Podcast: Zapata AI’s CEO & Co-Founder Christopher Savoie

SPACinsider-podcast-logo-300-x-200

Podcast: Zapata AI’s CEO & Co-Founder Christopher Savoie

Buzz around generative AI has been off to the races lately and that is exactly where Zapata AI has put many of its tools to work.

This week, we speak with Christopher Savoie, CEO and Co-founder of Zapata AI and Bill Sandbrook, Co-CEO and Chairman of Andretti Acquisition Corp. (NYSE: WNNR). The two announced a $283 million deal in September.

Bill explains how the Andretti team’s experience with Zapata AI on the competitive racetrack grew into the decision to take them public, while Christopher walks us through how Zapata’s number-based generative AI can create designs or scenarios for applications ranging from racing efficiency to financial portfolio optimization.

Chat GPT may be able to convincingly generate text and images, but could Zapata AI generate bridge blueprints, molecular formulas and new financial instruments?

Give it a listen



Nick Clayton (NC), SPACInsider: Hello and welcome to another SPACInsider Podcast, where we bring an independent eye in interviewing the targets of SPAC transactions and their SPAC partners. Buzz around generative AI has been off to the races lately and that is exactly where Zapata AI has put many of its tools to work. I’m Nick Clayton, and this week my colleague Marlena Haddad and I speak with Christopher Savoie, CEO and co-founder of Zapata AI and Bill Sandbrook, co-CEO and chairman of Andretti Acquisition Corp. The two announced a $283 million deal in September.

Bill explains how the Andretti team’s experience with Zapata AI on the competitive racetrack grew into the decision to take them public, while Christopher walks us through how Zapata’s number-based generative AI can create designs or scenarios for applications ranging from racing efficiency to financial portfolio optimization. Chat GPT may be able to convincingly generate text and images, but could Zapata AI generate bridge blueprints, molecular formulas and new financial instruments? Take a listen.

 

NC: So, Christopher, the pace of technological advancement in the AI space has been accelerating rapidly in recent years. How did Zapata start out back in 2017, and how has your approach evolved into the platform that you have today?

 

Christopher Savoie (CS), Zapata AI: We spun out of Harvard, Harvard University. My co-founder was a professor at Harvard in the Computational Chemistry Department of Harvard actually, where he specialized on quantum algorithms. So, that’s a very difficult, nerdy word to say, you know, we do linear algebra, the special kind of math for very, very hard mathematical problems. So, we were able to take that technology which we use in things like high energy particle physics, what happens in a collider and quantum chemistry, how does a drug molecule bind to a protein? These really, really high hard computational problems. And we figured out quickly that the kind of math that we use there was applicable directly to AI and specifically the kind of math that we do in generative AI, which we’ve heard a lot of lately with things like ChatGPT. So, the math that goes on behind that is a lot of complicated statistics and what we were doing at Harvard really is complicated statistics in the physics space, probably the most complex statistics that are around in the world. And so, we were able to apply the decades of experience that we’ve learned in quantum physics and quantum chemistry to this new world of AI. And so, our first patents in generative AI using quantum statistics were in 2018, less than a year after we had founded the company.

 

NC: And Bill, I’m curious about your process and finding your way to Zapata AI yourself. It’s interesting that the Andretti’s were already familiar with the company, but how did it come to the forefront as a candidate for you?

 

Bill Sandbrook (BS), Andretti Acquisition: Michael and I have been involved in racing and together for a number of years. And we started the SPAC, we were looking for mobility technology companies and we had done extensive research into many, many companies around the country in various aspects of mobility technology. EV companies. Charging companies, LiDAR companies, etcetera. But I’ve been a sponsor for some of Michael’s cars and I’ve been going to a lot of races and I had seen the Zapata guys and seen Christopher, got introduced to him and then kind of dug into the technology and started discussing with Michael that hey, this might be a really, really interesting company for us to try to target and do a deal with and deSPAC because of the knowledge the Andretti’s had with the technology, the success they were showing with their modeling and the ability to have all these other use cases outside of racing. And so, one thing led to another and Christopher and I and my team and his team started negotiating with the agreement of the Andretti’s. We tried to put a deal together that would merge this new technology for racing and its other applications, which would be a perfect fit for Andretti Acquisition Corp. to do a SPAC with.

 

NC: So, for those not aware, what is sort of the difference between generative AI and what we consider to be sort of the usual, you know, machine learning, some of those sorts of processes that other people might be familiar with?

 

CS: Well, that’s a great question. So, in machine learning up to now, which we’ve had around for decades, basically, it’s rote memorization. So, a machine can learn more than a human just because it’s a machine, right? So, you can store a lot of data. So, it can be really good at learning some patterns out of data. But the way machine learning had been done until now was basically to give a metaphor: I show you a thousand pictures of a cat and those thousand pictures are cat. So, if the one-thousand-and-first picture of cat is a purple polka dotted cat with blue stripes, that’s not a cat, right, to oversimplify. What generative modeling does is it generalizes a model of cat using some really cool statistics to do it. But this allows you, if you have a model of what is a cat that’s general like even a cat with no tail is still a cat and you have a model of what is a Picasso painting, if you have those two concepts now that are generalized, I can now say something like, ‘DALL-E, hey, draw me a Picasso cat,’ and it’ll draw a Picasso cat. Maybe with a triangle head that’s a little bit weird, but you can recognize it as a Picasso style cat. And that’s what’s kind of new and creepy and cool about this stuff is that it’s doing something that’s actually creative, actually generative. And it’s not just pictures or words. But it’s also, ‘generate me a bridge that works’ and this kind of thing that we can do for engineering. Or, in in financial services, ‘generate for me a good portfolio that has really good features to it.’ So, that’s what’s different from this. It’s not just regurgitating everything that we’ve learned, but it’s actually doing something creative and new and generative.

 

Marlena Haddad (MH), SPACInsider: And so, speaking to some of the use cases, what were some of the main use cases that you had in mind when you were developing the Orquestra® platform and how exactly did that bring you to competitive racing?

 

CS: So yeah, it was a bit of serendipity. So, we have been doing generative AI in a number of areas. I myself have previously worked on big data and AI projects at Nissan, Renault where I worked in enterprise architecture on these kinds of solutions in automotive, in mobility. So, when we met the Andretti team, they had been looking at how they could apply AI and advanced analytics to racing, particularly in the IndyCar Series. And we actually had some things that we could apply there. I personally had some experience in the automotive sector with cars and predictive analytics there. So, we were able to really come quickly to some use cases that could really utilize what we were doing for other companies, for other areas like in finance and energy to actually apply those to the specific area of mobility in the context of racing with and do some pretty interesting things with the numbers for generative AI. And the use cases there are not what we’ve mostly heard about in the last year or so of the ChatGPT world that we live in, where it’s mostly about words. Can I, you know, summarize this, or give an answer to this or that. What we have been doing is using the same underlying technology to do numerical analysis, which is another use of generative AI. So, in the case of Andretti, we were able to use generative AI to deep fake – deep faking a chat, or deep faking something on DALL-E with a drawing – draw me a Picasso painting of a cat. That’s the kind of thing that it does, but you can also deep fake other things that are more useful in industry, in car racing, in banking. Deep fake for me a better portfolio. What is a good portfolio? We can do that. We can generalize these things and then deep fake good things that are very realistic. So, that’s very useful as a tool to do analytics. Specifically, in the case of racing, we want to be able to predict things that happen with the car. Physical properties, those are really important, like how is the car steering around the track under these conditions? What are the dampers doing? What are these things… So, the physics of this really makes a difference in when do you pit the ca because we think the tires are going to go slower or run out. When do I make certain adjustments for the fact that we might have a yellow flag come out or not, predicting these kinds of things/ But predicting, you know, the physical attributes of the car, we have a lot of sensors on these cars – over 100 – but there are some physical properties that we can’t directly predict with the sensors. For example, the slip angle of a car when it’s going around the track, you would think, well, isn’t that the direction of the tires? Well, no, not really, because when you go left at 240 miles an hour in qualifying at the Indy 500, when you go left, the car wants to go right because of centrifugal force. So, your slip angles of those tires is very different from just the angle of the different sensors we have on them. So, what we’re able to do is once we can generalize things using the data from all the other sensors like, what does the steering angle look like going around that track?, we can say please deep fake to me kind of like talking to ChatGPT, deep fake for me, a picture of the slip angle of those tires going around this track with this car with this driver on a specific track. So, you get a very realistic thing just like you can say, DALL-E, show me a picture of, I don’t know, Diana Ross holding an apple. You can get that picture, right? You can do something like that and we can do this using data to actually drive very realistic scenarios of how would the car look if it did this? Or, if it did that? Or, it did something else. And we can do that with very accurate predictions of what the time series things would look like on the track. And that becomes very useful for making strategic decisions. When do I pit the car? When do I do this or that? When do I make adjustments? So, the cool thing about this is it’s not just about chat. It’s not just about marketing. We can do some pretty deep analytics using this generative AI technology. And that’s racing. So that’s one-use case, but ironically, we just announced a deal with a financial institution, Sumitomo Mitsui Trust Bank, in Japan, who want to use basically the same underlying capability that’s time series data going around the track – the math is all the same – to do portfolio management and trading management and management of how their clients, regional banks, want to do mathematics on their trading and predict different trades. Instead of predicting how a car’s going to go around the track, we predict how the market is going to react to certain variables. So, there are a lot of broad industrial mathematical use cases that are really useful, and we’re really excited about that. So, this is really bigger than just chat and ChatGPT at the end of the day.

 

MH: Great. And so just going off of that, what more can you tell us about some of those existing clients both inside and outside of the automotive space?

 

CS: So, the ones we can talk about are the ones that are in our public listing and the ones that we’ve announced, like the SMTB relationship, the banking relationship with banking use cases and there’s a press release out there on our website that you can look at and talks in depth about the kind of stuff we’re doing for them. And there is the Andretti relationship, where we’re doing cars and those car things obviously are applicable not just to race cars, but the kind of cars that we drive in the street when we have an electric vehicle, how are we using those batteries and these kinds of things. So, those kinds of things. It’s very broad across a lot of different sectors. But the problems that we’re working on are generally industrial problems, large math problems that really have business value for companies. And that’s what we’re really excited about. Yes, we can also do language. We have the capability with our Prose offering to compress using this quantum math language models and make them cheaper, faster, better than what you can do with a neural network-based model with older technology. So, we can do the language-based tasks as well. But we’re really excited about the numbers-based stuff. If you go and ask ChatGPT how to win the Indy 500, it’s not going to give you a numeric analysis. We’ve actually tried that and it says well, go fast and don’t crash the car, so they’ll give you common sense answers. But we can use generative AI to do numerical things that give you real accurate numerical results that are important in finance, in creating new and creative annuity products for an insurance company, for example. There are just so many things that involve this kind of deep math use of the same technologies that’s in something like ChatGPT to do some really useful stuff that has a lot of business value.

 

NC: And to that point, you know, one of your interesting collaborations that you’ve announced with IonQ, which SPAC watchers are familiar with as it did its own SPAC transaction back in 2021. I’m just curious and, you know, in addition to figuring out how you could, you know, together marry some quantum computing with your own generative AI, do they have any sort of pointers for just, you know, some of the opportunities or the pitfalls of the SPAC process and what their experience has been like with, you know, a couple of years in the public markets?

 

CS: Well, I think you’re looking at a good example of a company that has been able to maintain its price above where they went out on the SPAC and has performed very well compared to how the entire technology sector has generally fared since, you know, the Ukraine war and uncertainty in political events that have happened and the economy with inflation and the Feds moves, right? So, I think in general this shows that, you know, SPACs aren’t bad. SPACs aren’t terrible. SPACs are a mechanism for a young company to go public and that are expedient and, in some ways, less expensive than the normal way of going public, which is the normal IPO process. So, you know, I think that this is an example of how a technology company can actually perform well through the SPAC mechanism for going public.

 

NC: Yeah, and I want to get into the deal a little bit more, but I wanna … I have a few more points I wanted to get to on in terms of your operations and your business model as well. Just given that, you know, how new so many of these technologies and some of these applications are in the market, have you had any challenges in terms of pricing it for clients and how do you approach that given that some of these use cases are very new?

 

CS: Some of the use cases are new, but the businesses that we’re in pretty much are very mature. So, we’re in industrial areas. So, it’s not like we’re saying, ‘OK, let’s use ChatGPT to do something cool and new that that we don’t know about.’ We’re saying, ‘OK, if I can actually increase your gross margin as a financial institution by making your trades more effective, is that going to be valuable to you? There’s not a question there. And so, the value is actually known because these are, you know, older industries. We know how much it costs to do logistics for a company for example, and if we can optimize that and give them basis points or percentage points of more business value, that can be easily understood as to what the actual monetary value that is for a customer. So, we’re not having problems actually pricing this and you can see in our contracts that are public now that we’ve had multiyear, multimillion dollar deals for the most part. And, you know, the companies doing this, it’s not anything new. The generative aspect is new, but there are companies like Palantir out there, C3 AI and others who are basically performing in the same sector with maybe a different or older technology stack than the generative AI stuff that we’re doing. But this isn’t a new sector in that sense. It’s a new type of AI that has some advantages over the older versions of AI that we’ve been using.

 

MH: Got it. And many potential clients may be interested in dabbling in quantum AI, but how do you approach making your services sticky and well-integrated into their businesses?

 

CS: One of the advantages that we have is that we offer a software platform that’s based on Ray, which is what was used to train the open AI ChatGPT models. But we have an enterprise version of that software that integrates with a company’s LDAP systems, their security, their privacy and all that. And that’s really important to enterprises because they’re not going to very easily give up all their IP and all of their data and put that on some cloud without their security people understanding that. It gets even more dangerous when you take all of your data and you put it into a smaller model that has all the insights in one compact model that’s easier to steal than all of your data. So, for those reasons, security, enterprise capability, privacy reasons, when you’re dealing with people’s private information at a bank or something like that, this is really important. Or competitive information that you might have about your race car that you don’t want another race team getting. These are all important things that you don’t really have to worry about if it’s just, you know, consumers using a chat bot. But in these enterprise things, having a software system like Orquestra®, which is our offering that allows you to take enterprise security into account when you create, when you train these models, and when you deploy these models is really important for our customers. So, we offer, yes, the models, yes, the AI super geeky algorithms, but we also are able to offer a platform to do this in a safe and secure way that protects their privacy and their data.

 

MH: Definitely. And there’s been a lot of chatter of increasing regulations over AI, both in its use by companies and how public companies describe their AI tools. So, do you see any of that impacting Zapata in the future?

 

CS: Yes. For us, it definitely in some ways is a positive because we’ve been doing this for years. I mean the company was, as you noted, was founded in 2017. So, we’ve been doing generative AI for all of our existence, which is longer than, you know, OpenAI had even been a for-profit entity and we’ve been doing it in this numeric space. So, we’ve been caring a lot about people’s numbers, people’s privacy, people’s data and doing this in a sustainable and ethical way for a long time. So yeah, I think that kind of plays to our advantage because we’ve always been squarely in the enterprise space. As you know, we’ve announced in our public filings that we’re also a defense contractor as well. And so, when you’re playing with large energy companies, defense companies, this kind of thing, the stakes are really high. And even on the race track, it’s a life-or-death decision-making process there literally that you’re making. So, in these cases, you know, the ethics of your models, understanding your models, being able to produce things that are not hallucinations is really important. And so, I think that, you know, it’s good that that the public is starting to be aware and the press is starting to be aware of some of the underlying ethical considerations, privacy implications of this technology because it’s accelerating really fast and I think that the ethical side, the legal side, we really have to stay ahead of that. We really need to think about how these technologies are going to have an impact on many areas of society, cause it’s here to stay and we need to really be thinking about those aspects in our company and how we deploy our products for our customers.

 

NC: And moving back to the transaction itself, you know, you touched upon a little bit how it came together, but I’m interested in hearing from both of you just how did each of you decide that both, you know, Zapata from an internal standpoint that the timing was right to take this big next step into the public markets. And similarly, what sort of signals have you seen from the markets themselves that are sort of ready for some listed AI companies and how they’ll be received?

 

CS: Yeah. Well, we’re really excited just in general about being probably one of the first, if not the only pureplay generative AI company on the public market that we’re aware. There are other AI companies, but were really concentrated on generative AI and one that has this quantum capability behind it. We’re actually excited to have that out there as an opportunity for investors to be able to participate in our growth with. So, this is really helpful for us also in getting our message out there to a longer audience. If we had stayed private, we wouldn’t have as large of an audience interacting with what we’re doing out there. So, from a number of those aspects … it’s also as a public company, when we’re dealing with customers who are large Fortune companies, Global Fortune 100 companies and governments and folks like this, having the transparency of being a public entity, I think, is very helpful in the procurement process for us. It’s also great for retention of employees to have our incentive plans with public equity available to us. Our competitors are, you know, the larger tech companies in the world and being able to have, you know, the same kind of, you know, ability to be transparent and out there as a public company but also have access to the public capital markets is really a big deal for us.

 

BS: And from the Andretti side, when we embarked on this adventure, we were looking for mobility technology intersections, mobility technology companies. So, we vetted an extremely large number of companies in the EV space, in the electric charging space, in the in the LiDAR sensor space, in the cargo hauling EV truck markets, and, you know, when you look at Zapata being able to cross into mobility technology and race performance enhancement with its significantly larger amount of use cases into industrials, financials, etcetera, etcetera, it really was a force multiplier in our target or potential markets as opposed to just staying in an EV space or charging space. So, it really added a lot to, you know, to the offering of merging technology with mobility into multiple use cases. So, we couldn’t be happier with having come to this agreement with Zapata.

 

CS: And for us, knowing the management of the team because we’ve been a customer for a couple of years at the point that we made the decision to go forward here, it was an easier process because they knew our technology. They knew it worked. They knew what we could do. They knew our team. So, this made it, you know, trust is really important and that we could have that mutual trust and understanding of where we’re coming from really was a great capability. And I’ll add that, you know, this has been a really great relationship for us because of the Andretti name and what it means not only just in racing, but just in general in the world. It’s added a lot of credibility to us and marketing capability to us to be able to get to other companies, not just in mobility with the OEM’s and others, but just in general, with the public. ‘Oh, you’re the folks who do the race analytics for Andretti’ is a big statement to make and we can actually make that claim. And that’s a really big thing for us for a small company to have as far as credibility.

 

NC: And to that point, Bill, you know, one of the interesting aspects of a, you know, a SPAC deal is that, you know, as opposed to an IPO there can be different terms included for upside and downside protection and you are able to, you know, with your team go through the process of figuring out the, you know, the right equity valuation for the company as a whole and then looking at the market, all those different factors all at once, you know. So, I’m curious like what was your process in terms of weighing all those different things, kind of understanding there, you know, you may need some contingencies for redemptions, etcetera?

 

BS: Right. Now there’s a lot of moving parts as you infer. We kind of coalesced around to ensure that we had a successful deSPAC that the valuation had to be spot on. I mean that’s probably one of the biggest weaknesses of the SPAC market and the performance of many SPAC shares in that the valuation tended to be maybe overly optimistic. And so, we stayed laser focused and made sure that Zapata understood from our perspective that we had to get the valuation exactly spot on. So, I would say that more than anything, even though there are a lot of other moving parts that had to come together, I think the proper valuation is the foundation for the success of these SPACs.

 

MH: And some macro conditions over the past year have also put a lot of pressure on companies in Zapata’s space. So, does the company see some potential in engaging in M&A once the deal is closed?

 

CS: It’s really hard to hire organically machine learning experts even globally. We’ve had some great success compared to others in doing that and retaining some of the best folks in this space, but when we keep adding customers and demand, we’re going to need to increase that and doing that organically is of course a component of what we intend to do and we have done. But inorganic growth will be an opportunity for us. We can see that happening for sure, particularly if we can pick up things that are really strategic in verticals that we are having real growth opportunities come to us in things that, areas that we have announced that we’re already interested in like finance, like the mobility space for example.

 

NC: And given just how large the potential market for Zapata’s services is, how are you balancing, you know, your growth kind of philosophically between, you know, the need to really tap into it and get out there and be the first in a lot of these areas, versus, you know, the risks of overextending and some of those other issues there?

 

CS: Long term, you get judged as a public stock on your ability to create profit. I mean a share is nothing but a share of profits, right? So being EBITDA positive is the goal and increasing the margin of the company is definitely a long-term goal. Now, there is a sliding scale and there is a balance to be made as management to what market share do I give up in doing that, particularly at the beginning of a market like this. You don’t want to be a second runner or a third runner in this. You want to be the leader. You want to be the innovator. You want to be the first into a customer account in a lot of these cases. So, this is one reason why being a public company and having access to public capital markets is helpful when we do want to hit the accelerator and grow given the right value and the right timing for those kind moves. So, it is constantly something that we will look at as a management team and as a board as to, you know what the market is rewarding at a particular time. You can’t be tone deaf about this. The market over the past couple of years has not particularly rewarded risk and growth over profitability because of the cycle that we’re in. But that will change overtime. These things are cycles where we’re eyes wide open about that. And so, we want to be careful not to knee jerk one way or the other on that. We want to have a mind towards profitability and always being able to be cash flow positive when we need to be and when we want to be, but also be aware that we do need to take some risk in order to grow so that we’re not the second best, the third best, the fourth best into an account which is not what, I think, that the market wants to reward in a small growing company as opposed to, you know, a large incumbent. I think the opportunity here is in some of the growth and so we don’t want to be blind to that or tone deaf to that as well.

 

MH: Definitely. And then I’m interested to hear what you think is the application for your technology that may perhaps not get as much notice, but could be the most exciting.

 

CS: Yeah, I think that the numbers as a whole is not what people think about when they hear generative AI. They tend to think about, you know, pictures and maybe music even or words, particularly the ChatGPT types of stuff. But the cool thing is we can generate, we can generalize and generate more than just words and pictures. We can generate new ways to design a bridge, new ways to build s building, new ways to create a product, for example, an annuities project product that fits better to underrepresented populations who don’t get sold annuities. I mean, there are so many things that have business value out there that are not being talked about in the popular press because of the coolness of ChatGPT. I think that that hype kind of in some ways gets in the way of understanding what the true industrial and in some ways, boring, but really important things that this technology can do for society and for human beings. In healthcare, increasing, you know, our ability to deliver healthcare cheaply, affordably, more affordably to patients. In drug design, we have a paper out there about how we can use quantum enhanced generative AI to make a more, a better fitting drug-like molecule then standard AI can do. And to me, that’s really exciting when we can start doing things that really are going to have a greater impact on humanity that maybe won’t make it to the press. ‘Oh, wow, we get a cool molecule’ doesn’t really work as well as ‘hey, ChatGPT was able to finish my daughter’s homework for her,’ you know. That may sound cooler, but it really is the cool stuff here is, you know, creating a new drug, creating a new way to engineer a bridge or a safer way to engineer our transportation networks or something more efficient that will cut carbon in those transportation networks. Those are, I think, for me the fun things. The fun, boring things that really mean that this AI is here to stay and it’s going to be a part of everything we do as humans.