 Good afternoon, everyone's got that like getting ready for the evening energy. We have a fascinating panel talking about everyone's favorite subject, AI and early stage investing. I'm asking our panelists to go through and introduce themselves and then quickly talk about a recent, it doesn't have to be recent, but an AI investment they were involved in and maybe like a one sentence explanation of what the company does. So we start with you, Don. So I'm Don Stalter, Global Founders Capital based in San Francisco and we've invested in 250 plus businesses across the United States, a recent AI investment that we've made is in a business called Slope, which is a B2B payments business playing in the 125 trillion dollar B2B payments market, leveraging AI to do KYC, KYB and just to optimize all the payments processes. Very exciting company. Hi, I'm Corinne Riley, I'm an investor at Greylock focusing on B2B infrastructure and applications and yeah, the many AI investment related companies we've done. I work with a company called Base 10. Base 10 provides all the infrastructure you need to serve and deploy models in a scalable and more importantly, performant and cost effective way. Hey, everyone. I'm Natalie Vase. I'm a general partner at Spark Capital based in San Francisco. I just joined Spark Capital three months ago, so I'm going to talk about an investment I made at my prior firm, Amplify Partners. I was involved in an investment in a company called PostgreSQL, which brings hugging face transformers and all of your favorite ML models into the database, in this case PostgreSQL, which is usually the first database people get started with as app developers. So I'm very excited about that. And then just in terms of Spark Capital in general, two well known AI investments we did this year were adept and anthropic. So I was thinking, trying to think of like a really great metaphor for what you are dealing with right now, the three of you. So like, we're all like, we've lived in California or live in California. And so I was thinking about surfing and I'm a lousy surfer. But when you go out and you're in the water, sometimes you're just like waiting for a wave to hit. And then once you get it, it's like, if you get it, you ride it. It's amazing, right? And then sometimes you get a sequence of waves. And when I would go out, I just get hit bombarded by wave after wave and I couldn't even like get on my board. So I imagine if you bear with me for a second, that must be what it feels like for you now being an AI investor. Like, I can't imagine when you like introduce yourself at a party and then you're just like swamped by people. Natalie, can I'm just curious, can you like, how many pitches are you seeing on a weekly basis? How do you like filter out, that's also a question. It's like, how do you separate the sort of diamonds from the rough? I really appreciate your analogy. I'm a kite surfer. I'm a surfer myself. And it's not just about the waves, but also the conditions of the market or, I guess, the weather conditions that you're going out in. I think right now it's sort of a tsunami more than a set. And I think what we're sort of trying to figure out now is what happens when the dust settles. Every company, whether it's infrastructure, consumer, enterprise, has an AI angle. Whether or not their AI is like the fundamental piece of their business, I think that's what we're trying to figure out with every company that we talk to. What is the dust settling? I have no idea. Karen, do you have any idea? Is everyone sort of waiting for the next wave of companies to start raising their next investment? Is it like the next GBT model? Yeah, I think it depends on which layer you're looking at. So I think the infrastructure layer, when the dust settles, we'd be very different from the app layer. I mean, we're seeing huge outcomes in the app layer already these days. We have a portfolio company called Tome that helps you build presentations with AI. They see millions and millions of users in a very short period of time, versus at the infrastructure layer, I would say things are still shifting. And my opinion is they're going to continue to shift for quite a while longer. That doesn't mean we're not investing. Actually, the infrastructure layer is extremely busy right now. But when the dust settles, if it ever does, we'll really depend on what areas you're talking about. Can you help me, since I'm not as smart, and many people may not be as smart, define the infrastructure layer versus the application? What's a clear cut definition? Yeah, so we think of it in three layers. The model layer, where right now we have two investments at the model layer and the language models, inflection and adept, which in our opinion are two of the core private model companies right now. The infrastructure layer is any tooling that helps you serve, modify, deploy models within the company. And then at the application layer, it's anything that you can think of as from what people call your GPT wrappers to incumbents that are now adopting AI within the workflows they've already created. What do you, and maybe I'll turn to you, Dianne, what do you think, which one of those percent, I'm sure they all present different risks, but which one scares you the most right now of those three? I feel like this is a topic, a discussion that a lot of people are delving into and constantly discuss. I mean, there are quite a few concerns around how the application layer is going to be obviated by some of the infrastructure businesses out there. So maybe OpenAI will build a customer service app that kills all the other customer service apps that are built on it or whatever else. I think it's two early days to tell is my perspective. I'm an early stage investor. For me, it's all about investing in that really smart excited founder who has a vision, who isn't necessarily chasing kind of that shiny object, though, because there are a lot of those, and just trying to think about what the unit economics of that business will look like over time, and to really just help them build a real business. That's how we think about it. And what are the shiny objects? Like, I'm sure you're encountering companies I certainly have that just slapped on.ai to their name like six months ago. I mean, their businesses that were founded eight years ago who all of a sudden are AI founders. And I think there are ways to implement generative AI. There are sort of enablement strategies around it, which are fantastic. I think it's early days again. And it's OK for people to evolve in that direction. But again, I feel like 90% of founders when they start talking about AI don't really sort of delve into the use cases around what they're building necessarily. I think the first sort of big recommendation I'd have for founders is don't start describing that shiny object. Really talk about what you're building. What do you, one thing that comes up a lot in AI conversation is just the talent pool and how expensive the talented engineers. How do you think about that? And with your, I'll start to go back to you, Natalie. With your companies you're investing in, are you looking to see how many research papers there engineers have put out or their founders? Do you tell them to go out and spend as much money hiring the best? Or is it like save that money and have a really small team? Yeah, and that's a good question. I mean, I think there's an interesting point in time right now where talent for AI is particularly expensive. But you also have a lot of liquidity in the talent market because a lot of big companies have done layoffs. So it's actually a really good time as a startup to be hiring because there's a lot of people looking for jobs. You also have remote work, which is opening up the universe of places that you can hire from for your company. So in many ways, it's a great time to be a startup hiring. To your point about how we assess technology, what research we look at, I do a lot of investments at the infrastructure layer. And I think infrastructure in particular is led by research. And so even if you look at a lot of these foundational AI companies, you can trace back their work to research that was either at a company or at an organization like DeepMind or at a university. And so I like the title of this talk, Lab to Launch, because I think particularly in AI, research is leading industry. And I see that a lot in infrastructure. So I'm absolutely looking at the foundational research papers that people are writing. Who are the author of those papers? Where do those people go? Those are a great talent pool for these startups to be hiring from too. You mentioned DeepMind, which is in some ways the world's most well-funded research lab and in sort of has been for the past decade plus. Do you have that concern? Like, is it current? Do you have any, like, are there companies that you're like, OK, maybe this research you're doing is great, but maybe you need to start commercializing it? Or is everyone from the get-go now like have a commercial model? I would say, I mean, from the model, if you're investing at the model layer, right? First of all, we don't have a concern because we invested, we then backed Mustafa Suleiman, who was the previous founder of DeepMind and we've now backed him in his second venture, which is inflection.ai. We have a very large belief in the talent behind those networks. I would say the core question, right, when you're talking about AI talent is you have AI researchers, which some companies, like the model companies, aren't strong in need of, but not everyone needs a pool of AI researchers in their company in order to deploy AI. And actually, my belief is this is a big difference between the ML ops era, pre-transformer, and now, before you had to have this set of researchers that knew everything about ML in order to successfully deploy it within the company. And nowadays, the tools and the infrastructure that people are building are making it that any software engineer is essentially an AI engineer. And so I think the level of knowledge you need to have in order to use AI in your day-to-day engineering is going to be very quick to acquire for basically the existing cohort of engineers in anyone coming in the future. I'm not super worried about the talent from that perspective. Are you worried about valuations? About what? Valuations. I think valuations are in the control of the choices that us as investors make every day, right? I would say I'm more concerned about the companies that we are in. At Grillock, we make a very select number of nespans per year, helping them find value, find product market fit, and then scale up. I would say that's the core concern day today. Don, are you going to? I was going to say data talent is really important when it comes to AI in general. Being able to help build a data moat within a startup company and being able to leverage the time series data via machine learning to make great decisions as accurately as possible, I feel like it's the core of a lot of these businesses. And that's a skill set, that talent set that's been around for a very long time and that a lot of schools are promoting more heavily than before. At the University of Chicago, there's a new data science program that's very progressive, but that a lot of AI startups are hiring out of. So it's all about the data, I feel like. That's interesting. Your company was reminding me of the name of the company you mentioned. Slope. Slope, which is not necessarily like FinTech, right? It's a B2B payments business, so it operates kind of within a FinTech space. Now, the traditional way you think about it is like an AI company. Is that are you looking in the payments space or certain vertical applications? Is that where you think a lot more potential? So they're leveraging AI in order to better understand bank data, in order to process bank data more quickly, and also to do KYB and compliance. So it's KYB. It's like banking compliance effect. No, your bank. Exactly. KYB. And they're leveraging an LLM model. They call transformative to do it. And so they own that data. And they're building a massive team around it. You mentioned open AI before I did. But before, was it last Friday? When was it? It's only been like a week? Well, before that Thanksgiving Friday, before with the main question about open AI was it's relevant for you. So they had their dev day. Was it earlier this month or last? And there was all these jokes. And probably like, I think we even wrote an article about, well, there go all these eulogizing all these startups that were killed because open AI had just introduced this new product feature. And I think the same thing, that's not a new problem, right? For startups, it's like, how does Google, Microsoft, what if Google or Microsoft or Amazon or Meta does this thing? But it feels just because it feels like the pace of AI development is just moving so quickly. What do you do? I'm curious. Maybe you want answers. But are there any companies in your portfolio that kind of freaked out on that dev day? I think any company that's building in AI adjacencies was paying attention to that weekend. And I think it was just more representative of how quickly things are moving and how hard it is to predict the next day. So I think everyone was paying attention because everyone was asking questions about what this would mean for their company. And I think it sent ripples through the whole ecosystem. Oh, yeah. Well, that's it. I wasn't even talking about the SAM being fired and rehired thing. But just like when they announced their dev tool. Yeah, that too. Yeah, absolutely. All the startups that are building applications that now become a little bit easier to build with what they announced. But those startups can move just as quickly as Open AI can. We're really excited that Sam Altman actually led the most recent round in slope. Ostensibly, that builds sort of a defensive relationship kind of with Open AI. So he's supporting them. And I think maybe a lot of businesses are looking to get him involved from both a signal standpoint as well as a support standpoint. But fundamentally, it's really down to the teams, the operations, their ability to leverage data the right way. Was he candid with his communication in that investment? Sorry. He was very candid. But it was, for him, it was, I want to say, sort of a large angel investment more than anything else. So he's super operational. It's just a great signal. None of you would know exactly what happened down that weekend. You don't have any big reveals here. If anyone in the audience knows, feel free to find me afterwards. Feel like you're in this. But I mean, but it was, it's certainly, I mean, this joking aside, are any of your companies considered a nonprofit structure similar to Open AI, or? No. No. Would you advise them to do that now? I would say a precondition to being a venture capital invested company is that you're for profit. Typically, that's what we're looking for. Obviously, there is going to be exceptions in every case. But yeah, I mean, we've seen, over time, we've seen sort of nonprofit educational technology businesses come out of Y Combinator. And we've seen a variety of sort of nonprofits that shift to for profits. But you know, it's not something that we'll conventionally invest in. Yeah. I guess I'm curious, like, any thoughts about then? I mean, I clearly demonstrated that this, like, this world can change very rapidly. But I mean, you made a compelling point that, like, are you argue that your startups can all kind of out muscle, or they can move just as quickly as Open AI? Is there a point when the company gets, like, kind of, are you worried about it getting, like, big and bureaucratic and not moving quick enough, like, doing too many things? Is there, like, a mission drift problem with some of your companies, like, how do you feel, like, when you encounter that with some of the early-stage investments? My quick answer is that it's, I mean, at that point, right, once you're, we invest in Seed and Series A, primarily, right? So that's a core of our investments. And at that stage, the problem isn't, you know, do you have too many people in the building? If anything, it's a very lean team. You're trying to move quickly. The question that you're addressing, which is, you know, once you get to a larger state, right, how do you make sure that you're maintaining that speed of a startup, right? And not, and not quickly degrading? I mean, candidly, I think at that point, it's in the hands of the founder. And more importantly, where we have a hand is in making sure that the key hires that they're making at the executive level are the right people in those seats, right? And that's in particular a place where, actually, we spend a lot of our time with our portfolio companies is ensuring, hey, as you're hiring the head of sales, that's going to bring you from 10 million to 100 million ARR. Is it someone that can do that in a scalable way? Is that someone who can do that very quickly, right? Yeah, I mean, to add to that, you know, we love sort of co-investing with great funds on a lot of occasions, and building just really, really close relationships with founders, not necessarily taking board seats. But kind of through those relationships, you know, we like to be like the big brother or the big sister, you know, where, you know, maybe they're kind of having a big house party or something that the police have been called, but we can come in and we can advise them, you know, as to how to deal with the police and get out. That's kind of our approach. We're not there to, you know, come in and, you know, put them in jail or create some sort of an issue. We want to just be real good supporters right up until, you know, they can raise that next round and be successful from an operational standpoint, but not, you know, sort of put in full-on guardrails. Are any of your companies stashing GPUs in their basement or something? Like, how real is that scarcity? Definitely, they absolutely are. I mean, probably their attics, their basements, you know, everywhere. What, I mean, is it, are you, you're kind of like, just that's just a fact of life now that these companies have to spend, I don't know how much of their, like, early year investment on compute? Is it very, I guess it would vary by business model, but I assume across the board it's pretty hard. Yeah, I mean, the biggest problem that startups for us that I've seen are having is actually, I mean, of course there's a spend aspect of it. It's actually the access, right? Because, you know, if you're a large enough company, you can probably get access right now. If you're a very small company that has intermittent needs for these GPUs, you know, the large incumbents aren't going to give you that access. And that's why we have seen, I mean, there are a number of companies that are, you know, based on that premise, Together.ai, CoreWeave, RunPod, et cetera, that essentially allow you to use GPUs as you need them, and that's a great service to the startup ecosystem right now. I think most people, to my understanding, believe that the GPU crunch is going to end. In the next, the timeline is where everyone differs. Some people say as soon as within the year, some people say a year and a half or longer, but it will end eventually, at which point it'll be more accessible. And there's so many startups out raising money for kind of their new GPU platforms or GPU concepts. There are 50 million dollar seed rounds that we're seeing where, you know, the founder maybe has relationships in Taiwan with a big semiconductor company. So, you know, that gap is, you know, being quickly filled, it feels like, at the same time, you know, to your point on stashing GPUs and the basements and the addicts, like we do see it. We see people sort of love to geek out on those things and to try to vertically integrate or however you want to kind of describe their use case. I'm curious, I'm going to turn to you, Natalie, what, how early, is it, do any of your companies, like start thinking about regulation and regulatory risk? Honestly, not like at the pre-seed and seed, I would say. I think that's certainly going to be on people's minds as they are growing and doing go-to-market and trying to sell this, but I think in the really early product development, they're not necessarily thinking about it as much. And it may be because it sounds like you're not investing and say like, well, but like inflection, but it's definitely not early stage at this point, but like, they certainly have to wrestle with a lot of ethical issues around chatbots. I would imagine like that these companies are having conversations about ethical use of AI. Yeah, I think the core problem that most startups are just addressing is making sure that, you know, especially if you're building an app that is customer-facing or you're building infrastructure that people are relying on in order to interface with their customers, that you're doing it in a safe and secure way. And that is both a regulatory point, but it's also a security point in general. And I think candidly, I think we're just in the very early days of people figuring out what exactly that means, you know, what safeguards you need to have in place. There's a number of companies that are actually built upon this premise of making sure that as you, you know, are embedding, you know, the AI in there. Exactly. I was gonna say my fear with the regulatory stuff is how it affects startups. I think the big players are easier to adapt and to build in those, you know, processes. And I think it can be hard for startups to overcome that early. I was at Google in 2017 where every project was GDPR. And we had hundreds of people we could throw at it. And startups don't have that luxury. So I do worry about what the trickle down effect will be on early stage startups. But I think to Corinne's point and to Don's point, it's not always just about regulations, also about things that aren't regulated in law, but what is going to be your privacy policy, what's going to be your, you know, ethical, you know, guidelines for building the product. I think all startups at all stages think about those things. I write a lot about AI and so do many other people in the press. What, I'm sure there are things that really frustrate you. What do we get wrong about the technology and how it works? As in, yeah, what is like, what's the public perception of this sector that you think is like just dramatically wrong? I feel like people have gone in the direction of science fiction to a point where, you know, we're already talking about uploading consciousness, which, you know, is a really exciting theme. And, you know, it's great to let our imaginations run wild. But- You're not worried about Skynet quite yet. You know, for me, it's focusing on building a real business again. You know, it's focusing on, you know, what problem you're actually solving. We have this great company deal in our portfolio that does payroll and compliance, HR. And they're leveraging AI to do, you know, customer service and HR and they have this huge data set that they built around it. And that's a practical application of it. It's helping drive more productivity across the business, drive more revenue and everything else. I love that application. Have another business called Cedar.ai, which, you know, is software for the commercial rail space. And they have a computer vision technology that tracks the arrivals and the departures of the trains to make sure that they're on time. And that very simple use case is, you know, driven kind of meaningful value in the category. So I think, you know, those are the types of businesses that we get excited about right now. That's not to say that we shouldn't be excited about some of these massive visions. I mean, there are going to be some incredible businesses that kind of come out of those dreams. So there's like weird science fiction stuff kind of bothers you, okay. Anything else, do you guys have anything? I was just going to say, I think what people get wrong is that AI just happened overnight this year. The zeitgeist happened overnight, but the foundational things, the research that played into all the stuff we're seeing today has been in the works for, you know, a decade plus more. And I think what's happened is it's become consumerized, it's become accessible. And now, you know, it's in everyone's hands, everyone's household, but it's been an enterprise, it's been driving products we use for a long time, large language models, all of these things. So I would say it feels like it happened overnight, but it's been happening for a long time. Yeah, I guess the last thing I'll add is, the biggest thing people get wrong is that it's very much not done, right? So AI is very much under optimized in many ways in the way that it's actually being used in production today, and there's going to be significant changes coming out of these research labs, both in academia and these private labs, that are going to fundamentally change a lot of and improve, you know, these model architectures. And I think a lot of the infrastructure that's being built today is around today's architecture, and it's just always evolving, right? And so being ahead, being a couple of steps ahead of that is extremely helpful, both for founders who are using AI in their current products or who are thinking about building infrastructure, like staying ahead about what's to come from the research labs is extremely important, and I think people don't do enough. I imagine there are some founders, AI founders are aspiring ones out here. What, can you go quickly like through one by one, like what's the like thing that really irritates you when you hear a pitch, and what's the thing do you really like to hear? I'm gonna start with you, Naly. I don't want to put you in a spot. I think the best pitches are when you teach the investor something they don't know. I think a lot of venture firms pride themselves on having themes or thesis areas that they explore, but I think in reality, the best theses come from founders. So if you have a vision for how the world should be, and you feel like you can impact that, what's the unique insight that you have that maybe the investor doesn't hold? So I think sometimes the best pitches are ones where they surprise you and teach you something that challenges maybe a point of view that you held deeply for a long time. Anything that irritates you? Well, you don't have to say. I mean, I think there's sort of an interesting balance at the very early stages around really, really smart founders who are maybe great engineers, great scientists, and then sort of the great sales people who have a ton of charisma and who are able to raise capital. So at the very early stages, that balance is so key to AI founders. And as an investor, we have to sort of trust their technical prowess, which is something that we can diligence, which we can talk to sort of experts around and we can read research papers around, but sort of their ability to build a team, their ability to bring in like-minded people who are sort of at that same level, as oftentimes very challenging unless they have the right sales acumen. Cool. Grant, anything that you like to hear? I don't like to hear. Yeah, I mean, I would say the conditions for an investment are still very similar to where they were pre-AI or in the past, right? Meaning that, at least in enterprise where I spend all of my time, you're looking for people who know how to solve customer problems, right? And the fact that you're using AI in that process is an extra, it's a primitive that you can now use to your favor, right? And that's great. I believe everyone should be using it, but that alone isn't enough. You have to have that precondition of I am solving a customer problem. So that narrative of, hey, I understand this space, I'm passionate about it, and more importantly, I understand what the customer wants is continues to be the main thing that we're looking for. All right, I'm gonna do a quick lightning round. Just yes or no. In the next five years, one of the next big AI companies might be created in the next five years, yes or no? Yes. Yes. Yes. In the next five years, one of these AI unicorns won't exist. Yes. Yes. Which one? All right, that's for next time at Slash. Thanks, everybody. Appreciate it. Thank you.