TSD Summit Sessions: Artificial intelligence and the value of data with Jennifer Zhu Scott

Stop the World.

Everything you receive, everything, every piece of information that's been put in front of you has been highly tailored based on your likes and your dislikes. So we live in a world that's very edited in front of us.

Welcome to Stop the World, the ASPI podcast on security and international affairs. I'm David Rowe.

And I'm Olivia Nelson.

There's a bit going on in the world. That's why we use this podcast to take a breath, spend half an hour or so with expert guests to better understand the trends and foundations of major changes in defence, national security, technology, and foreign policy.

And this week we're pleased to bring you a conversation on artificial intelligence as part of our Sydney Dialogue Summit Session series.

Have I mentioned, Liv, that I'm keen on AI? We're just over a week away from the Sydney Dialogue and I'm a bit excited. So to get our listeners ready for the event, Alex Caples speaks with Jennifer Zhu Scott, an AI expert and founding partner of the firm InCapital.

They discuss the evolution and power of AI and how in the space of a decade we've gone from talking about how to keep governments out of cyberspace and technology to how we need governments to catch up and regulate for our safety.

That's right. Jen tells us how, thanks to rapid advances in technology, some big tech companies are more powerful than governments.

The real change, paradigm change, in my view, that's happening right now is that large tech companies are gaining nation-state kind of scale of power.

Alex and Jen also discuss the value of personal data, which is no longer limited to debates about privacy, but about who owns the data and who can profit from it.

They talk about personal data being used for profit by big companies and the need for individuals and corporations to regain control of their data. Both from an economics perspective and also to safeguard personal freedoms.

If you introduce this kind of economic value of individual data, people start realizing, okay, I actually want to get paid, therefore I have to manage and control my data. It's just a means for people to be more aware of the importance of individual data ownership.

Without any more from us, enjoy this podcast, which not only identifies the threats but takes you through some of the solutions, including the need for companies to build data strategies in the age of AI.

And it's over to Jen and Alex.

Well, welcome everybody. My name is Alex Caples. I'm the director of the Sydney Dialogue here at ASPI, and I'm with Jen Zhu Scott, founding partner of InCapital and a globally recognized expert on AI. Jen, pleasure to have you join us for the podcast series, the Sydney Dialogue Summit Sessions.

Thank you so much for having me, Alex. Great to be here.

Fabulous to have you. Thanks for giving us some of your time.

Sadly, you cannot join us in Sydney this year for the Sydney Dialogue, but fortunately we have this opportunity to talk about some of the things that are top of mind for you to today and to pick your brains on data and on AI, and I think on the changing nature perhaps of statehood and digital statehood versus nation states. So fascinating conversation to start to unpack, I think.

Let me start with data because you were really quick to recognize the value of data even before AI kind of had its breakout moment. And your TED Talk, which has obviously gone gangbusters since, You delivered that TED Talk in 2019, so you were kind of ahead of the curve, and you said that whoever owns the data owns the future. And in that, you made a really kind of convincing argument, to my mind at least, that people should be paid for the data that they produce and that essentially people and their data are a raw material that companies are using essentially to build profits.

The statistic that you quoted then was that 7 out of the top 10 most valuable companies in the world are tech companies that either directly generate profit from data or are empowered by data from the core. That's now sort of 5, 6 years on the advent of ChatGPT, the advent of various other GPTs and large language models. What does that stat look like now? Is that the same or have we moved on? Are we looking at perhaps even more companies that are generating profit from our data?

So 10 out of 10 most valuable companies that are all directly or indirectly profit from data. And for 10 years I've been saying that every company can become a data company because one day AI will become so cheap to use, so pervasive, that you have to build your data strategy. And I think that day is here. So even for traditional companies today, if you if you didn't have a data strategy and AI strategy, you should. And therefore, a lot of companies, even they come from traditional industry, they're very proactively building their AI capabilities so they don't become obsolete. So it will become reality that forget about top 10 companies, any company want to succeed, you have to have a data and AI strategy.

It's essentially core business now, really.

Well, it's a little bit like a spreadsheet, right? In 2000, do you wanna use a spreadsheet to manage your business or not? As simple as that.

Yeah, okay. That's very interesting because obviously also at that time you had talked about this idea that regulation, you know, always struggles to catch up. We know that basically much of the Sydney Dialogue is about the fact that tech moves faster than governments and governments are always in a position where they need to have to try and work out what's happening or what has happened and how to adjust to that in order to kind of preserve economy, prosperity, access, security, all of the things that governments are kind of meant to do under the social contract.

We've seen movement on regulation, but it's really been about privacy. It's really been about kind of making sure that people's personal data is protected online, individual kind of identity is protected online, less so about this idea of data ownership and you being able to control or make money from your own data. So are we any closer to the private data ownership model that you outlined back in the day?

Yes and no. Yes, in the way I think in terms of awareness. When I was, you know, first started writing about individual data ownership in 2017, nobody was talking about it. When I gave the TED Talk, it was a very strange and new idea. But today I think most of people realize how important it is to control their own data. Companies realize how important for them to control their data. Right now, you know, we, we are facing handful few companies building large language model. They are basically building a large language model based on the entire internet worth of data. Everything you ever show up online— image, voice, text, personal information, or your company's information— everything has been devoured by those large language models. And I really think, you know, the pricing in terms of, you know, pricing capability and also payment for individual data. That itself is not a mean to, to get people rich from the data you, you produce, but to introduce the economic model that allow individuals to see how important it is to control your own data. Because I think in the past 10 years, social media made general public very numb about privacy. Younger generation, they're putting everything online. You tell them privacy is very important, and they brought up in the world where every— everyone's oversharing. So for them to understand how important privacy, especially how deep individual data ownership is, actually serve as cornerstone for democracy and individual rights in an increasingly centralized, controlled by few large powerful tech corporates. It's hard for general public to understand that, but if you introduce this kind of economic value of individual data, people start realizing, okay, I actually want to get paid, therefore I have to manage and control my data. It's just a mean for people to be more aware of the importance of individual data ownership. I will argue that we have started to see this 2017, 2018, in the aftermath of Cambridge Analytica that people start to realize, even in the age of social media, the idea that's very essential for Western societies, which is free will, is an illusion. You don't really have your free will. Everything you receive, everything, every piece of information that's been put in front of you has been highly tailored based on your likes and your dislikes. So we live in a world that's very edited in front of us, and therefore I think In order to control that, to regain some of that freedom, you really, individual and corporates, really need to start to control your data. And also in terms of large language model, if you think about it, I think any of the corporates that start to subscribe OpenAI enterprise services will regret because, you know, why would I running my company with all the trade secrets, with all my customers' information, submit to a for-profit company, you know, their larger language model where my competitors will be able to access and use as well, right? So Samsung has already forbidden their programmers, engineers from uploading their codes to ChatGPT to check, as simple as that. So I think this kind of individual and corporate data solvency is more important than ever. You know, I did see this some years ago, but I think right now we are there and it's very pervasive in our everyday life.

Yeah, I mean, the emphasis to date has always been on that kind of, as you say, scraping the web and basically devouring absolutely everything in a fairly kind of indiscriminatory sort of way. The emphasis has been on quantity of data to train models and so on, without necessarily thinking about the quality of the data or even perhaps some of the proprietary aspects that attach to that. And that is that kind of corporate data that you're now talking about where people start to think about it as being proprietary.

Your Samsungs of the world have thought about that perhaps more than your individuals of the world. Social media has that strange kind of dichotomy I've always found out of social media, which is that yes, absolutely curated content and people are very, very conscious of, say if you're an influencer, very conscious of your brand, for example, and making sure that you protect your brand, but not necessarily seeing data in perhaps the same way.

So I guess my take on this is that, yes, corporates are starting to value that because they have that IP to protect or they have those secrets to protect. But perhaps we've not, beyond the Cambridge Analytica, the shock of people understanding the scale at which data is kind of scraped, not necessarily perhaps taking those lessons and applying them to their own individual lives or logins or Facebook, Insta, social media accounts. Is that your sense as well, or, or is there a groundswell out there?

Yeah, I think there are definitely— if I compare to when I gave the TED Talk 5, 6 years ago, the crowd, especially in the more technically fluent crowd, realize this is much larger than before. Also, I think in terms of the convergence between blockchain technology, which is very essential to realize the individual or corporate data sovereignty as part of the infrastructure. The technology is much more mature than 5, 6 years ago. 5, 6 years ago was still very theoretical, but now you can actually technically achieve some of those capabilities. So I do think that needle has been moved a little bit more compared to 2017, '18. Having said that, I do think right now we are in the rush to chasing the capability, dazzled, you know, completely bedazzled by the capability of what AI can do. And instead of thinking about why and why not, the how and what is much more seductive than the why and why not. And not only just individual, you know, in terms of be aware. I'm a parent, I have two daughters, and I talk a little bit about this in my TED Talk. I often see my friends— of course, naturally, understandably, you want to share, you know, information about— you want to brag about your kids, right? In the age of generative AI, do not post your kids' picture anywhere, you know, in public domain. And the stuff that people can do using AI, just use a little bit of image, is, um, scarily good. And I think, you know, this kind of having the mindset of Digital privacy is digital freedom, including especially for your kids, right? When they turn 18, they become adults, and if they decided to put themselves out there, that's their choice. And as, you know, underage children, parents have made— already made that decision, you know, put their information and pictures everywhere is irresponsible. So I think mindset like this need to be much more prominent in our society instead of everyone just being dazzled by, you know, what's the next version of ChatGPT?

Yeah, I mean, there will always be people, and particularly, I guess, in the innovation space, and we see this coming out of Silicon Valley and other, you know, where innovation in large language models is largely happening, people who will move fast and break things, who push that button just to kind of see what it does because they're very interested to understand the potentials and the limitations of the tech or the code that they've developed. For the model that they've developed. But what you are arguing there for, and I, I think you're not alone of course, is this sense of, well, yes, but what's success look like? What are we actually trying to get this model to do? What empowers or betters society, for example? And we're in catch-up mode. I mean, we're seeing governments now really working quite hard to understand AI's applications and its implications to leverage those opportunities, and no one wants to be left behind. But essentially, does that come with a sense that we're minimizing risks or maybe not even able to get our hands around some of the risks that attach to kind of this idea of exponential growth and just the next model, make it bigger, make it better, make it faster?

Yeah, I think there are a few things to unpack here. Number one is drop this kind of emperor complex. I think a lot of us have emperor complex. Something, something's scary, something's going wrong, we're always looking for someone high above and, you know, with that their power will be able to solve everything. In this case, regulation is very useful to set a floor. Regulation, as you have mentioned, by default is retrospective. Policymakers observe what's happened and then debate in Parliament for 8 months, and then finally spend 5 months to draft something. And by the time we introduce the bill, the world has already moved on. For me, I live and breathe everything that's happening in AI. I can't keep up with everything that's happening. It's breathtakingly fast. Everything's— the speed of innovation has been accelerated so much, even just in the past 18 months. You know, regulators try to put a cap on this, right? Regulators need to be more intelligent in terms of thinking about what are the floors I need to set in terms of for example, privacy, in terms of data ownership, in terms of large tech companies who shouldn't be able to just scraping, you know, every piece of information about your citizen, about your companies, and putting to their, their model, you know, what kind of guardrails we can put in there, right? So that's the first thing. Second thing, often in the policy world, in the kind of geopolitical crowd, the framework is always about Nation A versus Nation B, or, you know, what's EU doing, what's China doing, what's US doing. I think that's kind of a distracting framework. The real change, paradigm change, in my view, that's happening right now is that large tech companies are gaining nation-state kind of scale of power, and often sometimes is even, you know, much more powerful than the nation-state in many ways in, in the digital world. And they're kind of hiding behind this kind of good versus evil geopolitical framework chasing ahead. You know, it's happening both in U.S. and China side, for example, that tech leaders in both sides say, if we don't do it, China is going to do this. And China side is, if we don't do this, the U.S. is going to do this, and they are evil and they are dangerous. And I think that's really distraction. The real challenge here is that because of the world, the technology is going— we are living in the much more centralized digital world than it should be. And AI, especially generative AI that build a large language model, is way too centralized. Large language model, there are roughly less than probably 5 companies in the world can truly compete. And if you think about, for example, Elon Musk, you know, everything happened on Twitter is so important. The discourse happened on Twitter is so important. Every nation-state's head of state will have to give him a seat at the table because he runs the platform that a lot of your citizens are sharing their political views and being manipulated and, you know, receiving information however the algorithm decides them to see. So those kind of power is much bigger than any of the ideological framework that you know, can impact our world and our lives. So I think unpack that kind of old politics mindset to, to embrace this new digital world is also very important. So I think, you know, right now in terms of the AI world, what's missing is the alternative that's actually decentralized. You know, everybody's doing AI, it's very centralized. Yeah.

I mean, I guess that in part and parcel, heaps to unpack there, Jen, but you're right, this kind of digital state versus nation state idea that you've talked about previously, and I know is kind of front of mind for you at the moment, around vertical integration and hyperscaling and effectively 4 or 5 very large companies wielding the kind of power that nation-states wield. How do you, though, decouple that conversation? Because sovereignty is an idea that hangs around. I mean, inevitably we sort of see retreats into nationalism. You have multilateral arrangements and so on, but when when the going gets tough, immediately we saw post-COVID with supply chains and a global disruption, suddenly it became all about sovereign capability and sovereign supply chains and making sure that nations were resilient. And a lot of that kind of discourse that kind of came back, and it's enduring, and that is because this is how societies organize themselves, top-down from governments making those decisions and physical boundaries being kind of, for the most part, population boundaries.

When you're looking at a borderless kind of world where companies are wielding the kind of power that sovereign states wield, how then do you kind of square that circle?

Have you done any sort of thinking on the way in which you kind of force those two quite different worlds to connect?

Yeah, I think about this every day. You've done some thinking on it.

I think it's safe to say you've done some thinking on it because I was born and brought up in China, but I'm a global citizen.

My husband is Aussie. I married a Melbourne boy, and my kids are third culture children. And when people ask, where are you from, it's very difficult to give a very straight answer. My situation is not unique. A lot of people these days lead quite international life and to have connections beyond your border. So my thinking— I actually wrote an op-ed in FT a few days ago. You know, people people can search if they want to read. Obviously, the nation-state is not going to disappear, and the nation-state framework is very important, and, you know, for identity, for our society, you know, to protect the citizens and defend your border, etc. I don't think that's disappearing. What needs to emerge from that is to think about what transcends the border. If you think about AI as the most consequential technology mankind has ever come up with. One of, you know, I think genetic editing, etc., is also very consequential. But what AI could potentially solve, a lot of those challenges we're facing from climate change, from curing cancer, a lot of those problems, they are not Australian, they're not American, they're not Chinese, they're not European, they're just human. So when do we pause? And especially for people come from, you know, kind of policymaking background, have a platform to influence the political discourse of your nation-state, when have we kind of paused and think, hang on a second, we focus so much on, you know, you versus me, the versus us, what is our common ground here? You know, how can we find those challenges that we all share as mankind and collaborate and transcend that kind of ideological and national border. I think this is, you know, given the geopolitical climate today, what I'm saying right now probably sounds very naive and very idealistic. However, I insist to be idealistic because the truth of that we are all human and we all have the risk to get cancer. We're all gonna, you know, have future generations facing the challenge of the planet facing in terms of climate change. Because the geopolitical environment is challenging, that is why we have to make effort to find common ground, to collaborate on those issues. And I think once governments start to lead in the way that can transcend that ideological, typical cliché ideological framework, then the leadership is showing the world that there is a role in terms of governments, in terms of the, you know, regulation, because we are thinking about the total wellbeing of our species instead of, you know, what's within my border.

I mean, governments traditionally, I guess, perhaps often more short-sighted than we would like in the sense that they are beholden to, for want of a better word, their citizens, their stakeholders and so on. So they tend to be quite focused on making sure they're delivering outcomes and outputs for a very specific group of individuals. Even though intellectually, of course, we know that this is a borderless issue that needs to be dealt with. None of us can deal with this alone. It's exactly as you say, an incredibly difficult thing to get governments to think beyond, well, beyond their election terms very often, but also beyond their own borders.

And the sovereignty does kind of lend itself to this concept of competition.

It's going to be, I think, even more complicated, isn't it, Jen, by the idea that we could be moving beyond this idea of, you know, here's a group of companies that it's quite difficult to not control, but to understand, to ensure transparency, to make sure that the guardrails that they're imposing are the kind of guardrails we would like to see. We're in that state now, but we could be in the next sort of X number of years moving into a situation where actually it's the model itself, if we achieve kind of AGI, that's kind of almost the next step, isn't it? Where we're really, it's not even companies that potentially are controlling a model. It's a model that can teach itself.

Yeah. I think this is the moment I'll bring some hope to your colleagues. Please do. I think that the conversation about the fear about AGI, it's a distraction because right now With the current architecture in AI, we are very far from general intelligence. If you think about putting the simple comparison, you know, the human brain, we consume about 30 watts and we're processing about 100 trillion parameters at any given time. Right now I'm having conversation with you, I'm listening, you know, to your voice, I'm thinking about all these issues and data points, I'm absorbing the colors around myself, the types of data you and my brain and every human being's brain processing any given moment is infinite. And the most advanced model, large language model right now, so ChatGPT-4.0, they never announced how much they trained on, but it's been widely believed that it's about just close to a trillion parameters. And they consume roughly about half gigawatt to 1 gigawatt a month. So in terms of efficiency, there's no comparison. I think in terms of, you know, using large language model with this kind of very much brute force to achieve truly general intelligence is very difficult because you run into the bottleneck, which is energy, right? In terms of energy consumption. So the real hope here is actually much more efficient models. So this is a completely separate rabbit hole, but I will be very quick about this. Basically, the US regulation in terms of limit the advanced GPUs being sold to China market actually in a way pushed the global AI, generative AI development, you know, solve the efficiency problem from the China side. You know, the most efficient models right now are from China side because they have incentive to consume less data and use less compute power. And I think much more efficient model will have a chance for us to become more flexible intelligence rather than real general intelligence. Truly general intelligence like human being, be able to learn everything, is technically impossible at the current stage and also economically unnecessary. You don't need to have a robot that, you know, be able to understand how to play Go game and, you know, understand quantum physics and, you know, also be able to learn knitting.

Right.

So what's the point of that? So, so I think I wouldn't want to say that, obviously. But so, so I think the more likely scenario is a very advanced narrow intelligence that's relatively narrow but super advanced. Now we have no idea about, you know, how its intention, you know, where, how, how how it's formed. Even right now with the large language model, it's very much black box, right?

Yes.

And Topic most recently managed to map a little bit and bring a little bit of transparency and visibility in terms of which part of the data in larger language model actually made the difference in terms of the algorithm to understand or make sense. But mostly it's very much black box. That's why it's so data hungry. It wants to take as much as possible. That's not the direction where we will be able to get to AGI. AGI has to come from completely different architecture, fundamentally different, and somehow mimic the efficiency of human brain or animal brain. So I think from that point of view, don't worry about AGI. Worry about narrow intelligence that become more flexible in certain area. Worry about inappropriate applications in military. Worry about unethical combination of AI with genetic editing, for example, create this kind of artificial biological inequality between different parts of population. Those are much more real risks in the coming 10 years as opposed to AGI. If you pay attention to people who talk about, you know, the fear of AGI, often are the tech bros who in the same breath want to tell you, but if I do it, you will be safe. So, so I think it's a distraction.

That's a really interesting point of view and obviously one that we'll be canvassing at the dialogue, particularly really great point and one that is often overlooked, although increasingly just something we're going to have to look at square in the face, which is this idea of digital transition and energy transition needing to happen in parallel and effectively being intertwined conversations that very often have been had in different rooms by different people and needing to bring those two conversations together. So aside from my new and longish list of things to worry about that you've provided, which is useful, you're saying that we're probably moving towards that kind of bespoke data, local language models almost, designed around very specific parameters, and that those parameters or the things that they're intended to achieve may not be necessarily great, but we're not necessarily looking down the barrel of AGI in the next kind of 5 to 10 years. Yeah, perhaps ever. I shouldn't have put a time on that. Your view seems to be that AGI is just one of those things that's really going to be too difficult, too inefficient, too expensive to achieve.

The fear of AGI has been existing since the 1950s. Spanish.

So yeah, it's been just around the corner.

Yeah, so I think localized kind of agent is super important with a small model, especially when, you know, there are multiple players in the market started to work in this kind of a very data efficient, compute efficient model combined with the individual data ownership that you actually have AI that's truly private to you. I think that would be ideal. Whoever can crack that code will become much, much more powerful than any of those centralized companies. Centralized companies will keep telling you that because China is going to do it or US is going to do it, therefore we have to gain more power. And also because AGI is going to come to destroy humankind, so therefore I need to do this for the entire mankind. So don't be fooled by that, right?

So that I'm the safer of hands argument is maybe not one that we should be looking at or we should be looking at perhaps a bit more closely? You know, you live and breathe this world. Are there good examples of that kind of decentralized model that we should be perhaps paying more attention to?

Or, or 100%, I think watch China's smart EV. So obviously, you know, now China's EV is developing really fast. I was looking at some data that China spent 26 years to produce first 10 million smart EVs. The next 10 million, 70 months. So, so that's the scale. And in July, the data came out for the first time that new energy vehicles sales just surpassed the combustion engine vehicles. So those EVs are basically AI with wheels. And because, you know, ironically, China actually has very tough data privacy regulation, and it's very much formulated based on GDPR. And so companies in China, they are not as freely as, you know, for example, Meta or OpenAI, be able to just scrape the entire internet and have access to everything. Combining with this advanced GPU limit and combined with the smart EV development that's been so rapid, and I think the next very efficient, privacy-driven and processing local data, protecting individual data rights, that type of AI probably will show up in Chinese EVs.

Interesting.

I mean, that is a classic AI developed for a very specific purpose.

So that kind of model that you, you've talked about where it's really about AI for a specific outcome, but equally GDPR, this is because China's selling into these markets and needs to be compliant. Client, or is there another kind of reason?

Obviously, we know ironically China and the US and other nations are all struggling with this idea of how you control data and data privacy. They've got the same incentives or the same outcomes that they're after, but coming at it from different angles. China's control of data is often about social control as well. But GDPR, just to come back to that, is that because this is mindful of the market that your EVs are selling to, or is this because it's actually something that needs to exist within the Chinese ecosystem?

China's data privacy law came out unrelated to the EV development, and I think what we are observing right now is just a confluence of the rapid EV development plus the China domestic markets privacy data law, and combined with the GPU constraint and data constraint that as a result, I think the most interesting localized AI agent probably will show up in China's EEV in the coming 2-3 years. And what I'm seeing right now, I don't know if your audience visits China. I just came back from Chengdu and to, you know, visit— Chengdu is considered kind of second-tier city in China. There were millions cars on the street, but it's very quiet. The street is completely quiet because half of the cars are all EV, right? And you start to see blue sky in the city and the city's much, much cleaner as well. I think this movement is going to accelerate so much, but in terms of this kind of AI in the car, that's actually not such a confined use case because right now the human-machine interaction is very much between your smartphone and human. And smartphone human or your laptop with human is very much informational, right? You know, if you want to access some information, you want to process some information, it's not actionable. But when you have EV and when you have autonomous driving, that become machine-human interaction becomes actionable. And that combining with robots, you know, stay-at-home humanoids or not robots, and that's also actionable. So I think that's the direction we will see developing in the coming years instead of the AI just stay in your car and, you know, help you to close windows and turn up the aircon, etc. Not as simple as that. It's actually much more profound shift.

Interesting.

And that's a top-down kind of approach out of China, or is that something that's been developed at a regional level, state, city level?

In terms of the China ecosystem, often been portrayed as a very much a state-led. This does want to encourage, incentivize certain direction, or try to influence certain— or they also put a cap on certain direction as well. But it's a very capitalistic ecosystem. The competition is so cutthroat in China. So this kind of AI in EV, it's already existing in all the, you know, BYDs and NIO. It's just right now the The functionality is still a little bit basic, but because the development is so fast and the competition is so cutthroat, the market is growing so rapidly, and we will see acceleration of that kind of AI in the coming year, 2 years, 3 years will be rapidly changing and improving.

Is there a parallel outside China that we could point to?

And, and I suppose if there isn't, why isn't there?

Yeah, so outside of China, both in Europe, particularly in UK, also US, some of the companies have started to focus on privacy protecting from, you know, the company's DNA level. For example, You.com, which is a search engine that very much takes care of individuals' privacy. There's Xining AI, which is funded by Stability AI's former CEO and founder, Yimand. So He's working on decentralized AI. So I think this kind of local agent protecting individual data, protect your privacy, is a trend that is growing. It needs to grow a lot faster and a lot more. That requires individual's awareness, consumer's awareness, and consumer's preference to that kind of product offering rather than centralized model just focus on the capability rather than, you know, the construct of what kind of AI will be better for, you know, good digital society.

I do think that's a really interesting point, that kind of move away from that kind of default large language model or monopoly kind of offering. It requires users and consumers of products and services to be a little bit more thoughtful about moving to a local model wanting it, actually creating a market for that.

You do quite a lot of work around, I guess, what you would call tech for good. What are the projects that you are working on at the moment, or what's, what's kind of dear to your heart at this point?

So there are a couple of things. One is with efficiency, AI models that focus on efficiency consume less energy. I think, you know, right now we have so much excitement about large language models. And ChatGPT type of products. I predict that this bubble will burst quite soon, and large language models will become commoditized, and there are about 3 to 5 players in the world can truly play that game. And that kind of CapEx investment in a large language model will take decades to get back that kind of investment, you know, very low ROI at the moment. So I think Working on efficiency, focus on decentralization, and protect privacy. Those kind of companies and projects I'm working on. I'm also, you know, a few of us in the industry are very keen to form alliance in terms of promoting, providing alternative, which is decentralized AI rather than centralized AI. I think that's what the society, you know, going forward needs. We shouldn't put a cap on AI. A technology that's way too valuable, way too much potential, and it should be discovered. But we should be very mindful what kind of AI, what kind of architecture and construct we're putting forward in front of the world. And hopefully, you know, conversation like this will be able to move the needle in terms of the discourse in public sector, you know, how to think about those issues and not to be distracted by just a few distractions we mentioned earlier on in our conversation.

The nature of humans to be captured by the flashy and the exciting.

Couldn't agree more, Jen. Thank you so much for your time today. That's been a really interesting, certainly from my point of view, conversation and that move from perhaps the kind of things that we're thinking about where we're state-based, where we think about things in terms of ideologies, where we're looking at these big large language models and maybe we should be actually thinking about the decentralized approach, the bespoke kind of approach that is more about quality and less about quantity, and putting perhaps a little bit of an end goal on when we're developing models. What are we developing them for? Not just developing them to see what they might do.

Correct.

Jen, thanks so much.

We are sorry that you can't join us at the Sydney Dialogue, but we're so thankful for your time today. And have a good rest of your day.

Thank you very much. Have a great conference and hope I can make it in future. Thank you, Alex. Have a great day.

If you're interested in finding out more about the Sydney Dialogue, we've included a link to the TSD website in your episode notes.

And that's all we have time for today, folks. Thanks for listening.