The Exchanges

Q Maybe we can talk about the use case, right? So when we're talking about, let's say, a military base, What is the current state, right? If we're not using these micro reactors, what is being used today, and what's the trade-off there if we can't actually get to that future reactor?

A Current state at the military base is that they have backup generators. Like any, any site that has critical infrastructure, um, those backup generators will have diesel storage tanks. There'll be 40,000 to, to a 150,000 gallons of diesel, um, on those sites, and they'll only, only use it in a backup scenario, so it requires they put batteries all over the installation, and, uh, if there is an outage, they're typically gonna run out of that diesel. Um, especially if there's, you know, something like the, that, uh, Colonial Pipeline ransomware attack, where we lost an ability to move fuel in a huge multi-state area. Um, and they run out of fuel before their, their timeframe, which is, uh, usually a 14 day resilience timeframe. Um, so they've got a problem, and they're looking for solutions, and they're actually very interested in, um, both categories of microreactor, because those are the, around the scale of the base. All of the larger ones, an SMR or a gigawatt class, uh, reactor would be too large for them.

AI assessment note: “Current state at the military base is that they have backup generators.”

Q Right. Is this widely understood? Because, I mean, you said this was discovered in 1970. Why haven't we seen this translate into the classroom before?

A Well, there, there are a couple things there. It's been widely understood in the research community, and I think the last few decades in education reform has seen a, a huge lack of translation of what the research has shown to actual tools and solutions. That's been primarily dictated by the tech. So, so far, computing devices have mainly been on two D screens, and we were digitizing learning mechanisms, That were possible using the computational devices we had. And now with the advent of spatial computers, um, and mobile VR and XR technologies, we now finally have the natural interfaces where we can scale. Learning with your body. Learning using six degrees of freedom. Using a multitude of tactile tools before jumping to an equation which is very abstract.

AI assessment note: “It's been widely understood in the research community, and... primarily dictated by the tech.”

Q Let me just hear this from you. Why do you feel like this is so important for us to turn around?

A I think that math education has been attack, under attack for a long time of why do kids need to learn algebra? Why do kids need to know percents and ratios? And just let's, let's, let's set the record straight. Bob Moses had shared this in the, during the civil rights movement. Algebra one is a civil rights issue of our time. I'm a math ed person. So from, from my perspective, it's, it's essential. It's essential. Not only is it tied to, um, Salaries over time and, and, and earnings over time, it's of course tied to just the types of jobs that you have access to. You don't do well in Algebra I, you are not taking chemistry, physics, biology. If you are not succeeding in chemistry, physics, biology, you are not going on to the applied mathematics or the medical sciences. So you are just kind of cut off from a very large swath of jobs in our economy.

AI assessment note: “cut off from a very large swath of jobs in our economy.”

Q who are working on exactly that problem, figuring out how we elevate the classroom after it's been relatively stagnant, quite frankly, for many years. And to kick things off, let's really ground ourselves on where we are. Romy, maybe you can help us here. What is the data saying in terms of the student experience? What are kids really facing? How can we really get a better understanding of that?

A Thank you so much, Steph, for being here and moderating this discussion. As many of you know, the pandemic was a huge setback in education, and it exacerbated and widened already deep inequities. We know from data that came out last week that math achievement levels are the lowest they've been in this country since 1990. And literacy achievement levels are the lowest they've been since 2004. And the pandemic erased two decades of progress that we had made in education through a variety of different change initiatives. So we have a lot of work to do. The good news is that students, families, and educators agree that this is a moment for reimagining learning, both embracing technology, but also thinking about how we strengthen relationships in classrooms, how we better support student well-being to address the mental health crisis.

AI assessment note: “math achievement levels are the lowest they've been in this country since 1990”

Q new because it is a different product line. There are different considerations, whether it's being on the water or having to be in the air. Or having additional redundancy because you're in the air. And so I'll just kind of leave this open to the group. What have you had to consider and reinvent versus what has again, been able to be, you know, borrowed from the existing automotive industry?

A The, all the technology development around making the electrified powertrain super efficient, right? This is your electric motor connected to, you know, your batteries and All the high voltage systems that need to exist in the balance to make it work. And, you know, Mitch talked about reliability. Couldn't agree more. Electric is awesome because it is so reliable. That is an area that we probably all benefit from is just years and years and years of making it a super durable system. I would say for us, we, we had to create a ton of interoperability, um, systems that we didn't expect we'd have to build. And what I mean by that is Not all charging stations work with all vehicle types. You know, there's firmware, there's other, other, you know, battery management software within your vehicle, and then there's the utility. And if you think about creating a reliable ecosystem, all of us, you know, Mitch said power hungry. They're really power hungry, right? Our, our, our depot in Bethesda, Maryland, you know, draws five megawatts of power when we're charging. It's the largest electrified depot in the country. Running 200 school buses and five megawatts is like five hospitals coming online overnight. And so you need really advanced software that, that is managed in a way that we've never had to do in the past here in the States anyways. And that's interoperability between your charge…

AI assessment note: “we had to create a ton of interoperability, um, systems that we didn't expect”

Q say, it may sound silly to some folks, but you can imagine the human equivalent, right? So for Uber, it was literally calling up someone who then would dispatch a taxi, right? Or, you know, even Google Maps way back in the day, like before we had digital maps, you might even call someone or look up, you know, a physical map and say, hey, how do I get here?

A Steph, one of my favorite examples is, um, I'm a pilot and I, uh, fly a few different airplanes. A dear friend of mine, Richard Kane, built a company called VeriJet and uses the VisionJet. And the traditional aircraft chartering business is literally whiteboards and sticky notes where I'm taking this pilot and this airplane here, the fuel is here and it's like pathetic. So Richard, who's a brilliant computer scientist, built a platform on Verajet where the everything is optimized. Uh, he'll go to quantum compute later for, for doing even faster, but where the airplanes, the pilots, where the person is being picked up and dropped off and how they're being, All of it is fully digitized, and he used the vision jet, the SF-fifty, because it is massively connected. When the airplane lands, it sends all of its data of its, uh, systems check back to the AI. So you get humans completely out of the loop, and that's beautiful. It's efficiency, it's safety, and it's speed.

AI assessment note: “one of my favorite examples is... traditional aircraft chartering business is literally whiteboards”

Q the tone. It feels like listeners maybe don't need this reminder, but if we started with a couple statistics of where we are now, like almost as if the output of that funnel, of those actors, of those incentives, Are there any facts or statistics that either of you have run into that again, kind of just set the foundation or almost like the reality that we're in within healthcare?

A Yeah, I mean, I'll just start with like the, the headline number of healthcare costs and healthcare spending in our country, 4.3 trillion dollars every year, which is roughly about 20% of GDP is spent on healthcare. And I think the key thing there is not necessarily how much we're spending because, you know, if, if we were spending that much and getting amazing service and amazing outcomes, Then we would all want to pay more for it because we were getting value. But I think the, the challenge in our country is that our healthcare outcomes are actually getting worse. I don't know if you guys saw the stats recently on life expectancy, you know, decreasing over the course of the last couple of years, obviously largely driven by the pandemic, but the yield of any, of, of, of a given healthcare dollar spent in the U.S. is far, far lower than all other developed nations in the world, which is a sad, sad state of, of affairs. So I think that's probably like the headline stat, you know, that we should start from. And then, you know, related to what I said earlier, I'll probably call out additional stat, which is, you know, just the administrative waste and bloat in our system due to the fact that we have this complex trifecta of payer, provider, patient results in a tremendous amount of spend that is completely unnecessary and or just pure administrative overhead. So, you know, the sta…

AI assessment note: “4.3 trillion dollars every year, which is roughly about 20% of GDP is spent”

Q we use that term again, broken for quite some time. But it does feel like maybe certain things have changed more recently. Or at least there may be certain unlocks that allow founders to maybe step in and build within this intersection of healthcare and fintech. And so what gets you excited today about perhaps new openings within this arena and what may have changed over the last few years?

A It's rare that people talk excitedly about regulation, but healthcare is one of those domains where I think regulation can be a tailwind for innovation and category creation. Um, there's lots of historical examples of this. I think the most Traditional example that a lot of people point to is electronic health records, you know, did not really exist in adoption, major adoption, until, ah, the meaningful use, um, law came into play, where the government literally paid financial incentives to doctors to adopt digitized technologies for medical record storage. So that was really the sea change that, you know, drove, you know, so much of the digitization of our, of our infrastructure layer of healthcare. Similarly, right now, we have a number of regulatory tailwinds that are driving payment reform, payment related reform, I should say. And so, you know, we have things like a price transparency law that went into effect over the last couple of years that forced, uh, hospitals and insurance companies to publish their contracted rates. It was hugely controversial. There's still lawsuits in play. People are still pushing back, but the fact of the matter is we now have, you know, thousands of hospitals and hundreds of payers who have published all this data. Obviously they're publishing it in forms that make it very, very difficult to parse. And so entire companies exist to Actually, um…

AI assessment note: “we have a number of regulatory tailwinds that are driving payment reform”

Q would agree, you know, Julie, you used the word irrational, but I think other people may apply terms like clunky, maybe even broken. Um, but to your point, David, they may not know the depth of that particular industry and they may not understand why it might be broken. And so maybe we can kick things off there. Julie, where do you think the healthcare system breaks down the most?

A Yeah, you're absolutely right that it's pretty much all broken. Or, I mean, the other way that people oftentimes talk about it is that it's actually working as designed. Um, and a lot of that, you know, sends back to the way that payments are designed and, and processed exacerbates, I think, a lot of the, the challenges that we experience as consumers. I often take the, the system view to this space. I'm, I'm very enterprise focused in the companies that, that I primarily spend time with. And when you look at it from the system lens, you know, the, the core trifecta of stakeholders in our healthcare domain, very simply put, would be, The following. So one is you have obviously the providers, the actual doctors and nurses and all of your care providers and services companies that actually deliver care. They are obviously key stakeholder. You've got your payers, which are, you know, largely either insurance companies, private insurance companies, government funded insurance, and then, you know, arguably even consumers to some degree are payers themselves for the out of pocket component of what we, what we pay for healthcare. And then of course the third party being the patient themselves. And so the point there being that the people delivering the service And receiving the service are not the people paying or, you know, sort of privy to the payments flow of those services. And th…

AI assessment note: “that very bifurcation is what causes so much of the incentive misalignment”

Q the tone. It feels like listeners maybe don't need this reminder, but if we started with a couple statistics of where we are now, like almost as if the output of that funnel, of those actors, of those incentives, Are there any facts or statistics that either of you have run into that again, kind of just set the foundation or almost like the reality that we're in within healthcare?

A Yeah, I mean, I'll just start with like the, the headline number of healthcare costs and healthcare spending in our country, 4.3 trillion dollars every year, which is roughly about 20% of GDP is spent on healthcare. And I think the key thing there is not necessarily how much we're spending because, you know, if, if we were spending that much and getting amazing service and amazing outcomes, Then we would all want to pay more for it because we were getting value. But I think the, the challenge in our country is that our healthcare outcomes are actually getting worse. I don't know if you guys saw the stats recently on life expectancy, you know, decreasing over the course of the last couple of years, obviously largely driven by the pandemic, but the yield of any, of, of, of a given healthcare dollar spent in the U.S. is far, far lower than all other developed nations in the world, which is a sad, sad state of, of affairs. So I think that's probably like the headline stat, you know, that we should start from. And then, you know, related to what I said earlier, I'll probably call out additional stat, which is, you know, just the administrative waste and bloat in our system due to the fact that we have this complex trifecta of payer, provider, patient results in a tremendous amount of spend that is completely unnecessary and or just pure administrative overhead. So, you know, the sta…

AI assessment note: “headline number of healthcare costs and healthcare spending in our country, 4.3 trillion dollars”

Q we use that term again, broken for quite some time. But it does feel like maybe certain things have changed more recently. Or at least there may be certain unlocks that allow founders to maybe step in and build within this intersection of healthcare and fintech. And so what gets you excited today about perhaps new openings within this arena and what may have changed over the last few years?

A It's rare that people talk excitedly about regulation, but healthcare is one of those domains where I think regulation can be a tailwind for innovation and category creation. Um, there's lots of historical examples of this. I think the most Traditional example that a lot of people point to is electronic health records, you know, did not really exist in adoption, major adoption, until, ah, the meaningful use, um, law came into play, where the government literally paid financial incentives to doctors to adopt digitized technologies for medical record storage. So that was really the sea change that, you know, drove, you know, so much of the digitization of our, of our infrastructure layer of healthcare. Similarly, right now, we have a number of regulatory tailwinds that are driving payment reform, payment related reform, I should say. And so, you know, we have things like a price transparency law that went into effect over the last couple of years that forced, uh, hospitals and insurance companies to publish their contracted rates. It was hugely controversial. There's still lawsuits in play. People are still pushing back, but the fact of the matter is we now have, you know, thousands of hospitals and hundreds of payers who have published all this data. Obviously they're publishing it in forms that make it very, very difficult to parse. And so entire companies exist to Actually, um…

AI assessment note: “healthcare is one of those domains where I think regulation can be a tailwind”

Q alluded to earlier, which you're giving, in this case, the company access to that information so that they can utilize it more effectively. Are there any other themes that you'd call out here in terms of how founders can look at this industry and say, ha, there's this data opacity. Let me solve that. Or any other gaps that you see that are maybe also just waiting to be addressed?

A So many. So we actually wrote a piece, um, a few months ago, uh, called Payviters Unbundled. And so what we did was we looked at the biggest companies in the healthcare space and even in, in like markets in general. Most of them are large insurance companies that also have a provider can put into their business. So they're called Payviters because they're sort of vertically integrated across Insurance and, and care delivery services. And we sort of did a breakdown of what are the drivers of their, of their business models? What are the kind of key components of them? What are opportunities to basically do what they're already doing, 10 X better as a startup and directly compete, but then also articulated a number of underserved areas that, you know, those incumbents are not really paying attention to where there's sort of white space opportunity for startups. And so, you know, a couple of examples there, core insurance products are, you know, Obviously a place where there's a lot left to be desired in terms of everything from user experience to the cost, the set of services that you get as part of an insurance plan. But we also recognize, especially from, you know, seeing insure tech play out outside of healthcare, you know, it's a really hard business to build and a really hard business model to get right. And so that said, you know, we think there's a huge opportunity for sor…

AI assessment note: “we think there's a huge opportunity for sort of neo carriers, so to speak”

Q any A-C-C-Z fund. For more details, please see A-C-C-Z dot com slash disclosures. There are so many different areas of technology that people are talking about today, whether it be AI or longevity science or rocketry, and those are all very exciting, but marketplaces is something that A-sixteen Z keeps coming back to. And so Connie, I want to hear from you. Why are marketplaces still so important to study?

A We are obsessed with marketplaces, and we have been since our inception. A lot of that comes from actually our partner, Jeff Jordan, who himself has been an operator and a pioneer in marketplaces, holding leadership roles at eBay, at OpenTable, and because of his experience, we have all been trained to believe that marketplaces are truly phenomenal, phenomenal business models, and there's a couple reasons for that. We are in the venture capital business, so we are looking for outsized returns. Many of the biggest returns historically have all been marketplaces, and this is because marketplaces generally share two traits that are incredibly valuable and attractive to investors. First, they generally have a very strong theory of defensibility. A marketplace, the larger it gets, the harder it is to disrupt. Let's think about the biggest marketplaces today. I mean, Craigslist, It's still a vibrant marketplace. And you look at this website, and you're thinking like, these are links. This is just text. There's no photos. There's no videos. You don't need it, apparently, because that marketplace has liquidity. It has traffic of buyers and sellers. It has traffic of a renter and landlords, and so you still get lots of matches that are happening. People try and pick it away, but it's still working, right? So theory of defensibility is very strong with marketplaces, which is something we…

AI assessment note: “Many of the biggest returns historically have all been marketplaces, and this is because”

Q Another example of this is Teemu, right? It's not on the list because it's not a private company, but that thing has grown massively. What can we learn from Teemu as it's kind of risen in the last year or so?

A For sure. So Teemu doesn't have video. They don't focus on video, but they also don't need it for the types of products that they're selling, right? Teemu is a monster in terms of its growth. I mean, Olivia and I chat about the stats. We're tracking it all the time. It is It's insane how quickly this company has grown largely, um, catalyzed by paid ads, but it has continued to grow. And the repeat purchase based off the credit card data that we're seeing is strong. Now, what I think is really interesting and the major nugget to learn from Timu is product discovery personalization. Also, gamification of shopping. They have lots of coupons and games that you can even play. I remember Olivia was playing like some fishing game inside of Timu to win some free slippers, and it's just something fun that we do, right? It makes shopping fun again. But going back to that personalization, which I think is a very new thing for marketplaces and commerce in the U.S. in general to adopt. When we go on Amazon, we all go to the search bar, but that's not how we shop in real life. When I go to Costco, when I go to Target, I go in to buy one thing. I leave with like 10 things, right? Usually more. And the natural way that we shop in real life is we get inspiration as we go along. And the reason why Timu is even more powerful than replicating a trip to Target is because the more I browse it, the m…

AI assessment note: “the major nugget to learn from Timu is product discovery personalization. Also, gamification of shopping.”

Q any A-C-C-Z fund. For more details, please see A-C-C-Z dot com slash disclosures. There are so many different areas of technology that people are talking about today, whether it be AI or longevity science or rocketry, and those are all very exciting, but marketplaces is something that A-sixteen Z keeps coming back to. And so Connie, I want to hear from you. Why are marketplaces still so important to study?

A We are obsessed with marketplaces, and we have been since our inception. A lot of that comes from actually our partner, Jeff Jordan, who himself has been an operator and a pioneer in marketplaces, holding leadership roles at eBay, at OpenTable, and because of his experience, we have all been trained to believe that marketplaces are truly phenomenal, phenomenal business models, and there's a couple reasons for that. We are in the venture capital business, so we are looking for outsized returns. Many of the biggest returns historically have all been marketplaces, and this is because marketplaces generally share two traits that are incredibly valuable and attractive to investors. First, they generally have a very strong theory of defensibility. A marketplace, the larger it gets, the harder it is to disrupt. Let's think about the biggest marketplaces today. I mean, Craigslist, It's still a vibrant marketplace. And you look at this website, and you're thinking like, these are links. This is just text. There's no photos. There's no videos. You don't need it, apparently, because that marketplace has liquidity. It has traffic of buyers and sellers. It has traffic of a renter and landlords, and so you still get lots of matches that are happening. People try and pick it away, but it's still working, right? So theory of defensibility is very strong with marketplaces, which is something we…

AI assessment note: “marketplaces generally share two traits that are incredibly valuable and attractive to investors.”

Q to be talking about all things audio, creators, communities, social. So why don't we just start out because both of you have been podcasting for the last couple of years on your Good Time show, which started in 2020. What made you start that show? Was it the fact that you wanted to build a network? Was it just for fun? What was going through your mind at the time?

A Kind of a happy accident. So this is 2020 December, kind of the thick of the pandemic, and we're all working from home. We miss seeing our friends. We used to go home to India once a year, and we didn't do that that year. And, you know, we basically thought, wouldn't it be nice to kind of host like a virtual dinner party kind of thing? Because we used to host all like these founders at home. Uh, once a month or so and just have conversations about building companies and startups and stuff. And we just didn't do that in person anymore. So we thought, you know, we, we could maybe host this as a live audio session. We started this on Clubhouse and that's kind of how the whole show started.

AI assessment note: “we basically thought, wouldn't it be nice to kind of host like a virtual dinner party”

Q the perfect manifestation of how technology always is like much better in one direction when it's invented, but there's always like these things that need to be iterated on. And so are there other things worth knowing about whether it's These negative prompts, whether it's these glitches that are still in the matrix. What would you call out from your, again, many hours of being deep in, in these tools?

A I think it depends on the model. One example was when DALI came out. It's not very good at understanding that it's drawing things in a square. So if, if, if you're drawing a person, it's often going to have like its feet and its head cut off because it's, it's seeing those in portrait photos. But one thing you could do with DALI is you can actually upload like an image To, like, do variations of, and if you upload an image that's just like a little white border, um, then it knows that nothing can go there, and that, that kind of encourages it, forces it to kind of think inside the box, if you will. But then, of course, you have now tools like Midjourney, who've been, like, iterating on their text-to-image model, like, a lot more aggressively than OpenAI, who Understandably, I think maybe have some other things in the, in the, in the cooker, you know, which have now grown that into the model itself. So when you type things in, it knows it's a square and actually it would sometimes do quite clever things in order to fit it in that space. So if you ask for kind of like a group selfie of three people, you know, on something like Dally, that's going to be cut off at the end because you're used to seeing someone taking like a disposable camera photo. Whereas It's clever enough to know that one of them kind of needs to be standing behind the other or like leaning in from the side. So …

AI assessment note: “One example was when DALI came out. It's not very good at understanding”

Q there Certain learnings, maybe the eighty-twenty approach of becoming a good prompt engineer in terms of things that you think are, are really valuable to understand. Maybe it's the prompt length. Maybe it's using certain modifiers within your prompt. Maybe it's just like a, a framework for thinking about prompting. Is there anything that's surface that you think would be really valuable to someone who's just starting out with prompting?

A Oh yeah. Like I think if you've never used one before, like the best Way to explain how they work is to always like describe something as if it already exists. Imagine that it's an image in some kind of downloadable clip art library, or it's, it's on, it's on a photography gallery, and you know, someone's written underneath. Oh, this is a fine example of a modern, uh, photography shot. And those are the kind of descriptions that you're trying to kind of mimic to tell these kind of tools. What you're looking for. And it also gives it like a natural sense of why these tools are bad at some things and, and the kind of prompts that don't really work because you never, you know, if there's like a, let's say some, uh, like an archive image of some women celebrating on the steps of a church in 1972, it will have that kind of caption, but they never go, the woman on the left is wearing a yellow hat. The woman on the right is wearing, you know, they just don't go into that because you can see it. So ironically, they often describe generally what the image is about, but not like how you would draw it step by step. And that's why these tools are less good at saying like, I want this thing over here and then that thing next to it and then something on top and that thing should be much bigger because that's in the real life. That's not how images are described in language.

AI assessment note: “the best Way to explain how they work is to always like describe something as if it already exists.”

Q the perfect manifestation of how technology always is like much better in one direction when it's invented, but there's always like these things that need to be iterated on. And so are there other things worth knowing about whether it's These negative prompts, whether it's these glitches that are still in the matrix. What would you call out from your, again, many hours of being deep in, in these tools?

A I think it depends on the model. One example was when DALI came out. It's not very good at understanding that it's drawing things in a square. So if, if, if you're drawing a person, it's often going to have like its feet and its head cut off because it's, it's seeing those in portrait photos. But one thing you could do with DALI is you can actually upload like an image To, like, do variations of, and if you upload an image that's just like a little white border, um, then it knows that nothing can go there, and that, that kind of encourages it, forces it to kind of think inside the box, if you will. But then, of course, you have now tools like Midjourney, who've been, like, iterating on their text-to-image model, like, a lot more aggressively than OpenAI, who Understandably, I think maybe have some other things in the, in the, in the cooker, you know, which have now grown that into the model itself. So when you type things in, it knows it's a square and actually it would sometimes do quite clever things in order to fit it in that space. So if you ask for kind of like a group selfie of three people, you know, on something like Dally, that's going to be cut off at the end because you're used to seeing someone taking like a disposable camera photo. Whereas It's clever enough to know that one of them kind of needs to be standing behind the other or like leaning in from the side. So …

AI assessment note: “One example was when DALI came out. It's not very good at understanding”

Q to call out a few and get you to quickly explain what this is, how it works, because I think many people are maybe familiar with DoNotPay's parking ticket support, things like crafting a cease and desist letter, but let me just give you a few examples that I was surprised by. So one of them that I saw on your website Is free raffle tickets. What does that mean?

A So there's this obscure law in America that says that, um, you can enter any competition for free. So you hear on the radio, you can pay to enter, but they always bury in the terms of service that if you mail an obscure letter to an obscure address, you can enter for free. So we've built a product that automates free entry into any competition. So you just paste in the terms of service, or you can even just choose from Competitions that our bot scrapes on websites and it enters it for you for free. And this is the thing with do not pay. There are so many laws that help consumers that no one has the time to actually follow through with, and that's a great job for technology.

AI assessment note: “we've built a product that automates free entry into any competition.”

Q necessary and crafting them almost like a template. And then a third generation, which is using AI, which is not just a template, but really bringing in all of the prior examples for a Particular letter and crafting something net new. How do you think about which use cases require step one, step two, or step three, or how are you assessing where to apply some of these new technologies?

A So we started at step two. And so what we did was we said, well, all of these letters that you had to pay a lawyer 300 dollars for, we're going to automate with document automation, which is a very simple technology, but it's something that hadn't been done for these types of letters we were doing. What's really exciting about step three is that sometimes the companies respond and the governments respond and step three with AI allows us to respond back instantly. So instead of helping someone with a hundred dollar dispute with Comcast, we can now work on 10,000 dollar medical bills because Comcast or United might let that hundred dollars go, but no hospital is just going to let 10,000 dollars go just because of one letter. And so you really have to have that conversation back and forth. And so what we're working on now is actually communicating life, um, on medical bills specifically, and I can go into the laws we're using around that. That will be kind of our first major AI product, um, in Do Not Pay.

AI assessment note: “instead of helping someone with a hundred dollar dispute... we can now work on 10,000 dollar medical bills”

Q Something that OpenAI has also done, though, is had humans as part of the training process to really hone it in. Are you also doing that? Are you bringing in lawyers who are specialized in these areas to take a look at the output from your models to hone them in or improve them?

A Yeah, we have some of the best lawyers in the country helping us, like at Wilson Sonsini and other firms. The AI has been very problematic for us, so there's really two problems with it. The first is that it's dishonest. So with Comcast, for example, it says things like, I've had three internet outages in the past 24 hours, and that might be a good way of getting a refund from Comcast, but it's not true, and it can cause liability issues. And in court, if you lie, that's a crime, and it could send people to jail. The second issue is that it talks too much. With Comcast, or even in a courtroom, there are some questions and human speech that don't require a response. It's like a rhetorical question. So imagine the judge says something like, um, hold on, let me take a look at the case. Uh, AI models would often say, okay, thank you. And by the way, I'm innocent for like these six reasons. The judge will just get annoyed. You have to build other AIs to decide whether to say something, let alone what to say. And so you have to really retrain it and also build these guardrails. Otherwise things get out of control.

AI assessment note: “we have some of the best lawyers in the country helping us, like at Wilson Sonsini”

Q That's, that's a good perspective. I think another amazing thing that I've seen, the technology can be cross-border as well. So Typically, a lawyer is specialized in one very particular area, maybe in one jurisdiction. Can you tell me a little bit more about how you're thinking about the potential for this technology to support people, not just in your region, but really anywhere?

A Yeah, the benefit of AI is it can read 10,000 documents and produce an answer in seconds, and not even the best human lawyer can do that. So I think the example you're talking about is a consumer rights issue relating to a timeshare. We had a elderly consumer in Boston. They got suckered into a timeshare actually in Mexico and do not pay got them out of it because there's a five day law in Mexico to cancel timeshare contracts. And so things like that, where you can help consumers in Boston, uh, for issues in Mexico, most lawyers in Boston don't know about Mexican law, but AI does because you can just feed it all the timeshare cases and it comes out with a way out. And the same is true actually for Comcast. These generic AI models like, um, DaVinci from OpenAI, they're, they're not actually that good at the law. What we've had to do at DoNotPay is you feed it documents. So we say, we feed, fed it all of these outcomes from the templates in many years of operating. And you say, based on these thousand documents, write X. And then it becomes much better than just a generic usage of any of the models.

AI assessment note: “things like that, where you can help consumers in Boston, uh, for issues in Mexico”

Q in a way, because after running a company for many years, the reaction of most founders is, let me take a break. And your reaction was actually, no, I see this opportunity. I want to go all in on Sandbox. And so tell me a little bit more about that. What was the vision you had in 2016 that made you say, hey, I need to do this right now?

A I think it was a fear of missing out. You know, the painful lesson of missing out during the mobile era. Was, uh, really ingrained. It's like, man, I wish I was there earlier. So when VR first came about, I was like, okay, this is going to be a consumer headset. Oculus is going to launch and HTC Vive is going to launch. We have to get there early. So I just decided to dive down and just go with it. You know, I think what made it easy for, uh, me and my team at that time was also, we knew how to build games. We built games from multi-platform. We knew how to build in Unity, which was used to build in VR. And I also had a few folks left in Blue Tea, and that's how everything started.

AI assessment note: “I think it was a fear of missing out.”

Q you're finally hitting the traction you're looking for. And this is one of the reasons I love your story is that there was just so many roadblocks that emerged out of nowhere. And the next one for you sounds like it was the pandemic, which is pretty fundamental to your business because it's an in-person business. And so tell me about that. How did you stay alive? During this period.

A Yeah, it was tough. People would say during the recession, companies will lose revenue by 30 to 60%. For us, we lost a 110% of our revenue, because not only did people, we couldn't open any of our stores, so there was no sales, but we also had to refund everyone that booked. And the outcome was that, was we had a great team, but, you know, we had only, we were trying to raise that Series B during the early 2020, and we had about three, three months of runway left. So we had to let go most of the folks. We let go about 80% of our team. That was very difficult. There were a lot of great people at the team that we wanted to keep, but we couldn't. So how do we make it work? I think, to a degree, it was the same thing that I've done early in Sandbox, where we sat down and try to get all the stakeholders to agree to fight on. I mean, it was much bigger this time. It was, um, getting the investors that backed us. Like, hey, this is what we're gonna do. This is how we're gonna survive it. Do I have your support? Getting the people that remain in the team to get the support. Getting the secured creditors and unsecured creditors. Like, hey, we need to, um, make sure Sandbox have a long enough runway so that we can all survive. And just creating a very big plan out of that. Now, the challenge was we couldn't convince everyone. And because of that, we had to go through chapter 11. We neede…

AI assessment note: “we had to go through chapter 11. We needed protection to get through that”

Q in a way, because after running a company for many years, the reaction of most founders is, let me take a break. And your reaction was actually, no, I see this opportunity. I want to go all in on Sandbox. And so tell me a little bit more about that. What was the vision you had in 2016 that made you say, hey, I need to do this right now?

A I think it was a fear of missing out. You know, the painful lesson of missing out during the mobile era. Was, uh, really ingrained. It's like, man, I wish I was there earlier. So when VR first came about, I was like, okay, this is going to be a consumer headset. Oculus is going to launch and HTC Vive is going to launch. We have to get there early. So I just decided to dive down and just go with it. You know, I think what made it easy for, uh, me and my team at that time was also, we knew how to build games. We built games from multi-platform. We knew how to build in Unity, which was used to build in VR. And I also had a few folks left in Blue Tea, and that's how everything started.

AI assessment note: “I think it was a fear of missing out. You know, the painful lesson”

Q you're finally hitting the traction you're looking for. And this is one of the reasons I love your story is that there was just so many roadblocks that emerged out of nowhere. And the next one for you sounds like it was the pandemic, which is pretty fundamental to your business because it's an in-person business. And so tell me about that. How did you stay alive? During this period.

A Yeah, it was tough. People would say during the recession, companies will lose revenue by 30 to 60%. For us, we lost a 110% of our revenue, because not only did people, we couldn't open any of our stores, so there was no sales, but we also had to refund everyone that booked. And the outcome was that, was we had a great team, but, you know, we had only, we were trying to raise that Series B during the early 2020, and we had about three, three months of runway left. So we had to let go most of the folks. We let go about 80% of our team. That was very difficult. There were a lot of great people at the team that we wanted to keep, but we couldn't. So how do we make it work? I think, to a degree, it was the same thing that I've done early in Sandbox, where we sat down and try to get all the stakeholders to agree to fight on. I mean, it was much bigger this time. It was, um, getting the investors that backed us. Like, hey, this is what we're gonna do. This is how we're gonna survive it. Do I have your support? Getting the people that remain in the team to get the support. Getting the secured creditors and unsecured creditors. Like, hey, we need to, um, make sure Sandbox have a long enough runway so that we can all survive. And just creating a very big plan out of that. Now, the challenge was we couldn't convince everyone. And because of that, we had to go through chapter 11. We neede…

AI assessment note: “we let go about 80% of our team... we had to go through chapter 11”

Q new game every single day, and they'd all be amazing. And there'd be this endless suite of Games that people could try out, but obviously there's a resource constraint. And so tell me a little bit more about how you think about the rollout of new games and also the amount of resources it takes to build a new game. I think you have six right now. Is that correct?

A Yeah, that's correct. So generally, it takes about 12 months to build a title. Internally, three months of pre-production, um, nine months of production work, but we also work post-production to make sure things look, uh, goes according to plan or any changes that we make while it's live to the audience. So the way we think about content is, in the beginning, we have to build everything ourselves, because it doesn't exist. But over time, you know, other people can build on our platform. That if you're a VR game developer, over time, you can find it more lucrative to build on Sandbox than other options. So how we scale location is a factor of that. The more locations we have, the more tickets we sell. The more tickets we sell, the more third-party developers interested in us. So in the beginning, we built a lot of locations ourselves, and we've proven this to be a very lucrative retail business model, and over time, franchising will really kick in And really bring sandbox all across the world. Once you have that, then you can start needing developers that, you know, with their unique skill set to bring that. Either local content or IP branded content they're really good at, or genres they're amazing at, and just create this ecosystem.

AI assessment note: “generally, it takes about 12 months to build a title. Internally, three months”

Q word fun. Um, and after I learned to code, I saw it in a different light. And I think that's really important is this idea where it's like, how do you make coding fun? So what are your thoughts there? How have you been able to design Replit to actually enhance people's creativity and make them want to come back and make them see this skill in a new light?

A So I think most things that are fun tend to be devoid of a lot of drudgerous routine work, right? You know, when, when you're playing a video game, you're not like building the video game every time or setting up the TV or, or, or doing some like rote IT task, right? When you are, um, doing a sports hobby, you're in the flow, you're doing the thing you're excited about doing. The problem with coding is that a lot of the maintenance around the development environment and the packages and the integration of all the different components, Was the thing that engineers were spending most of their time doing. The moments they were coding, they were an absolute bliss, but those moments were actually very little in terms of the, if you think about the pie charts of what it meant to work as a programmer. So the first thing that Repli did is remove the need to do all this setup. That in itself made programming a lot more, uh, fun. And then add the collaborative aspect. Like a lot of what we find fun in life has to do with other people. We're just social animals, right? And so like, if I can share my program with you with just a link, that's really fun. That's what makes Figma fun. That's what makes Any other collaborative tool fun is that I can just like send you a link and you're, you're in there with me, or you can play it and try it out. And then finally, there's like a lot of like exp…

AI assessment note: “So the first thing that Repli did is remove the need to do all this setup.”

Q That's awesome. Let's move on to the next one. Someone has used Replit to build over 50 projects. Now you can use your own definition of project here. Um, these don't need to be full blown startups, but have you seen someone build that magnitude of projects on Replit?

A Yeah, I, I think that's, um, that's fairly reasonable number of projects. So there's a six-year-old developer. His name is Ray Han. He's one of our most prolific programmers. He actually built a bunch of interesting projects where he reverse engineered How Replit works. And he built, he built an unofficial API. One of his unofficial APIs is a security program that searches people's repls to find discord tokens. So a lot of people build discord bots on Replit and they copy and paste the tokens in clear text as opposed to putting them in our encrypted service, the secrets manager. And so he would Find those tokens. He would invalidate them because discord has service to invalidate those. And then he would send them a notification. He would say like, Hey, like we found that you've exposed your token. Uh, so that's one example of a project he made. He w he was, and Scylla is one of the most prolific, uh, bounty hunters. And he built one of our earliest bounties, which was A startup that wanted to build like a stable diffusion based t-shirt generator. So you would share a t-shirt based on a prompt and then get it printed and sent to you. Like there isn't one week where I don't see Rayhan producing like a new piece of software.

AI assessment note: “Yeah, I, I think that's, um, that's fairly reasonable number of projects.”