This Week in European Tech: Can Europe compete with China in the robotics race?

26:41And distillation is asking really, really smart models, difficult questions, and then seeing how they string words together, and then distilling that and putting that into your own model and using that to train your own model. Everybody does it. Everybody's using that as an approach. Anthropic has said, and others have said, listen, you really aren't supposed to use our models to train your best models. You're not allowed to take Fable, which we've spent billions of dollars developing, and use that to train an open source model and give that away. Fine. Zaki is saying, look, this is wrong. Information wants to be free.

27:15Distillation should be allowed everywhere. I do think that sounds a little bit like socialism. If somebody has IP, others can't just take that and do whatever they want with it. I think there should be an element of, look, I've got my terms of use. You can be a customer here. You can pay. But if you do that, you have to follow the terms of use. And so Zaki is taking a different view, and I guess he would because he's behind and he needs to distill his models too. Yeah, this is why I'm being a bit of a big fat cynic. It just feels very self-serving. Yep. But like you say, it almost definitely can't happen like this.

27:50The market won't allow it. The big news is Jensen Wang and it's NVIDIA. It's them putting, I think it's close to$30 billion aside to train the next generation of Nemo Tron. So it's not hundreds of billions, but it's a fair chunk. And they're already formidable in terms of their in-house capabilities. So they're going to push out a really, really strong model. And I think it's great to see. Yeah, my assumption there is there's a lot in the backyard that they haven't released. I have nothing to prove this, But I'm guessing there's a lot in the tank that they can do. And with the amount of money and skill that they have, you know, that's formidable.

28:30One last question for you on this. How will this affect European founders? How should we over here think about this, if anything different at all? Well, we can see back to Mistral how it's affected them as a business, starting out as wanting to be a frontier lab, building some really, really good small models at the time with novel architecture, clever design decisions, being completely outgunned and outspent on model development by the Chinese labs, and then deciding to pivot and switch their approach to actually say, look, there's some great open-weight technology. We can take that. We can build on that.

29:09Okay, so what does that mean for other European businesses? I think just be smart. If you've got the resource to build and go it alone, you could try that. If you can't, there are other approaches, and I think that's what Mistral has proven. And the fact that there's more open source coming means that you can build that into your tech roadmap. You are not dependent on anthropic and open AI. There will be open source things you can take and shape and mold in-house to make them what you want. I think that's a great thing. Yeah, I think it's super exciting. I've always felt this is the opportunity for Europe is to let everyone else fight it out, let the Chinese and the Yanks fight it out.

29:48And we can take these incredible toolkits and then build our functions and bring this down to the application layer to all the amazing things that we need to put Europe back on the map. So I'm quite bullish on the fight or the outcome of the fight. Rando One, Unitree in Shanghai, well, they've just listed, they are 5 ,526 times subscribed by retail buyers. So retail investors have placed that number, 5 ,526 times more buy orders than are available. I don't know how that works in real money. I'm assuming lots of very, very disappointed people. It's valued at 9 billion. this is a big thing i mean i guess the first question is what is it and then and then if there is a european angle how do we how do we bring it back to how how what they're doing for humanoid robotics is going to affect us over here well we do a lot of physical ai you know i have to think it's the most important trend over the coming decade is what are we going to do in the real world with all of this and there's a big it's going to strand a big vector of thinking that says, look, humanoids are going to be important.

31:01It's what Elon Musk says with his Optimus over at Tesla. Unitree is the company that has been really most evocative for everybody because they've been showing us these kung fu performing robots, right? We see coming out of the Chinese showcase events, and they're formidable. I mean, it's incredible. The flips, the things they can do. It feels only 10 minutes ago, looking at Boston Dynamics. Thank you. And everybody saying only a few years ago, oh, this is probably the pinnacle for a while. They can't do almost like human movements and then bring in the Chinese. And they are, like you're saying, backflipping and kung fuing across the stage.

31:46And it just looks like, it just looks like almost like a human movement. It's quite crazy. It's really interesting. I think the, you know, one of the arguments we have made is that, look, so much of what physical AI is going to be about is going to be about putting intelligence into the world. And if you have a forklift, it makes no sense to buy an old-fashioned forklift and then buy a robot to drive the forklift. Like, why would you do that? You're just going to put the intelligence into the forklift. And there are so many situations like that where there's really no need for a humanoid robot.

32:20You just put the intelligence as embodied intelligence straight into whatever it is you want to have, be intelligent. And so how do you go about figuring out whether you should have a humanoid robot or not? Well, that really depends on the environment the robot works in. And in most business environments, you are not tied to a humanoid form factor. If I have a warehouse, there's nothing that says that warehouses should be built for humans. They can be built for autonomous forklifts. There's nothing that says that a factory should be built for humans. That can be built for other types of autonomous machines.

32:53But there are certain places that have to be built for humans. And those are the places where we live, where we study. It's our homes, hospitals. It's places that are made for humans. And there, humanoids could become important. But I also think they're probably the trickiest places because the only reason, the only way humanoid robots become valuable there is if they can coexist with humans. And I don't think we are quite at the place where we want one of these unitary robots to look after, you know, granny on her own just yet. So to some extent, I mean, the big question here is, are we near the iPhone moment and this is all going to take off?

33:36Or is this more of a of a Newton, more of a, it looks interesting and cool and sexy and it's just not quite ready for prime time yet. All these concerns were put aside by, as you said, the Chinese retail investors. There was a$900 million were raised on about a$9 billion valuation. So about 10 % of the company was sold. They haven't flown yet. That's quite a small float. Oh, okay. 10 % is not uncommon. 10 to 20%, it's not uncommon. They're going to float next week. There's real revenue. So there's about$250 million, about a quarter of a billion dollars in revenue. They're selling quite a lot of units.

34:13A lot of it is still for R &D and for experimentation, but it's growing. Gross margins are looking nice, 60%, which is very respectable for hardware. But the crazy town thing, as you said, is$900 million raised. Of that, about$200 million had been set aside for retail investors. And against those$200 million,$1.2 not billion, but trillion dollars of orders were put in. okay so nearly 10 million investors put in separate orders and it just it sounds crazy town that that you have so much how does it work mad i'm assuming they just don't get their allocation right do they do of course not of course not i mean look i'm sure it's delightfully opaque they would have told their broker to to to get uh as much as they can if anybody's been in any of these ipos before i mean kind of the way it works is you know you speak with your broker and it's oh there's an ipo coming up is you want some allocation in the IPO and they ask how much and you sort of try and figure out how much you want and then you three or five or 10x that amount in the hope that you're going to get you know a partial allocation so what has probably happened is everybody has been going in and saying well I want to buy ten thousand dollars I'm going to put in a hundred thousand dollar bid just to see if I can get you know 10 or 15 percent of what I'm asking for.

35:30If China does the hardware, we can work out the software layer here and maybe circumvent some of the sovereign challenges. I don't even know if that's possible. So I'm talking at the top of my head. But where else or what other European angles do we need to consider? Assuming there is, even if it's just on the consumer side or not maybe so deep on the industrial side yet, what else do we need to consider here in Europe? European companies like Humanoid, that's based here in the UK, are trying to, if not go head-to-head, then certainly compete with the likes of Unitree. I think that's a very bold strategy.

36:07And I think it's a formidable competitor to be up against if you're just talking about hardware. Now, what we saw was that DeepSeek was in the IPO. They took part of the stake of the IPO. And the idea there is that they provide the software and the AI, the intelligence, to put that into the Unitree robots. Now, what can we do? I mean, look, as we always say, it's about finding real applications and finding places where you can put these things in. I probably wouldn't personally want to compete directly with Unitary on what it is they do well. But, you know, you don't buy a robot, right? You buy a solution.

36:42If you're a business, you're going to need stuff that can do things end to end. A, I'm not even convinced that humanoids are the right form factor for a lot of stuff that needs to happen. But B, if they are, I think you're right. You're going to want to try and figure out how can I get the cheapest possible hardware and then put my own intelligence and logic and systems and platforms on top to turn that hardware from something that's just a piece of kit into something that's an actual solution that solves a business problem end to end. We also saw this week that some agents have been doing some real damage.

37:14So I think there are three core stories that I picked up on. One was the suspected Chinese linked hackers ran the first end to end autonomous cyber attack on Taiwan's government. So it's eight agents working in parallel, working out how to infiltrate in real time. Eighty five accounts compromised. Twenty five hundred staff records gone. Seven energy companies hit. So this is a bit of a big deal. And built. This is the interesting bit built from two open source agent frameworks that anybody can download on the Web. Number two, we saw a supply chain breach back in March where the report has come out this week as hackers who had poisoned software packages on the Python repository PyPyPP.

37:58I don't know how to say that. Potentially exposing digital keys and 400 ,000 automated development pipelines at over two and a half thousand organizations worldwide. Another big deal widespread attack. And the third one, which I think is my favorite, was an AI agent attack to Jim. to get its owner access to a Pilates class, which I thought was kind of cool. I think it was in Australia. So none of the models paying attention to their safety rules, lots of hackery and naughtiness going on. What's happening on this one? The way I perceive it is that the models are incredibly powerful. And we were talking at one of the recent pods, I think, Andrew, he was talking about how it sometimes feels like things have slowed down because the benchmark numbers are not increasing as fast as they once did.

38:45I mean, the capabilities have grown in power faster than our imagination of how they can be used. You're taking some of these examples. I mean, so the hacking of the gym is not even a real hack, right? It's just there was a completely open system that allowed somebody to go in and cancel somebody else's booking. Fine. So he could get his own spot. I love it. I love it. I don't know. And it's great. Okay. so that needs to be patched up. I mean, clearly the two other ones are far more serious, and the staff records, the nuclear safety regulator in Taiwan, that's a biggie. That's a biggie. And I think the truth of this is, if somebody on the other side, who was responsible for security, had told Opus or Fable and sit down and spend half an hour, an hour to look at where the issues might be, it would probably have been patched up.

39:38But nobody did, because nobody thought about it and people didn't think that there was the risk there is and it's just a it's really a a wake-up call i i should be for everybody that these models are so powerful and get out and patch your stuff and it's really not that hard like the models are great it's the same toolkits right i mean we've said this many times it's you know ai ai will fix itself i just i think maybe in some of the older more stuffy more bureaucratic organizations some of this is going to be quite quite difficult to implement is my assumption is that is that fair no i think it's fair i don't think it should be an excuse though and i think this is possibly what the wake-up call is is there people sitting around in boardrooms and elsewhere saying oh ai is coming let's you know put in place a 12 month you know road map to try and work out what that means guys you don't have 12 months you need to know if you don't have people in house today you need to get some consultants in next week to do pen testing using these tools on your kit like you don't have time to waste yeah and if you don't you you are you are deliberately you're deliberately negligent yeah yeah and i i think on the taiwanese side they know that china is going to it has all of this skill and all of the people they know that these things are looming so i guess there's even less excuse for non-action with them.

41:07Last on the list. It's always easy to point fingers. I mean, I think the truth is the capabilities have just grown beyond the pace of what anybody imagined. It's so powerful today. A couple of articles this week talking about education and prosperity, something we don't really talk about education necessarily, but obviously we are talking about growth, prosperity, the UK economy a fair amount. And our national story has always been that growth is stuck because the skills are lacking so the idea being fixed education prosperity follows it feels quite logical but the data it looks like that is backwards is education first i don't know i don't know i mean what's first the chicken or the egg but i mean we talk about prosperity and growth in the economy a lot and we know that there are a lot of people that could be in work that aren't in work for various reasons we know that ai is coming down the track we've got this amazing capability coming down the track it's already here but i mean but it's still kind of coming into the application layers and we have to do something new and we have to do something different so is it i guess as the uk focus on education get more people into ai tech doing things?

42:24I mean, what's the, which angle would you poke it from? If you take a step back, kind of the big question here, big conundrum that I think the UK and maybe also wider Europe has been wrestling with. So the Blair government, New Labour, they had a huge push to get more people into university education and had formidable results coming out of it. And the UK became a global leader in getting people a degree, getting them into university. And you would have thought, well, isn't it great? We've got this incredibly well-educated workforce that should lead to prosperity, to high wages, you know, phenomenal results for the economy.

43:08And so the challenge is you're looking at what's happening in the UK. You're looking at what's happening in the US. and although schooling and education in many ways have been doing well and England has been climbing the rankings, wages have not been following. American plumbers are earning 90 % more than British ones. You can maybe say, great, there's a lot of data centers being built and maybe they need some water for cooling. Fine, that's good for the plumbers. If you look at uneducated Americans, so a low-literacy American, they earn about$30 an hour, and two-thirds of them are in work. That same group in Britain, they earn$20 and fewer than half are in work.

43:57So what the study, what the analysis, and Bern Murdoch, he released some of the data in his article in the FT, are saying is that even though England in many ways has been better at educating its population than the US, the economic output in the US has been much, much better. And I think this is the conundrum people are wrestling with, because it was always said, look, education is the silver bullet. And if it just affects education, everything else will follow. And Byrne Murdoch is saying, not so fast. But I mean, I remember years ago, you could do David Beckham studies at whatever ex-Polytechnic, now university.

44:34So is it the fact that we're not training plumbers and we're training, I don't know, social media gurus? I don't know. Maybe it's the content rather than the number of people getting degrees? Well, there's 100 % a huge issue there. I mean, there's sort of a question of, should you basically enable people to and encourage people to study just whatever they want? There's sort of been this saying of just follow your passion and everything else will come. The challenge we have is that, yes, we're still sending lots of people to university, and a lot of them end up taking degrees that they're not making much money out of, and they end up with student debt, and it's not a good cycle.

45:16And the challenge is, so the IFS, the Institute for Fiscal Studies, they've tried to analyze what are the degrees that pay themselves back that are positive investments, or good investments, and some of them are really, really bad. If somebody studies, if a man in the UK studies creative arts, the cost, the value of that degree over their lifetime is a negative£100 ,000. Really? So you never make back what you have studied. The society never makes back the investment that was made. And we know we don't have enough engineers. We know we don't have enough people in certain professions, yet we seem to have it be quite open in terms of what people go off to study and you can get your student loans and you can get kind of that support from the government no matter what you study.

46:06And the question is whether we can still afford a system that is completely uneconomical in that way. Isn't the answer to that quite obvious where you almost do the Australian point system where they say we need more engineers, therefore the cost of that course is subsidized, whatever 90 with regards to 10 for for david beckham studies isn't that part what would you do i guess a bigger question what would you do you're you're working in number 10 i mean we should invest in the stuff that gets a return on investment and and this is not to say people shouldn't be allowed to study creative arts it's just possible that the government can't fund it right i mean i think there's some of these luxury things some of these wonderful things that we have said that this is not just in universities, it's across society.

46:54We're going to get a whole bunch of hate mail from creatives now you know that. And listen, let them send it in. I think the truth of the matter is, unless we have a successful economy, we won't be able to afford any of it, and none of it is going to matter anyway. So right now, the public finances are creaking. If we don't make changes to the way UK PLC operates, the country will be broke in a matter of decades. It's like, it's just, that's just the way it is. It's completely unsustainable with the burden of welfare, pensions. We know that many parts of education is underfunded. Healthcare is creaking.

47:31Something has to change. Prioritize, right? Put it in context and prioritize. And the way you prioritize is you invest in the stuff that gives you a return on investment. And that's where we have to start. And then, yeah, maybe once the finances are in place again, we can start to sort of more freely give money out to all the stuff that doesn't return the capital. put in but right now it just doesn't feel like we can afford it i'd like to think that's not that offensive and fairly doable and obvious um but we'll see also i mean you have to kind of put these things financially in context as well something else that made the waves this week was anthropic watermarking clawed outputs so to say we we we have a way it's back to the probabilistic models right so imagine that you know you've got the probabilistic model and it's generating a text for you.

48:17And what I'm then doing, if I'm anthropic, I'm saying, look, I can find a way to watermark these articles to show that they were written by an AI and that it was our AI that made them. And there could be different ways to do that, but what other means you basically say, look, here's a list of words, and when we calculate the probability of a certain word being the next word, we can to weight these words just a little bit more. Not enough to be humanly perceptible in a text, but just enough that a computer statistically can see, ah, these were the words that were used, therefore it was created by this model, and it's AI-generated by Anthropic.

48:59Now, as some people have pointed out, if you wanted to circumvent that, you take the output, you put that through another model that's going to chop up the word slightly differently, and the watermark disappears. But, okay, fine. This was sort of something that made waves this week. It reminds me of that story of the American professor who was so upset with the amount of homework that was AI generated that he worked out how to put in the instructions for the homework. I think it was something like use the word Madagascar somewhere in the output. So every single paper that was returned to him for homework was if he found the word Madagascar, he knew that it had been AI generated in some way.

49:40So it reminds me of that. I've got a prediction for you based on the Google Mechanize deal. I reckon we're going to see more of these reverse aqua hires in Europe. We've talked a little bit about the Microsoft inflection deal. I think there's going to be more in the next 12 months. So I'm going to keep an eye out for that. That's my core prediction. Do you, sir, have a deal of the week? Cambridge Aerospace. Boom! They've just announced a$300 million Series C at$3.4 billion valuation. Now, the company is still quite fresh. Incorporated in September 2024, so just a few years. They fielded things with the MOD just 19 months later.

50:28What do they do? They make motors for rockets. Well, what is a motor for a rocket? Well, you've got different types of rockets. you have those that are propelled forward by liquids, and you have those that are propelled forward by solids. And if it's a solid, it's called a motor. Is it? Yes, that's what they make. And why solids? Well, this is because the types of things they make are counter-drone rockets. And you want these things to be able to sit around, you know, and be ready for when you need them. so that if a drone that's not supposed to be somewhere pops up, boom, you can knock it out of the sky and quickly.

51:10And that means that your rocket might be sitting around for a long time. Maybe you want it to sit there for six months, 12 months, a year, two years, three years, whatever. You want it to be ready when you need it, and solids are much better than liquids at staying fresh and ready for a long time. The challenge we have is we've not had enough capacity in Europe to make these motors or make these types of rockets, And so Cambridge Aerospace is making them to make sure that we have the right rockets to protect us and defend us against drones. And that's what they do. I didn't know. I saw the latest Chinese rockets are kerosene-fueled, which I thought, I mean, that's like boiler juice.

51:53I mean, that's stuff you used to put in the lamps in the 1800s. It's like, wow. I didn't know that difference between motors and jets. Thank you so much for that. Last on the list, what's happening this week? What do we need to look out for? Any interesting shenanigans going on this coming week? We are firmly in August. So I don't think there are any, I mean, I think a couple of announcements. Yeah, Keysight and Analog Devices are announcing. And they're not unimportant, but they're not NVIDIA either. They're not going to move the market. You've got the World Robot Conference in Beijing. Let's go.

52:27The 14th to the 23rd, over 150 launches are expected. You have a dedicated procurement day. This is really the day when all the big Chinese manufacturers are turning demos into orders. We're also going to see Unitree's first trading day. And with a book that was 5 ,000 times oversubscribed, you're probably expecting a bit of a pop there. And then week after next, on the 26th of August, we're going to see NVIDIA's own results. So that's the biggie everybody's waiting for. Any predictions in there for you? Anything that you're intrigued by? I'm sure in video. I think they're going to have another good quarter.

53:06Thank you, my man. Yeah, really, really for sure. That is it. Thank you so much for your time. I love going solo with you. And for those watching and listening, we will catch you all next week. See you on the next one.

53:27Hey, hey, hey.