S12 Bonus: The Human In the Loop Bottleneck: Moving Beyond Static Models to Unleash Autonomous, Recursive Self-Improving AI with Kunal Bhatia, Co-Founder & CEO of Hexo Labs

[SPEAKER_02]: So I think the next most important thing for ourselves from proving system is to solve problems that are its own bottleneck.

[SPEAKER_02]: And interestingly, solving problems towards energy and compute are probably the most important bottlenecks that AI systems face themselves today.

[SPEAKER_02]: And that's why some of the partnerships we're working on and we've been doing are with labs that work on energy and compute.

[SPEAKER_02]: So we work with quantum physicists, we work on new energy, technology, we work with scientists across these domains.

[SPEAKER_02]: getting the AI to solve problems in these domains is one of the important strategies we've taken towards actually building the technology as well.

[SPEAKER_02]: I'm Konal Bahatia, I'm the co-founder and CEO of Exo Labs.

[SPEAKER_03]: This is Code Story.

[SPEAKER_03]: a podcast bringing you interviews with tech visionaries.

[SPEAKER_03]: Six, six months moonlighting goes.

[SPEAKER_03]: It's the last and all of the backgrounds who share what it takes to change an industry.

[SPEAKER_00]: I don't exactly know what to do.

[SPEAKER_03]: It doesn't go as to get right.

[SPEAKER_01]: who built the teams that have their bad company is its team's help each other, which is proud of our team.

[SPEAKER_03]: Keeping scalability top of mind, all that infrastructure was up there.

[SPEAKER_03]: Yes, we've been fighting it as we grew up.

[SPEAKER_03]: Total waste of time.

[SPEAKER_03]: The stories you don't read in the headlines.

[SPEAKER_01]: It's not an easy thing to achieve.

[SPEAKER_03]: To get yourself a deficit of off, try to begin to ride the ups and downs of the start-up line.

[SPEAKER_01]: Need to really want it.

[SPEAKER_02]: Not just about technology.

[SPEAKER_02]: All this and more on code story.

[SPEAKER_03]: I'm your host, Noel Appart.

[SPEAKER_03]: And today, how Kunal Boutilla is focused on building self-improving AI.

[SPEAKER_03]: To the point of enabling AI to build things themselves.

[SPEAKER_03]: Kunal Batia is originally from India, but moved to the Bay Area to start building his company.

[SPEAKER_03]: He admits there's quite a contrast between the two places, but originally he was from Bangalore, which is like the Silicon Valley of India, so the professional transition felt familiar.

[SPEAKER_03]: He's worked in AI for 12 years, and is on his third company in the AI space.

[SPEAKER_03]: Outside of tech, he's married with a three-year-old daughter.

[SPEAKER_03]: He and his family love to go on hikes and be outdoors.

[SPEAKER_03]: Gnall and his team have been researching AI technologies within their current venture.

[SPEAKER_03]: In particular, they were focused on building self-improving AI.

[SPEAKER_03]: Beyond that, they started building and thinking about how to build the AI platform that builds all other technology.

[SPEAKER_03]: This is the creation story of Hexo Labs.

[SPEAKER_02]: So, we're a research lab coming out of stealth with this launch that we just made.

[SPEAKER_02]: Our focus is on building stealth of growing AI, and the interesting thing is, what is stealth of improving AI?

[SPEAKER_02]: We're trying to build the end game of the AI trend itself, right?

[SPEAKER_02]: Where is AI going to?

[SPEAKER_02]: It's basically going to a point where AI starts building itself.

[SPEAKER_02]: Now, how did we get here, right?

[SPEAKER_02]: So interestingly, me and my co-founder started hexo as a company, which is doing consulting services.

[SPEAKER_02]: We used to go out to hold bunch of customers asked them what kind of AID you want, built out.

[SPEAKER_02]: We would just build it for them.

[SPEAKER_02]: And in the process, we are customer pipeline grew so much and we weren't able to deliver at the same speed.

[SPEAKER_02]: And that's one of the questions I asked my co-founder is that, what if an AI could build AI, right?

[SPEAKER_02]: We're building all this AI for all our customers.

[SPEAKER_02]: What if an AI could build this itself?

[SPEAKER_02]: And that's where the genesis of this idea and direction came from.

[SPEAKER_02]: We've both been in the AI space for over a decade now.

[SPEAKER_02]: So I think we knew what the limitations of the technology are and what where it's going to.

[SPEAKER_02]: And I think now where we see the trend going as well, right?

[SPEAKER_02]: The bet is on it, on actually on the technology itself.

[SPEAKER_02]: Can we build the technology that builds every other technology in the future?

[SPEAKER_02]: Because it's like, we think of it like the last invention that humans are making.

[SPEAKER_02]: Because if AI can start building AI and building itself and it can self-improve to solve any problem, you actually have something that can outperform us on almost every task.

[SPEAKER_03]: Tell me what the NDP for hexo labs is then at.

[SPEAKER_03]: That the platform to build all the rest of the, all of the rest of the technologies, what does the NDP look like for you guys?

[SPEAKER_03]: How long they take to build and what sort of tools are you using to bring it to life?

[SPEAKER_02]: So we just launched our paper.

[SPEAKER_02]: It's called self-improving AICA.

[SPEAKER_02]: This is actually in a way an MVP.

[SPEAKER_02]: Because given that we're a research lab, we're not like most of the other enterprise AI companies where we have a product and we sell it right now.

[SPEAKER_02]: Right now we're more into core research.

[SPEAKER_02]: And we're doing research around how self-improving AI systems will work and how it will be built.

[SPEAKER_02]: So this is our first paper that's come on and we have a whole bunch of other work that's going on which we'll get released in the next few weeks, a few months.

[SPEAKER_02]: How does the MVP of this look, right?

[SPEAKER_02]: I think the bet right now is that if AI can start building better versions of itself, can it start solving hard problems that that's the goal, right?

[SPEAKER_02]: If it can solve any hard problem, then I think the challenge is not so much in what the product looks like it's more like how do we pick the right problems for it to solve.

[SPEAKER_02]: So we're already actually working with some of the top labs in the country and globally as well to help them solve some of these hard problems.

[SPEAKER_02]: So we work with a couple of labs at Stanford.

[SPEAKER_02]: We work with Lawrence Livermore.

[SPEAKER_02]: We work with UC Santa Barbara.

[SPEAKER_02]: We work with University of Oxford.

[SPEAKER_02]: We work with researchers in all of these places, right?

[SPEAKER_02]: And we're actually getting our agent to work on some of these extremely hard problems that some of the research has worked across.

[SPEAKER_02]: quantum technology is across energy, across particle physics, across even AI research and biomedical research as well.

[SPEAKER_02]: So interestingly, I think AI being able to solve some of these problems is I think the real impact or the MVP we look at.

[SPEAKER_03]: So you've got that MVP, which is the white paper, the paper that you published and where you're headed there.

[SPEAKER_03]: How are you planning to progress and mature this?

[SPEAKER_03]: What are the next steps now that you've released that?

[SPEAKER_03]: And I'm curious about to like in that question.

[SPEAKER_03]: I'm curious about where you would, or how you would go about building your roadmap for self-improving AI.

[SPEAKER_03]: And then also the AI that builds all the other technology.

[SPEAKER_03]: How do you take that forward?

[SPEAKER_03]: How do you decide what the next most important thing to do is?

[SPEAKER_02]: So I think the next most important thing for a self-sumpturing system is to solve problems that are its own bottleneck.

[SPEAKER_02]: And interestingly, solving problems towards energy and compute are probably the most important bottlenecks that AI systems face themselves today.

[SPEAKER_02]: And that's why some of the partnerships we're working on and we've been doing are with labs that work on energy and compute.

[SPEAKER_02]: So we work with quantum physicists we work on

[SPEAKER_02]: new energy, technology we work with scientists across these domains in some of these top universities and getting the AI to solve problems in these domains and these extremely bleeding edge frontier domains is one of the important strategies we've taken as towards

[SPEAKER_02]: actually building this technology as well.

[SPEAKER_02]: So the AI actually works like some of these ads work with our AI systems to solve their problems and inadvertently probably unlock new IP that could itself be monetized potentially in the future.

[SPEAKER_03]: I'm curious about teen gone, I'm curious about how you built your team and to do something like this you have to have the right people at the helm, I'm curious of what you look for in those people to indicate that they're the winning horses to join you.

[SPEAKER_02]: So we do have a whole bunch of researchers AI researchers on our team from multiple universities from Stanford, from UCSC, from Oxford, from the top universities and top labs where AI is being built and research.

[SPEAKER_02]: While we do look for technical expertise, we also look for some of the softer skills and some of the more nuances around how we hire around attitude or some of these people.

[SPEAKER_02]: Are they people who think they are willing to take bets?

[SPEAKER_02]: and have the grip to stick it out.

[SPEAKER_02]: I think that's one of the most important things I think in an early seed startup is not so much about, of course, technical expertise matters and the skills they're bringing on the table, but startups are a long journey, so are they in it for short term, slips are they in it for just the name that, okay, hey, I work in a startup, probably in it for the mission that you're in, right?

[SPEAKER_02]: I think that's some of the things we look for, and I think in this environment it's very hard to just do a couple of interviews and figure out if a person is a right fit.

[SPEAKER_02]: We actually do short-term projects like maybe two or three week projects to someone to actually assess if there's a mutual fit, right?

[SPEAKER_02]: So that's one of the ways we hire.

[SPEAKER_03]: I'm curious about how you have approached scalability here.

[SPEAKER_03]: I'm curious how stories around scalability and if... how you've approached this in the beginning, how you thought about it, and then also have there been interesting areas we can, or you expect you will have to fight this as you grow.

[SPEAKER_02]: at the moment, the bottleneck to scalability is actually access to compute.

[SPEAKER_02]: So we do have partners in some of the hypiscs who we are working with to gather more compute, because any kind of self-improving AI system needs a lot of compute.

[SPEAKER_02]: Just to run small experiments, they're probably used thousands of dollars of GPU expense.

[SPEAKER_02]: You need a lot of compute, right?

[SPEAKER_02]: And we have directed our AI systems to target extremely high value problems statements across some of these technology that I mentioned to you.

[SPEAKER_02]: Getting access to that compute is super important and being able to apply it.

[SPEAKER_02]: So I think the scalability challenge is more from the scalability of the backend systems, right?

[SPEAKER_02]: Not so much about the front end of the business at the moment, even on the front end, we actually have a community of about 800-plus researchers in the Bay Area already.

[SPEAKER_02]: We've been running something called a Frontier Research Club for the last couple of months where we have researchers come and present.

[SPEAKER_02]: Our pipeline of researchers we can work with are significantly high.

[SPEAKER_02]: At the moment, we are just prioritizing problems that AI should be solving for itself, for it to be able to really take off into a self-cooling system.

[SPEAKER_03]: So, obviously you're doing something very forward-thinking and I'm certain there's a lot to say here, but as you step out on them, balcony, you look across all that you've built thus far, all that you've accomplished thus far with hexo labs, what do you most proud of?

[SPEAKER_02]: I think our ability to stick it out through the early days.

[SPEAKER_02]: My co-founder and I met a little inorganically compared to most other startup relationships, startup founder relationships.

[SPEAKER_02]: Most of the stories you hear are founders who went to school together or work overtime because they're a previous company they worked for.

[SPEAKER_02]: My co-founder and I met through a VC fund.

[SPEAKER_02]: a VC fund introduced both of us and then ended up investing in our company and that's how we started in Garus the ground and it's been more than four years now we've been together and stuck it out through the early phase when we were people who didn't know each other as well but just we're aligned on some remission and on values with the kind of company we want a bit.

[SPEAKER_02]: From that point onwards to here, be able to shape it up into what we're doing is something I would say I'm pretty proud about.

[SPEAKER_02]: There would be a lot of these frameworks and templates that people have for how startups should be built, how co-founders should be made.

[SPEAKER_02]: I don't think those templates are really true, I think.

[SPEAKER_02]: You have a unique story and keep walking down your own path and overfitting or retrofitting templates onto your own story is something that would just frustrate you.

[SPEAKER_02]: as a founder.

[SPEAKER_02]: So I think just trusting that your story is your story, I think that's the main part.

[SPEAKER_02]: And I think that's what we really proud of.

[SPEAKER_02]: That was come this far now.

[SPEAKER_02]: No, we really need to take about what we're trying to do as well.

[SPEAKER_03]: I'm curious, can all of this early, right?

[SPEAKER_03]: But I'm curious if there's, if you can tell me maybe about a mistake that you and the team have made and how you responded to it.

[SPEAKER_02]: a mistake we made in the early days was listen to too many experts or people who didn't have enough skin in the game and who weren't in our position and had a lot of advice for us.

[SPEAKER_02]: I think one of the things that was that also took us into many different bugs which were dead ends was just listening

[SPEAKER_02]: At the end of the day, this journey is ours, the context is ours, the insights are ours, I think our ability to trust our own instincts is super important.

[SPEAKER_02]: And to just give you an example, there was a time I think when we were trying to very early on when we were doing our consulting services, we actually had a different thought on how we would like to build a research lab.

[SPEAKER_02]: And I think a whole bunch of people we spoke to shot it down saying the big labs would do this, the big labs would do that ABC would do this and to be honest nobody's done it and it's been like three years since then and we could have been the ones doing it and I think that's one of the things that when I looked back I think that was the biggest mistake we did was not trust ourselves and listen to others.

[SPEAKER_03]: This will be really fun, especially for you guys and this early days is really long-tuned.

[SPEAKER_03]: I'm curious about what the future looks like for HXO Labs for.

[SPEAKER_03]: What's your building, how you're approaching it?

[SPEAKER_03]: Your point of view, all those things, and where the industry is going, where's the future look like?

[SPEAKER_02]: I think the industry is moving towards a place where they're going to be a few big labs, which have access to super intelligence and self-proguring AI capabilities.

[SPEAKER_02]: And I think this is an extremely powerful technology.

[SPEAKER_02]: almost at the scale of same of nuclear, right?

[SPEAKER_02]: It's that powerful.

[SPEAKER_02]: And the transition that's happening in the AI space right now is not just a platform or technology shift, right?

[SPEAKER_02]: It's not like the transition between personal computers to smartphones or the cloud or one of these precious.

[SPEAKER_02]: a societal e-box shift that's potentially happening.

[SPEAKER_02]: On the scale of like how we move from being hunter-gatherers to farming society to industrial society and now what's the next society we're going to be, right?

[SPEAKER_02]: I think it's at that scale.

[SPEAKER_02]: most people are not comprehending what that actually means and how do you actually build the kind of mindsets and the frameworks to actually operate in such a traumatic change.

[SPEAKER_02]: And in that world and what's assuring in that world is basically some really advanced AI.

[SPEAKER_02]: which a few big labs are building and anything else you try to build on the way it just gets rolled over by them.

[SPEAKER_02]: I think a few weeks ago we saw the news that Lord Anthropi came up with blood design and Figma's shares just dropped.

[SPEAKER_02]: This was an anthropi going after Figma.

[SPEAKER_02]: It was an agent, Lord design agent which went after an entire industry.

[SPEAKER_02]: So what we are going to see in the next few years is entire industries get wiped out because the single agent can do all of the work that the entire industry does.

[SPEAKER_02]: So it's going to be a little hard path from a societal perspective and from a jobs perspective but I think a very exciting future from what those capabilities mean because every individual will have access to

[SPEAKER_02]: I think there is enough space to take a counter narrative to what to the big labs, and that's what we're doing, we're saying that the big labs is the ones for building super intelligence and they will provide it to everyone, but there is space for sovereign super intelligence if you can actually afford that sovereignty.

[SPEAKER_03]: Okay, Kano, let's wish you who influences the way that you work.

[SPEAKER_03]: Name a person or many persons are something you look up to and why.

[SPEAKER_02]: It's actually hard to answer that question.

[SPEAKER_02]: There are a lot of people I look up to and when I look at some of the decisions they've made and the journey is it's very inspiring.

[SPEAKER_02]: There are a whole bunch of names around entrepreneurs, people who've achieved the pinnacle of their fields.

[SPEAKER_02]: But I do think there is a limitation with actually looking up to someone like you can look at them and say that okay, wow, they've done something amazing.

[SPEAKER_02]: But at the end of the day, it's so contextual, right, because the next Einstein will not be figuring out the same equations that Einstein did, or solving the same problems that they did, the next seed job won't be building in the iPhone.

[SPEAKER_02]: So it becomes very hard.

[SPEAKER_02]: that if you try to copy someone, it just doesn't work for you.

[SPEAKER_02]: So it's interesting to look at some of these people and say, hey, look, they're done from cool stuff.

[SPEAKER_02]: But I think looking at some of the qualities that they invite.

[SPEAKER_02]: You realize that you realize very quickly that the main quality they invite is that they really just had self-belief and it wasn't so much about anybody else.

[SPEAKER_02]: It was them figuring out how they can make their visions true and trusting their gut at the end of the day.

[SPEAKER_02]: So I think it's a really hard question.

[SPEAKER_02]: I mean, there are a whole bunch of people that I look up to, but at the end of the day, I think it's about really about me trusting my own gut.

[SPEAKER_03]: Okay, could not last question, so you're getting on a plane and is to think next to a young entrepreneur who's built the next big thing.

[SPEAKER_03]: They're jazzed about it.

[SPEAKER_03]: They can't wait to show it off to the world.

[SPEAKER_03]: Get me sure off to you, right?

[SPEAKER_03]: Good on the plane.

[SPEAKER_03]: What advice do you give that person having gone down this road a bit several times?

[SPEAKER_02]: In today's time, any of the playbooks we've seen over the last decade or two, do not apply anymore.

[SPEAKER_02]: all the playbooks that a lot of the startup advice that you see ten point tech which your problems, your solution, your MVP or traction, all that the standard 10 slide tech template and the frameworks that people applied to building SaaS companies and companies in the last 10 or 20 years do not apply anymore because we are facing a tremendous change with AI, right?

[SPEAKER_02]: It's not just, again, just repeating that it's not just a platform or technology shift.

[SPEAKER_02]: which means that the hunter-gatherer to agricultural transition took a couple of millennia, the agricultural to industrial transition took a couple of centuries, and the next transition, which is from the industrial to the super intelligence societal epoch, is going to take maybe a couple of decades, which means that you really have to think asymmetrically, all your bets have to be asymmetric and not so much point solutions,

[SPEAKER_02]: In fact, there is an 99% chance, if you were to come and pitched me, there's a 99% chance that anthropic or Claude will disrupt your business, right, or open AI, one of these AI's.

[SPEAKER_02]: AI will do the job before your entire company is trying to do, or your entire industry is trying to do.

[SPEAKER_02]: So, in that world, what will still hold true?

[SPEAKER_02]: So our ability to really question what will still hold true in that kind of a word is super important and I think not many people are not enough people are thinking asymmetrically right now.

[SPEAKER_02]: This the amount of progress that happened in the last.

[SPEAKER_02]: decade will happen in the next one year.

[SPEAKER_02]: And in the next 10 years, if we were to just leanerly extrapolate the last 10 years to 100 years, that much progress is going to happen in 5 or 10 years.

[SPEAKER_02]: 100 years worth of progress is going to happen 5 or 10 years.

[SPEAKER_02]: So the human mind is not trained to think as exponentially.

[SPEAKER_02]: And I think we need to start training ourselves to think of what exponential curves look like.

[SPEAKER_03]: I'd spend tasks at advice.

[SPEAKER_03]: I could all thank you for being on the show today, and thank you for telling the creation story of HECSO Labs.

[SPEAKER_03]: Thank you so much, Norow.

[SPEAKER_03]: Gridging here.

[SPEAKER_03]: And this concludes another chapter of Code Story.

[SPEAKER_03]: code story is hosted and produced by Noah Labhart.

[SPEAKER_03]: Be sure to subscribe on Apple Podcasts, Spotify or the podcasting app at your choice.

[SPEAKER_03]: And when you get a chance, leave us a review.

[SPEAKER_03]: Both things help us out tremendously.

[SPEAKER_03]: And thanks again for listening.