Louis Lehot Talks AI, Machine Learning, and Generative Technology

Brett Waters: Mhm. It,

Speaker B: It's.

Benjamin Levy: Mhm.

Brett Waters: Well, good morning, good afternoon, good evening, wherever you're joining from. Uh, and welcome to the second in our series of live stream panel discussions for investors and startup founders. Uh, I'm Brett Waters, I've been in Silicon Valley my entire life, immersed in the world of entrepreneurship, innovation and venture capital. And joining me as co host this morning is Louis Lowe.

Ahmed Reza: Hey Louie.

Louis Lowe: Good morning, Brett. Thanks for putting this together as always. I like the little jingle, uh, for everyone who doesn't know, I'm Louis Lowe, I'm a startup lawyer, uh, here in Palo Alto, a few, uh, stone's throws from Stanford University. And uh, my mission in life is to help founders and investors, uh, form new businesses, finance them, scale them for growth, and have a great outcome.

Brett Waters: Nice. So Louis, it's hard to believe, but it was only six months ago that most of us had never heard of ChatGPT before. Um, and then in January they had 100 million users. Um, so it really kind of, more than anything I've ever experienced in my Silicon Valley Valley career kind of seemed to come out of nowhere and all of a sudden, uh, appears to be changing everything. Um, and you know, one of the main things that we need to know, Louis, is will chatgpt replace lawyers?

Louis Lowe: Um, certainly going to replace, uh, many things that we do. And I think it's going to increase the accuracy of what we do, the efficiency of what we do. I, uh, think it makes us better. So I'm certainly, uh, out there competing in the marketplace as somebody who uses more, uh, technology tools than less.

Brett Waters: Nice. We've got a terrific panel, uh, of experts joining us this morning. Um, uh, Ben. Hey.

Benjamin Levy: Hey. Thanks for having me today.

Brett Waters: You bet. Give us the brief background on yourself.

Benjamin Levy: Um, sure. So Benjamin Levy, co founder and general partner at uh, Bootstrap Labs, an early stage venture capital firm, uh, focused on applied AI technology since actually the first seven years, so probably a bit before everyone else out there. Um, we've been investing in deep technology for much longer than that, but really focused on applied AI for the last, um, seven years. Um, feature of work, mobility, health, digital infrastructure, financial infrastructure, and energy and climate. Um, so really looking forward to this conversation here as we've seen things evolved, uh, since the early days to where we are today.

Brett Waters: Yeah, and Walter, what do we need to know about you?

Walter: Uh, well, my background is in the eye, like most people in the panel, I think. And uh, I did my PhD on computational, um, semiotics, that's the storytelling abilities of language, uh, models. I've been an entrepreneur for a very long time. Together with my wife, we were involved in four IPOs, two in America and two in Europe. And I'm now enjoying the summer of AI in the beautiful campus of Stanford for my fifth year. And, um, and I've recently been, uh, become the curator of the new Tet event, which is called Tet AI, which will be for the first time October 17th. And my wife is the producer. She's the boss, actually.

Brett Waters: The boss.

Benjamin Levy: So be nice to them if you want to be speaking over there.

Brett Waters: Yeah. And, uh, Dipanweta. How are you?

Dipanweta: Hi. I'm doing well. Thank you for having me, Brett. Um, so I'm joining in from Washington, D.C. i'm the CEO and co founder of Sourcero. Uh, we are a leading medical analytics platform for life sciences products. Um, and our customers use us to gain more insights into how their products are performing once it's in market and then drive better decisions around both market retention and expansion. At the end of the day, we live because patients can't wait to make sure that they get the right treatment at the right time. And of course, we use AI to accelerate the decision making and really augment our, uh, users and that's our philosophy to using any cognitive technology.

Brett Waters: Yeah. Yeah. Alex, what are you all about?

Alex Babin: Hi, everyone. Happy to be here. I'm Alex Babin, CEO and co founder of Zero Systems. Uh, joining from not so sunny, Campbell, California. It's not so sunny here, um, but, um, uh, Xero is orchestrating an operating system to build, run and deploy AI, generative AI applications for large, uh, enterprises. And we focus on knowledge workers like lawyers, accountants, consultants, uh, financial specialists and

Benjamin Levy: so on, so forth.

Alex Babin: Because we believe that, uh, as we're saying, AI will not replace people, to replace people who are not using AI. And, uh, people should be focusing on doing what they're good at and what they like doing at the same time. And everything else should be delegated to AI.

Speaker B: That's what we do.

Brett Waters: Excellent. And Suman? Oh, he may be frozen. So let's go to, uh, Ahmed.

Walter: Hi.

Ahmed Reza: Thanks so much for having me.

Brett Waters: Give, uh, us the brief background of yourself.

Ahmed Reza: All right. My name is Ahmed Reza. I am the founder and CEO of yobi. I've, uh, been in the AI space for a while, but not as decorated as some of the other wonderful folks on this call. Uh, we're working on building, uh, synthetic agents. Um, basically started imagining what the future of business communications might look like, uh, a few years ago after transformers are out. And, you know, attention is all you need. So the AI geeks got to work. Um, it was really hard to explain to people what we did, but thanks to ChatGPT, now everybody is like breaking down our doors for synthetic agents and customer service, sales and marketing

Brett Waters: and. Suman, uh, can you hear us? Not yet.

Louis Lowe: Well, let me tell you who Suman is.

Brett Waters: Okay, perfect.

Louis Lowe: Like uh, Ben, ah, is an investor who's solely focused on very early stage uh, AI businesses. And uh, he has a, a new fund that he's investing from and is oftentimes the first check into a new business. Uh, and uh, he's based here in the Silicon Valley and, and got a fantastic network of AI companies and I look forward to his perspective on what the opportunities, uh, and challenges are both for companies and investors in the market.

Benjamin Levy: Louis, I think he's, you're hired. After that he's going to hire you for all his intros.

Brett Waters: Oh, nice job. So, so Ben, you said something pretty interesting, which is you said that you feel like since ChatGPT became a household term that it's easier to explain what your company does. Is that what you said?

Benjamin Levy: Um, not. But I'm happy to jump on this. Look, you know, I think Walter, you know, wrote the paper early on on this, but it's kind of interesting, right? I built an AI company in 2000 and I would say some of the things we do today are not very different. But um, Genai is kind of, I like to call it a global consciousness moment. Nikolai, my co founder, like to speak about another Sputnik moment. Um, people around the world have realized the capabilities and now everybody um, has maybe not a better understanding, but at least uh, uh, uh, is touching the product and is able to see it when before that frankly was embedded deep into the code and most people were using it without understanding it or even seeing it. So I think from that standpoint we've made the world aware of it, uh, for good and for bad. Um, we can talk about some of the challenges that it brings along with it. Um, you know, we've been investing for a long time in AI, I think one of the first V in the us um, so we always thought about narrow systems and human augmented intelligence is where value creation was really taking place. Um, we've been investing in large language model type technologies for know, seven um, plus years. So not news to us. We've been, you know, having companies using this sort of platform for a little while. But again the value is still to be determined in my book in that, you know, we call the applied AI, um, as Our approach to investing and I think it still stands true for gen. I think people are still searching for product market fit for this gen core technology. Um, and I think the context in which you apply it would be fundamental. The different uh, health as we just heard somebody mention, uh, or sales and marketing. I think what this strategy can do should be trusted for will vary widely. So I'll pose there because I'm sure everybody has a point of view on that.

Brett Waters: Yeah. Anybody else want, anybody else want to jump in on this

Dipanweta: on generative AI? Yeah, I'm happy to speak to it a bit. Uh, you mentioned health and I think for us we are working or we're supporting users and customers who have been working with a lot of data in a high tech environment, AKA life sciences, drug discovery. We've seen some great applications of AI in this space. But when it comes to the operations of those organization and actually optimizing and scaling up how they operate, that's really where they're still to be transformed. Um, and what's very interesting about our customer base is just that, is that they're starting to realize what kind of quote unquote superpowers this sort of technology can give them in scaling up and making their, their operations more efficient, making their operations more data driven, having more nuanced customer base in a real time basis.

Alex Babin: Uh, I wanted to add something, what Benjamin said, uh, Ben, you just compared it with a Sputnik type of event. Uh, I would say more of a uh, brother's right. First flight, we're in the air, we're flying 300 yards. Uh, it's a great achievement on its own in a mindset less on the technology, more in the mindset because we've been doing it for over four years. Uh, but what we see right now is that enterprises and customers, they actually don't need that paper plane yet. They need Boeing 777 with all the whistles, uh, and bells and the cocktails to be served in first class. So we are in this transition moment between while we are in the air to like okay, where is my ticket and where's my seat in this, on this plane?

Benjamin Levy: And many will crash and burn in the process.

Brett Waters: Many will crash and burn in the process.

Alex Babin: And remember this old saying, uh, which is partially joke, partially true, that at the dawn of Aviation, CTOs were typically the test pilots. That's how the industry ran out of bad ctos really quickly.

Brett Waters: I like that.

Ahmed Reza: If I could add just one, I can't help but in the beginning of the uh, aviation race um, it's interesting that you bring that up Alex. There were all of these companies that were uh, racing to build uh, their first airplane. And we should take this lesson very very seriously. The guys that got the most, that had the most PhDs and all this, they never launched anything. It was two bicycle mechanics. Right. They're very much um, practical folks that ended up building the very first airplane and ended up like changing the field. So we are very early in this uh, AI race and I think uh, uh, some people are already calling winners. It's similar to uh, the airplane titles like oh, these guys are the best funded, have the most PhDs. That's not necessarily the case uh, as history has shown.

Benjamin Levy: Yeah that's, I think we like to think of DeepMind acquisition as a, you know, generative AGI acquisition if you will. There's a raw talent 400 million maybe. And then you look at Mobileye at the time was a $12 billion acquisition difference between general AI and applied AI. Right. When you're able to apply it in the context and the market and to give it all the bells and whistles product market fit is so key. I said recently nobody buys your software because you have AI.

Alex Babin: That's actually correct and uh, you have a very good point. The most well funded companies in uh, a dominant aviation actually funded by government and other organizations failed. And two guys from a garage actually built it because they figured out how to do it safely. The problem is, fundamental problem is that all the test flights were taking like we've been taking from a high uh, buildings or um, high level. And when the plane was crashing and it's inevitable at the very beginning people were killed and they actually were slowed down. What these guys in North Carolina figured out uh, is that to use sand dunes as a safe way to test the wing and find the right geometry so it's safe to fly. And they were iterating and iterating and iterating and they were not putting the engine on the plane until they figure out how the wing should look like geometry should look like. So that's I actually what I see right now happening with large uh, organizations, they do a great job on kind of sparking the whole thing. But also open source community, these guys are testing and testing until they put the engine off. So it's uh, incredible to see those uh, parallels between those two.

Benjamin Levy: But I think the alignment of incentives and risk reward were also a lot closer back then. You fail, you die. Here people can try a bunch of things and they don't really are the recipient of the byproduct, you can literally

Alex Babin: run out of engineers.

Louis Lowe: Well, we've got founders and investors from all over the world, uh, telling uh, us where they're calling in from in the comment area. And one of the questions that I keep getting over and over and that I have when I meet new founders, uh, in the business that I do is how do you take an idea and turn it into the first dollar of revenue? Because I think that's challenge number one is it's not because there's technology that there's a product. And each of you founders has found a way to turn your technology into a product. And then I'm going to follow up on that with how do you take it from the first dollar of revenue to the tenth dollar of revenue? And, and Alex, I think, um, I'm going to start with you because your name starts with an A. Um, tell me about the first sale you had for Xero and how you turned your technology into a product.

Alex Babin: Um, it's a great question, Louis. Uh, and actually it's been talked about many times, never changed since then. It always starts with beachhead, uh, market or the market, the small enough market you can win over and then move to another market. Um, and in our case it was legal. We went into uh, selling into legal because lawyers, they are knowledge workers and then amount of work, manual work they do and the V like and the cost of time spent on non billable stuff, you know that better than anyone else is incredible. So it's about the ROI on the next day. So we started building AI products to help lawyers to get, to be more productive, do more with less and get ROI faster. And then from that we started expanding to other uh, areas. But the biggest uh, advantage of that and growing the business from that niche market was number one, legal is referenceable. If it's good for Louis, it's good for everyone else. In the top, uh, 100, uh, of fortune, uh, 100 companies. Right. Uh, because uh, legal firms, they have security standards, they are risk averse. And if you can sell to legal, you can sell to anyone. So the, this M and of course the data, incredible amount of data to train AI on and then move to another vertical. So I would say fundamentals never change. You start with a small market that can be referenceable to other verticals and expand and try to win it.

Louis Lowe: Um, Dee, how long have you at Sourcero?

Dipanweta: Um, lots of things. Um, I definitely echo many of the things that Alex said for us. The other driver was this kind, the kind of Content our users use technical, unstructured medical, often in a regulated environment, did not lend itself well to some of the tech available before. So, you know, as cognitive technology evolves unto itself, and particularly in our case, transformers were a big driver of why we chose to do what we do. Uh, we started addressing, uh, problems in life sciences. But I want to be very real. Our customers buy us because we supercharge their workflows. Uh, they're excited that there is AI under the hood and they're excited about what AI can do for them. And certainly ChatGPT has brought more people to our door. It's, you know, my commercial team tells me it's the best BD tool on earth, but once they're over that excitement of uh, oh my God, this could be cool. The question is, how do I use it? Where does it fit in my workflow? Am I, is compliance going to kill me about it? Right, that's the many slips between cup and lip. And so we have been very, very, very, very focused on actually solving the business challeng choosing where to apply AI, uh, to generate our first dollar in the 8th and the 10th and so forth. But that's really the key. Discover the business problem.

Benjamin Levy: Right?

Brett Waters: Uh, one of the mantras that I preach to my Stanford students is great startups begin with a problem worth solving. Uh, and right now I think generative AI is this cool thing that everybody's talking about. But to what extent are entrepreneurs putting it to use on problems worth solving? So Steph was a little bit freaked out by all of your aviation crash metaphors. Uh, and I'm just glad I'm not getting on a flight this afternoon after all that discussion. Uh, but Steph asked the question about when is it safe to fly AI?

Alex Babin: That's a great question.

Benjamin Levy: Uh, I actually know Steph. Thanks, uh, for the question, Steph.

Alex Babin: I think Steph should not be, uh, scared of, uh, flying AI right now. It's just about what plane you're boarding. So there are companies that have been doing AI long time, uh, before I remember explaining to my clients what LLM is two years ago. Now everyone and they dog know what LLM is. Um, but uh, it actually about what? Uh, under the hood. And literally the uh, enterprise clients don't care if it's a black magic or AI. They just need roi and they need ROI as fast as possible and also to be secure, dependable, scalable and so on, so forth. So it's uh, there are a lot of companies that are just rappers around chat GPT. I Would say that type of thing is not very much scalable, and we'll see. A.I. bicycles, A.I. sandwiches, A.I. every.

Brett Waters: Um, you know, we need. Alex, we need an AI juice machine that sits on your.

Alex Babin: I think there was one. They failed miserably, by the way. You know, took a little juice, uh, out of it.

Benjamin Levy: But Pilot has been out there for a while. You've been flying with AI without looking at it or thinking about it. Right?

Louis Lowe: Exactly.

Benjamin Levy: It's already there. It's just not completely autonomous from takeoff and landing. But.

Brett Waters: So Orly asks kind of a related question, which is just differentiation. You know that, um, suddenly everybody has AI in their. In their tagline, right?

Louis Lowe: Yeah.

Brett Waters: So what, you know what. What is. What. What is the differentiation? You know, what makes one AI company more potentially successful in another?

Benjamin Levy: Um, I'm thinking we got Suman online. Suman, maybe you want to jump on. That's early. You're on mute right now. But, uh, if you unmute yourself.

Brett Waters: Welcome.

Benjamin Levy: Zoom in.

Louis Lowe: You got to unmute, buddy.

Benjamin Levy: I was hoping to give him a bit more airtime.

Brett Waters: Yeah.

Benjamin Levy: You know what we need?

Brett Waters: We need AI that knows how to unmute you.

Walter: I quite like it.

Benjamin Levy: We still don't hear you. Some reason. Suman.

Ahmed Reza: Yeah, I think the folks in San

Speaker B: Francisco

Benjamin Levy: and jumping on the differentiation. You know, obviously, we've seen a lot and there's a lot of noise in the market. Um, look, like Alex said, if you are a wrapper around chat GPT, that might become an issue. I mean, we're asking ourselves these questions. We think LLMs will be generalized and commoditized, and they're going to be a lot of different flavors out there. Um, one of the big challenges we see right now with Genai is hallucination. So in context where hallucination is a fantastic thing, great. But if there are divergent models that will never converge and two people get different answers for some questions, I think we're society built on truth or trying to, um. And that is a challenge, um, in the enterprise world for sure. If you're trying to create a new scenario for a movie, fantastic. If you're trying to create an original outreach for a prospect you don't know and you don't have too much to lose because your brand might be trash because you're saying the wrong type of email, that's okay. So it's empowering the small guys in ways that they are careless and carefree. Um, but for the bigger guys, there's different challenges. I think. Um, like we heard earlier, they're expecting a Fully flown, fully secured solution. And most people don't have that. We have a company that has worked on that for years and we believe they have a great solution for the enterprise using these sort of models. But, um, different conversation when it comes to differentiation. Look, we always ask what is a proprietary part of what you do? Do you have access to unique sets of data that others may not? Um, are you creating widgets that capture data that doesn't exist in the wild? Um, are you able to learn from these and create vertical applications, often narrower systems that are extremely good and better than anything you can find off the shelves? And we've seen our stuff time and time over again, beating the big brands in very narrow situations because they have not only this narrow focus, but also this ability to build a solution to the problem. Not just models for the sake of models or technology for the sake of technology. Um, and then we can talk about go to market. But since this is a panel that probably loves prediction, here's my first prediction. I'm saying we're going to see the first billion dollar company with less than 10 employees in this decade. Uh, um, because we not only go from a world of AI first companies, which is what we've been focusing our time and investing in, to AI first companies that are AI enabled and therefore will be able to launch sales, marketing, manufacturing, recruiting, all these things with virtual agents. I'd love to hear what Ahmed is working on. Um, and that's scary and disruptive, let's put it this way. But it's exciting too.

Brett Waters: Ahmed, do you want to jump in?

Ahmed Reza: Yeah, certainly. Um, I guess cat's out of the bag there. So I have a digital clone of myself that's interacting on social media on my behalf.

Benjamin Levy: There you go,

Alex Babin: Talking to us right now.

Ahmed Reza: Yeah, so actually we're working on meetings. We'll be able to do it very soon. Um, so yeah, I figured, how do you make money? Build things that make money. Sales, marketing and customer service. Things that cost me a lot of money. So, uh, in my last company I did sales myself. I basically made my millions using AI, right? And the first time I met Louis, I said, louis, I'm a pretty crazy guy. I use machine learning to bootstrap a startup to millions in revenue. And I've never been able to figure out how to get a second salesperson to do what I do. Um, I, ah, was the only salesperson. And uh, uh, being a computer geek, I'm not that great with people. So I wish I could just clone myself. And then I was like, maybe I can. So I have a deep fake of my voice. Uh, and, uh, we have, uh, this engine called the goal oriented conversation engine that's able to be a better SDR than anything else I've seen. I'm overbooked. Uh, if you look at my social media, if you look at our inbound, uh, we were at near, uh, zero revenue last year. Uh, and in less than six months we've blown through our second million arrangements. And it's insane, like just being in the driver's seat watching this thing, it's like, damn, this is real.

Brett Waters: Yeah. Are you talking to the real person

Alex Babin: right now or a digital quantity? Reveal yourself, please.

Ahmed Reza: Yeah, the clone would get much better answers.

Brett Waters: When you, when you can create a deep fake that'll participate in, uh, you know, live stream panel discussions, then we'll be impressed. So let me ask you guys a question. One of the criticisms of large language models is that it's kind of a brute force, guess the next word approach, right? So it's not super, super elegant. Um, I guess I have two questions. One is, is that an accurate way to characterize it? And then second kind of, where do we go from here?

Benjamin Levy: M. Walters, you want to take that?

Walter: Oh, yeah. So, yes, um, we're just waiting until the Net chats commercial was finished. Um, so. Well, it's not so easy as uh, uh, as we say, because now people always, you know, like, we made many mistakes in the past, you know, since Chomsky to now, uh, and perceptron and the different layers and the inability to understand, you know, reinitializing of, of LLMs. But, uh, now that we are where we can basically most LLMs have like a trillion words, um, which is an interesting, also an interesting question. I, I think I know where they come from. Um, and, but these words are now, if you put them together. Um, so they basically operate in a, you know. No, let me, let me rephrase that. So lately I've been thinking a lot about, uh, how we, uh, educate our children and how we educate our large language models. You know, when kids are like, uh, uh, two to three years old, we already treat them as diffusers and transformers. You know, like, uh, we already tell them like, uh, you know, draw something. So that's the diffuser, you know, like, which colors are you using here? But these are not the right colors. You know, like we're prompting them into at the same time we are telling them like, right, uh, house, you know, what is the next letter? You know, and then we try to, to make them you know, and have ah, a sense of language. Because the, the mission of our pre cortex is to predict the next second, you know, it is the reduction of surprise, you know, otherwise we will not survive. And that's also what we tell our children. Look, look right, look left when you cross over, you know, stranger danger, you know, like we, we teach them all these things. Now it is true that language model, dots, you know, uh, uh, birds was in both directions. Now the language model goes you know in a transformer to the, to the right. It, it predicts the next letter but it also predicts the next word and it also predicts the next sentence in the next paragraph and the next chapter, you know. And, and this is all in context learning with some prompt engineering which will now be because I hear, you know like people, people are obsessed with the no code. You don't, you don't need to know anything anymore, you know, like uh, but that, that will quickly change because we've been there before. You know, people are now already you know, uh, doing function calls and uh, and injecting a lot of Python in their prompts.

Benjamin Levy: So I think a lot of parents are really believers in reinforced learning though.

Alex Babin: Really resonates with Walter said Sometimes I feel like my kids are ne Networks train on YouTube videos but uh, and Netflix. Uh, but actually what Walter touched is really interesting. Um, and uh, of course large language models as we know them and Transformers are critically important but it's really commoditized very uh, much right now. Ben said before uh, and just wanted to kind of touch on a different topic. We all talk about what the state is right now. We're not talking about what the future is. For example, we are working as part of our deep tech thing. We're working on Action Transformer. So what Walter is saying, predicting what uh, the next step for the user would be in the process of uh, performing their work. That's a pretty incredible piece of model. Um, we already have working uh, of course we can predict 1, 2, 3 steps when the user uh, goes through the process to accomplish a business task. But in the future it might be much more which is in the very beginning of that process. But it's incredible if uh, people think that large language model feels like black magic. Oh you should have seen this kind of things. And uh, there are so many possibilities opening up when you start combining those two things and there might be third opportunities uh, showing up. So I think future for large language models is going to be not in one large model rule at all. But the family of Smaller models that are highly specialized, that kind of being connected by a larger model to do kind of a general reasoning, but also those specialized models like Action Transformers and others to actually provides much more than what we can do right now.

Benjamin Levy: The really looks interesting right now, if I may share. In 2017 at our conference we had one of our advisors starting to map a path to AGI and when we ever get there and do we want to get there, a different conversation. But what was interesting about that roadmap to AGI is that we went from narrow discrete AI systems which was very much 20, 17, 18, 19, those narrow systems to a uh, vision of constellations of systems of AI systems and I think the coordination of these constellations of narrow AI systems and orchestrations of these systems which uh, is very much what we're seeing with kind of baby AGI and autogpt and other things. Right. This is where we're going in coordinating discrete, really strong models in different areas. To start coordinating these things is where I think we're going to have a leap of, you know, an order of magnitude improvement in productivity. Um, and what the impact of that will be on society. Again, I think you need to be a uh, thoughtful investor.

Walter: Yeah, I agree. What we're doing actually I really like baby AGI, you know. And uh, so it's now uh, sort of called the auto GPT category, you know, like. And uh, I really like it. Although you had to have to handcraft these agents a little bit now. But it will be better. And now certainly with the introduction of function calls it will be better. But um, you know, these agents get stuck. These agents get stuck in, in decision making or like they're waiting for somebody. But you know, like this is the same with students with employees, with uh, you know, with executives, you know, they get stuck, they wait for somebody else, then you have to either give them a pep talk or threaten them in a way, you know. Um, and, and uh, the same thing we are now trying to do with these agents. But it's an interesting thing, you know, if you are now setting up a company and we added the, you know, um, uh, Ben, you said about the tin wrapper, you know, um, uh, the thin, the thin wrapper around LLMs. Is that really a company? Uh, well probably it isn't, but it can be a, you know, it can be a lifestyle company, you know.

Benjamin Levy: Right, right, exactly. And we're seeing a fair amount of those. I mean you go to millions of dollars in revenue. Can you become a category defining company that's much challenging Much more challenging.

Walter: And uh, so quickly.

Brett Waters: Hate to interrupt but I just want to see whether Suman's audio is working. Working now. Come on.

Speaker B: Is it working?

Louis Lowe: Perfect.

Speaker B: Well, so I had about 20 minutes of prepared remarks. Get in there. Now just to hit one point, I think uh, a few of us have seen these different paradigm shifts. So you know, first, uh, one for me was the Internet, right? And there was a period of time where people talking about the widgets and then there was uh, you know, 2009, 2010, um, people talking about the cloud. Um, and so I think we're in the same type of era. Like if you sort of stay away from talking about the macro things, I think um, there's just a lot of experimentation that's happening in this broad AI space. And so very kind, um, of top down trying to predict what's going to work, what's going to be safe, what's going to kind um, of be the killer app. It's really hard and I think you need to have this portfolio approach, um, where you know, through trial and error you're fortunate to stumble into companies like Zero, right, Or uh, Sorcero. And uh, and there's just, you know, a couple anecdotes. I think Alex had a social uh, media post where he had this like runaway suitcase. It was like an AI suitcase and it was like runaway and like. And it's very true. It's, it's. I think through, through trial and error we're going to find these like edges, use cases where things don't work, but we are focusing on the ones that do work. And when these, you know, we discover these AI white spaces and we find these companies that are creating products that do work. I think the upside is so tremendous and I think that's, that's the excitement, that's the enthusiasm that I think people are feeling. So, so I think, um, anyway, so that's, that's a bit of an anecdote on how I look at kind of the space right now and where we are, um, and how a lot of the companies that we're seeing fit in actually worth unfolding.

Alex Babin: Suman, you mentioned something that is ah, anecdote but actually true story number one. And second, it's a whole topic to unfold. So you mentioned this suitcase, uh, with uh, AI technology and the tracker that the person who was in the airport, um, um, wanted to be in a luggage and uh, they forgot to load it into the airplane. So the guy who got into the airplane and airplane was taking off and the suitcase, uh, was trained to actually avoid obstacles. So it found its way to the tarmac and as a video of the suitcase chasing the airplane. And that is funny, but actually creates a lot of problems because that's potentially a um, situation where it can have a collision with the airplane or whatever. So AI is uh, at the stage where those safeguards and guardrails and everything should be in place because as many people say, it's like putting the nuclear reactor in the middle of the city, pushing the button to see if it works. It might be, but what if it doesn't? So uh, Walter brought up example of baby AGI. How about chaos GPT another example when uh, the same uh, free agent was trained to actually create chaos and uh, destroy things. And the first thing it tried to find how to build a nuclear bomb. And when we couldn't, the second thing it did, it went to open the Twitter account and start tweeting the misinformation. So it's a um, weapon, um, of mass destruction. So there's so many components in this worth unfolding as well.

Dipanweta: Yeah, I think what we're finding as well is because AI is so relevant now and because it's in everybody's homes, our customers who might not have ever used even barely software before are going to come to us and say I want AI. They're not even asking for a solution, they want to buy AI. So uh, in the enterprise software space there's also a risk of salting the field because there'll always be just about enough people who will sell you AI, which will do nothing for you, waste many tens, if not hundreds of million dollars in several years, and will also get a set of enterprise buyers really skeptical about this new fangled tax. And you do see that happening over

Benjamin Levy: and over again, poisoning the world. Yeah, it's like Walter mentioned, I enjoyed the summer of AI, but let's remind ourselves they were winters and more than one.

Dipanweta: And so you know, just having that hyper narrow focus on what is really deliverable with what we have today with safeguards, with measurements, with benchmarking still remains extremely important. It's slow and it's patient, but it needs to be done. And at the end of the day it still needs to fit into someone's workflow to change it and for them to keep buying it. They might buy it once because it's hype, but they're not going to keep buying it because if it doesn't do something for them. So sure, the first gut. But I'm, I think we're also really looking at what's the second cut of that going to be? What's going to keep them coming back? And it's unlikely to be the hype.

Brett Waters: So, so dependent. I talked about second, uh, cut. And many of you have kind of used the analogy of aviation crashes and nuclear bombs to refer to the fact that we're in a zone right now where we're not sure what works and what doesn't work. And then Suman said that for him, as an investor, it's about taking the portfolio approach where you got to spread your bets around because you don't know which one of these things will end up being successful. But Alex and. And Ahmed and Dipanweta, uh, for those of us who are entrepreneurs, we don't have the advantage of being able to make, you know, have a portfolio. We. We need to go all in on something and build a company around it. That's one of the many ways. Many ways in which VCs rule the world is that they have the advantage of distribution of risk.

Ahmed Reza: Yeah. When you look at what's in front of you right now, I think it's a safe bet to just bet on a Bengali.

Dipanweta: I'm not gonna say no.

Ahmed Reza: All right, there's the investment pieces, romance. I just realized, like, there's three Bengalis on Nai, uh, here. Imad Mushtaq, who's also Bengali. Estivale. I was like, wow. So I've been talking to more folks, and this, for me, honestly, from. From a human perspective, my, uh, mom has never known what I do. You know, for all she knows, like, I'm in the mob or something. She just goes, what do you do? There's no buildings with her name on it, you know, and you're not a doctor, so you're obviously not successful. Uh, so. So this is the first time she understood what I was doing because it speaks Bangla. Uh, and I had a long conversation with, uh, my mom and with certain aunties, and if there's any saving grace for AI, it is that our aunties will kill us if we don't build, like, the best AI that brings home aids.

Benjamin Levy: Well, I'm, um, going back to what Walter said, uh, the vision of the future said, well, that's interesting. Maybe the, you know, everybody wants to know what the. The skills necessary for the future. I was like, well, what if you were to say that you were a therapist for virtual agents, because they have their breaks down, they have waiting, they need their pet talks.

Walter: Well, it's a very interesting, uh, at Stanford, we have Simulcraft of Human Behavior, where we put 25 agents in a city called Smallville and we give them identities and then we study them with 100 moderators, uh, students. And these agents have their, uh, job, they have their wives, their families. They go to work, they come back, they reflect and, and we can even get in there as an inner voice and uh, direct them or embody them. Uh, we have to.

Benjamin Levy: Can you make a summer camp for. My kid is 14. I'm sure he's gonna love this.

Walter: I think for investors and for um, entrepreneurs now, everything is going extremely fast. So as an entrepreneur you always have to go a little bit in the future. Not too much, you know, and try to time it. But timing now is impossible because we are now, uh, you know, every, every hour something new appears. So what can you do so that you can set up a company that is future, uh, powerful, but that now also survives, you know, without being taken over as a sort of a feature inside OpenAI, you know. So I think uh, that um, uh, and that's why I like baby AGI, because most, uh, CEOs and I do, uh, you know, I talk to ah, quite a number of, of boards now for the first 20 minutes of every board meeting is now AI, you know, and um, so they. And let's take a smaller, uh, a smaller public company, 2,000 people. Well, the CEOs probably now have already found out that they should have 10 teams of four people and fire all the rest, you know, because, um, and, and refactor their, uh, their complete thinking and, and rethink, uh, you know, like David Deutsch said, every point is now a boundary point. You know, like the frontier is everywhere in that company. You know, like, so, you know, like what you have to do, I, I think is the following. Um, you do something very far out in the future, and far out means a year. Uh, you know, like baby AGI or something like that. At the same time, in order to do baby AGI and in order to set that up, you need to know what, in context, Learning is, what LLMs are, you know, the complete orchestration platforms. Uh, so you can have clients from day one. Because I believe that nowadays if you are working in AI and you don't have a client, and a client means a paying, paying person. Uh, if you have a client and that makes you progress so much, first of all, you have a delivery to do, you talk about that client, uh, and you learn from it. You keep the code base, you build on the code base, and so you orchestrate your Code base because you are going for baby AGI. So I think it's now bootstrapping is, you know, because the AIs are there and we can generate other agents than to do stuff and kick them in their ass when they are stuck.

Brett Waters: So Louis, you are not an IP attorney, but you have some partners who are IP attorneys. So, you know, how about this whole concern about, you, um, know, the large, uh, language models are basically sucking a bunch of other people's thoughts off the Internet and repackaging them and spitting them out. And uh, so Orly asks the question about, you know, if I adopt, if I have a company and I adopt some of this technology, is there copyright risk?

Louis Lowe: Yeah, no, it's a great question. And it's the subject of some pretty tough, uh, litigation right now. And I think the best example of it is Getty Images, uh, which makes its pictures available online. But if you want to download them or save them or use them in a presentation, you've got to pay a fee. And uh, obviously many of the um, visual, uh, AI engines are, are receiving these Getty images into their flow and then they're, and then they're creating thing, they're generating AI, uh, images on the other end of the engine. Which, which then begs the question if, if these copyright, um, protected images were the inputs into the AI engine, do they have some right into the outputs of the AI engine?

Brett Waters: Exactly.

Louis Lowe: That is not a question of settled law. And, and uh, I've been asked by many investors whether we would give a legal opinion, uh, on, on the, the IP infringement characteristics of a particular uh, AI model. And um, they're fascinating mandates, uh, that you know, we examine on a case by case basis.

Benjamin Levy: Brett, I mean, I give you. We thought a lot about that as well, Louis, obviously. And um, you know, one side, I mean, interesting conversation, but we all went to school, we all read books, and if I try to write an essay inspiring myself from Victor Hugo or anyone else, uh, am I plagiarism? Is that plagiarism? No. Can someone sue me for it? No. Um, now, where these image made available and shouldn't have been made available or were used in a way that was not permitted. That's different conversation. Exactly. You know, and I think I can see a day where maybe OpenAI would be subpoenaed to list every single source you know, they've used to train their models. Um, you know, we're not there. They'll fight it to death. But um, yeah, at some point people that have a lot of money and feel Their business is being disrupted will throw a lot of money at the problem because it's there. Is there everything? Right? Um, yeah.

Walter: Look at the solution. You know here the, the question is where do all these trillion words come from? You know and is there copyright on these words? But uh, you know the uh, and it's basically the answer is very simple. It is some is that you know like some is that is Lip Gen and Saiha up. They are in Russia, you know, they, they copy everything and you know all the students are using, and all the academics are using them because why, you know we write these articles, we do the research and then uh, we uh, we go through the examiners and then when the article is published we have to pay uh, a ridiculous amount to Elsevier. You know, so everyone is using these trillions of words in, in the sum is that uh, databases of Libgen and uh, and Sci Hub. You know it's a, it's a well known secret. Uh, so that's where it comes from.

Brett Waters: So, so Sudipto asks an interesting question which I, I think the essence of his question is if you're, if you're a company who wants to incorporate more AI, uh into your offerings for your, especially if you're a non tech company, do you need to build an internal team around this or uh, use uh, external AI providers? I think is the essence of his question.

Benjamin Levy: I don't know.

Dipanweta: I'd love to go ahead.

Brett Waters: Yeah.

Dipanweta: The experience we've seen. So um, many of our target market have fairly sizable AI data science teams internally. Some of them were set up years ago with the intention of building tools, bots, other things that would address their business needs. What most of them discovered is they're not product companies. So yes, certainly, you know, every once in a while there's a vaguely usable model, there is a chatbot that does a thing, but it definitely does not map to the amount of investment and the pace of development even a tiny little software company could come up with because they are a product company. Right. So that is certainly becoming a realization in the market. I would say what we talk to our customers about is you should have folks on your side who understand technology and who have the ability to evaluate and frankly ask the right questions. You know, whether it's security, usability, ROI question, but maybe not try and make the build decision for most things. So there needs to be a marriage of technical understanding with business need understanding on the buying side. But we generally see that if somebody thinks that they're going to build their own LLM and actually a potential customer told us that we're thinking about building our own LLM barely knew how to respond to that. We generally say no, actually, uh, use

Benjamin Levy: the product like Walter. We spend some time obviously in advising boards and talking to very large corporations about what's happening in the market, marketplace and with AI and, and the analogy I like to use for people because you know, we heard someone talk about, you know, Internet and cloud and all this big wave of disruption. I think back to what you were saying, Walter. I don't know if we can build companies that will last hundreds of years anymore. I don't know if that's where the future is going to be like. But the peak of the market and the waves are higher each and every time. They build on top of each other and you can monetize them in a shorter period of time as well. It's like the movie industry. The only reason it was taking so long to monetize the movies because of the rights of the contract, not because we could launch everywhere on every screen across the planet. And so you think that analogy maybe will create a lot more wealth faster and then it move on to something else. I don't know about the cycles, which a lot of LPs would love, right? They hate the fact that VCs is a ten plus year business cycle. Um, but, but the analogy I make for corporations and for them to understand the disruption they're facing is this. If you made a lamp, if you manufacture a lamp lamp for hundreds of years, that's how you've made your core business. You manufacture lamps and all of a sudden you have to learn to sell it online. Great. You have that, you manage, it's the same lamp. Your supply chain didn't change. The way you manufacture didn't change. Now you and mobile came along. You use mobile, you learn to sell on mobile versus Internet, then you use VR because it was cool. You can visualize your lamp before you even buy it. Great. Again, didn't disrupt the core business. But now you need to manufacture a smart lamp. Well, what does that mean? What it means you need to have people in house that understand AI. You need to actually maybe change your business model to give the lamp for free and save money on the energy and split that with the user. You need to rethink your entire core business and that could be worth for product or service. So to answer, you know the question earlier, you need to think about that. You need, what does your core product and service looks like when it's AI, embedded AI, smart Right. Because again that's what's going to disrupt you. Right. And that takes in house capability as well as adopting innovation faster. For everything else that's not core to what you do.

Brett Waters: That's good, that's really good.

Speaker B: And just to double click on that really quick, a great book. There's a lot of AI books out there now, but a book that predated the whole LLM and kind of chat GPT. It's called uh, Competing in the Age of AI by Marco, uh, Ian City and, and premium Lakhani HBS professors. Um, and I think it lays out a pretty good framework for how if you're an AI company or not AI company, how you should kind of think about that. It touches on a number of points that Ben mentioned. So I definitely call that out.

Louis Lowe: Um, Ben and Walter and Suman, you've each uh, invested in a pretty broad swath of AI companies in the market. And uh, I think we have an awesome panel of entrepreneurs who've actually taken a technology and brought it to market and gotten to way over a million, in many cases many millions of dollars of annual recurring revenue. Um, but there are a lot of AI companies out there that might be um, hyped and that might be in the market looking for capital, who've raised lots of capital and burned lots of capital. What do you see as the big challenge for those companies to, to, to succeed and make it to the next level? And, and what are you seeing in the market as the kind of the next big obstacle for AI uh companies to overcome?

Walter: Well I think it's uh, it's always the same thing like uh, you get your uh, uh so you know VCs like us and like uh, you know we give early, you know, early seat money and uh, we give them money and then we, we wait a year and in that year we want a proof of concept, we want the product and we want clients and we want to see growth. And entrepreneurs take that money and they are convinced that they are going to be there at the end of that year. But then they got lost in traffic, you know like in uh, you know like they are, they're uh, you know, you know especially start to use all these fancy words from finance and, and super voting rights and by the man by the end of the year then they don't have a uh, product or they don't have a client or they don't have any growth and, and then they come back and they say can you help us fire our co founders? So it hasn't changed it's been for 30 years like that. It's not with AI. But this changes.

Benjamin Levy: No, I mean I go back to the same thing I tell founders, look, as there's no shortcut in life, you need to create value so you can raise $20 million over a weekend on $80 million pre. If you're well known founders that investors know and believe you're going to build the next big thing. But that gives you just a bit longer Runway to create a value that's going to be 2x that. And so in the end, again you need to deliver and create value. If not, you're going to fail miserably. I mean, you know, I still have LPs that asked me oh, uh, you know, can we invest in OpenAI? I'm saying no, that was a foundation to begin. Which by the way that's more of an R and D organization. A lot of that money is, is on stream shoe budget. Uh, but for them it doesn't care. They don't care because guess what, Microsoft makes all the money on the cloud. This is a distraction. They're selling picks and shovels to people to process in the cloud and make money there. Right. So you got to look beyond, you know, the image and perceptions, the opportunities

Walter: in the market are now all the large corporations, um, that normally, you know, a corporation is large enough, has money enough to hire people to do that internally. The problem is that all the people they want to hire want to set up their own company.

Benjamin Levy: You know, we love that as VCs.

Walter: Yeah.

Brett Waters: And that's what has made Silicon Valley great.

Walter: Yeah.

Brett Waters: Guys, we are just about, we're just about out of time but I'd love to quickly go around, around and hear from each of you a particular use case that you're excited about. Um, some use case that you think really is going to be profoundly valuable that AI, generative AI, uh, will provide. And Suman, we'll start with you since you got cut off at the beginning. Give us an example.

Speaker B: Well, I'm a shameless promoter of the companies that I invested in. So perfect. I'm obviously excited about Zero Systems. I think, um, generative AI and broader AI applications within the enterprise based on proprietary data that companies own and want to manage. I think that's a huge white space, green field. Um, and the company's been amazing in its growth. So that's one. The second is a company I'm with, ah, Benin. It's a company called Hayden AI and I think that one's tackling computer vision use cases that we didn't really talk a lot about in this chat, but I think once you have that data, there's a lot of interest things you can do with it in the city. And uh, that's one that nobody would have thought of, uh, probably in the front end and it's just evolving amazingly well. So I think those are two and then a third that I'm interested in is in the, in the financial vertical. So if you're, you think about companies like Bloomberg and all the data they have, I take that analogy with ChatGPT and all the finance information that's available, I'm on the lookout for an interesting company in that space. Um, and so you wanted one. I gave you three.

Brett Waters: Excellent, thank you. Under promising and over delivering. That's what it's all about. Dipanwetta, how about you? What use case do you see that you're super excited about?

Dipanweta: So there, you know, we use AI services across many of our product features. There are three that we're seeing is really exciting. One is supporting the generation of plain language summaries or patient lay summaries which usually accompany a clinical trial report or a poster. So making sure this medical information which is out there, which is mandatory, is more accessible both to patient advocacy groups and physicians. Really, really critical, much more scalable, doable with some of the generative AI features. We also are supporting our customers in uh, understanding or finding indications on next best action. So I've processed all these insights from all of my data, from all of my data sources. What is it telling me about how do I handle dosage better or what it is telling me about a trend emerging in an evidence gap with a physician. So we see these two places as very, very exciting within our product portfolio.

Brett Waters: Awesome. Alex.

Alex Babin: Actually there is not just one case I'm excited about, uh, especially when talking about AI, but the pace at the, at which those use cases appear. And uh, recently we just discovered that one of our largest uh, clients, it's one of the largest insurance companies in the United States, they were applying our solution, one of our AI apps that we've built on top of our engine, to one of the business cases of theirs. And the savings are $3 billion a year, just blew my mind away. And we never thought that's possible, but that's how those use cases work. And this uh, this happens during, uh, during the process of discovering use cases and building more use cases. So I think there would be a lot more potential to be discovered across the whole industry and many other clients as well to see how all of them gonna be disrupted? There's no one use case that would dominate everything. It's so many of them that it would be hard to choose which one is much more important.

Brett Waters: Okay, Ahmed.

Ahmed Reza: I think it's an extremely exciting time to be alive, especially if you're a geek. Uh, this is like imagining what computers would be before computers came about. I think computing fundamentally will change as you can speak to a computer in a much more natural way, which is where, like, we're leaning into the synthetic agents so that you can mentally model another person right, as your interface. So that's what we're really thinking. Um, uh, that the large language models afford us is this ability to speak naturally to a computer. There's tons of compute resources, and I think it's going to be transformational at the level of the industrial revolution.

Brett Waters: Nice. Walter, you told us about Baby AGI. What else?

Walter: I think for me it's all about media and entertainment.

Brett Waters: Media and entertainment. Um, and you're in Malibu right now, right, Walter?

Speaker B: Yep.

Louis Lowe: Yeah.

Brett Waters: Here we go.

Walter: Because, uh, this were, you know, this, uh, you know, in.

Brett Waters: In.

Walter: I'm not talking about the zip code, but in Hollywood, in. In the sense that, uh, um, you know, media and entertainment industry is still an industry where people do business by phone, they text and they send you DocuSign, you know, like, uh, there is no real computer science optimalization there. So think about scouting. You know, like, think, uh, about scouting for locations. Think about, uh, props. You know, this can all be software. You know, like, I, uh, want to see a movie in the future that I can prompt. And I say like, okay, instead of, uh, Tom Cruise, put me in there. And, uh, you know, and, uh, and uh, this will all be possible. The composability of music and of. Of music and of movies are, you know, a couple of years away. In two years from now, actually, the Grammys just announced the Minimis, you know, their new. So AIs, um, can, uh, only win if 20 of humans collaborate with them.

Benjamin Levy: Okay, it's a. Ready Player one, here we come.

Brett Waters: All right, Benjamin, you get the last word on this.

Benjamin Levy: Uh, what? Thanks. Um, I had the benefit of listening everybody else, but here's what I say. I think one area that I'm very passionate about is empower people to reach their full potential. Which is, I believe, what we try to do as venture capitalists with founders. And to do that, you need to give them access to knowledge. And so acquisition of knowledge is interesting and important as human, as society. And so we invested seven years ago in the founder of in the core technology behind Alexa and he built a new company called Brian. And this is not a gen AI per se. They are using geni elements and large language models to sensitize and summarize information accurately. And so I think you need to start building solutions with the intent to be trustworthy, to be transparent, um, to be, you know, remove as much many biases as you can because that's the world we want to see. And when you take this kind of technology and you map it to any sort of information that enterprise has and able to deliver knowledge in an instant, I'm not saying delivering link to information that you need to read and assimilate. I'm talking about sensitizing knowledge and assimilating faster with source to a truth. I think that is going to just unleash billions of dollars in productivity and that's really exciting to be part of.

Brett Waters: Very nice. Louis, take us home.

Louis Lowe: Um, I'm just so excited, uh, and motivated by this conversation, uh, Brett, and thank you so much for bringing together, uh, four rockstar founders and uh, uh, three rockstar investors who uh, are bringing this to reality. Uh, I couldn't be more optimistic, Brett, no matter what happens in the world when I'm surrounded, uh, by people like the folks that we brought together today. So, uh, thanks for doing it. Brett, over to you.

Brett Waters: I totally agree. Uh, this has been a great conversation. I think we could go on for hours.

Speaker B: Hours.

Brett Waters: Um, but I'm sure all of you have better things to do. Uh, Louie and I, all we have to do is go out to lunch to have some tacos and we're good. Thanks everybody. Really appreciate it.

Benjamin Levy: See you soon. Thank you everyone.

Louis Lowe: Thank you.