47. Trade Restrictions will Allow China to Out-Innovate US AI Companies – Tech Field Day Podcast

AI innovation and restrictions. Where are we at with ai? Is the future here yet, or are we still waiting for those real applications?

Join us in the Tech Field Day podcast. Welcome to the Tech Field Day podcast, where we bring together a group of IT technical experts to discuss a single idea about key concepts in the industry. This podcast features a variety of perspectives from members of the Tech Field Day delegate community, and is often recorded in association with one of our events.

Tech Field Day is part of the Futurum Group, and this podcast is also published on our sister company Site Techstrong tv. On this episode, we'll be discussing whether China will out innovate u US AI companies because of the trade restrictions imposed on it. But before we get to the discussion, let's meet who's on the panel today.

com. com and Disco Posse, uh, for all social media. Hi, I'm Justin Warren.

com. We are a boutique, uh, technology analyst firm and consulting, And of course, I'm Alistair Cook. I'm an event lead here at TE Field Day, and Cloud Field Day is, uh, one of my events.

And, uh, in that space of looking at cloud and ai, we're thinking about these restrictions on China trade restrictions that success of US governments are placed on China with the objective of, of curbing China's ability to use US developed technologies. This is not something new and I think we're gonna continue to see through the, uh, current, current, um, current administration in the United States. We'll see more of this use of trade restrictions and tariffs to control, uh, less friendly states and possibly more friendly states as well.

One of the things we definitely see is that adversity breeds innovation. And so there's kind of a bit of an unexpected side consequence that despite the fact that China has no access supposedly to high performance, modern GPUs, they're able to create, or a Chinese company was able to create a whole new large language model by standing on the shoulders of the US companies and maybe using some older technology to complete the innovation. I think it's quite telling that the restrictions have led to more innovation.

We see this all of the time. The more you restrict something, the more people find ways around those restrictions, and this is how you end up with real innovation. Innovation.

Uh, we see through the innovator's dilemma that continuing to do the thing that you are able to do in the past gets in the way of choosing to do something new. And so being forced to do something new leads to a lot of innovation. And Justin, this particular was, was a thought that had had come to the top of your mind.

What are you seeing in terms of what's being achieved by Chinese, uh, development and, uh, developers and Chinese companies, even though they're being restricted from access to all of this US innovation? Yeah, it's, it's pretty interesting, uh, given that I, I live in Southeast Asia, so, uh, reasonably close to China, uh, when there's certainly a major trading partner of Australia. So we are, we're in an interesting position, uh, relative to other countries in the world.

Um, I think that people underestimate China quite a lot, partly because China doesn't like information about itself to, to kind of leak out quite so much. Whereas, um, countries like the US are a little bit more open. But I, I think that that leads people to think that China is what is only what they see.

So China has a reputation for, uh, sending out cheap electronics and, and, you know, low grade plastics and, and so on. That is to misunderstand what China does internally compared to what it exports to the rest of the world. China is a high technology nation.

Um, they have extensive high speed rail. They have done a lot of, uh, fairly advanced manufacturing, uh, of, you know, for building and construction for, for example, we saw evidence of that during the COVID-19 pandemic where they were standing up hospitals in weeks. Uh, the kinds of things that are kind of unimaginable, uh, in other countries or, or certainly very difficult to do.

And part of that is that it's simply a very, very large company, uh, very large country with lots of people in it. So they have lots of resources to draw on, and that includes the, in the, the smartness of their people. So if you've got a billion people, you can probably find half a dozen who are quite good, um, in the same way that the US with 300 million people can field a pretty good basketball team.

So I think that what we're seeing from China is, is yes, we, we have these external constraints and China has for a very long time been trying to build up, its its internal capability so that it's less dependent on other countries. And what we are seeing here is basically the results of decades of effort in trying to learn from other people, but then be more self-sufficient as well. I think it's interesting to compare their approach with the recent deep seek stuff to what OpenAI has done, which is OpenAI has tried to continually just raise billions and billions of dollars and pour that into CapEx and bigger is better.

And because, like you said, China doesn't necessarily have access to the latest and greatest chip sets or the, the size and volume of the data centers, they have to make do with something a little bit smaller, which does force them to find innovative and new ways to train models and implement ai. However, at least as far as Deep Sea goes, it seems like they've released what they've done in an open source way, which almost removes that competitive advantage. Because if you're not keeping, if you're not keeping all this innovation under wraps, then you're just giving away the, the game to open AI and the rest of the companies that can adopt your techniques with the faster chips.

This is one thing that I, I was always stunned by when we say like, open source, is it a good business or a bad business model? And this is always as a technology and the rate of and pace of innovation in open source, it's generally seen that like, ah, open source means that anybody can contribute and that's different than just being open. So I believe that the openness of llama, you know, from, from the meta crew probably drove a fantastic amount of innovation in what they were able to put back into the foundational model.

And of course, they've got this incredible base of data with continuous use of their own platform, OpenAI, you know, or the one company that shouldn't have the word open in their name of course, but are they really materially different in the results that are generated from those foundation models because they're not open? And then when we get deep seek, there's of course a lot of controversy about where the training data came from. And really the truth is providing your weights and biases and providing the open source model does not necessarily really tell you like, this is gonna be better than other things.

'cause it may be open source, but is actually openly contributed and actively open source, which means that our everybody in this AI community feeding back into it where that flywheel effective innovation starts to happen. There's definitely a lot of that, the hidden, the hiddenness that goes on behind that. We, I almost believe that the open sourcing of it is a bit of a, you know, don't look behind the curtain, uh, because they're really what led to this moment of us being open is the part that you don't want to be open about.

And it's just a bit, we see this in the international business community, it's tough, you know, geographically we're all different politically, we're all very different. And so we're seeing this, uh, open enough, uh, to appear open, uh, is my opinion on, on what deep seek really is. And I'm a little worried about what got us to the point of being open.

Yeah, open source isn't a business model. Um, I, I've been researching and talking about this for a long time. Um, the open source is, is an approach to doing technology.

Um, it's, it's not the only one. What I, what I see in this particular deep sea is, is it's a competitive move in a similar fashion to, for example, Kubernetes. So releasing Kubernetes, uh, from, from Google, which was based on the internal system they use called Borg.

Uh, I, I think, and we broadly agree that this is a competitive move against some of the other cloud competitors that they have so that they can, uh, I guess strip out some cash from you. You've basically changed the market and the amount of money that is available to be spent by anybody. So if you have other advantages and you can deliver this at very low cost and with an extra bonus of getting other people to work on it for free, lowers your cost too.

That changes the entire market dynamics. And I see something similar here where we have, if we have an extremely low cost competitor, enter what is a really, really expensive industry, very, very high capital costs, and yet we have someone coming in saying, well, we can do this with very, very low capital costs. Um, and now by the way, anyone can do this with quite low capital costs.

Um, how are you all going to make money in this market? Exactly. And we don't really care so much about making money in that, particularly even maybe that area of the market.

We see that kind of competitive tactic all over the place. It's not just limited to technology and it's not just limited to this idea of proprietary versus open source. Um, I think we're seeing a real, um, rethink of what open source means compared to how you make money and seeing it as a tactic and not one that you always have to choose.

I think for a long time, actually probably the last 10 to 15 years, thanks to VC subsidies, a lot of companies have open sourced technology that they probably shouldn't have. And they were doing it just by copying other people without a real understanding of why are we doing it this way? Is there a benefit to us?

Is it a benefit to society generally? What is the purpose of doing it like this? I think that's a good thing that we actually think about this a bit more clearly and look at, okay, is open source really the right choice here?

Uh, or should we be using more proprietary technologies? You really hit this thing of like transparency and open are as a technology, open source is one thing, but transparency of the operations very different. Uh, that's very not open with deep seek.

And I, I think you've kind, Justin you've kind of described strategic open sourcing in a way that feels like weaponized open sourcing to me. It's that you, you use open sourcing of something and some parts as a way of attacking an adversary and, um, deep seek, uh, doing this, open sourcing some parts as an attack against us capitalists, uh, society and, uh, the, the massive investments. And is this an attack at those massive investments that people have made in their AI infrastructure in United States And Ned, do you think people have made good decisions about these massive investments?

Have they exposed themselves to a risk of a, an open sourcing attack here? Um, well the example of the open sourcing attack with Kubernetes is a little bit different because Google wasn't making money off of Borg and, and still doesn't make money off of Kubernetes. That's, that's not how they make money and how they make profits for, so for them to open source an existing project and then provide continual support for it didn't really impact their bottom line.

'cause they make money off of ads that that's how Google makes their money. I don't, there's not a corollary to that with deep seeq that I can see where they already have some way of making a ton of money and now by open sourcing they're, they're kicking out the legs of a potential competitor or a disruptor. Um, they're trying to be the disruptor to a certain degree, but I still don't see the path to profit for any of them.

Certainly if you haven't had to make the deep capital expenditures that OpenAI has made or, or Microsoft has made, you don't have to worry about paying that inve investment back to whoever is, is funding you. Um, what I think whoever who profits the most off of this is actually companies like Microsoft and Amazon that are not necessarily tied to a particular model. They have built up, uh, an infrastructure and a programming interface wherein you can work with multiple models and select the model that works for you.

And ul ultimately they're the aggregators in this situation, not open AI or deep seek. So if one model becomes too expensive or cumbersome, you can switch to another model and still continue to use Microsoft or AWS's or somebody else's interface to work with that model. So I would rather be an an aggregator in this situation than one of the model creators.

Yeah, we've, we've seen this dynamic in, in previous innovations as well. I mean, people like to talk about railroads is, is one thing, but uh, people, if you read the actual history of how railroads, um, happened, there was an explosion of people who were getting into this brand new technology. Um, it was really capital intensive.

They built out all of these, um, little custom railroad things and there were heaps of them around and loads of them went bankrupt, uh, because it, they just couldn't make it work. Um, and then of course it flipped to the other way and we ended up with a few monopolists, so who had, uh, total control over specific rail lines and we had to break them up. Um, so I, I think we have some possibly sim similar dynamics here in that the benefits of railroads are actually brilliant and flow on top of having the technology available.

So the fact that you can, you can make, for example, a railroad available very cheaply, uh, and broadly to everybody means that if there is value in having railroads, then everybody gets that and you can build value on top of those. I think what we're seeing here with, uh, someone coming out with what the large companies like OpenAI say, this is brilliant game changing technology that everybody should have. It's like, but oh my God, we've managed to do it really cheaply and that's bad for some reason.

I think that possibly, uh, illustrates where the motivations or incentives lie in this particular industry. Uh, and that's what I think we should really be focusing on is, look, is there actually benefit in this technology? What is it good at?

And if it's good at doing that, surely being able to do that cheaply and efficiently is a good thing. I wanna swing us back to the original premise a bit, which was around the restrictions on China as what led to this innovation that if there'd been no restrictions on what could be exported from the US to China, maybe they wouldn't have made this innovation. And in the same kind of vein, we look at what's happening in, in Russia, they've been sanctioned by most Western countries for the last three years, and yet the in particular weapons that they're building still contain products from Western economies.

And so these restrictions tend to lead to ways of getting around restrictions, whether it's innovation in new areas that are, are more cost effective because you don't have access to the cheaper things from a different market or whether it's ways of getting around those restrictions, still getting the cheaper things, still getting those, uh, high performance NVIDIA GPUs shipped to places that are on band lists. Does the sort of excessive restriction in control lead to more innovation on those controlled places? I, I think just as a general premise, restricting policing restrictions on how someone can approach a problem forces them to be more creative.

I think we can just say that unequivocally, uh, we've all experienced this when you know you didn't have the right part and you're trying to fix something in your house and you're like, well, I'll just rig something together and it turns out to work really well. I mean, but that's not every time I to go to Justin's point where you had all these failed mini railroads. Yeah, like 90% of the time your fix is not gonna be great.

But that 10% of the time when it actually leads to innovation or a new way of doing something, it's fantastic. And that's what we tend to remember is the successes, or at least those are the people who rise up to the top and you're like, oh, look how successful they were by innovating. So yeah, it will absolutely force innovation and that's a good thing.

I I feel like we have to remember how absolutely early on we are in this whole AI adoption process. I mean, just identifying the opportunities is we're still in the infancy of that. I mean, what open AI came out with chat GPT two years ago, two and a half years ago, and we're still trying to identify good ways to use that.

0. And I'm not even sure what version we're in now. So it takes a long time to find product market fit and we are nowhere near that with AI at this point.

So I think innovation probably is the name of the game and we have to remember how early on we are in this whole process. I think if you look at what the deep seek R one as a specific thing, we were latched onto this because of course we compare it against GPT-4 oh and, and and all these ideas of, of reasoning and that innovation that led to the feeling that we are actually approaching real true A GI style reasoning, like you said, go back to first principles. Why do we go to first principles?

'cause we say let's strip off every, everything, let's start with every constraint. If we had every constraint, how do we approach this? You go with first principles and because there's an embargo in place and there's lack of access to the same level of resources we have over here, I think it genuinely did cause them to go back to first principles in the same way that, you know, we look at how GRS first, uh, whatever they, they called the, you know, the, the giga super, you know, whatever cluster is that actually built in 120 days where even Nvidia is like, it'll take us about two years to do this ourselves, and they did it in three months.

So the that those just eliminating the feeling of restrictions is there. And I think that the Chinese technology, you know, they've, they're an innovative bunch, even if it's innovation by copy, they're very good at that and they've built a whole ecosystem and an economy around being under that. And then the other hidden factor is that we do not have the transparency into the actual capital spending that led to what they did.

So that $6 million is probably the same way that I say, yeah, I'm paying, you know, 30 bucks a month on, on Netflix. What I'm really paying is a lot more than that for the things that actually allow me to get that 30 bucks. So I think that's the, there's a lot of hidden numbers behind it, but truly I do agree that that innovation had to happen because they had no choice.

And when you're left with no choice, as Damon John says, you know, the power of broke, it's pretty inspiring when you're close to the zero. Yeah, necessity is the mother of invention as the saying goes. But I, I like to focus on the necessity bit.

Um, not everything is necessary. Um, like there are quite tight restrictions on having AC access to nuclear fission material because not everybody needs to have a nuclear generator in their house. It's not really necessary.

So people don't tend to innovate their way around those restrictions. Um, other things that, like the technology is there, we've had what blockchain for a very long time. Um, mostly blockchain is good for GIT and money laundering and crime.

It hasn't really taken off as a technology because it's, most people don't really see it as that necessary. I I think we're still in the early days of figuring out whether generative AI is actually necessary or how necessary it is and how often. Personally I think that it's useful in very, very small number of cases.

Um, and it's incredibly useful for, uh, generating spam email that I seem to receive lots more of, I dunno that we've actually seen enough use cases for it being something that people in their own lives determine being necessary. So I, I agree with, with both, um, Ned and Eric that yes, it's, it's very early days. We'll see whether this thing has any legs that are justified in, in doing it, but if it is worthwhile finding ways to do it faster and cheaper, that is pretty much how innovation works for every technology.

So you, you have a breakthrough and then you figure out, hey, can we get this to work at all? It's like, yes, okay, now it works. Now can we make it good?

Um, for some, in some things where you, you have, uh, really tight restrictions on things for, you mentioned Russia and, and a few other cases like the, the necessity there is very high and you can get away with something which is just functional, like barely works. Okay, maybe that's enough. But once you move past that, you move into things like, look, I I'm not satisfied with something which is just barely functional.

I want something which actually looks nice, which works well, which doesn't harm me as I'm actually using it. That gets me the result quicker and in less time. And that's, that's generally what we do After we've had the breakthrough, we figure out how to refine it.

And that's really what I'm, I'm looking for now is how are we gonna refine this technology? How are we gonna find a way to actually get it to give value to more people? And how do we reduce the cost?

How do we optimize its use so that the cost of using this to answer questions goes down and therefore the applicability to more questions comes down. This is a, a relative of givens paradoxes and the, the easier it is you make to UA resource to use, the more of it will be be used. Uh, we definitely need to head I think, towards that optimization where instead of requiring massive amounts of resource to complete, uh, some generative AI function, assuming a generative AI function is useful to you, we can apply it with a much smaller amount of resource.

Of course, there's also predictive ai, which is what I typically describe as being production ai, uh, the old fashioned machine learning, or as, um, Justin would say three, uh, linear regressions in a trench coat. Um, but it is, this is how most production AI has run for many years and it continues to be the, the generative AI part has some, some use, but, um, yeah, not, not as generally applicable I think as the, uh, as the predictive ai, right? I I want to know what's gonna happen in the future rather than, uh, generative AI largely tells me about what's happened in the past.

And while there is a linkage, um, it's not quite the same, I don't think, Yeah, we, we have yet to see what the pickax is for coal miners story is in the greatest successes. Like we're seeing a lot of them and I, and we're also not seeing a lot of 'em that are actually already existing around us, but they're not newsworthy necessarily or they haven't stood out enough. And it's, it's fine.

Like we're, in a weird way, we fight over the models of like, oh, which is the best model? 7%. Like we can't get faster.

The foundation models are similar. They, the foundation model itself, while being a core part of it is not as important as to all of the other things. They're gonna use that foundation model to really create that next layer innovation, which is sort of the the second, second order innovation.

And I think that's, as you say, Alistair, this is the stuff that we've yet to witness and we're not sure what it is. And Justin, you talked about that, like beyond generating more bloody spam, you know, have we, have we done something? I don't think anyone has said what they've done yet 'cause they're afraid to say that they're doing it, but I think there's actually a ton of innovation that's occurred that we are just, we're immersed in it so much that it's like asking a fish, how's the water?

It's like, ask anybody, how's generative AI affecting your infrastructure? Like it's it too late, I can't tell. Well, thank you all for joining us today on the Tech Field Day podcast, but before we can go, uh, where can people connect with each of you to continue this conversation and maybe learn a bit bit more about what you think around AI and innovation and regulation?

Um, where can we find you, Ned? com or I'm very active on LinkedIn, so connect me, connect with me there and send me a message. And I'm similar, uh, disco Posse all over the social media.

Uh, Eric Wright is harder to find than Disco Posse. Uh, LinkedIn is actually probably the best place to catch me these days. And, uh, yeah, thank you.

I'm a Canadian, uh, by birth, so I put the A in ai. There you go. Off to you Justin.

Oh, that's a terrible pun. I love it. com.

We have a weekly newsletter called The Crux, so feel free to subscribe to that if you want, uh, more of my thoughts injected directly into your brain. And of course, those thoughts are not AI generated. There's a natural intelligence in Justin's delivering that crux to me.

You can find me Alistair Cook Online, uh, also as Dez, you can find me also on, on LinkedIn fairly often and on a, a bunch of different parts of the Tech Field Day and Futureum world. Well, thank you for listening to this episode of the Tech Field Day podcast. If you enjoyed the discussion, please subscribe on YouTube or your favorite podcast application so you don't miss a single episode.

Do consider giving us a rating and a nice review. This podcast was brought to you by Tech Field Day, home of IT experts from across the enterprise and a part of the RUM group. com/podcast or view us on text on tv.

Thanks for listening and we'll see you next week.