Kiwi co-founded world‑builder hits $2.5 billion valuation

WEBVTT - Kiwi co-founded world‑builder hits $2.5 billion valuation A young New Zealander is now at the center of one of the biggest bets being made on the future of artificial intelligence, with Odyssey AI's three hundred and ten million US dollar Series B capital raising round, putting doctor Jeff Hawk in the same fundraising league as the world's leading frontier labs. In this episode, you'll hear how a kid from Auckland who once worked on autonomous forklifts, ended up co founding a Palo Alto and London based AI lab now valued at US one point four or five billion dollars and is backed by the likes of natural Capital, Amazon, AMD, Nvidia and Google's chief scientist Jeff Dean.

I'm Peter Griffin, and this week on the Business off Tech, yet another tale of a really smart kiwi doing pioneering work in the world. I'm Peter Griffin and this week on the Business of Tech, yet another tale of a really smart kiwi doing pioneering work in the area of AI and raising serious capital in the process. Odyssey AI is not trying to build yet another large language model. Instead, Jeff Hawk, the company's co founder, is focused on world models AI systems that learn from site and sound to understand and simulate how the real world works over time, predicting what happens next and letting us interact with those simulated futures.

It's foundational tech for robotics and gaming today, but the same capability could end up underpending everything from healthcare and education to defense and scientific research. The scale of the bet On Odyssey is remarkable. This round was led by Natural Capital, with Amazon, AMD Ventures, GV, EQT, IQT and others coming in alongside high profile angels like Gary Tan, the CEO and president of y Combinator. AWS has also signed on as Odyssey's preferred cloud provider, with the lab set to lean heavily on aws's trainium chips and other accelerators to push the limits of these compute hungry world models.

Jess journey up to this point is a classic deep tech story, but with a distinctly global twist. He grew up in Auckland, studied mechatronics, engineering and computer science at the University of Auckland, then cut his teeth and a local startup building autonomous forklifts back in the mid two thousands, well before AI was a venture buzzword. From there, he moved into research in the US and UK, culminating innoctant at the Oxford Robotics Institute, and then several intense years at autonomous driving company Wave, where he was effectively the first technical hire and helped grow the company from zero to so a multi billion dollar valuation, alongside fellow Kiwi Alex Kendall, the co founder of Wave, that background in autonomous vehicles is really important because it's where many of the earliest world models were actually developed.

At Wave, DeepMind and Microsoft Research in Cambridge, researchers used vast amounts of driving data to train systems that could model how complex real world environments evolve over time, work that Jeff is now generalizing at Odyssey beyond roads and traffic to encompass robots, virtual environments, and much more. Jeff explains to me why he thinks world models are the missing piece of intelligence that complements language models, not competes with them. Odyssey is targeting two big initial markets, robotics and games and robotics.

Jeff argues that world models can help solve the open endedness problem, the insane diversity of real world in environments that makes it impossible to capture every situation a robot might encounter in data. Then there's the sheer amount of compute required to make this work, which is why Odissey lined up Amazon, AMD and Nvidia, the big chip players, as key partners and what Chef calls a very global race for AI infrastructure. He describes the lab's current models as roughly at a GPT two era in world model terms, already shown clear commercial potential, but with this new funding aim at getting to the chat GPT moment where the technology becomes widely deployable and demand explodes.

So stay with me for a fascinating chat with Odissey AI co founder Jeff Hawk about world models, the brutal realities of deep tech entrepreneurship, raising serious capital from some of the world's most sophisticated investors, and why he's bullish on New Zealand's role in the AI revolution. Jeff Hawk, Welcome to the Business of Tech, all the way from London. How are you doing? I defy well, thank you, thank you for having me.

What a week it's been for you, Jeff. I mean Odyssey founded in twenty twenty three, so in the scheme of things, not that old for a startup really has embodied I guess the whole rise of generative AI, that whole period. So we're going to get into exactly what Odyssey does and what attracted that amazing sort of capital from the likes of Amazon and others, a lot of capital three hundred and ten million US series b ray, so that's fabulous what that will enable you to do. I just want to dial back for people who haven't heard of you.

Some people may have heard of your friend and colleague Alex Kendall, who's another key We doing amazing things with Wave, which has raised well over a billion dollars in capital now, so done incredibly well and forged a relationship with Uber and Nvidia and others. So that's going very well. But tell us about your origin story back here in New Zealand. Sure, so.

I grew up in Auckland and did my undergrad there as well, studied mechatronics engineering and commuter science at the University of Auckland and then worked for a few years in a startup developing autonomous forkliffs in the mid two thousands, which was a definitely different world in terms of where I think the New Zealand venture ecosystem was at, but never lessa was where I cut my teeth and worked out what I like doing. From there, I basically ended up moving on to research.

Effectively, you could think of my background as being researched in deep learning and autonomous systems. So ended up moving to the US, did a research Marsters there and then worked a few startups, and then moved to the UK where I'm based now to do my doctorate, and then also end up having a very good five years building Wave from zero Toserato a few billion dollars. When you say it like that, it seems so easy, But man, what a what a tough journey that. It was not?

I can promise it was no. And it strikes me that you've been around this sort of the autonomous driving scene for quite a long time, you'd have seen some major changes there, and artificial intelligence obviously being at the heart of it. We've seen the likes of the Waimo, We've seen what Tesla's been doing as well as others, and then what yourself and Alex doing in a slightly different direction with Wave. But it must have been fascinating journey just seeing how quickly that technology has evolved.

Yeah, it's certainly made a big change from when I started my career it was very much sort of more sort of classical algorithmic methods for decision making. And then midway through my career and so maybe the mid twenty tens or early twenty tens where that flipped and effectively most of the methods that people were looking at became learned and they learned I mean essentially you're trying to observe data and try to work out what the right outcome is in a ater driven manner.

So that shift really happened in the twenty tens. These days, I would say it's probably consensus that most things should be learned or sort of AI, but that certainly wasn't true ten years ago. So yeah, it's been a pretty pretty pretty major sea change. So tell us about your relationship with Alex.

He was I think he studied a little bit after you at the University of Auckland and then he ended up at Cambridge. You went to Oxford and the Oxford Robotics Institute, which has a really great reputation. Yeah. Yeah, he was a years after meet Orchands, so we never everlapped there, but we sort of ended up in sort of maybe sister labs adjacent labs between Oxford and Cambridge and it's a relatively small world.

So I ended up joining Wavers essentially the first technical person after the two co founders, Alex being one of them, and ended up being responsible for leading ticket Wave and growing the company. So it was definitely. A thrown in the deepend in terms of working out hard to build a company. But I think I think we did okay together.

He did very well together. And you know what's really striking about both of you as you're part of co founder teams that have taken on capital and grown massively, taken on people. Any reflections on what you see and Alex it's made him a really effective leader and any traits you think that you sort of share if you're going on your founded journey that's become very much a high growth company I. Think is I think to be successful in a deep tech startup, which is effectively the type of startups that we're both building, you have to sort of navigate a balance of making sure that you are incredibly intellectually honest with yourself about and make sure you really understand what you're building.

It's not an area that you can sort of go in lightly without quite a lot of domain understanding bush massa heart because you also have to then be able to sort of tread this line and also selling where the future will go in order to grow, to bring in talent, to bring capital, and I alwast to say that incredibly well, I think you need a lot of resilience and grit. Inevitably, things won't work. There'll be something that drops out in terms of a deal that's being negotiated. There'll be something that doesn't work.

Research, I mean probably eighty percent of research at least doesn't work. So you spend a lot of the time being punched in the face metaphorically, and you need to get through that. You know, these kind of companies are not built in one or even two years. These companies are built in five or ten years.

Even five years would be incredibly quick. The thing I like about deep tech as a category is that there's a lot of sort of residual value, often the sort of the primary risk of taking on as something that is not possible today that you're working out how. To make possible in the future. That's exciting.

I think personally, a lot of I guess intrinsic motivation of the sense of discovery and also get from a business perspective, it means you're typically very well positioned in the future and that you have a very defensible position as a hopefully generational business. And it's exactly the area I think that we're seeing more focus on here in New Zealand. We had a reputation as being you know, the SaaS player you know was zero the end and sequence and others like that did very well with business to business software.

But increasingly more money is going to deep take for exactly those reasons. When you have deep domain expertise, that's where the real value is going to be. But it can be a long slog, it can be as you say, five to ten years capital intensive. But that round you've just done a strandred ten million dollars.

I mean that gives you serious capacity, including the relationship with Amazon to use their tranium chips and that sort of thing, to really accelerate your R and D and your product development and actually bring us to critical mass quickly. Yeah, so we're very grateful for the one coming together. We're grateful for having the backers of amazing investors. Our focus from here is basically using the capital to get ourselves from this point of I think of a sort of GPT two in terms of the error of where language models were that sort of early niscent models.

You're seeing a lot of potential. You can see whether they lead. You don't have to squint too hard to work out the commercial applications, but they're not yet ready for mass commercialization. And sort of what we're focused on from a capital perspective is working out how to get ourselves from where we are today, where the front year of the techs, to where we need to be for that point of mass adoption, looking for that sort of chatty beauty equivalent of explosive up tech and demand.

So you're a frontier AI lab, but it's worth pointing out that you're not doing the typical sort of large language model thing that Anthropic and open AI and Google are sort of in a race to outdo each other on correct you're focused on what's called world models. Maybe take us into exactly what that means. Yeah, so they're sort of if I take I sit back and I look at sort of the field of sort of AI, R and D as a whole, there's kind of two broad themes that. Are going on in parallel and where one of them.

So the first theme is might have heard of this idea of recursive AI development, So there's labs new labs like recursive Superintelligence. Ineffable probably fits this category. And these are companies that have raised sort of similar or even large amounts caple than we have, so that they're purely focused in the domain of language, and that's great. The second majors category of major AI research bets at the moment is where we fit in world models.

So think of this as saying, I'll view if there was kind of this missing form of intelligence where language is great, but it's almost like a filter, right, it removes information in our lived experience of the world, and so you're never going to have quite the same degree of information or reasoning capability as if you were able to learn more more natively from the more fundamental senses. So the question is how do we learn from really cit and sound as a starting point, And importantly, you could think of this as learning from observation and interaction with the world.

So that's what we do. So these are foundational models, meaning that they affect many to many different industries. Like LMS, they know they're not just a single vertical they're designed to suit a broad category of many many industries. You could think of them as being trained to be a simulation of the world.

So you're effectively saying, well, I know something in terms of what I see about the world, what potential futures will happen from here. So if I show you sort of a video or an image, effectively, what we're doing is we're asking the model what comes next. And what you're doing is that each. Time you're asking it that question, you're saying you're giving it some information about a potential change so you can steer it.

So for example, it's not just sort of rolling forward of simulating what might happen, but you're steering it with influence, which is pretty compellent. So you can kind of think of this in the way what it outputs. Rather than outputting a stream of text, it outputs a stream of pixels you can interact with. You're dealing with objects, how they move, potentially, how you interact with them.

I just had a few keys who came back from China last week and they were on the podcast, and now we're talking about the robotics revolution that's going on in China. You know, the hardest thing really is to get that sort of lived experience that you can lay on top of a robot. So they're actually literally putting cameras on people walking around to try and get as much data, people walking through factories, through houses, so that they can actually realistically map sort of that onto a robot.

And you know, particularly for unstructured movement and that sort of thing. So I guess you're doing a similar thing for anything. It could be virtual world, it could be a gaming world. Also, I guess robotics is a key use case for what you have.

Absolutely, So games and robotics are the two main nuceries we focus on mostly. In the spirit of prioritization rather than anything else. There are a ton of other things so that that can be done with it, but we think they give us a sort of a good coverage while while we're developing the technology to touch on robotics specifically, think of this as like the eighteen years of learning how the world works before you might go and learn to drive from right, If thirty hours of driving instruction, that's the equivalent of your robot fundation model development.

But you're coming into that with eighteen years of user machine learning term pre training of working out how the world works. So that analogy polled. Maybe another one is imagine you're sort of going to trade school. You need to learn to be a I don't know, plumber something like that, but you're going into that with a whole lot of experience around how the world works.

So that category problem is what we bring to the table for robotics. So this is effectively the base model or base intelligence that the robot fundation model developers can build on to make increasingly capable robots. The specific problem that we solve is one of generality. So robotics has a problem of getting data, which you mentioned.

The challenge is getting diversity of data in many respects. You want to have all the crazy things if your robot encounters an elephant. How many examples of ecocentric data are there with elephants probably very few, And as a consequence, you need to have this sort of intelligence you bring to the table to work out how. To generalize more effectively.

So the research strongly supports using these world models as this initial free train backbone that the robots use. My bet is it's going to be the single biggest influence on robot performance in the next five years. Yeah, and we need those breakthroughs if we're going to have optimists and exactly and others, a humanoid robots in particular being useful and bringing this error of abundance that Elon Musk keeps talking about. This is exactly the problems that need to be solved.

I think I read somewhere about your technology being used to help people interact with sort of video streams. How does outfit? So you could think of this as sort of modeling a stream of pixels. This is true in robotics, is true in games or anything really that has a stream of pixels, so it could be advertising, education, anything.

So the way it works is basically the model has some information it keeps around. You can think of this as knowing sort of the recent history or local context, as well as knowing some information further in the past. Think of this as sort of global context. And basically what it's doing is this sort of simulating forward from that.

So effectively you can take take a video and effectively simulate forward what might happen given that base video. So as you see fairly general fundational technology. Absolutely, Yeah, and like large language models, you need high capacity computing to to develop these models, to train these models. Hence the relationship with AWS.

Yeah, computers are a very important part of any frontier lab stars for absolutely everyone. When we're very lucky to have the active involvement of. Free of the major ship chip developers today, Amazon and Obvious One, AMD and Nvidia. We view it strategically, so this is something where if you're proactive about getting both your training your own prints working well up front, it gives you the ability to sort of navigate the infrastructure build out.

The challenge we've seen is particularly given, but it's a very competitive space where you can get GPUs around the world for at a given point in time can vary and that's likely to be true for the coming years. So from our perspective, it's advantageous to have a positive relationship with all of these of all of the companies developing these new AI accelerators and make sure that how our tech works. Well with them. So you're based in London, You've got a team in London, You've also got a team in palle al To.

Yeah, correct, So half. The tech team is in Palato, half the team is in London, and we also be a third office in Zurich. This is quite a common spread and a lot of the sort of frontier labs where it's a sort of in large part sort of around concentration of aal. I guess the key thing I've spoken to Alex about that is getting people who understand this area of you know, we're robotics and AI meets real world environment.

There's a lot of people out of the universities and that that the likes Deep Mind that have done a lot in around large language models and transformer technology in that. But yeah, what's it been like getting the talent you need to grow business? It's definitely it's always a fight. I think the good thing or the advantage we have really is because we're a very obviously clear positioned is a new category of technology.

It's often top researchers like they gravitate to some of the hardest problems and where they think the big research opportunity is. And as I mentioned earlier, this is sort of one of the two main research pools at the moment, so that helps. I won't claim it's easy, it's inevitably stop. Talent is expensive, is often hard fought, but I think you need to focus on finding people who really want to solve this problem at this company and prioritize that.

That means they stay for a number of years, and you need that to be able to make the kind of research breakthroughs we were looking for. If you have high churn of your team, it makes it very difficult to actually make those advances. So yeah, we protest finding people who really want to solve the problem of foundation world models and make sure we're working with them. Interesting to see.

You know, you've got in q TEL on the cap table there, so you've got links into the US intelligence community and you know you're able to simulate rich interactive environments. So I guess we're talking about things like drone swarms and that sort of stuff. So yeah, interest in your views on the potential dual use of this technology and sort of how you manage that, you know, when it potentially could be used for simulating war zones and that sort of thing. Yeah, escifically, market we expect will become increasing importance as we grow.

We've seen this player in language models today. I think the thing that's really motivating it is in AI generally is people are seeing the capabilities of what what this can do and hence it expands beyond save your commercial industries and to introduce in government as well. I mean, I would be amazed if this was not the way that world models played out as well. And as a result, we want to make sure that we have the right connections and links to make sure we're able.

To work well with within across government. And I guess playing a key role in the safe deployment offered as well. You know, we've heard it so much about hallucinations and large language models and the power of them, things like mythos and fable and restrictions on who should be able to use those things. I guess is the exact same sort of tension is playing out with world models as well.

When you're simulating and allowing people to interact with the simulations of the real world, it's got to be really accurate. Yes, it does. It's definitely earlier as the technology than language models, so we are I mean, temporarily we're a couple of years behind, you know, where language models are to day in their sort of guess production capability, but aim is very fast and we're seeing this mature rapidly. So hence, within a year or two, I think we'll be in the same kind of spot as where language models are today in terms of and needing to take on work out how to navigate some of these challenges.

It's interesting. It's slightly different but very similar in philosophy to what Wave is doing, you know, with its approach to autonomous vehicles. Having spent all that time on autonomous vehicles yourself, I've sat in a Weimo a few times now, and it's amazing how quickly it becomes second nature to you when you're in San Francisco and suddenly you're flicking through your phone forgetting that no one is driving the car. So do you think we really are on the cusp, whether it's Tesla or Uber.

With Wave doing partnerships, we are going to see massive option around the world. Is that really the first way that millions and millions of people are going to interact with autonomous vehicles? Is it will be second nature soon for us getting in a car where there's no no one driving it. It's definitely coming.

I think the debate is always a round timing effectively the two macro bets are going on in the industryal macro strategies. Weimer is effectively going full driverst and then trying to scale out of cross markets and where you've kind of been the orthogonal bit of saying, well, actually let's go horizontal across all markets, then go move to driverless, and I think the jury is still owed exactly how that will shake out. It's an enormous market and I think the absolutely wily space for multiple major players, so I do expect within certainly directionally, this is absolutely correct.

The timing is always the thing that's often quite hard to gestion in deep tech. So I think it's a question of how quickly sort of the autonomy rollout can take place, add to level of performance that is actually useful, and then it'll be an interesting question of what point people want to give up maybe their private, private their owned vehicles. I personally love Cushing Way most I am a regular user in San Francisco as well. Annoyingly, they cut off freeway access so it's a little bit less useful for while they deal with a few technical issues there, but it's amazing.

It's an amazing technology, and I think it will be a pretty profound change or profound addition to how we think about transport. I like it, but it's certainly quite nice to better zone out and relax and basically be shuttle twet you need to go. There's that and you know there's a lot of structure to our transport networks and that so there's a lot of really which data and you can drive those routes millions of times, which these companies have done. When you have a robot wandering around a care robot or in a hospital maybe dispensing medications, as the guys in China saw recently, you know, so much more complexity involved there, which is I guess is what you know, the likes of Odyssey is really trying to tackle that randomness and chaos and unpredictable nature of what humans do or anything that moves around a physical environment.

Yeah. In AA research, people sometimes call this open endedness as a problem area of like how do you structure AI to deal with like anything in the world, And that's a very very very. Long list of potential things you might be able to deal with it. Interestingly, autonomous driving was where one of the early world model one of the labs where early world models were developed.

We've been the main one the two others that were developing them in parallel were deep Mind and mcsoft research in Cambridge, and I think we were very fortunate to have a lot of data because it's relatively easy to strap up sensors and computer on top of a car and drive it around because you know the car will work, so you can scale up data volume. It's expensive, its operationally complex, but you can do it. So we were fortunate of a lot of data and therefore could build these narrow, singular to main world models.

The challenge is basically exactly as you say, it's how do you extend it to all of the sort of amazing environments where your office will be different from mind, the sort of our flats or kitchens will be different. So, as a result, if you need robots or any kind of physical interaction with the world, it is a combinatorially complex problem where you get increasing diversity and it makes it very hard to generalize. So if you're then trying to get data from all these environments that is attached to the robot morphology that you care about, that's an incredibly difficult problem.

I mean, my personal view, you can probably tell where I'm spending my time. Is that Rather, the way you have to solve this is by this sort of broad foundation world model that you. Then bring into the table to help solve these problems. Do you see world models as sort of a step We've heard a lot about large language models getting better and better and better in scaling and scaling towards artificial general intelligence.

Are world models stepping stones on the way to that as well? Or are they more specialized sitting off to the side of large language models to say, well, I. Think it's a stepping stone. I think there's are complimentary with language models, not directly in competition.

I think in time they'll merge, certainly a few years out, and that we technically don't know how to do this yet, but directionally we don't consider ourselves an agi BET as a company, but I think there is a good chance we might accidentally find a good way there. Maybe it's a sort of a parallel. The percentage of your brain that is actually devoted to language is not actually that large. Something like forty percent of this is based onto your visual processing and your visual cortex.

Don't cut me on the exact numbers. I'd have to refresh my memory, but it's that kind of order of magnitude, and I think that's true, right, we've been focused in some intelligence and the domain of language as an industry. It's amazing, there's incredible things, but we're still there's this missing piece of intelligence that needs to be. Brought into the picture, and that's the problem where we're working to stop.

It's a really good analogy. What's the big sort of scientific or technical hurdle that sort of keeps you up at night that's really gonna unblock progress in this field. But there's lots of them. Probably yeah, that's a that's a beandounding you want to go.

So there's kind of two that are sort of the main ones that are certainly I think quite big, hard unsolved problems you still yet to be sold problems. One is multimodality, so that means in machine learning terms, this means effectively added not just pixels but also audio as well as pixels. It's a surprisingly hard problem, although I think that's one where we're making a really good progress on and we have early models that do this. It's a surprisingly hard problem to simulate forward audio and think of a site and sound jointly, where sometimes if you're predict predicting like a singular frame, think about this as like fifty milliseconds of audio.

You actually don't have that much information about audio to deal with, so you might only have like half for phoneme in terms of information to deal with. So it's quite a difficult problem to make work to high quality. I know we have audio that exists in sort of in AI today, but the way they structure it is it's often dealing with much longer sequences before you try to generate it, particularly if you're trying to simulate build a simulator with this, it's quite difficult. We're making good progress, so I'm very bullish on it.

The bigger long term problem, which is a bit out there, is what's called learning by experience. So it's the idea that this is maybe dipping into causal machine learning a little bit too much, but most machine learning works by predicting correlations or for observations of the world and world models. As a field, it actually came from model based reinforcement learning as a topic, and the whole idea behind that is you need both a learned environment model a world model, and you need an agent that is basically interacting with it, and these two pieces both play a part in it and there's very clear evidence of if you learn a world model.

You can think of this as learning the rules of the game or the rules of the environment, which means that you can then build a better agent, which there means that helps you then build a better world model. And the idea of the sort of learning through experience of your world model and agent is I think one of the bigger long term topics. But that's fairly a far out as a big topic in terms of. Science, but is certainly a wee wee off in commercialization.

What's the reason for the name Odyssey? Like many things, I think it's a product of the number of things we liked the name. The inspiration was two thousand and one Space Odyssey and Homer. I think in both cases it sort of describes this, you know, I mean a respix that describes a deep tech startup quite well.

Right, it's a long journey, right, This is a commitment for quite a long time. But you know, with a successful conclusion, we liked it. We found it so a big kind of aspirational and I guess a bit from more practical terms, we saw a route to getting the appropriate domain names and Twitter handles, which is something that does feature more often than you might think and startup company names. It's such a great story.

It really epitomizes that that journey through an incredible world that is at Odyssey exactly exploring worlds right, which we're going to see next month Christopher Nolan's version of it. That's going to be amazing. But Jeff, just to finish off, I just wrote about this in Business there this week, taking a potshot really at the government around our digital skills training. We're but lackluster, particularly around artificial intelligence.

But then I look at you know, and I listed all the names of yourself and Alex and the guys from Antioch and super Bass and Shane from Deep Mind. I mean, we're obviously doing something right that we're developing all of this great talent. Sure, you guys all went off to some of the best universities in the world, but there's clearly we're creating curious minds here in New Zealand and doing something right with our education base. I guess, you know, I would like to see more of that value and that innovation captured here so that we can grow it from New Zealand.

You think that's becoming more realistic as our capital base sort of improves, and you have Kiwi's who have done really well overseas maybe to invest time and their own resources back into the benches here in New Zealand. Yeah. I certainly think New Zealand is all the right ingredients for it, and I think increasingly the distance matters less. I mean, even speaking personally, as a company, we're a distributed company, and I think the reality is that most companies that go global, there's no way around us.

Every company that succeeds will be a global company. So I don't think from that perspective of New Zealand is at a disadvantage. I think the education base is really good, and I think having a broad pool of people who've seen and lived this, who have strong attachment to New Zealand. Speaking personally, I do I think it's good for overall, good for the country and what it could mean for the growth of the tech industry.

I do think seeing more examples of what is possible personally. Speaking personally, that was what I don't think I realized what it was like to build a company like Google until I actually moved away from New Zealand. And I think in many respects of it is that kind of experience of realizing what this sort of ceiling of ambition can look like as something that is the main thing that I think returnless called I don't know what returning founder is it can contribute. So yeah, I personally bullish on prospects in New Zealand overall.

And the more people we can get building companies and less speculating on the property market and more trying to build new productive businesses, I think they're better. And look, we've just had the space x ipo, which is essentially an AI company. Now we've got Nthropic and Open AI coming. There's just huge capital flowing into it.

Does it feel a little bit bubblesh to you or are you a true believer in the future of this industry and you know this capital will find, you know, a really really good outcome for people and for society. I think both can be true. I think it is as certainly as Frothy is certainly punching in terms of evaluations, SpaceX being the obvious one. In most cases, this.

Is backed up by revenue in a way that I don't think is, which is where I'm not sure i'd quite call it a bubble where the growth that we've seen and adoption of AI and the utility is you know, there's there's substance behind this, and I think that is one thing that is a bit different from maybe what we're seeing in previous sort of textise. So I'm very bullish on it. Overall. I think it's great that Opening Eye and Thropic are going public.

It's good it recycles capital through the ecosystem. It means more money flowing back to you know, that cash will be recycled and so back to growth funds, so well then invest in new near labs, maybe like Odyssey. So it's how the ecosystem works. So personally, I'm very bullish.

I think it's a good thing for society. It will definitely mean change. I mean, again speaking personally, the tech industry likes to disrupt itself first, and the change in software development from AI in the past six months has been material to the point where I mean maybe as an anecdote of one of my colleagues took a cou few months out on printer leave, came back and told me that he didn't think his job was the same anymore. And I think he's right.

It's a huge change, and so it will be change, But I think overall it's a good thing in a man of respects. It's sort of the human the human desire for discovery of building tools, and you know, improving our improving our society is something we've done for thousands of years. And I don't see how this is a needed It's the next hammer, the next. Wheel, exactly, and standing in the way of it doesn't make sense at all.

I guess you buy into this idea if the error of abundance that comes with AI, where you know that one extreme Sam Moltman, and that is saying, well, you'll only need to work if you really want to work, because the efficiency, if the robotics and the AI and the embodied AI and all of that will be so good it will create this surplus for society. Maybe I don't think I'm quite as much as sort of a futurist as he is in that respect, but I think it will certainly result in a structural changes in industries.

I mean, we're already seeing that to some extent, mostly for the better. Again, it doesn't mean that it sort of won't be sort of, it doesn't mean you only end up a sort of disruption or some sort of change that sometimes can be painful in certain parts, but cumulatively it usually results in improvements and better functioning companies with more ability. To go after things. I mean respects.

The big superpower of AI is democratization of intelligence, and that you have many previously problems that would have required a team of one hundred can realistically be tackled now with a team of five to ten. I think that's amazing superpower. I mean, it's great for smaller countries like New Zealand and in many respects. And sometimes there's an argument of is it sort of a more polarizing it more of an equalizer.

I think in time it will become more of an equalizer. I'm sure it will be a bit turbulent, but I think in time it will be. That's what That's what we'll land. What a time to be a startup firn around it as well an entrepreneur.

Jeff, it's what a journey you're on. But congratulations on the Rays. That's incredible, and thank you. They have all those esteemed sort of companies involved, and nice House in a small capacity involved as well.

So the key WE connection is there. So good luck for the journey ahead, and thanks so much for coming on the Business of Tech. Thank you very much, appreciate your timped. That's it for this episode off the Business of Tech.

Huge thanks to Jeff Hawk in a very busy week for taking time out fresh off Odyssey's three hundred and ten million dollar US rays, for walking us through the emerging world of simulation driven AI and to reflect on what it means for robotics, gaming, the future of work, and even for a small country like New Zealand. If you enjoyed the conversation, please follow or subscribe to the show, leave a rating or a review, and share the episode with someone who's trying to make sense to where AI is heading next.

You can find more of my coverage of AI policy and the business of technology in my weekly columns and newsletters at Business Desk, and we'll have links to Odyssey's latest research and that massive funding round announcement in the show notes. Thanks so much for listening. I'll catch you next time on the Business of Tech.