.

And an arms race totally failed me at the moment. That's why I say that led to a battle. But the point is there, right? The point's there, yeah. A dreadnought or a Sputnik? What is it? I think it's neither, actually. And one of the things I said is it's indicative of what I was saying is true, that there's plenty of room in there to improve how we're doing this. But this deep-seek innovation is a brilliant packaging.

of, of ideas that were known and are out there. My way of sort of thinking about it is what we, what we're seeing at the moment coming out of the U S is like these hot rods. They're like put in a V8 Hemi and let's see how it goes and damn the fuel efficiency. But, but what we're seeing from deep seek is like you open that hood and it's like a Mercedes AMG and you're like, wow, look at how they fitted that intercooler in there. The artistry

of what they've done from a hardware perspective is kind of mind blowing. I should say I'm not implementing these models at the moment. It's beyond where my group can be. I understand terms that are in there, but I can't sort of say exactly where it matches the state of the art. But when I read the paper, the thing that's singing through is a group of people that know how to operate these GPUs. There's two things I find fascinating about them.

One is the sort of organization and distribution of the load on the GPUs, which I think is showing a skillset which is just not present in the type of people we have in the UK. And I would argue probably a lot of the people, it's not a PhD machine learners skillset. This is someone who got gnarly with the hardware, someone who cares deeply about how these things connect together. In that car analogy, it's like...

Not only are they like just kind of moving engine parts around, they're kind of taking the pieces apart and reassembling them. Yeah. Is that? Yeah, yeah, absolutely. And they're doing it in a way that like the person who's watching the YouTube car channel is really like, wow. You know, there's an artistry to it. Right. In terms of the way that they're leveraging and understanding what these components can and can't do. So there's that. And then the other thing that we have no clues on, I'm not sure if it's been spoken about at all,

I'm mind blown how they must be operating organizationally. So there's two forms of organization. Organizationally of the data and the compute on the system, but organization of that team. Because anyone who's been in this space knows you get team sizes of about 35.

And you start to hit problems in terms of communication within the team distribution. You know, the Mythical Man month is the original piece on this. They're operating a team of like, it seems around 100 or so, or a couple of hundred, something around that. Which is just a bit too large that you've got communication overhead problems between the team. So there's the communication overhead problems on like the hardware side.

But then there's how they're operating organizationally, which I don't have any clue about and I've become immediately fascinated by because I'm thinking if I'm running this team, how is it that I'm managing these different skill sets that they clearly have, which is, first of all, state-of-the-art machine learning, understanding what to do. Second of all, how to deploy it operationally. And third of all, how to run experiments relatively efficiently online.

And these are some really interesting skill sets that are not the traditional someone did a PhD at Berkeley or under Yann LeCun or whatever else. They're skill sets that are emerging as a new profession in some sense. And just coming from the perspective out here in Silicon Valley where we're special snowflakes out here, we are leading the world toward the golden age of AGI. Don't worry, we got this, everybody.

When you think about what this potentially means, because obviously everybody's going to reverse engineer this and try to integrate it into how they are doing this and dramatically reduce costs and increase efficiency.

What do you think this means for AI and its development? Here we are in 2025, two years removed from the chat GPT moment. I think it's super interesting, but maybe there's a comparison. It's like when Detroit is realizing that Japanese cars are here to stay. There's a piece of that in it. It's not exactly that. I don't think it's Sputnik. Why is it not Sputnik? Because Sputnik is...

At that time, Eisenhower isn't that interested in space. There's no shared organization. It's not a priority.

the missile program is distributed across multiple different sections, the Army, the Air Force, so on and so forth. There's no shared drive to get into space. There's no Project Stargate announced by the president two days before. Yeah, so this is exactly the opposite. There's been a lot of thinking. There's been a Project Stargate announcement. There's been massive budgeting. So it's more like,

Collins, Armstrong and Aldrin are on the way to the moon, you know, in their command module. And like George Jetson just goes past overtaking them. Like going down to plant our flag. Like, wait a minute, there's already one here.

They build a Saturn V and then someone's like coming past like an assessment. Exactly. It's something a bit more like that, which is like, which I find absolutely fascinating because I think, you know, coming back to your point around AGI, I think what's going on is AGI vaporware.

What we're seeing from Silicon Valley is the same move played again and again that goes all the way back to Microsoft with Windows One. That vaporware pattern is, I don't think it's a conspiracy, it's just second nature to Silicon Valley. The fact that everything is being driven by PR and very large scale investments coming largely from the trade surplus that you get in the Middle East-

and that money is being bet on projects that are therefore by definition going to succeed. That pattern of operation, there's something we talked about, Danny, before. It's like, Sam Altman is not an AI expert, but we treat him as if he is. He's an expert in building large companies. Elon Musk, not an AI expert. Satya Nadella, not an AI expert.

But we treat their words as if they're AI experts, and that's feeding the AGI vaporware. What they talk about becomes the truth on Wall Street, this combination between Silicon Valley and Wall Street, and is increasing these valuations.

to this notion that these companies are going to dominate this market, which is a dystopian notion in itself. The idea that any one large company would dominate such an important critical market is troubling. Yeah, and I think we were talking before you got on, I think there's a dividing line between big tech and little tech. Little tech is over the moon about this because they're like, oh, we have access to effectively what is effectively going to be a free, very powerful model that is, you know,

costs 30 times less to run, that we can build any number of apps, services, products, whatever on top of. I mean, just to pause that, I mean, 6 million is still a lot. This is one of the weird things about the distortion field we're in. 6 million is still a lot to train a model. You know, it's now considered a little, but it's still a lot of money. It's like still out of reach of the academic ecosystem. And the reason I talk about the dreadnought moment is because what happened in this ecosystem is people

The diverse open innovation ecosystem came up with a lot of ideas and was industrialized to some extent. So a lot of academics were employed in Google, OpenAI, whatever else, and were allowed to continue engaging in academic work in a diverse way. And as soon as one of those models put its head above the parapet and started looking promising, well, then industry could do something that academia could never do, which is say, oh, double down on that. Yeah. That's double the money we put into it. It's double the data. It's double the compute.

Now, academia could never have done that. If that had happened in the pure academic ecosystem, there would have to have been a period of learning about why it's working better, some refinement of what to do and how to exploit that. Things would have been much slower, but you might have got the same progress maybe in five or six years' time, but with a developed understanding of what was working, and I would argue

likely a much more efficient model, much more efficient than DeepSeek is where I'm going. And that didn't happen because

Open AI and other large tech companies saw this as a, you know, they were raising money through PR. That's how they got funding, by press release. And by beating the state of the art, by doubling down on this, they can stay ahead. They can make the press headlines. They can continue their raise. And then that feeds the ecosystem. They go larger and larger. So there was no stopping to think, which you would normally expect in these types of development. There was like, here's the furrow. We're going to plow it.

And that failure to stop and think, given that we don't understand all the ways, you know, the theory behind this sort of stuff, is kind of like, in my mind, it's like doubling down on the newcomer steam engine, the early steam engine that was like, it's like, we've got so much coal, we don't have to care.

Right, right, right. And no one's stopping, think, should we have a separate condenser? Should we do this? Should we do that? Right? And the rewards have come with that. And then that's been the mantra. And that's also the moat. So all of those things have conspired together to create a narrative that I suspect isn't true. You know, I might be wrong, but it just feels incorrect. And so me calling out the dreadnought moment is that

There's probably an entirely different way of doing this that involves much more interaction, that is much more intelligent in how data is acquired. And that's kind of what this moment opens, right? Well, I think that what this shows is that even within the way that they've done it so far, that they've missed efficiencies. So even under what they were doing, even under the approach they're taking. So to extend the car analogy, it's kind of like what I'm sort of saying is, look, there's a different type of vehicle.

you know, maybe this is a steam car and actually there's an internal combustion engine or a jet car or whatever else that is available. You can get even more gnarly with the hardware. Yeah, but it actually involves going away and thinking about the engine totally different way. And, you know, Yan actually in that panel, I was on that panel with Yan LeCun, I refer to it in the clip. He talked about it as being we're in the propeller stage and there will be a jet age. Yeah, and Yan LeCun is the head of...

Meadows AI efforts. This episode of the Times Tech podcast is sponsored by Vanta. Let's talk about something that might be keeping you up at night. Cybersecurity. According to Vanta's latest state of trust report, it's the number one concern for UK businesses. And that's where Vanta comes in. Whether you're a startup, growing fast or already established, Vanta can help you get ISO 27001 certified and more without the headaches.

And Vanta allows your company to centralize security workflows, complete questionnaires up to five times faster and proactively manage vendor risk to help your team not only get compliant, but stay compliant. So stop stressing over cybersecurity and start focusing on growing your business in 2025. Check out Vanta and let them handle the tough stuff. Head to vanta.com forward slash the times tech to learn more.

Because when it comes to your business, it's not just about keeping the lights on. It's about keeping everything secure.

Get unlimited talk, text, and data for just $25 a month with Boost Mobile forever.

After 30 gigabytes, customers may experience slower speeds. Customers will pay $25 a month as long as they remain active on the Boost Unlimited plan.

Thumbtack presents the ins and outs of caring for your home. Out. Uncertainty. Self-doubt. Stressing about not knowing where to start. In. Plans and guides that make it easy to get home projects done. Out. Word art. Sorry, live-laugh lovers. In. Knowing what to do, when to do it, and who to hire. Start caring for your home with confidence. Download Thumbtack today.

You said before we click record, you didn't think that the UK could have created this. Why not? You know, initially when I first heard it, my initial feeling was...

Because of the mistakes we've made over the last two or three years that we missed out, that we could have been there actually. And the big mistake is going all in on safety, right? And undermining the open source ecosystem, which played directly into big tech's hands. Now, if you look at UK policy today, it's exactly the policy that they inherited.

It hasn't changed one bit. And even if you look at the opportunities plan, yes, it's a refinement of things that were in place when Sunak took over. What then happened was the rumor is that Sunak was actually shown a deepfake video of him saying something he'd never said. And that got him so worried that he basically, he did a total pivot on his way to the G7 saying it's going to be all about guardrails.

without actually understanding what UK policy was. And what's happened is that pivot has dissipated over time. But that pivot to safety, like I understand that from a policy perspective, but what did that mean for like folks like yourself on the coalface? Yeah, like in a kind of bunch of your students or people working for AI startups in Britain could create problems.

model like this for a fraction of the cost which just wasn't possible yeah or do they need to can we just like all right deep seek is here it's number one in the app store it's open source it's open source and i mean you mentioned earlier before we started recording you're doing this work with local governments and here we have you know you have andreessen calling this but nick moment but he also called it like an immense gift

that you have this very cheap, very efficient, very powerful model that anyone can use. And yes, it's Chinese, which is a part of the question. But like when you're talking about Keir Starmer and others talking about like this AI age and we're going to like integrate it right into the heart of government and extract all these efficiencies, this feels like a gift. But I don't know if one, if that's correct, and two, if the fact that it's from China

How much that matters? No, I think these are all the right questions. And the problems, you know, the reasons why the UK veered away from open source is still there. It was a geopolitical security issue. But, you know, a piece you didn't play from that clip is we played into the hands geopolitically of China by banning high performance compute systems.

Because it was going to, I mean, that's what I said a year ago. Well, this is just going to drive innovation. I mean, a colleague of mine was quoted in the Times as sort of pointing out that necessity is the mother of invention.

but constraints are the midwife. Yeah. And, you know, it's Henry Chevrolet. And I think that's very apt, right? So you put those constraints on. So why doesn't it happen? I think there's a more interesting question. Why doesn't it happen in France? Because in this moment that UK moved away from its strength area of open source as a result of this concerns about safety, valid concerns, but I think an error, France went all in. Macron went all in. He went all in because Yann LeCun was telling him that's what to do.

and had a significant influence on French policy. And thus you have Mistral. Now, so I think there's a more interesting question. We're not, is why didn't France do this? I find that question more interesting because it's clear to me why UK hasn't done this. And you've done the counterfactual, but the counterfactual is France in some sense, because they made very sensible decisions. Now, UK could have been further ahead of France because we have a stronger base in open source when these sort of issues came up.

But we then didn't go all in on. And, you know, the investments that they've made in France, investments in public AI are sort of in the hundreds of millions, right? They're way ahead of where the UK is with any investment. So I think the question is, why doesn't it happen in France? And I do think it comes down to this point I was sort of making earlier that having read the paper and looked at some of the details of what's going on, there's some really, you know, it's like...

why doesn't Britain build Mercedes? Or why doesn't any country build? The quality and the tightness of the engineering you need and the beauty of it. I think there's a lot in there that is beyond my even ability to appreciate. Yeah. I was at OpenAI this week and I was talking to one of their folks and I was like, what are you going to tell your kids? In an age where we're sprinting toward this idea where we're going to have machines that are better than the best human at any cognitive task.

And he was like, well, I don't have kids, but I have a brother who's 12 years younger than me and he's just getting out of high school and he's really into woodshop, woodworking. And he's like, that feels like something that's going to have some staying power. You know, in a world of AI, I'm doing a woodshop. I just like, you know, sharpen your tools because that's all you've got.

And that's what my book's about. The Atomic Human is saying, well, what's left? What can't you separate? What is the inflation proof? So, and actually, and this isn't in the book. Neil, do you have a woodshed? Do you have a woodshop? What's going to be your value add in this world? Well, here's...

An answer that I think is relevant to this sort of economic argument, right? One thing I would say is potentially anything that's measurable economically, quantifiable, and that's what economics is about, quantifying the value of human labor, right? Anything that's measurable to a first order approximation, I'd say, well, that's going to be replaceable. Because if we can measure it, we can probably watch the humans do it. We can get the data. I'm not saying it's replaceable next year, but within 10 years.

What that means is that, and the notion in the atomic humor is every time the machine does something that we used to think of uniquely us, it's cutting into us like Democritus's atom. So then the question is, where do you get if that's the stuff that can be done? Well, you get to the stuff that's not measurable. You get to the pieces of humanity that we find difficult to quantify. We get to the teacher that chooses to spend an extra five minutes with a child that's upset.

and finds out that child's being abused. We get to the bus driver who pauses to allow the elderly lady to cross the street rather than trying to finish their route because they're being measured in that way. We get to the nurse that spends the extra 10 minutes with your mother-in-law in hospital because it makes her comfortable.

And none of those things are measurable. And they're all at the heart of what it means to be human. I'm going to have to... I'm getting looks. You're getting looks. You're like, shut up, this rubbish. What is this thing about humans? We've gone off the rails a little bit, but it's great. Katie, did you have a wrap-up question for Neil? No, I didn't. I was just going to say thank you, but... You were just going to say, well, most of that's unusable. Great. Well, I'm going to...

you know, sharpen my tools in my woodshed and level up on that in hopes of, you know, I'm still stay relevant in this new world. Seek some deep stuff. Exactly. But for all of you, a reminder, any questions, thoughts, suggestions, send them to techpodatthetimes.co.uk. That's techpodatthetimes.co.uk. Cool. We'll see you next week.

This episode of the Times Tech Podcast is sponsored by Vanta. Let's talk about something that might be keeping you up at night, cybersecurity. According to Vanta's latest State of Trust report, it's the number one concern for UK businesses, and that's where Vanta comes in. Whether you're a startup, growing fast, or already established, Vanta can help you get ISO 27001 certified and more without the headaches.

And Vanta allows your company to centralize security workflows, complete questionnaires up to five times faster and proactively manage vendor risk to help your team not only get compliant, but stay compliant. So stop stressing over cybersecurity and start focusing on growing your business in 2025. Check out Vanta and let them handle the tough stuff. Head to vanta.com forward slash the times tech to learn more.

Because when it comes to your business, it's not just about keeping the lights on. It's about keeping everything secure. This episode is brought to you by Progressive Insurance. Do you ever think about switching insurance companies to see if you could save some cash? Progressive makes it easy to see if you could save when you bundle your home and auto policies. Try it at Progressive.com. Progressive Casualty Insurance Company and Affiliates. Potential savings will vary. Not available in all states.