WNNR Andretti Acquisition Corp

 

Filed by Andretti Acquisition Corp. and Zapata Computing, Inc.

Pursuant to Rule 425 under the Securities Act of 1933 and deemed filed

pursuant to Rule 14a-12 under the Securities Exchange Act of 1934

Subject Company: Andretti Acquisition Corp.

Commission File No. 001-41218

Date: February 8, 2024

 

 

Full Transcript of TheStreet Action Alerts PLUS Podcast with Christopher Savoie, CEO Zapata AI

 

Intro: Welcome to another episode of the AAP podcast. I’m Chris Versace, lead portfolio manager for TheStreet’s AAP portfolio, and I'm pretty excited about this episode of the podcast. And I say that because we all know that through 2023, artificial intelligence - AI - was a very big topic particularly for the magnificent seven group of stocks. And as we've moved into 2024, we've heard from Microsoft, Amazon, Meta and others that they're going to continue to invest in AI because of the disruptive power that it has, the new business models that can be brought forth and how it's going to change the way we work. Well, in thinking about all of that, I'm very happy that I was able to sit down with Christopher Savoie, the founder and CEO of Zapata AI. Zapata has been working with AI for industrial applications since 2017, and Christopher Savoie shares how his start in molecular biology and working with machine learnings for sequencing gave rise to the formation of Zapata in its use of rich analytics with real time data and large language models to address customer specific problems. Our conversation turns to generative AI, with Savoie explaining how it can drive productivity with diverse examples, ranging from engineering and construction to financial products like annuities and even race car driving strategies. We then discussed why generative AI is the latest in a line of tools, much like the internet, that will alter how we work, driving productivity along the way. Savoie also shares why GPT 4 may not be the product some are hoping it will be and why multiple, smaller language models working together, not one model “to rule them all”, will help companies address their specific problems and needs. Now, please sit back and get ready to enjoy my conversation with Christopher Savoie, the founder and CEO of Zapata AI.

 

Chris Versace (CV), Action Alerts PLUS: Well everyone, I am super excited to have Christopher Savoie join us from Zapata. But before we get started, just a couple interesting tidbits about Christopher. Not only did he found Zapata, which is, of course, focused on the industrial AI, but he also spent time at Verizon, Nissan, and as you'll find, he has an interesting history with a Apple iPhone feature known as Siri. Christopher, thank you so much for joining me today.

 

Christopher Savoie (CS), Zapata AI: Thank you for having me, Chris. Appreciate it.

 

CV: So, you know, I'm very excited to talk to you guys because, you know, with Zapata, you're really focused on AI and, of course, there's been a lot of conversation about that, a lot of buzz, if you will, really over the last year. And I think we're starting to see an inflection point where companies are moving from the hope-ium of AI, right, to actual applications. We're hearing more about realized productivity and I want to get into that with you, but if you could, give us a thumbnail sketch on your background. I gave it, you know, way less than that, but it is impressive.

 

CS: Yeah, a lot of five civil words in the background, there's probably … I started out as a molecular biophysicist by training, working on immune molecules. Structural biology is another word for that. Basically, taking molecules that are important in medicine and figuring out how they're structured and how they work. Did that in the wet lab, meaning doing mass spectrometry and things like this like lab experiments. I was a lab rat for a while with a white coat on in that early part of my career and in Graduate School. But quickly got into the notion that biology is really about sequences, which means it's about sequences of letters, letters of amino acids, letters of DNA. I'm sure all of your audience has heard about that, you know, and it really became pretty obvious in the 90s when I was working on this stuff that understanding biology, understanding how to create biological drugs, all of this kind of stuff is really about understanding the sequences, right? It's about understanding what sequence does what, and this became about, you know, letters, letters, alphabet letters. You know, they're nothing but like … they're just letters like any other letters, right? And the sequence of those letters matters. You know, you put a certain sequence of words, it's in English, and you put, change the same letters around and you got Italian, right? So, really it is about understanding and doing the analysis behind that. So, this is where my interest kind of burgeoned in AI and machine learning of really understanding sequences of letters initially for the purposes of biology, but as you've kind of alluded to, later on for the uses of understanding the sequences for other purposes. And we've seen that recently with these large language models with ChatGPT. It's about understanding those sequences of letters and then producing outputs that are creative and new with this generative AI.

 

CV: Yeah, so help me understand how you went from that to help developing some of the bases for Apple’s Siri product and then moving to Zapata.

 

CS: So yeah, so I was in Japan and doing my biological research. I had been working on machine learning applied to vaccine design, if you want to call it that. So, finding out which epitopes, which sequences of letters will give you the most immunity or something or less immunity, or different immune reaction. Really trying to understand what's behind having this sequence versus that sequence. Having this letter combination versus that letter combination. And another graduate student, Babu Ramanujam, who now runs AI over at Cognizant, was working on artificial life and this agent-oriented kind of work. Having agents that that symbiose, that talk to each other and are kind of artificial life. And by combining these two concepts, we came up with the idea that we could use an agent-oriented system to parse language without worrying about the grammar. So, this engine would see something in the input that it likes, another agent would see something else in the input that it recognizes and likes, and the two agents would talk together about what to do about it. It’s am oversimplified, overly, you know, dumbed down way of talking about it. But basically, it was about taking this sequence analysis and distributing it over a bunch of in a multi agent system. So, in some ways, each one of these agents had what we would call today a small language model, recognizing

 

 

a little bit of the input that it understands. So, to give you an example of that, if you had something like play the DVD player, or play the Blu-ray player, or something like that, you would have a DVD agent that knew DVD; you'd have even a Blu-ray agent, that knew Blu-ray. But if you just said play, it would be able to talk to the Blu-ray agent and the DVD agent and say you got two options here and would be able to create a menu, which one you want? So that's kind of the whole basis behind it. It helps reduce the ambiguity because you have these multiple systems that can kind of talk to each other and resolve the output.

 

CV: So then, you know, if you're going down that path and we've seen, I guess, voice agents or virtual assistants kind of come about, why the pivot or you to AI and Zapata?

 

CS: I think, you know, it's all been about machine learning of one type of another or another, right? It's about using machines to take the data that's available and then understand inputs and create outputs. You know, that's really been the, I guess, the focus of my interest in my career throughout the years is, OK, whatever the input is, it could be a sequence from biology or it could be, as we're doing with Andretti Global, we're doing racing, which is a bunch of inputs and data from sensors on cars. To me it's all input and you want to take that input, understand the world using this machine learning in a way that kind of humans do. We memorize things and we, you know, combine different things and then we get new insights on that data, but we can do that at a much larger level with the computer because we can store more information at one time than one human could keep in their mind, you know. Try to keep an entire iPhones worth of telephone numbers in your head and memorize them and then regurgitate them quickly. I can't do it. I can't even remember my own phone number nowadays.

 

CV: I was just going to say my, you know, my memory because of the iPhone, particularly for phone numbers, is simply horrible now because I think names or nicknames and I press the button, you know, whether it's a call or to message somebody off it goes, you know? It’s far, far from when I was a kid and I had to remember, you know, streams of, you know, seven, sometimes 10 digits. So …

 

CS: Yeah, I don't know my kids phone numbers if you ask me now. I don't. I know where to get them on the iPhone, right? And that's the thing. So, computers can store that information and readily access that information better. So, they can do these some of these rote memorization tasks and these learning tasks better than we can because they can have more information from more different sources and understand insights across these different variables better. And to me that's powerful. And that was powerful for biology. It's powerful for understanding analytics and auto racing, for supply chain and all of these other kind of industrial uses of this kind of capability.

 

CV: Well, let's get to that in one second because, I think, the vast majority of people, you know, they've read science fiction. They might have been familiar to some degree with artificial intelligence or AI but for a lot of folks, it was really almost a year ago when NVIDIA CEO’s kind of came out and, you know, talked quite a bit about the future of AI and what it would mean and the amount of processing power and why, you know, their GPUs – and GPUs from AMD and others – would be in high demand. But what’s interesting to me is Zapata was formed years before that. So, when we think about that, what did you see at the time? Has AI developed the way you thought it would? Is it accelerating? Help us kind of understand your perspective on what's happened in these last few years?

 

CS: Yeah, I think this gets to kind of the origin story. So, you know, we spun out of Alán Aspuru-Guzik’s lab at Harvard University where he was working on this quantum math, these quantum algorithms. And a lot of times when I say that, people think, ‘Oh well, oh, so quantum computers, things that will run’ … Yeah, eventually they'll run on quantum computers. But what we're talking about is really high-end math, linear algebra. This is a problem when I talk to investors in the normal public because if you say you're a linear algebra company, people have no idea what you're talking.

 

CV: I think, you know, the reality is … I doubled majored in math and economics, right? But it's been a long time since I've had to, you know, think on the math side of things. I think a lot of people, you just traumatize them with the word algebra.

 

 

CS: Oh yeah and then I start saying things, ‘Well, what I mean is matrix multiplication.’ Then they go even further with their eyes rolling and they're like, ‘Well, get me out of this conversation, please.’ You get the allergic reactions to math happening pretty quickly. And then you say you want to clarify, ‘Well, what I mean is quantum math.’ Now we're in, you know, Mars territory. And they really, they're looking at me like I'm a Martian and I have five eyes and come from outer space. So, you know, but that's in essence what we're doing is better math. I think if I wanna dumb it down, it's linear algebra. That's what we do. Linear algebra, what is that? Matrix multiplication. That's what GPUs do really fast. So, you know, we were using linear algebra in that lab, which is a chemistry lab to understand chemistry. Well, that doesn't sound very hard, does it? Well, until you realize that chemistry problems are these big massive statistical problems in the quantum space, right? When we talk about what is quantum, well, it means that this electron in my fingernail here is not just in my fingernail, but it's also quantumly on the other side of Andromeda, on the other side of the universe. It's less there, thankfully, than it is here. But I also, when I have to consider the statistics of where that electron actually is, I have to consider where it might be in the universe. And that's a big equation. That's a tough math problem. And then I have to take the electron next to it and say, ‘Well, that may be correlated with that electron, but it also might be on the other side of the other side of the universe, right?’ So, this is the kind of huge math problem that chemistry is all about. That's why you need these quantum physicists chemist people to be doing these problems to figure out whether a drug is going to dock to a protein.

 

CV: I gotta tell you, Christopher, sounds tricky.

 

CS: It's very tricky. It is. I mean, quantum is these huge massive statistical problems. So, why was that relevant? OK, we're going to take this math. We know how to do that to some extent. We can estimate how this molecule … and if it gets up to the size of something like methane, you need a supercomputer or bridge to really calculate these energies. It's really massive computing. It's one of the more massive computing problems that we have in the world is doing quantum chemistry. You know, pharmaceutical companies use massive amounts of compute to try and make these estimations, right? So, we have this high-end math. We know how to do that math pretty efficiently, not perfectly accurately. Someday we'll have quantum computers that will do it perfectly accurately, but we're not waiting around for that. We need to do some of these calculations today to figure out how to make the next drop, right? So, where do I take that math and apply it? Well, one answer is chemistry, of course, right, we can do that better. OK, as a company, let's go and talk to chemistry companies about how we might apply that math. In typical chemistry companies like BSF or in other chemical usage companies like BP (British Petroleum) and their downstream chemistry and this kind of a thing. How do you burn things more efficiently is a really important problem, OK? But really, chemistry is just one use of this math. And when we started back in 2017, before there was generative AI, at least as a concept – ChatGPT wasn't around; none of this stuff was around – we said what mathematical problems look like chemistry? And the answer that quickly came to us was generative AI. And this is, you know, 2017-2018. Our first patent in that space was applied for in 2018, less than a year after we started. But mind you, I think at the time. OpenAI was not a for profit company. It was a it was a nonprofit entity. It wasn't a business. There wasn't even a concept of ChatGPT out there yet. So, we didn't say, ‘Oh yeah, let's go do language models and blah, blah, blah, blah with this math’ because it didn't exist. But when we started talking about, well, we're doing generative AI with this, people looked at me again, like I came from Mars because they're like, ‘Well, generative AI? Wait, isn't that deep faking? Isn't that, like fake pictures of cats?’ And then it would be like investors would ask me like, ‘Well, what are you going to do with the business model for fake pictures of cats? We don't get it. Was this NFT's? Is this … what are you going to do with this stuff?’ No, we're going to do industrial problems. ‘So, you're gonna have industrial problems that are fake pictures of cats?’ Well, no, no, it's not about pictures. And it's not about just chat bots. It's about doing industrial problems like we are doing. It's the same technology, but applied to other areas. And we were not interested honestly in language even though I had a background in doing some of those early applications in natural language. That was not like our first thing. That's a B2B2C thing. That's consumer. That's chatbot. And it's hard, you know. I did it back in the day. It’s hard to get really right. Siri's still not right. You know, it doesn't always give you the right answer, right, nor does ChatGPT for that matter. So, it wasn't for me a thing that, you know, we needed to apply to language because it's kind of ambiguous, has some problems. But we could use that technology to do other things like predictive analytics, things that I had done at Verizon, Nissan and other places where there are some smaller problems, if you will, that are

 

 

much more pragmatic that could use generative AI. And so that's how we started out was to actually going after the numbers type of problems like the stuff we're doing with Andretti with sensor data from racing, or what we announced recently that we're doing with SMTB, Sumitomo Mitsui Trust Bank, in creating, trading and investment strategies for a bank using numerical data from the markets. These are not language problems per se. So, we started doing that and then, you know, a couple Novembers ago, ChatGPT comes out and everyone is thrilled by this. And they're like, ‘Wow, this is it. We've got a general AI that knows everything.’ Except it can't do simple math and get it right. So, that really was kind of the moment where we said, ‘OK, let's do language too.’ And we've also, as you know, announced our Prose product, which allows us to do this much more efficiently using the quantum math that we came out of than the previous generation of technologies that these other OpenAI and other models like Llama are based on, which use a previous version of the technology.

 

CV: OK. So let me ask you two questions because you've kind of given one or two examples of what you said as industrial AI. But for the listener, how should they think about those types of applications? You know, I think twice already you've mentioned using sensor data for racing and you mentioned this SMTB example, but is there a more general way to think about it?

 

CS: Yeah, I think if you think about what is generative AI, how is that different from any of the AI or machine learning we've heard of in the past decades? Well, what's new about it? What's so creepy about this kind of human like, why are we so impressed with what ChatGPT can do, right? It's because it's actually doing something that we consider to be kind of human, which is creativity. It's generative, meaning generating something new. And what is that? What is creating something new? Well, machine learning until now has … so let's take the picture of a cat because I get it all the time. A picture of a cat is what? What is a cat, OK? Machine learning until now was I show you 1,000 pictures of a cat and that is the definition of cat. And if the 1,001 first picture of a cat is a blue polka dotted cat with green stripes and no tail, that's not cat. But a human that has this generative modeling capability built into the way we work would look at that and say that's a cat with no tail and it has a pink stripes or flowers all over it or whatever, you can do that. So, what has generative AI been able to do? It's been able to generalize a model, a statistical model – now we're getting back into the math of it – a statistical model of what is cat. So, what is a cat as a statistical model means it's kind of fluffy, no pun intended, but it's not, ‘This is cat.’ If you asked a machine learning model until January before AI came around, ‘Draw me a new cat,’ it would take the eyes from that one, the tail from that one, the paws from that one. It would regurgitate it and say that's cat. It would take whatever it's learned and regurgitate it. But that's not what the generative AI is doing and that's why it's so kind of cool. It actually has a model for what is a cat. So, it can actually draw a cat with no tail. If you go to DALL-E and you say, ‘Draw me a cat with no tail,’ it'll draw your cat with no tail. Why? Because it knows what cat as a concept is. It's generalized the concept of cat. And it's also generalized the concept of what a Picasso painting is versus a Monet painting, or a Monet painting. So, you can actually go to DALL-E and say, ‘Draw me a Picasso cat.’ And you'll get a Picasso like cat, right, with the maybe triangle nose and that kind of a thing. So, what is kind of cool about this new technology is that it's using these statistical models to come up with a generality. A generalized version of cats so that you can do these kind of malleable things where you're doing something actually creative. So, you're generalizing the model and then you can actually be creative, expressive, within that model. You can do interesting things like a new cat and blah blah blah. So, why is all this interesting in industry? Well, what does an engineer do? Well, it has a model for bridges. Good bridges, bad bridges. Ones that won't work because if you use this math, the bridge is going to fall down. Not so good, right? So, there are constraints here, but it has a general idea. The engineer has a general idea of what a good bridge is, right? And as, you know, the engineer has a model for that, and he or she can then say, ‘OK, given that model and giving this river that I'm creating a bridge over, what's a good bridge there? Let's be creative within that.’ OK. It could have this; it could have that. It could be this color; it could be that color. Some of the basics have to be the same or it's not bridge anymore. It's a disaster. But what the generative AI can do that humans can't, just like memorizing a lot of phone numbers like we talked about, it can have 1,000 bridge ideas in an instant, right? A thousand different ones that that engineer hasn't even thought of. It can kind of color outside the lines a little bit better.

 

CV: So, I can, I mean I understand what you're getting at here. But I could see applications for home construction, for new building construction, right, anything where you can kind of take a lot of history or models as you say and, you know, once the model has, once the AI has ingested them, then it can start to create new designs. And that's where the productivity comes in.

 

 

CS: Exactly. And this can be, you know, you think of creativity. Well, maybe engineering is the easiest example. But if you think of someone designing a new annuities project for an insurer, it's the same thing. Like, ‘OK, how do I create a new financial model that will work well for this demographic or that demographic that doesn't work well for our other demographics, right?’ A new product because I'm not going to invest as much because I don't have as much money. So how do I help that maybe underrepresented population take advantage of annuities, which would benefit them, but we're not selling it to people in certain socioeconomic spheres or whatever. So now I can start to branch out and be more creative and think about different models better using the same kind of models. So, creativity isn't just the engineers and the builders and whatever, it's the financial modelers and other people creating other products.

 

CV: Yeah, you know, it's funny. Until you just gave that financial example, I was thinking, ‘Wow. It's easy to come up with a lot of design aspects.’ And then you think about industrial design, not just buildings, but cars, bicycles. I mean, all, all, all sorts of physical things. But you're right, you can take that to designing products essentially.

 

CS: Yep. And strategies like in the racing. What are we doing there? Well, if I, you know, basically deep faking. Deep fake me a cat. Deep fake me a new bridge. But I can also say, ‘Deep fake me a route around that track or a race strategy pit stops at certain times that will do better, right? And it will give me five different strategies.

 

CV: And I'm just thinking that I don't know if you're a fan of pickleball or not, but over the weekend they had the Grand Slam and it was Agassi, Graf, Sharapova and John McEnroe. And I'm thinking, ‘Wow, imagine if John McEnroe had industrial AI?’ They could say, ‘All you need to do is hit the ball here and it will land exactly right next to the line.’ He would kill it every time.

 

CS: Absolutely, absolutely. And this is what's making us superhuman. I don't think we'll remove the experts or, you know, the pro tennis players out of a job or the equivalent bridge designers out of a job. But what we will be able to do is give them more creative tools. In music, we may be able to give them different melody ideas that they hadn't thought of, right? Different rhythmic ideas that they hadn't thought of, you know, just by giving other influences. So, you know, I think that this is augmentative. A lot of us talk about, you know, how this is going to, you know, reduce our workforce or, you know, change … it will change how we do our jobs. But I really am an optimist that that this is actually going to help us in all of these creative aspects of our economy and our society.

 

CV: So, you're not as worried as some of the doomsayers are about the eventual robot workforce, and people will be sitting home for a couple days a week. To use your words, it it's another tool. The Internet is a tool. Microsoft Office, whether it's Word, Excel, PowerPoint tools, that's how you're thinking about this.

 

CS: Absolutely. And you wouldn’t say, ‘OK, we have a tractor, let's go back to oxen, right?’ You know, did tractors remove jobs from farmers? No. They made farmers more efficient, right? You still need somebody to figure out what crops you're going to plant, when and where, and, you know, that's more of a marketing thing than anything, right? So, you know, if you're keep producing tobacco after the lawsuits, you're not doing so well as a farmer, right? So, I think it's augmentative in that sense and not really subtractive. We're going to find out how to use these tools and there's going to be a little bit of have and have not in a capitalist society. Some people are going to figure out how to use those tools better than other people quickly, and they're gonna have an advantage over people who don't use the tools and don't adopt them.

 

CV: So, what industries or sectors inside of other industries do you think are most ripe for, let's say, positive disruption through this technology?

 

 

CS: I'm actually really interested in, you know, the industrial folks who are not so digitalized yet. And because they have these processes that are still people oriented in a way and they haven't yet digitalized. It's not the digital industries in some way. They're adopting AI, they're kind of AI native from the beginning, they've been doing analytics forever. So, the people who have done digital transformation quickly and early and were early adopters are probably the places where we're not going to see as much of a change and as much of a delta, ironically. It's going to be places like logistics companies that literally today still have a person with a clipboard saying I think I can fit one more pallet on that ship or that truck. And literally that's how it's happening today. People don't think … they think, ‘Well FedEx has this big computer that they're using, a supercomputer, that gets my package from here to there and they've computed everything along the way.’ No. That's not really how it goes. There's actually a person counting pallets going into that truck.

 

CV: So, if I may, if you think of a company like Amazon with the power of AWS that is investing in AI, and you think about their core business, right, which is digital shopping and moving things around from one location to the other, that could be a big positive for them.

 

CS: A hundred percent. Yeah, absolutely. And maybe they will be because they're, you know, within the same business structure, they do have the capability. Maybe they will push that. We've seen them want to push drones and AI and all this. We've seen some of those images of what the future looks like for them. And I think that, you know, one thing is, you know, the technology drones and this kind of thing and whatever, but the logistics piece really isn't different from FedEx right now. Nobody has revolutionized that. I think it's going to take a kind of a SpaceX motion where you say, ‘I'm going to blow up literally the idea of how we're going to do rockets, right,’ and blow it up and start over with like cause would you build a logistic system that just uses people counting pellets going onto trucks if you were to create it to in the world of AI? Probably not. We're going to need that kind of movement to really disrupt these industries in a way. And I think that's a huge opportunity. I think the people who get this will do it, but it is going to take, you know, these kind of, I hesitate to say it, but Elon Musk-y kind of attitudes and bets to give …

 

CV: It's always easier to start, you know, fresh, right? Because otherwise you're tracking, you know, back and forth or trying to transform your business and that that takes time, right? You know, there there's that old saying, you know, about turning a tanker around. It just takes a lot of time and then all of a sudden it just starts to whip and it happens a lot faster than you think. So, I do think you're right on that. But, you know, for folks that are looking to create new companies, tap into venture capital, that could be very exciting.

 

CS: I think the disruptive stuff is going to be and if you don't disrupt yourself and you want to stay with oxen because you've got the biggest herd of oxen, great. But when the tractor comes along and John Deere takes over, you're going to be out of luck.

 

CV: Is there any area where you see a lot of this – this is one of my favorite words, hope-ium, right? – where the hoping about AI may … it's a little ahead of itself, right, that like we are just not there yet. Expectations need to be dialed back if you will. Are there any areas that you see that?

 

CS: Well, I think, you know, the idea that GPT 4 is some kind of general intelligence that can answer any question for anything and do it accurately has been overhyped. You know, we're going to need multiple models, smaller models, working together that get the answer right for this to really work. To think that the one model that rules them all is going to happen is a little bit naive and we can see that, you know. We hear this in the market already with CEOs who've found out that ChatGPT finished their daughters’ history homework. Then they come into the board room one day and they ‘I want all of my people to use ChatGPT and tell me your ChatGPT strategy for blah blah blah blah for answering me and my FP&A analysis. I want to ask it five models of my next quarterly results and tell me what the answer is going to be.’ And that's clearly not what it can do. So, there has been some over expectation because of the things it can do that are quite surprising and quite human that that I've tried to describe, this creative aspect of things. So, we think it's a human. We try to treat it like a human and a very smart human that knows everything and it's not. It's not omniscient. It's based on the entire internet, including the stuff that's wrong on the internet. So, it's really not, that's not the best tool for that. It can still do a lot of things really well, but it's not a panacea. It's not a be all, know all type of thing. And what this leads to is, you know, wrong expectations even on the use case level like ... And it was really kind of funny … so, we're doing the analysis and predictive analytics of how, say, a slip angle will be on a track for a race car. And so, we're doing this with Andretti. And so, I

 

 

showed up at one of the races one day with Eric Bretzman, who's the head engineer over at Andretti. And he came to me – he's like, he's smirking. He said, ‘We can fire Zapata now because we have OpenAI and I asked ChatGPT how to win the Indy 500 and I have the answer.’ And he pulled out his iPad and he said, look, ‘Look, see, we've got it. Don't crash the car.

 

CV: Drive faster.

 

CS: Yeah. Yeah. Well, qualify in the top ten was the next one. So, you get these kind of common sense answers and whatever, but not, ‘Yeah, find the best slip angle so that you don't degrade your tires and blah blah blah blah.’ And, you know, that's just not what it does. So, I think that there's been this overhyping that it's going to give you the ultimate answers. It will be good at generalizing things and creating summaries of things and finding general knowledge, but most of the knowledge that that CEOs really want in their companies, the devils in the details of the real answers, right? And you really need the real analytics. Generative AI will help in those predictive analytics, but it's not this big language model thing that's going to actually do that for us.

 

CV: And I assume that is where Zapata comes in.

 

CS: That is one of the places where we can help. You know, we are, we can help, you know, reduce the size and the cost of these language models. I mean if you look at some of the things that have come out in the popular press recently about the energy costs and the estimations that those are the cost of this is going to take, you know? So, if you're a CEO of a company that's touting its ESG policies and you're spending a bunch of money on a model that's really inefficient, that is predicted to contribute 3.5% of all total global energy for AI, you're really kind of got a little bit of a problem there, right? So, I think for our role in this is to use the quantum physics stuff to reduce the size of the models to make them more efficient so that we don't run into having some kind of ESG problem with energy like we have with Bitcoin mining and this kind of stuff where it's kind of obscene the amount of energy that we're using. That's great for NVIDIA and NVIDIA stock, but probably not so great for our Earth. And certainly not great for the bottom-line cost wise. So, we can help with that and we can do some of the problems that are more numbers oriented than language. And I think it's going to be a combination of language and numbers to get the kind of answers that people really expect to get out of this kind of generative AI capability.

 

CV: Wow, I got to say you framed the whole conversation quite differently than I think a lot of people were probably expecting that you would, but I always think that's a good thing. And Christopher, you've been so joining us with your time, but before we get out of here, anything we didn't talk about that we should?

 

CS: No, I think it just does come back down to if I were to reemphasize, you know, that cost matters in this thing. That they're getting these models right, getting them accurate, and also making them secure is really an important thing for enterprise. You know, it's one thing to have a chat bot that answers, you know, homework questions or whatever. It's another thing entirely when you're storing people’s medical records or their personal information in a model. And I think that we really haven't got a solution yet for the kind of security and privacy aspects around this and the kind of enterprise things that we need to really worry about here. We haven't thought about that a lot. One of the major things there is that you take all of the data in a company, say all of the bank’s data, right, and that's got a lot of personal information. And you just … that's really hard to steal though. Stealing all of the bank's information would be a massive hacking problem, right? Trying to steal all of the gold in Fort Knox. Maybe you get a few bars out, right? But the problem we have here is we're taking all of these insights and we're condensing them into a really tiny model that has all of the important information in that bank in one model that can be more easily stolen. So, I think we really need to start thinking about – and we're thinking about a lot in in what we're deploying – that model security. Because we're actually creating a problem that I think in enterprise, a lot of people haven't thought about, which is this privacy aspect and the social aspects of that.

 

CV: I was just gonna ask, it sounds like – and I thought about this quite a bit – but it sounds like AI could be a very positive tool for cyber attackers.

 

 

CS: Absolutely. Makes the thing that I'm stealing smaller. So, I could have all my money in 5,000 bank accounts, or I can put in one gold bar in my kitchen. And that's kind of what we're doing here.

 

CV: OK, alright. Well, Christopher, I'm gonna reserve the right to call you back in about six months because I imagine that between now and then a lot more is gonna happen about AI. I mean I am reading how it will design my fashion in two years and it can do all sorts of other things. And I think we will want to tap into your thoughts and understanding in the coming months.

 

CS: Yeah, thank you much. I really would love that opportunity. This stuff is changing every week.

 

CV: Excellent, excellent. Alright, well, that is our podcast for today. Thank you, Christopher, and thank you listeners for tuning in.

 

 

FORWARD LOOKING STATEMENTS

Certain statements included in this communication, and certain oral statements made from time to time by representatives of Andretti or Zapata Holdings, Inc. (“Zapata”), that are not historical facts are forward-looking statements for purposes of the safe harbor provisions under the United States Private Securities Litigation Reform Act of 1995. Forward-looking statements generally are accompanied by words such as “believe,” “may,” “will,” “continue,” “intend,” “expect,” “should,” “would,” “plan,” “predict,” “potential,” “seem” “seek” “future” “outlook,” and similar expressions that predict or indicate future events or trends or that are not statements of historical matters. These forward-looking statements include, but are not limited to, statements regarding projections of market opportunity. These statements are based on various assumptions, whether or not identified in this Current Report, and on the current expectations of the management of Zapata and Andretti, as the case may be, and are not predictions of actual performance. These forward-looking statements are provided for illustrative purposes only and are not intended to serve as, and must not be relied on by an investor as, a guarantee, an assurance, a prediction or a definitive statement of fact or probability. Actual events and circumstances are beyond the control of Zapata and Andretti. These forward-looking statements are subject to a number of risks and uncertainties, including changes in domestic and foreign business, market, financial, political and legal conditions, the inability of Zapata or Andretti to successfully or timely consummate the proposed business combination of Zapata and a wholly owned subsidiary of Andretti (the “Business Combination”), the occurrence of any event, change or other circumstances that could give rise to the termination of negotiations and any subsequent definitive agreements with respect to the Business Combination; the outcome of any legal proceedings that may be instituted against Andretti, Zapata, the Surviving Company or others following the announcement of the Business Combination and any definitive agreements with respect thereto; the inability to complete the Business Combination due to the failure to obtain approval of the shareholders of Andretti, the ability to meet stock exchange listing standards following the consummation of the Business Combination; the risk that the Business Combination disrupts current plans and operations of Zapata as a result of the announcement and consummation of the Business Combination, failure to realize the anticipated benefits of the Business Combination, risks related to the performance of Zapata’s business and the timing of expected business or revenue milestones, and the effects of competition on Zapata’s business. If any of these risks materialize or our assumptions prove incorrect, actual results could differ materially from the results implied by these forward-looking statements. In addition, forward-looking statements reflect Zapata’s expectations, plans or forecasts of future events and views as of the date of this Current Report. Zapata anticipates that subsequent events and developments will cause Zapata’s assessments to change. Neither Andretti nor Zapata undertakes or accepts any obligation to release publicly any updates or revisions to any forward-looking statements to reflect any change in its expectations or any change in events, conditions or circumstances on which any such statement is based. These forward-looking statements should not be relied upon as representing Andretti’s or Zapata’s assessments of any date subsequent to the date of this Current Report. Accordingly, undue reliance should not be placed upon the forward-looking statements.

 

 

IMPORTANT ADDITIONAL INFORMATION AND WHERE TO FIND IT

In connection with the contemplated transaction, Andretti filed a Registration Statement, which includes a proxy statement/prospectus, with the SEC. Additionally, Andretti will file other relevant materials with the SEC in connection with the transaction. A definitive proxy statement/final prospectus will also be sent to the shareholders of Andretti, seeking any required shareholder approval. This Current Report is not a substitute for the Registration Statement, the definitive proxy statement/final prospectus, or any other document that Andretti will send to its shareholders. Before making any voting or investment decision, investors and security holders of Andretti are urged to carefully read the entire Registration Statement and proxy statement/prospectus and any other relevant documents filed with the SEC as well as any amendments or supplements to these documents, because they contain important information about the transaction. Shareholders also can obtain copies of such documents, without charge, at the SEC’s website at www.sec.gov. In addition, the documents filed by Andretti may be obtained free of charge from Andretti at andrettiacquisition.com. Alternatively, these documents can be obtained free of charge from Andretti upon written request to Andretti Acquisition Corp., 7615 Zionsville Road, Indianapolis, Indiana 46268, or by calling (317) 872-2700. The information contained on, or that may be accessed through, the websites referenced in this Current Report is not incorporated by reference into, and is not a part of, this communication.

 

PARTICIPANTS IN THE SOLICITATION

Andretti, Andretti’s sponsors, Zapata and certain of their respective directors and executive officers may be deemed to be participants in the solicitation of proxies from the shareholders of Andretti, in connection with the Business Combination. Information regarding Andretti’s directors and executive officers is contained in Andretti’s Annual Report on Form 10-K for the year ended December 31, 2022, which is filed with the SEC. Additional information regarding the interests of those participants, the directors and executive officers of Zapata and other persons who may be deemed participants in the transaction may be obtained by reading the Registration Statement and the proxy statement/prospectus and other relevant documents filed with the SEC. Free copies of these documents may be obtained as described above.

 

NO OFFER OR SOLICITATION

This Current Report is for informational purposes only and shall not constitute a proxy statement or solicitation of a proxy, consent, or authorization with respect to any securities or in respect of the Business Combination. This Current Report shall also not constitute an offer to sell or a solicitation of an offer to buy any securities, nor shall there be any sale, issuance, or transfer of securities in any state or jurisdiction in which such offer, solicitation, or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction. No offering of securities shall be made except by means of a prospectus meeting the requirements of Section 10 of the Securities Act or an exemption therefrom.

 

 

 

---