Kleiner Perkins at AI House Davos: Leigh Marie Braswell on the Future of Generative AI

Welcome, everybody. I hope you've had a very interesting VEF so far. We're coming to one of the last days, but I'm very excited to have all of these great speakers on the panel. And the topic, for those of you that didn't follow it, is will AI create the next tech giants? Most of you in the room will just say yes. But let's discuss it. And maybe from our side, a bit of background, right? From— I've been building companies, maybe like 30 in the last 10 years of my career. And why did we want these VCs also on the stage together with a great entrepreneur was because with S2S, we're fostering also the next generation of investors where students invest in students. For the Americans in the room, it's kind of like dorm room funds. So where are my NextGen investors? I know there's some here. Okay, there. There's a few here. There. Okay, great. Because we know that not only to scale up startups takes fantastic founders. Who are founders in the room?

Okay.

It also takes the money and the backers to get you there. And that being said, I kind of want to start introducing our great panel. So Leigh Marie, you're a professional athlete. She's going to tell you in a bit why. And you've also been like one of the first engineers at Scale AI. So, you know, you know AI like, I mean, you grew up basically doing this since your whole career. And you work now for one of the biggest and baddest funds out there. And I think your password to get into their internet, if you're ever Visiting is really making history. So maybe tell us a bit how you went from one of the first engineers in Scale AI to actually making history with Kleiner Perkins. That would be very nice to hear more about it.

Awesome. And wow, what an intro. Thank you for that. And thanks for everybody coming to the panel too. Yeah, so first I'll explain the athlete comment. It's mathlete. So I actually grew up doing competitive math, and that's honestly what kind of got my career on the track that it's on. But the transition from engineering to investing— so I was at Scale for 4 years, was one of the first engineers there, and then eventually became a product manager there. COVID happened, and this was my third year, and a lot of my friends used the shelter-at-home time to start companies. And so these are people that I think are very special, that I would love to work with. And so I started helping them hire engineers and then also just helping them with product and ML strategy. And so I started angel investing. And I made, you know, I don't know, you know, 7 angel investments over the course of that year. And I just realized how much fun it was to, like the early days of Scale, go 0 to 1 with these people and learn about all these different things and have this unique vantage point that as an investor, you just get to look at, you know, AI broadly or something like that. So that was the motivation for then trying out venture. I spent a couple years at Founders Fund, and then now recently newest partner at KP.

Yeah. And you've kind of founded— or I guess you guys both joined around the same year. You joined Kleiner Perkins last year, is that correct?

Correct.

OK. That bridges me kind of nicely to Luna, because Luna joined Google Ventures last year. Now, from Leigh Marie being a mathlete— by the way, over playing poker, she met the Scale AI founder. So that's what happened. So in case you want a good poker game—

If you're ever in Los Angeles, Okay. But yeah.

Brings me also to Leigh Marie. Actually, we figured out that I know your husband. He studied with me. And that was kind of fun, full circle, because when I first met Luna Schmied, I was like, Schmied? And she's like, I'm Canadian. I was like, okay, interesting. But you started your career really in investing. And I also thought an interesting question for you, because you've worked with a few major funds and also with SoftBank. And now you're with Google Ventures. And Google Ventures is kind of like this very interesting vehicle that they're backed by Alphabet, right? But they're totally independent. And so my question to you was, like, what brought you into this interest of investing and investing especially in AI? Because Leigh Marie has a technical background. You're the classical investor here.

I am the classical investor. But the fun story is actually I was born when my parents were still in university in Canada. And my dad had the unfortunate misfortune of being born in the wrong time. And so he was— while I was born, he was in the late '80s, early '90s doing a master's degree and research in neural networks. And so I actually have grown up around technology with both of my parents. And my dad is now in his late '60s trying to do a startup. So if anyone's looking for a co-founder in his '60s, I've got someone for you.

Go diversity.

So I've spent my entire life around technology, ended up going into a more traditional finance route, started my career over at Morgan Stanley and Warburg Pincus, very traditional finance where I learned sort of the fundamentals of investing. But always wanted and had an itch to go back into technology. And so I ended up going into the founding team of the Vision Fund back in early 2018 and 2019, invested in both GM Cruise and in Neuro on the computer vision and autonomous driving side. Incredible companies to work with and loved a lot of what they're building. In fact, I actually have backed former autonomous founders as well as they build in the next version of generative AI. Most recently was with Summit Partners and their venture fund, where I continued to invest in technology. And at the time, I think it's funny because we called it ML before. Once something was solved, we called it ML, and then everything future-looking was AI. And now everything that was ML is now being called AI. So I think there's 2 versions of AI. There's ML, and then there's generative AI. And there's a lot in between that's still not solved yet. And so I've spent my entire career really in and around all 3 of those. Your oldest industrial companies have ML teams that are working on things, and I've worked with those over both the investment banking and the private equity side, and then moved over to the more next-gen technology companies over between SoftBank, Summit, and now GV.

And maybe I think both of you have to spill the tea, like which companies are pretty exciting about in your careers that you've invested in. I would love to hear that afterwards if we all agree. But last but not least, And just, you know, to the men in the room, hi, Thomas. Nice to have you here too.

Thank you for having a man on the panel.

Yeah, it was important. I was collecting all these great—

Giving out the quota.

Yeah, giving out the quota. That's what happens when you let me organize. So Thomas, you have had a very interesting career yourself. So he actually built a company and sold it in computer vision to Apple. And then you flipped your career around, and at Apple, you actually did strategic investments. Investments in Europe, as I understand it. And maybe you can also share a bit, like, how you've helped maybe one very famous founder get his first cash to found his next very famous company. And he actually then went to herd sheep, right? Is that correct?

Goats, actually. Goats.

Goats.

Excuse me, goats.

A goat farmer. And now you're back into building your next business. So maybe Tell us a bit, what were you doing at Apple and how was it? And why did you say, I'm going to go and herd goats and start my company again? Sounds a bit like an interesting story.

Yeah, I can talk about the goats, but not about the time at Apple, of course. Never comment, never confirm. As a matter of fact, I was just living in Cupertino. But a little bit more on the background, I was just out of grad school. Many of you guys are actually from ETH or TUM. Who likes TUM more than ETH? Oh, there you go. There you go. We got a couple of them. Thank you. Thank you.

Thank you.

So after grad school, I founded my first business, Mateo, in augmented reality back then. And I'm seeing that a lot. It was nothing but a research idea. We bootstrapped the business from my friends from the VC space. Bootstrapping means making more money than you spend. Very unfamiliar. concept in your world. I did that for 10 years, basically almost like an engineering office in the augmented reality and virtual reality space. We sort of liked that, but unfortunately, the financial crisis back then almost killed us. So we said, okay, let's get some venture money in. That was my first contact with the fine industry. I was driving up and down Sand Hill Road in California, 50 meetings, no term sheet. I understood, okay, this is an industry you really have to understand. It took me another 4 years to actually get it I signed the— I'm an engineer, I'm sorry. I can't talk to business people. So in 2012, we got funded by Atlantic Bridge, a company I'm representing sort of as a Frühstücksdirektor, a breakfast director.

Breakfast director.

We became friends over the years. We grew the business significantly and then basically gave up when the big American tech wave was coming. Went into augmented reality. Sounds like a sad story. It's actually not that much. Financially, it was halfway solid. I onboarded at Apple in an engineering role. Super exciting, cool company. But at the end of the day, big tech, right? And so inside Apple, I switched to the dark side, as I call it, the business side.

No offense.

And licensed technology. And then Jonas is in the room, Jonas from Aleph Alpha, actually. actually met him at that time. And then I became a goat farmer, investor a little bit on the side, and realized actually managing the money for other people is super fun, but not for me. So I jumped back and founded a new business in Munich now and San Francisco, a company called Rampler AI, which is a whole lot of fun, but no revenue yet. And we can talk more about that.

But bootstrapping.

Bootstrapping, always. No, actually, we are funded. Oh no, you're funded. Yeah, this time around. We're funded.

Okay. Well, I mean, talking now, really all of them have like really fantastic backgrounds. So I'm very happy to have you here. And we want to have this discussion. And I loved it because we decided to actually have this panel with this title before OpenAI had this whole Sam Altman chaos. And Microsoft just said, okay, we'll take 600 people or what it was. And we can just Big tech basically said, just come to us, no problem, we'll actually swallow you. So now, kind of opening up to my first question, and I'm going to go for Luna on this one.

All right.

It's like this interesting thing that we at the moment have the view OpenAI is huge. But if you look at it in numbers, I still think, you know, a lot of these startups that we're discussing as scale-ups relative to big tech are still small. So my question to you is actually, do you see these AI scale-ups really accruing this value that the big tech has? Can they take it on? And also in big tech, are the incumbents— these incumbents, are they like— who's going to win the race there? How are you dealing with that? Maybe, you know, for us, we all follow the news probably in the room, but we don't follow the internal wars and thoughts maybe that you have as an investor. So what are your thoughts on who's going to accrue more value, the AI startups or the big tech? And like, what is happening in the back between big tech?

Yeah, I'm happy to kick this off. So I sort of divide the world that I look at in generative AI in sort of 3 buckets. There's obviously many more, but 3 main ones that I look at. And the first one is foundational models. And so I think that's a category that requires so much capital to get off the ground. I think of it the same as AWS, GCP, sort of the cloud infrastructure layer, where you need so much capital to get off the ground that it makes sense that there's not going to be that many winners because they need so much compute and so much capital. So I think that's one area where big tech has an advantage, right? They've got the capital, they are dedicated to it, they're working towards it. And so maybe there's one or two independent players, but it is a hard route for a startup to go to Sand Hill Road and sort of get together all your term sheets to get to hundreds and billions of dollars.

Mm-hmm.

And so that category is one that at GV, we have not invested in, we're not currently looking. But then I think of the other 2 categories as very similar to what we've seen in SaaS. So there's sort of the infrastructure DevOps layer where there have been many $100 million businesses and then several billion-dollar businesses that have been built. I think it makes a ton of sense for that infrastructure layer to be rebuilt for AI businesses. And there's a lot of interesting things that we're looking at right now. I would say that right now, it's still early days, right? We don't know what the infrastructure layer is gonna look like, 'cause we don't know what the application layer is gonna look like. And so there's a lot of really interesting businesses, a lot of really interesting approaches. And I think it's just too early to say you haven't created a $100 million revenue business in the last 2 years yet. I think that's all gonna come as people figure out what their infrastructure layer looks like, what they want an additional tool for.

Hmm.

And the final category where actually we spend a lot of our time is the application layer. And I think— so big tech is an interesting way to put it because I think of incumbents in 2 ways. There's big tech, the big tech companies, but there's also a ton of great technology businesses that have houses all up and down the main promenade that are in some way an incumbent. And they've got an application layer that's already built out and they may have a data advantage. And so I think of differentiation very likely not to be on the model side of things, but more likely to be on the data side and the application layer side. And so I think where data really matters, you have a data advantage from one of those companies, and you have an application layer that isn't going to change fundamentally. I think if we're going to call those incumbents or big tech, I don't know, they have a good shot at winning those.

Mm-hm.

But there's other areas in which, you know, data needs to be cobbled together in different ways that people haven't seen yet, where the application layer and the way that you interact with the type of software is going to fundamentally change because of AI. And I think those are incredibly interesting places. for new startups to be built. So I think the answer is yes and no.

Yes and no.

If we wanted to boil it down to 2 words.

If I can add to that.

Yeah, please.

Actually, I fully agree. Big tech, what they're doing right now with open sourcing most of the LLMs, giving it out to the community, that's great for startups. That's actually building the startups.

It's like a donation of big tech.

Exactly. But the donation doesn't come for free, right? What's exactly happening is that the startups are being pushed upwards, and the application layer where you can do a whole lot of things is actually getting very, very thin. If you look at OpenAI, their assistant app store, whatever you would want to call it, the actual IP generated on that layer is very, very small. And big tech is giving the stuff away because they themselves have regulatory problems with it. So they're basically saying, okay, let's give it out to the open source community. Apple, the Ferret, for example, or Gemini, whatever. They want to test the market knowing that they have the data and the compute to jump into it if it's really becoming commercially viable. But to get a positive spin on it, it's actually good for the startup scene. And at Rambler, we're actually using everything from Segment Anything, Llama, whatever we can get our hands on. Because yes, it's prohibitively expensive to train yourself. Yeah.

Yeah, that's one of the big barriers. But the other thing is also they don't take responsibility. I mean, LLaMA just puts it out there and has fun. And you ask them, how did you train it? Well, maybe we're not going to share that.

Right.

But I also think, like, leading back to you, Thomas, I have like a question, like, for the founders also in the room, right? We're changing kind of the value creation within an AI company. Like, I don't know, Leigh Marie, I mean, Kleiner Perkins has literally invested in everything from Facebook, Amazon, and Google. So you built big tech. But I think the models are changing. So what is creating value in these companies at the moment? And, you know, like, my question is always, will maybe these software companies be driven more into open source, or will it be a more closed-source network? Because you kind of led on to that question.

Mm-hmm.

I don't know who in here is building an open-source code company in AI. Okay, is somebody building closed? Okay, closed. They are too closed. Little hands going up. Maybe you can, you know, it's a question.

Yeah, I mean, I think, you know, AI is not AI from a business perspective, right? If you're in the medical device industry, then AI is a little add-on, a little feature to a device, right? What do you do in that space? Data, data, data to get your accuracy higher. And this is your core value. In this more generative AI space, what I'm seeing is actually that the semiconductor layer is vastly underserved. It's all running on relatively strong compute platforms, and obviously NVIDIA is leading the space, but there will be a lot of sensing devices, compute devices, which have to run AI, which have a much, much smaller energy footprint. And then of course, you know, if you're in whatever generative AI, the good idea is always to have real customers. It's a vertical, right? If you're doing whatever, and I think that plays well with what you said, if you have access to the customer, the customer probably doesn't care that much what's under the hood, but you have to have access to the customer. And that's exactly the hard part. And right now in this GenAI space, obviously big tech is trying that as well, right? We have Microsoft Azure running generative AI. They do approach customers, big customers, small customers, and sell it as an additional service to Azure. Over to you.

Well, I'm going to give it back to somebody new. You kind of opened up about, you know, what it takes, what the growth components are. And Leigh Marie, you obviously joined Scale AI, so you're the one I'm going to ask, how do you scale an AI company? What are your learnings from an entrepreneurial engineering perspective that you had? And also, I think, how are we taking this forward in 2024? Like, what are you as an investor perspective looking for in growth strategy? And how do your portfolio companies go about growth strategy?

Absolutely. So I'll just sort of kick it off with, I mean, the rate of change of ML over the past, let's call it, you know, year and a half.

Yeah.

I mean, Copilot beta came out early 2022. ChatGPT is like a year old, a year and a few months. So things are just changing so rapidly. Like, even compared to— so Scale was founded in 2016. And sort of the first market that Scale went after, honestly, we just kind of listened to who needs training data the most. And back then, it was all the autonomous vehicle companies. And so Scale, to bootstrap early growth, focused on this, what many people call niche segment, but it turned out to be extremely profitable. And to Scale's credit, over the years, both when I was there and then after I left, they've been extremely dynamic. So always listening to where do people actually need training data? Where can we use this core engine that we've built? I think Scale's true moat is it is really hard to build a labeling system that scales up, pardon the pun, but that continues to scale up as you need tons and tons of data. And so they've been open to working with the government. And then recently they've seen a lot of success providing RLHF data for these LLM companies. And so if I were to extrapolate from Scale's journey, maybe a lesson today, especially today, is that, you know, pick an initial wedge that you start with, be the best in that wedge to get distribution, but then be open to expanding or pivoting or what have you because things are changing so rapidly. even more rapidly than they were changing when I was at scale. And so just kind of like figuring out, you've got all these incumbents, they already have distribution. Can you build a much better product than them? And then can you use that to start to get some distribution of your own? And can you maybe learn some non-obvious things about the market or where the market's headed that you can ultimately use to beat out incumbents?

Yeah, to Leigh Marie's point, I think what's really interesting right now is that everyone has these massive budgets for experimenting with LLMs. And everyone in their earnings calls wants to mention LLMs as many times as they want, as they can. But it's just so early, right? If you look at all the studies about LLMs actually in production, having, serving a real use case without sort of humans on the backend and whatever else to sort of double-check on everything, it's actually quite a small number. So that was one of the things that I wanted to come to West to sort of hear from people on is sort of the long-term budgets for things within generative AI and what that ends up looking like. And so if you're a founder, it's easy to say, well, I got like 5 customers, but you may just be in their experimental budgets and things may shift in terms of how they want to actually deploy in production long term.

And I would agree, right? If I would add a multimodal LLM to the whole mix, right? Right now it's relatively, in the B2B space at least, it's relatively limited. What you can do with that. It's literally for the marketing and the PR department. That's what you can do in B2B use cases. This got to change, right? And the information is more than textual, right? You have video information, you have all kinds of other information. And in the isolated use case, like repairing a pump or assembling a shelf or something, you have to be highly specific and way more accurate than with what an LLM does for the marketing. department. And that's a huge field actually for the startup domain where you're highly specific to the use cases and you build it out just way more efficiently and become extremely verticalized.

Well, I think it's also interesting when you're discussing like now what are the also entry barriers. So, you know, we're kind of comparing can these companies become the next big tech and what's going to happen back there. So maybe A question as well to all of you, but I can direct it at Leigh Marie because I'm looking at you, is like, where do you also see like this defensibility of startups? Because, you know, Kleiner Perkins originally invested in Facebook, and Facebook had this huge lock-in effect through social media, right? So I'm like always curious to see like what is the lock-in effect for these AI companies? And, you know, we're talking a bit abstractly about it. Please, both of— all of you go ahead and also Give an example from your portfolio that you can talk about. I think this is also interesting to put substance to matter here.

Totally. And that's a great example. I mean, classic network effects. Something gets better the more people that you have on it. So you wanna bring all your friends, and that's a form of defensibility if you can really scale that up. Yeah, I continue to go back to the same book over and over again. And if you haven't read it and have your whole company read it, I highly suggest it. 7 Powers. It is a classic book. And that is one of the 7 powers. And as a startup, you need to have at least one, probably more than one if you're a very successful growth stage startup, to create a defensible business. And so, you know, I get very wowed by an AI demo, and it's so easy to just kind of like, it's so easy to just be wowed by all the demos these days, but you need to fundamentally go back and say, how does this business find one of these 7 powers? So some other examples of like, you know, AI companies with 7 powers. When you think of image generation, you probably think of Midjourney. They just have this incredible brand in the space. And you'll see all these competitors pop up with great models, but Midjourney is the company that seems to be like capturing all the mindshare. Or you have somebody like OpenAI that all these talented researchers are going to. That's the cornered resource, 7 powers.

Mm-hmm.

So there's just, there's a lot of, and then all these copilots, there's switching costs to get off of them because if the copilot really learns how you work, do you really wanna switch to another copilot? Probably not. So there's just a few examples of these, but I highly suggest you sort of read the book if you haven't already.

Who has read the book? I actually have it. Okay, good.

Yeah, I've read it multiple times.

Okay.

I suggest all my companies read it. But yeah, so it's a cool— it's a good framework to think about defensibility given all of the stuff that's going on.

Yeah. Okay. Well, I mean, I think another dimension that I see now with with AI is that— or maybe I'm going to ask this as well again back to Thomas. We always think of the companies also in terms of FTE. But we're in this— AI basically rationalizes away FTE. So maybe carpenters one day will earn more than software engineers. I don't know if this could be true. But my question is kind of like to you as well, and also maybe to Luna, is like, can big tech basically Will, you know, they have like 30— how many employees does Google have?

I actually don't know the answer to that. Okay.

I'm just going to say 30,000 or something.

1,000 less.

1,000 less. Okay.

Okay.

1,000 less. But, you know, and OpenAI has maybe these 600. So I'm kind of like interested to also talk about the dynamic. Where are we going when we talk about big tech? Will big tech, the next, Although big tech have that many employees or not, or how's that dynamic going to look like?

I mean, once again, big tech is just fine what they are doing. They're just commercial companies. Obviously, they want to be dominating the market. Yes, they want to build monolithic monopolies, and they'll continue doing that. And, you know, the guys in the conference center over there, they should discuss that. It's a political discussion. More on the On the actual usefulness for society, right, as an engineer, right? Oppenheimer, we all watched Oppenheimer last year, same classical problem. Is someone inventing stuff responsible or is the one using it responsible for all the societal effects? Once again, another political question. My personal opinion is systems get more efficient. We drive more efficient cars than 10 years ago. We have electric cars. Same thing applies to AI. So there's a bright future for all of us. There are jobs galore and everything. On your statement of the carpenter, I actually see in this scenario actually very bright future for everything which is highly skilled, trained in combination with AI. And I talk around this a little bit. If you get the accuracy up of the LLMs by using data which is tightly curated to the use case, then, you know, you can, in combination with augmented reality, for example, superimpose the guy how to make his job more efficiently. And to do some shameless marketing for Rambler, of course, this was actually the starting point for us. We said, okay, we have the capabilities of multimodal LLMs, we have basically a system that blends that into your eyes, let's combine the 2 altogether. One more thing, one more thing, 30 seconds, right? For the European entrepreneurs, and we're in Switzerland, okay, let's— I should restart. We're not in America. For the EU27 entrepreneurs, we have a very interesting situation that we are starting to regulate AI extensively in Europe. And that's a good or a bad thing? Once again, talk to the guy in the conference center. But from an entrepreneurial thing, I actually see the chance if you're over-compliant with this freaking EU regulation that is coming up, it creates a moat in itself. Because the US guys, until they discover that, you can have already customers onboarded, you're compliant. And, you know, the EU legislation, for what it's worth, not advocating for it, but it will disseminate into the world, same as the GDPR. So, the advice to founders is That space actually super hard and creates a moat against the American startups.

Okay, so that's another lock-in effect. Maybe it's in the book as well, I'm not sure. But it's an interesting one because I think a lot of founders dancing around regulation eats up a lot of time and money. But it's good to hear that EU is avant-garde and hopefully good at it.

I didn't say avant-garde. I said we have to live with the pain.

Okay, we have to live with the pain for the— okay, we'll live with the pain.

opportunities that come out of regulation. So I'm London-based, and when I moved to London, I think one of the first things that I was focused on within Europe was where regulation has created industries that I think are actually ahead of what exists in the US. And 2 of those are fintech and climate, where regulation in Europe and the UK— and because now apparently they're separate after Brexit— and Switzerland and everywhere else, the regulation has created a really robust ecosystem where startups can thrive. And so I think that'll be the same here, right? I don't— again, I think it's early days, so I don't know exactly where that is. And if any of you know and has a crystal ball, please come to us afterwards, and we'd love to hear that. But it always does create opportunity as well. Everyone always thinks of it as a forcing function and limiting companies, but it does create opportunity.

And Luna, because we were once talking about the investments that you're doing with GV, and something struck me. She had the comment in saying, you know, We don't have that many big AI companies yet, so we're moving our investments in really early. So I was just curious to hear from you, like, how are— how do you think about this? I mean, it's quite a lot of work to find early-stage scale-ups, right? Or potential scale-ups, let's put it that way.

Yeah.

How are you guys going about it? And what are you kind of targeting at the moment? You saw— you talked before about application layers. But I'll also be curious to hear a little bit more how you see this from an investor perspective, and especially the founders in the room that are raising would love to hear this too.

Yeah, for a variety of reasons over the years, GVs moved a little bit earlier. And so I'd say our sweet spot is probably similar to you guys, which is around that Series A. So a late seed to a Series B. And we've gone from sort of the scale-ups. We put an entire fund back in the day in Uber Series C, which was a very famous investment that we made. to earlier in the journey. And we've created sort of value-add resources to that point on sort of helping companies in that area. I think that if you're interested and excited in generative AI, it makes sense to invest early right now. And so we've invested both, like I mentioned, in the DevOps and the infrastructure layer, as well as on the application side, thinking through, you know, this— I haven't read The 7 Powers, but I think it's the same as how we think of things, which is The moats and the way that you build businesses hasn't fundamentally changed with LLMs. You are building with AI, absolutely. But the moats that you build in a startup have not changed. I don't think there's a single new moat unless you say you have a model moat, right? Which is just foundational models, maybe. And so we think of investing in startups exactly the same way that we have. We've invested in AI for quite a few years, actually. And so, have invested everything from Samanova on the hardware side all the way up to the application layer.

And Leigh Marie, I mean, you're— how's Kleiner Perkins thinking about this? What are you all going for? Because— or where do you see the next tech giants also arising?

So, I want to underscore something you said, I think, earlier. Some of the hottest, quote unquote, GenAI companies today are companies that have been around before generative AI, but that added it in Very quickly evolved to they were thinking about ML internally and are now these incredible applications. So we've been involved with Glean for a really long time. So they're an enterprise assistant tool.

Yeah.

And they were created way before sort of the LLM revolution or whatever you want to call it. And they've been able to incorporate that technology into their product and have seen incredible growth as a result. And a lot of our companies, yeah, similar thing. We have a seed investment in this company called Ambience that essentially listens in on patient-doctor conversations, automatically transcribes what's going on, and then inputs it into downstream systems. They were also founded, you know, right before all this happened and have been extremely receptive and they're a very strong technical team and able to add it in. So when you kind of think about like, I guess incumbents in GenAI, a lot of them actually are like, you know, even Series B startups.

We don't think of them as like old companies. There's an incumbent layer here that are just startups that were built 4 years ago.

Yeah, or 3 years ago.

Exactly, or 3 years ago.

So we're already in some gen AI companies. And then over the past year, have invested in a lot more.

Well, that's also hopeful. I have heard both of you— OK, we have one investment in Switzerland. That's at GV, right?

Yep. We invested in Scandnet.

OK, Scandnet. OK, so Swiss startups here.

Yeah.

Kleiner Perkins needs one.

We do.

We do. But I also want to have this final— so I didn't quite mention this at the beginning. But we're also going to try— because there's very interesting people in the room at WEF. So we're going to— I'm going to like have one final question to them, and then the stage is yours. So start thinking about your question. Okay. Now, where will the next tech giant be, Thomas? Is that— where do you see it as an entrepreneur? Is it going to be Rambler AI?

Of course.

That's a layup question.

Where's my final layup?

That's the smartest question I've had for the evening.

No, I mean, honestly, Like, really honestly, I think this AI, LLM hype bandwagon will, you know, move on in 2024, right? I mean, it's a little bit overhyped, right?

Yeah.

And in this very moment, I'm not that old, right? 20 years in the industry, right? What will happen is that the engineers move in and the social science majors will move out, right? Unfortunately, also the VCs will move out at this point.

It sounds like an ocean.

But this is really when it's happening. And once again, I have not a negative view. I'm doing it myself, obviously, right? But I have a very, you know, like solve a problem kind of approach. And this will, you know, I mean, this will become very hard in 2024. I mean, and so for the entrepreneurial side, because, you know, I'm also representing an investor, but at heart I'm an entrepreneur. There's these 3 basic simple rules, which is effectively praise for impact every day. What do I mean with that? If big tech moves in into a certain topic and you've been at ETH, well, that's a problem to start with. But you're at TUM and you did research for 10 years on a specific sensor, you started, you're building it, You have your first success, big tech moves in, you're gone. Brace for impact. Second one is enjoy the ride and stick with it, right? Building companies takes forever. It's not in 2 years, right? And the third one and most important one always applies is an order a day keeps the venture capitalist away. That's just the basic rule of building a great company.

venture capitalists away? And maybe from the 2 partners here in the room, what is—

No offense, by the way, right? I'm just saying, right?

When the order doesn't come in and venture capital is needed, what are the tips for the upcoming investors, but also the startups in the room, Leigh Marie, in AI?

Yeah, I guess actually pretty similar. I mean, that was some really great advice. I think Brace for Impact is a great just sort of When I look at the best founders that I work with, they're just constantly ready for whatever's thrown at them and able to change and to handle it effectively. But yeah, I think, you know, I don't have AI-specific advice. I think ultimately it's what are you special at? Like, what are you very good at compared to other people? And then how do you build a differentiated business leveraging that? It ultimately starts with a founder, set of founders, and, you know, there's usually some cool tech involved and then— at some point you have to build a business. So that's the oversimplified advice.

Okay. Well, it also said stick at it and be ready for impact. Luna?

I guess brace for impact is the takeaway from this panel. No, I think again, like the fundamentals for building a business with AI are very similar to the fundamentals of building a business over the last 10 years. Nothing shifted with what businesses look like. It's a fundamental shift in technology that you're building with and you need to be nimble and be able to adapt to it. But building a business is not fundamentally different today than it was 10 years ago.

So it sounds like we're not in a major shift that everybody's afraid of. It sounds like we're doing business as usual, just with a very new powerful technology as well to a standpoint.

I mean, there's the question that you hear a lot, which is, oh, are all these jobs going to go away because of AI? And if you look at sort of history and how technology has impacted overall jobs, If you were to believe that hype every time you would hear it, we would have like 60% unemployment at this point with every single technology that was supposed to take away all the human jobs. And it's not. And so I don't think it's fundamentally different from every past innovation. Like the assembly line was supposed to kill 30% of jobs or whatever else, and we seem to still have the same unemployment rate that we had then.

That's a hopeful note. And now I'm just going to open it up. So I didn't explain this, but we're gonna— please raise your hands for questions, because what I'm gonna do is normally at WEF, we have so much great people in the room. I want you all to pick in on this question and ask a few questions yourself. So Leonie's gonna throw you the box, okay? And please speak into the microphone, 'cause we're voice recording this. Okay, 1, 2, 3, who's gonna be the first catcher? Okay, Arno from S2S, yay!

Oh, we're not throwing it.

No, we're not, okay, thank you.

Okay, so everyone, So I heard a lot of love for TUM and ETH. I'm also going to give some love to EPFL, where I'm a PhD student in artificial intelligence and also active as a student investor at S2S Ventures. So my question is for actually the 3 of you. There's the example of Isomorphic Labs from Google— I mean, from Alphabet— that's active in London and Lausanne. And the question there is, how do you avoid cannibalizing on startups that Google Ventures invests in? OK, there's an independent part. But, and also, how do institutional investors like Kleiner Perkins look at that, as well as startups building something themselves and getting them as investors? Okay, so basically the question was towards IP theft from big tech, to hyper-simplify. Yeah, that's a tough one. So honestly, big companies— I'll answer that for you, okay? Don't you worry, okay? Don't you worry, okay?

I'll take this one.

Actually, big companies are not mean. They very rarely steal something.

Okay?

But there is this concept of the unaided memory. I don't know if you heard that, but basically you signed it a million times. It's in these NDAs. And basically because you give them an idea, they will probably, with the might in resources, talent, do something similar, right? You know, honestly, it's not much you can do about it, right? And doing information hiding and not showing them the great things you developed and whatever doesn't help you either. But generally speaking, big tech doesn't— well, at least Western big tech, I should be more specific— Western big tech, it doesn't necessarily steal. I've never seen that. That's at least my And once again, I was sitting on the other side. I can tell you the contracts in the Western world, if they are negotiated well, are basically honored by all of them, right? But they do it regardless what you say or do.

Yeah, again, I don't think things have changed with AI, right? It's not fundamentally different. And you could have said 10 years ago that big tech is going to kill every new startup that comes up because they have so much capital. And that's just not fundamentally the truth. And so There's always going to be room for startups and for innovation. And if anything, the existence of the models and the open source models actually lowers the barrier to start something new because it's been— the playing field has been evened out. And so I don't think that there is any sort of, you know, disadvantage coming in as a startup more so than there was 5 years ago.

I guess maybe a slightly different viewpoint. I mean, I think especially the past few years when we've seen these big tech companies either do very official partnerships, investing in these companies, or just having some sort of relationship, whether it's an integra— as easy as an integration or as complex as like you're sort of going to market together or something like that. It's like, it's just part of the investment decision. So it's the pros and the cons of this relationship. And sometimes the pros of the relationship like really outweigh any potential cons. Like if you're getting— you know, tons of compute resources, which is actually, I mean, even though we do have open source models, you also have to have a lot of compute for certain types of models. So there is a high barrier to entry to some of these new types of businesses. So it's just something that, yeah, we honestly like in our risk we have to go into, maybe this could happen, but maybe the partnership is worth it.

Maybe that's a good time for an ETH plug too on the compute resources.

Yes. Oh, okay.

Just to tee you up there.

Thanks, Alex. Okay. Yeah. We have like a lot of great GPUs if you haven't heard.

So EPFL and few Swiss.

It's called Swiss AI. Some of you may have heard about this project. So yes, we have some heavy compute power, and please connect with the EPFL and ETH people in the room, and they can tell you more about it. Thank you, Luna. I have one question in the back there.

Thanks. My name is Marco Rossi. I work with ISO. I direct ISO International Standards. I used to design semiconductor processors that run your iPhones for the last 20 years. So everything AI, whatever the layer, needs to run on something, right? We haven't yet had like molecule run some AI, like processors in the end and data centers. Now there is a perception it's very much commoditized now, except for eventually transition to quantum computing, but still far away from being mainstream. So do you think that semiconductor underlying AI application computing and sensing is by and large commoditized, or you see opportunities there for some disruption? I actually do see a whole lot of opportunity. And if I want to give you an example, right, the space we're in, augmented reality, everyone wants these smart glasses. Meta has them. Google doesn't have them anymore since last week. Look that up on TechCrunch. But anyway, the challenge with that is you want a perception stack which is running on a minimal, minimal power envelope. And especially in the inference stage, there's no hardware out there. And this is the smart glasses space, but it'll be any washing machine, anything with sensors connected to it. And this space is not commoditized at all, and it won't become commoditized anytime soon because the requirements are way too specific. That's at least my opinion.

Yeah.

Oh, you can— is there—

Leigh Marie, in the back? I agree with that. And then separately, I'm really excited about— I mean, this is maybe a little bit adjacent, but I think there is a cool opportunity to build a software company here too. Because when we think about, like, yes, you on your MacBook with all of your compute, you can run any model you want maybe. But think about all the other devices that also would benefit from running really large models. There's probably a great software company also to be built Partnering with hardware companies for you to quantize models or run the latest, greatest, but very small model on your computer. So there's a few that I'm—

They just launched like this. I don't know. We have a 16-year-old at home and he just like downloaded this. I forget the name of the model. They can use it remotely on the laptop. And I was impressed on how good it is.

So there's a few projects in the space. There's GGML, which is quite popular. There's a version of LLaMA you can run on lots of different types of devices. And then there are a few other cool startups too in the United States— LM Studio, Ollama. But these are all very early-stage companies. I think it's a cool space.

Good. Who wants the ball or the— no questions. Okay. Is that true? Okay. Hannah. Hannah, for— I love this. It's always, it's always like a bit daring, but I always think we have so many interesting people in the room. Okay, Hannah, go. Thank you very much.

Um, so I mean, in the venture scene, you're always placing bets, right?

And I think you've talked on a structural level and trying to see like the big pictures, but I'm going to be asking you to place bets now. So if you could bet on one future tech giant that will be created as a result of GenAI, what would it be and what do they do?

Thank you.

I guess, can I ask a clarification question?

Oh, yeah.

Like—

sorry.

Do they have to be below a certain size right now? What do you define as a tech giant? Is it you want to see the most growth in the prediction, or are you just like a company that's going to be like the absolute largest company?

Okay, it's getting technical. So, I mean, it can be something you already have your eye on, but it could also be a field you think, like, here there will come one for sure doing this. So, like, feel free to interpret it freely.

Interesting.

I'll start. For me, it's easy. My own company, Ranta.

We all knew that.

Look us up. Like us on Facebook. Right?

Of course.

I'm pretty old on Instagram, whatever, Bumble, whatever.

TikTok.

Oh, that's it.

That's it.

There we go.

Sorry, we all thought there was more to come.

I'm happy to go through.

Super self-centered.

I'm happy to talk through a category that I think is really interesting, which is if we think about where we are in the economy right now, a lot of tech companies are focused on customer success and retention and expansion instead of net new because they're thinking about capital efficiency. You know, maybe Maybe some people are even thinking about bootstrapping, I don't know.

I'll explain it to them.

I think it's a really interesting category in that it's been created relatively recently with Gainsight, which has built a beautiful company. But there's so much more in terms of data sources that are available now, if I think through the moats that we talked about earlier, and the application layer in terms of how you interact and how you manage customer success may change. And so I think there's a wide open space for someone to create something new for B2B and B2C businesses. And so I think that's going to be one of the big things that comes out of this because it's one of the areas in which you had— maybe you had the data, maybe you didn't have the data before, but it was just too much for you to handle and to sort of get insights out of. And I think we have everything that we need now to build an incredible business there.

I think if I had to pick a category, it's kind of the assistant automation category. I mean, it's interesting. There's a lot of a ton of different companies in that broad space now. There's some that are sort of more general enterprise like Lean, which is already quite a big company and a Kleiner portfolio company. And then there are some that are smaller and more specific. So we just invested in Harvey, which is a legal assistant that I am very optimistic about. I talked about the medical assistant. I think so it'll be interesting seeing like, and then I think there's also a huge opportunity past Copilot to do something with developers and not just assist them, but fully automate parts of their work. And make it such that many more people can code than people can code today. So I think that there's just like a ton of room there. It'll be— I'm curious to see how many companies, big companies, will come out of it, or how much consolidation there will eventually be. But that'd be kind of the general area that I would be— that I'm keeping my eye on.

Okay, well, I like that takeaway. So to answer the question, sounds like we're going to be building some next tech giants in some different spaces. And I want to thank you all very much for being here. Leigh Marie flew in from San Francisco, Luna from London, and you came straight from Bavaria and need a beer now.

From TUM.

Okay. And please, I would like to just raise a hand of applause for all of you and thank you for having you here. Yeah. Yes. OK, we're done. For the startups in the room, we're now in room 2. The ones that are invited knew it. OK. Thank you. Good. We'll get up.

All right.