 Welcome to SuperCloud 6. I'm Hawaii Xu, AI and the Cloud Executive in Silicon Valley for a very long time. Today, I'm the panel host for AI Founder's Day. In this session, I'm very happy to have my friend, Joshua, Founder CEO of Do Not Pay. You want to tell us about the story about why it's Do Not Pay, because everyone loved the name. So tell us about yourself, and then why and how you get here. Thanks so much for having me. So I'm an accidental entrepreneur. I started my company Do Not Pay by accident. When I turned 18, I moved from England to study at Stanford. And I got a huge number of parking tickets. I think about 30 parking tickets between the ages of 18 and 20. And I couldn't afford to pay these expensive tickets. And so I had to find out other ways to get tickets. So this is going to be an example of innovation out of necessity. Exactly. And I learned something remarkable, which is if you know the right things to say, you can save a lot of money and get out of your tickets. And so I built Do Not Pay just for my family and friends. But it turns out everyone in the world hates parking tickets. And I went from 10 cases in my first week to 50,000 cases in the first few weeks. And that's what made me realize that the idea of helping people with their consumer rights is bigger than just tickets. And so fast forward eight years, I've been working on Do Not Pay and expanded to hundreds of areas of the law. So you started a company after you became an EIR intern at Greylock or during that? It was already a website, but there was no business model. It was a completely free public service. And I worked with amazing VC firms like Greylock to turn it into a real company. So Greylock helped you to sort of steer from an interesting project to an interesting company. Exactly. And I think what all the VC's told me is with every business, the question is why now? Why hasn't this idea been done before? And with Do Not Pay, the answer was that there were companies that helped people with tickets, other companies that helped people with delayed flights, still other companies that helped people with bank fees. But there was not one company that did everything in a subscription model. And that's why I wanted to be different. And that's how Do Not Pay has seen some success where others have failed. So you started this company around the 2016? That's right. So this was before the AI craze. This was when if you had a decision tree or a rules-based system, it seemed like AI and very sophisticated. These days, obviously we're using much more advanced technology. So when you started a company, decision tree plus rules were sufficient to do Do Not Pay. Exactly. How was the effect in the sense that it didn't work? Or did you feel like, hey, I wish I had better technology? Like what was the state back then? It worked decently well. So it would go down a decision tree and ask questions about the person's parking ticket. So it would say, were the signs too hard to read? Or was the parking base too small to fit your car in? And then it would go down the nodes and eventually it would match you to a template letter, take down all the details, put those details into a letter and send it off to the government. The problem that we had with that is the technology was not idiot-proof. So people would come and say they wouldn't know which option to press on the decision tree. Or they would come with a problem where we didn't perfectly craft the options to match that problem. And that's why AI is much more useful. And one other thing is sometimes it's not just tickets, we do many other things since then. The governments or the companies would respond and if you have AI you can respond to what they're saying. So it's not just about sending a template and hoping for the best, it's having a conversation with these governments and companies now. So when you started the company 2016, it was mostly decision tree matching some rules and it helped people to come up with a letter and then send to whoever and that's it, right? But you are saying that ideally I wanted to be able to have a conversation with whoever I'm talking to. So that's why you need a more sophisticated AI. So when did that happen? Is that a Czech GPT moment or when GPT was released in 2020? Like when did you feel like, wow, this is a much better AI for what we wanted to do? I think when GPT-3 came out in 2021, everything changed. Before that the models weren't good enough to hold a conversation, but from my perspective, GPT-3 was the first that could hold a real conversation for what we were trying to do. And that's when we started building it into our products. One other thing that helped us a lot is that COVID probably shifted the legal system for 10 years. In the past, a lot of things you had to go in person to do like fill in tax forms and all of these things. But with COVID the government changed the rules so that a lot of stuff could be done online now. Yeah, you mentioned to me before that, you know, like I don't know whether that's because COVID or not, like literally you can just craft a letter and they're using some API from the United States post office, right, and they send the letter out and then get it certified or whatnot. Is that because COVID, they provided those API or the API was there but IRS didn't want to acknowledge it. So one example is in Silicon Valley we have to file these tax forms called 83B elections, which is when you start or an investor in the company, you have to declare that investment to the IRS to minimize your taxes going down the road. And before COVID the IRS said that we do not accept digital signatures. So you had to literally sign it with a pen like the Stone Ages. But when COVID came in, they changed the rules so that they said that documents like that could be signed on DocuSign or with other e-signatures. And so that meant like many other legal forms, you could do this stuff online now. And then now that you can sign it online, it's easy to build an integration with a postal service to send the document automatically. So the API was there before but the IRS didn't want to take your digital signature before. Exactly, and also you saw this with a lot of big companies where they wouldn't even allow you to cancel your subscription online. They would force you to go in person, which should be illegal but it's not in some cases. But with COVID and those things, those rules were thrown out of the window. That's right. Everything shifted online and our mission is to give people consumer rights online. And so our toolbox expanded dramatically. So from your perspective, pandemics actually made a more impact to do not pay or do not pay customers than the Genitive AI or Charger BT made an impact. I think it's a combination. I'm so lucky I've been an entrepreneur long enough to see the world change so that our business is even more exciting. So both are big, you know? Yeah. And then just to, you know, let's dig a little bit on the technology. The, what is the Charger BT? You said a 2020-21, you know, it's a big deal. Did you like almost rewrite your system? Now that you have the Charger BT or the GPT API? Or, you know, there are a few corner cases. It's a very hard to do. Now you extend it and edit a more module. Like what's the impact to your entire tech stack? The biggest thing was it expanded what was possible. So one product that we wouldn't be able to do without Charger BT is something we call AI bill negotiation. So in this, the AI logs into someone's utility account like Xfinity and starts chatting with the big company to negotiate their bill down. And Charger BT will say things like, I've been a very loyal customer but I've had internet outages and I'll switch if you don't give me a discount right now. And then the Comcast agents will respond and oftentimes the big companies are using AI as well. So the two AIs are chatting with each other to negotiate someone's bill. And that would never have been possible five years ago. But it's happening today? It's happening today. And we've been doing it for a few years now but what's really exciting is in the next few weeks we're allowing the consumer to watch their own AI bots negotiate on their behalf. So it's almost like an entertainment aspect as well. ESPN for saving money where you can watch bots fight for you. Wow, ESPN for saving money. I like that tagline, that's good. So ESPN for do not pay for saving money but tell us a little more about all the use cases. You started a company for do not pay the parking tickets. Now you do way broader than that. So maybe what are the categories of the things or more examples, more interesting use cases? I think the biggest shame with AI is it's typically used by big companies and governments to rip people off and steal people's privacy and things like that. So across our 200 products, the under-arching theme is we want to give power to the people. So we help people cancel subscriptions, get refunds, negotiate bills, file insurance claims, get out of fees and parking tickets. And it's all of these areas where big companies and governments know that individuals don't have the time to jump through all these hoops and fight back. I joke that we don't need AI to cancel a subscription but America is such a broken country that unfortunately you do need AI to cancel subscriptions. So one example is with magazine subscriptions, they make you chat with a person just to cancel the subscription. You have to go on the New York Times website and chat with an agent to get out of your subscription and that's a good job for AI. No one has time to wait on hold for six hours to save like $10. And so that's a good job for these robots. And then so far, I'm assuming do not pay only allows you to do things that's, you know, you can log on to the website and then do things. Not quite like making a phone call yet, right? Not yet, but there are things we've integrated with the real world where it's not just online. So we have an integration with the postal service like you mentioned to send mail. AI can send faxes, believe it or not, people still send faxes to do some stuff. It can dial phone trees. So we have one product, it will phone up the DMV and it won't actually talk to anyone. It will just press buttons to get you an appointment on the DMV automated phone line. So we are moving to interact. Okay, sorry, in that case, your bots actually call the DMV. Yes. And then through the prompts, you know, making certain selections and it's a lot, hey, you know, I wanted to have a disappointment on April 1st. Yes. It's already on. And it will press the buttons and work with, it's a similar situation. And without human doing anything. Yeah, you have one robot talking to a phone tree robot. And six months ago, I think it wasn't possible for AI to hold a phone conversation. The latency is too much, but now you're seeing all these demos and exciting products where the AI can respond with voice in real time. And so we're working very hard on phone bots as well to help people. So what is the large language model behind the scene? Is that open source models or frontier models from Anthropic, open AI, Google of the world? We're using a combination of models, but I think GPT-4 is the most helpful for what we're doing. There are some things that are too empowering to people. So they might not be approved by open AI. So we use open source models for that, like GPT-J. So what do you give me an example? For example, maybe we want to be really aggressive to a company and get big discounts or something. And the GPT-4 would have refused to do that? Well, we value it so much that we want to stay on the right side of the terms of service. Rarer, right, but GPT-4 sometimes would have. One example is contacting the government. So we have products that contact the government and there's an open question as to whether you can use GPT-4 for that. Got it, got it. So it's sort of in the gray area. Yes. So not as much as GPT-4 model refuses to answer the question. It's really, do you want to use it? About the limitations, yeah. The limitation of it. Yes, and I'm a big AI optimist, but these models aren't perfect. The biggest problem that we've seen is it makes things up all the time. And so one example with, on the build negotiation side, it will say, I've had five internet outages in the past 24 hours for the consumer. And while that might be an effective tactic from a liability perspective, you can't have the AI lie all the time. So we've had to tell it in the prompt, stick to the facts, stick to the truth. And we actually have another AI that kind of checks if it's hallucinating. So we have one AI watch another AI to talk to another AI. And that's the future of AI negotiation. So that's interesting. So basically, AI may inevitably hallucinate but you have another subsystem checking on the result all the time. Yes, you have to have a watch tower. That's fantastic. So in terms of the, everything is on the cloud, I'm assuming your tech stack, right? You know, calling. What is the complexity of it? For instance, right? You know, hey, I wish my, I wish the GPT model to be far better. I'm waiting for that day or, you know, llama or whatnot is the open source model enough. But I just figure out how to deploy like on your engineering side. What are the tough problems are you solving? Codebase has gotten way too complicated. And we see it as actually a chain of agents, not some AI and some not AI. So one agent will send a letter in the mail. One agent will record what the company was saying with OCR and scan the letter. And it's like a teamwork exercise of all these different automated agents. And it's very complicated. It's all in the cloud and things are passing one task off to the other. And I think there's two schools of thought with AI. The first is that AI will be really useful because the contact window will be huge. And then the second school of thought is the contact window doesn't have to be that big. You can just have lots of agents in past tasks off. And I think we believe in the second that it's an agent-driven future. Agent-driven future. So you are not quite waiting for models to have larger contacts window or the efficacy to improve. Of course, they will improve, but that's not what you're waiting for. You feel like you are going to work with whatever you have. And then it's not just a short-term patching or work-arounds. It's actually sustainable for years to come. That's your view. That's what we think. We think of it like a lot of ants, like carrying an apple. They team up and they manage to move the apple. One ant is useless, but together they can move the apple. So when you look at a GPT-4, GPT in the future, GPT-5, GPT-6, it's still ants, maybe bigger ants. They're still ants. Exactly. You want lots of ants to work together to lift the apple. Yes, and we're very experienced. In 2015, the ants were very dumb, but now they've gotten a lot smarter. So we're just happy they've gotten this far. And we're not holding out hope that they'll get smarter. I think they will, but we can build really useful stuff with the current models. So from that point of view, do you feel like you are going to be very excited when, or anticipating next generation with a big, large model? Yeah, if they come, that's great, but I plan to do no matter what, that sort of things. I think the biggest thing is the features of the model. So the most exciting thing for us of GPT-4 was multimodal capabilities. So a lot of the law and consumer rights is about evidence. So imagine if something can look at a medical bill and say, what's the error in the medical bill? Or look at a sign for parking tickets and say, why is this sign getting you out of the ticket? And so multimodal is really exciting for us. But in terms of the model... So you're already using the image or video recognition to as a part of the evidence? Definitely. So you ask your customers to upload those pictures so that you can make sense out of it? We pull them from Street View as well, Google Street View. And then the other thing I'm excited about is latency. So to do phone conversations, you really need quick models and even synthetic voice models to generate voices and audio quickly. And if latency goes down, you can accomplish a lot more in the real world. So I'm much more excited by features and latency than I am by them getting smarter because it doesn't take a genius to negotiate a Comcast bill. Interesting. What about cost? You mentioned the latency and then, you know, what about the cost, the actual dollar cost? Are you worried about it? How big is the... How big part of it? We've got a great business model where we work for our customers. This is a subscription business. We have hundreds of thousands of subscribers. And because of that, we're not too worried about cost. We've been very encouraged by the fact that it's gone down 90% and I think it will go down further. But right now it's not a big issue for us. Well, it's not a big issue today. It's going to be even less an issue moving forward. Because, you know, we all know that AI infrastructure cost, the marginal cost will go down, right? Maybe even several auto-magnitude lower in a few years. Yeah, and there are some things that we haven't done because it would be too expensive. So imagine AI bot that scans every website you visit. That would probably be too costly if you want to use a good model like GPT-4 because people will visit hundreds or thousands of websites a day if you're a power user like me. So maybe that would be too expensive. What's the use case to scan all the websites? To see how you're being ripped off, maybe scanning terms of service, cancellation policies, privacy policies, things like that. Before I tell you that, hey, can you cancel this thing? You actually remind me, hey, for this thing you are paying too much. Is that proactively telling me? That's right. We want to go from retroactive. So a customer saying I want to refund to proactive saying, hey, I saved you $50 and you didn't even know about it while you were sleeping. And I think once cost goes down, you can run a lot of these scenarios in a lot more detail to get to that outcome. Wow, wow. So, hundreds of thousands of subscribers, you are providing value to them. What's next? What to expect, right? You have 200 products, meaning that 200 potentials you save me money. What's next? We want to take all of the data that our subscribers give us and figure out ways to make the money and save them time. So one product we're working on right now is an AI that scans emails and looks for legal settlements you're a part of and you didn't even know about. Which I got those from letter, mail, I just don't have time to deal with it. Yes, so it will find it and claim it on your behalf and hopefully one day you'll wake up and say, oh wow, it got me $100 that I didn't even know about or even $5 that you couldn't have the time to claim. Interesting, so how do your subscribers justify paying you this monthly fee? What's the latest? It's $18 a month. Right, so what is the, how do they feel like, hey, every month I need to fight with people? What's the typical mentality people who are subscribing to your services? So this is a huge product challenge for us because the use cases that we fight are very episodic. They don't happen unless you're a bad driver like me, the average person only gets a ticket once a year. So what we do is we bring people in for use case like tickets or getting a refund and then we show them other ways to save money using our product. So it will look at your in-flight wifi receipts and get you money for the in-flight wifi because it never works. And the ROI is typically 3x the subscription cost. And then you feel like you are expanding use cases. That 3x will be 5x, 10x. Yes. And it's also 1%, 99% cheaper than a legal lawyer to fight for these use cases. We even with a parking ticket, if you ask the lawyer to help you with a parking ticket, it probably costs $100. Right, right. Very interesting. And then what is the median or average amount that you usually save people? $50, $150, what is that interesting amount? It varies, but in some cases, the average is about 400 a year. 400 a year, I see. So, but 400 a year is already more than your cost to pay, do not pay. And we also save people a lot of time. So, we cancel people's Equinox subscriptions, which is very popular in San Francisco, where we are. And because it's very hard to cancel? Yes, and all these tech people subscribe to Equinox and our software will cancel it for you. And Equinox subscription is cheapest plan, is maybe 300 a month. So that can be thousands a year that we save people. So, from that point of view, do you feel like what do not pay is basically your personal assistant? Yes, it's an assistant for fighting for your rights. Another thing we do is we help people with privacy. So we delete that data from being sold and all the data brokers. And you can't put a price on that. Right. And we discussed the cost, that sort of things. You know, you told me one thing that I thought that's pretty amazing. I was asking you, hey, when are you going to raise the next round of money? Well, we're very profitable right now. We actually have more money than we've raised in total. So we pay dividends back to our shareholders. So we're not raising money anytime soon. And my response is, you know, you are the first one. I've ever heard that not only not raising money, but also giving dividends to the shareholders. So you might be the first company I've heard, at least for me, hundreds. I think one of your investors told you that among hundreds of companies in the portfolio, you are the only one doing that, right? So what does that tell people? Like, you know, I could read it, you know, different ways, right? One is, look, you know, Jenny and I, it's hard to be profitable, vast majority. They are not making money. You know, life is tough for application developers that is today for the near future. But the other way to look at it is look, you know, here's an example, right? Here's a role model even, right? And then we should look at that role model and then that's where the world should be, you know, for the next few years. How should people think about it? I think we're entering a new paradigm where you can have a large outcome with a very small team. I think Sam Altman said that there will be one company one day that's worth billions with one person. We're not quite there at Do Not Pay. We have seven people and seven contractors, so around 14 people. But with AI, you can automate so many things like customer service and even product development. You see startups like Cognition building AI engineers. So you can achieve a lot with very little and that's a new model for operating thanks to the technology. Right. What's your advice to founders, entrepreneurs, or people who are getting into Gen AI today? What's your advice to them? My biggest advice is just get started. I could never have imagined that Do Not Pay would be what it is today and I wouldn't even know that people would like it. I just built it for fun and solve for a problem that you experience because at least if you do that, you have one customer like with me in parking tickets. And then there are so many, so much hype going on, right? Whether you call it hype or not debatable, but there's so much energy, right? NVIDIA just announced a much bigger GPU this week. Literally, right? The entire industry invested $100 billion last year, probably another $100 billion for buying GPUs, getting the space and the cooling for the data center, everything. When are we going to see the return of that $100 billion investment? Do Not Pay is a profitable, that's great. But do you see, how do you see the rest of the applications, right? How to think about it? I think it will eventually be a commodity and there are a lot of great products in society that make really helpful for society, but they don't make any money, like airlines. And I worry that all these big tech companies competing will compete away the profits. So I wouldn't quite short NVIDIA right now, but I do think that there will be a lot of other big, big and small competitors who will come in and compete with them. Right, but that's sort of for NVIDIA and then it's competitors, right? What about applications? So there's all this buzz in Silicon Valley about AI company this, AI company that, but I think every company will be an AI company and the most important thing is building a useful product. And what we're seeing is, AI is benefiting incumbents the most. So like with Notion, I was a user of Notion for years and now they have Notion AI, so I use Notion AI. And I think similarly with Do Not Pay, we were fortunate to have our customer base and template business before this came in. And so adding AI to existing things that work, I think is a better bet than just speculating on these AI companies. So what does that mean? Life is tough for the very, very new entrant? Well, there are of course a lot of opportunities because there's some big companies that move very slowly and don't adapt. And if you don't disrupt, you get disrupted. And that's what keeps me up every night. How can we, because when Do Not Pay started it just generated letters, but now you can just ask chatGPT for a letter. So you constantly have to move with the technology and that's what we try to do. That's a very good point, right? You know, eight years ago, I don't imagine how I would have talked to a chatbot to say, hey, write me a letter. I would have give to this company to fight it for this, you know, parking tickets. Nowadays, I'm assuming that, you know, chatGPT is pretty good at it, right? So what chatGPT does is commoditizing some of the services you are doing. So how do you think about this? I would have two responses to that. The first would be that there's some stuff we do. It's too anti-authority for the big companies to touch. These big companies are partners of Microsoft and Google. So Google is never gonna build a robot that fights with Comcast because Comcast is such a big advertiser on Google that it doesn't align with their interests. And then the second thing I would say is that there's a lot of value in the integrations. So AI is just language and language is very important, but the most value is actually building the robots to do the clicking to submit your letter or build the robots to handle the chat or build the phone line robot that actually dials the numbers and has the conversation. So AI and LLMs is just the first step and you need to connect it to the real world to be useful and there's a lot of value to be created in the integration layer from Do Not Pay and also many other enterprise and consumer startups. So a lot of the complexity is at the last mile stage, integration or whatnot. Yeah, there are billion dollar companies that help people configure their Apple devices even before AI and that gives you an idea to how much integration work there is in general. So one last question about the use cases, you can help me to find the parking tickets or whatever the subscription. Do you expect that you also would help me to book airline tickets or do some other stuff? Going on vacation, hey, do this and that or get a ranting ski equipment. How do you see those use cases? The answer is no. And I think that there are two types of agents that will work in AI. The first is positive sum agents. So like you described, an agent that books your flight, rents your car, et cetera. And the reason they're positive sum is it's a win for the company and it's a win for you. It's a win for you because you book your flight and it's a win for the company because they get revenue from you from the ticket. And positive sum use cases I think will be commoditized. Siri will do it, Google will do it, OpenAI will probably do it with the GPT app store and other things like that. And there will be a lot of competition. On the other hand, there will be adversarial agents where you're fighting and the interests of you is different than the interests of the person that agent is communicating with. And that's things like canceling a subscription, negotiating a bill, things like that. And I think with that, there'll be a lot less competition and a lot more specialized agents and companies like Do Not Pay. So we want to stick to the second group because no one else is doing it. Right, so, but me as a consumer, do you expect me to have this agent to book airline tickets, rent the car? You know, when do you think? I think within the next six to 12 months, at longest. Really? But there is one interesting thing which is I don't know if the chat user experience is actually preferable. I think people like clicking buttons. Yeah, yeah. Well, one is chat. The other thing is I just say something and then it will figure out, I don't even need to stay at a chat, right? I just issue a command. That's true. If it's command-based, then I think it will work. I'm talking about command-based. Do you think it would happen in six to 12 months? Like, because I'm actually a little bit skeptical just to give you the context of my question, right? Because, you know, it probably will book some tickets, you know, here and there, but is that the tickets I really want, right? Because I'm picky with even if I'm not picky, my kids are picky, right? So there are all those things, like I don't know I can trust the agent to book my next vacations airline tickets in the next six to 12 months, but I would love to hear more of your thoughts on this. I agree with you. I think that consumers, there's a lot of friction in these use cases. For simple things, you can say, send my girlfriends and flowers and AI will take care of it. And I think that is six months away. But to the point of the user experience, I think people like seeing all the information and there's so much context to these decisions. There's all of the options. There's what you're thinking. And AI can't read your mind yet. And so that's the biggest bottleneck. Maybe with Elon Musk and Neuralink, if it connects that to the AI, then it'll be. Okay, that's the one missing part, right? Neuralink. That's the one missing part, yes. Other than that, if we can solve that problem, you feel like you can progress. That's the reason is we don't have enough state, enough context so that you are asking too much for the agent to book wonderful flights because you know what the picking is, right? Not the agent. Yeah, and over time, it might get enough data, but it will take a while. Okay, cool. So before we close off, I would still like you to imagine, right? You know, what the world would look like, you know, once, you know, we have GPT-5, 6, you know, the marginal cost for AI, and then gen AI is going to decrease further by some amount of magnitude, right? You know, what the world would resemble? I think there are a lot of people being paid a lot of money to sit at computers and type out documents like the legal profession and accounting profession, and AI will make all of those professions a lot cheaper. And it'll be great for consumers because everyone will have a robot helping them with various different things in their pocket, but people who are overpaid and do charge a lot for doing very little should be worried about transitioning and their jobs. And then when do you think that that would happen? I think it's already happening. I think that you see Clana, they're automating a lot of their customer support and they don't want to say that, you know, it's replacing jobs or anything like that because it's very controversial, but this stuff is making people more efficient, so you need fewer human beings. Okay, so those are the lots of different professionals. What about developers? Do you think AI will replace developers? You know, we have this many developers, you know, in three, five years, do you think we would have this many or we have a lot less? We will have a lot less and there'll still be room for the elite developers to oversee the systems and they'll have more money than God, but then there'll be a lot of the lower level developers where AI can just complete their tasks from start to finish and that's very worrying. It will make everyone more efficient though, so I think people can also embrace AI and then see the benefits of it. What is your advice to people in the grade school or in college, right? Or, you know, people applying for the college, should they learn coding? Should they, you know, think about a developer as a career? I think in a world where AI replaces all knowledge, the thing that matters is socializing and meeting other people in real life, so you see jobs like investment bankers, I actually think will do even better with AI because they'll still need to be able to deal makers and so I would encourage everyone to just socialize and meet as many people as possible because that's the one thing that can't be replaced with AI. But in the world that we are heading to, right, there will be a lot more bots, a lot more agents, so even if you're an investment banker, you might be doing a lot of work interacting with bots. Yeah, there will also be jobs that didn't exist that will be created. Like 20 years ago, we could never imagine that a YouTuber would be a professional job, but now it is or an SEO expert. Similarly with AI, there will be jobs where you're creating fake data to train AI and there are actually already people who are like RLHF jobs that didn't exist even two years ago. So I do think there'll be new jobs and new opportunities. It's just about people transitioning from the old world to the new world. So that's data labeler, right? That's one new category of the jobs. Proper engineering, maybe another category, right? Yeah, the AI could even create fake jobs for us if you imagine a video game where we're all fake lawyers in a video game powered by AI. So I'm sure it will work itself out. Wow, that's fantastic. So thank you so much for giving us your view, right? Do not pay the tech stack, how to evolve the firm, the traditional AI to the generative AI and 200 use cases, right? Hundreds of thousands of subscribers already and it's a profitable business. I quoted this one in a few panels, even just today. CNBC had an article in December. 2023 was a great year for generative AI. Lofty profits for Nvidia and lots of experiments for everyone else, right? So sounds like for you, it's definitely not an experiment. And do you see that we are getting past lots of experiments in 2024? I think so. I think this is the year of applications. So you're going to start seeing this more and more in the real world and I was ordering DoorDash the other day and I was talking to an agent, so AI agent. And so every day, new use cases and new applications for big businesses, but hopefully with companies like Do Not Pay for consumers too. So when mobile or internet came, there were a few mobile native or internet and native applications came out and then totally changed the way we live and then the way we work, right? Do you think, where do you think we are? We are kind of first inning or how do you think about it? I think that we're overhyped and underhyped at the same time. We're underhyped because there's so many applications that can be built now with the technology and so there's a lot of value. We're overhyped in that AI is nowhere close to AGI or anything like that and it's got a long, it's just a statistical model right now. I asked chat GBD to add up a few numbers recently and it got the answer wrong. So AI is not reasoning at the level of a human yet and so it will take a while to get to that AGI moment. And it's just another tool, right? Yeah. And then when do you think AGI moment of where we arrive? I think there needs to be a fundamental breakthrough because the way these models are developed as you know is they just feed in every piece of human literature, every website, Reddit, et cetera into this model and it just predicts what's the next word that will come next but you need a new approach to actually get to more intelligent AI models and I'm sure they'll do it because there's so much money being thrown at the space but it's- So when AGI finally arrives, it's not going to be the transformer being the architecture? Exactly. I think they will need something beyond transformers. So give me your prediction on AGI arrival year. There's so much money being thrown at it. I would say within 10 years, AI will be more intelligent than all humans but I'm not sure about a definition of AGI or not but it will- That's Joshua's definition. Exactly. Wow. Joshua prediction in AGI 2034 or by 2034. Yes. Thank you so much for coming to my panel. This is a wonderful conversation and I think our audience will appreciate a lot about the evolution, the journey you have gone through. Thank you so much and thank you for watching SuperCloud 6.