Profound, Resistant AI & OpenAI’s Economic Pitch | E2071

Hey everybody. Welcome back to this week in Startups. My name is Alex. I'm Alex over on X and I'm also on LinkedIn and all those other fine places. And today on the program we have a quick look at the two segments from Mark Zuckerberg's interview over the weekend that you absolutely must need to know. We have two founders of Twist 500 companies on the podcast today, profound and resistant AI. Then we're going to talk a little bit about the Biden chips rule and what OpenAI is saying about future AI development here in the US and, and if time allows, we'll do it just a little bit at the end about the Indian IPO market and how one company could help kick open the 2025 IPO calendar.

This week in Startups is brought to you by Northwest Registered Agent. Starting your business should be simple. With Northwest Registered Agent, you can form your entire business Identity in just 10 clicks and 10 minutes. From LLCs to trademarks, domains to custom websites, they've got you covered. Get more privacy, more options and more done. Visit northwestregisteredagent.com twisttoday Paddle self serve software is global by default, but selling to every country is complicated. Paddle's merchant of record manages payments, sales tax chargebacks and refunds so you don't have to. Wherever you want to take your business, go there with paddle. Go to paddle.com twist to get started with your exclusive listener fee, free period. And Squarespace. Turn your idea into a new website. Go to squarespace.com twist for a free trial. When you're ready to launch, use offer code Twist to save 10% off your first purchase of a website or domain.

Now, to begin, I want to start with just a little note on crypto prices. We have talked more about crypto here on Twist in the last couple of months than we have in the last couple of years, I think, and that's because things are happening. However, while there was a rapid appreciation in the value of major crypto tokens like bitcoin in the last couple of months, we have recently seen a bit of a retread. If you take a look at this chart here from Yahoo Finance, this just shows you the price of bitcoin in the last day. As you can tell, there was a little bit of a flash crash in there to just about $90,000. Now, if you've been holding bitcoin for a long time, that's not that bad. However, if you bought recently when it was over a hundred thousand, you're probably licking your wounds a little Bit. All this is to say that there does appear to be some profit taking or perhaps a little bit of a pullback in sentiment. While the incoming administration is widely expected to be a bit more crypto friendly than the Biden administration, I do think that people are now probably selling the news a little bit after buying the rumor to use the old financial saw. One other data point on crypto, I did just take a look at the total value out there of USD Stablecoins, essentially Tether Circle and all their friends, and they are doing exceptionally well. So even if we are seeing some price declines in the value of Bitcoin and other coins, I think the underlying fundamentals, if you will, of the crypto market are still accelerating and are pretty healthy going into this year. All right, now I want to talk about Mark Zuckerberg. He went on the Joe Rogan podcast, and if you're a bit like myself, the idea of listening to three hours of a podcast is a bit much. So, so what I did was I went through the transcript and pulled what I think are the two most interesting and important bits for it, just to go over those so you know what everyone's talking about over on X and in other technology circles. So to begin with, Mark Zuckerberg, of course, CEO and founder over at Meta, nay, Facebook, he had this to say about Apple. I think it's very apropos. Let's roll the clip.

I don't know, I mean, at some point I did this like, back of the envelope calculation of like all the random rules that Apple puts out. You know, if they didn't apply, like, I think, you know, it's like, and this is just Meta. I think we like make twice as much profit or something. And that's just us. I mean, it's like all these small companies that like, probably can't even exist because of the taxes that they put in place. So, yeah, I think it's. It's a big issue.

So part of the context for that clip was Mark discussing Apple's overall business success and how in the era of the iPhone, the company saw a lot of growth. However, more recently, iPhone sales, much like smartphone sales around the world, have slowed. You might even say stagnated. So if you're Apple and your cash cow is no longer growing, well, what do you do? You turn to other things. For example, large fees on the App Store, which is part of what Mark was talking about there. This has been a pretty potent issue in and around the world of technology for the last couple of years. People are concerned that Apple is being a little bit too precious about its walled garden, and that could frankly lead to a diminishment in the profitability and therefore the viability of many online business models. Models. Now, I don't think Meta is in any danger of running out of money. They can afford to both invest in AI and reality labs to the tune of several billion dollars a year, but at the same time they like to make more money. And if Meta was able to double its profit, it'd be worth a lot more. So a question we could ask ourselves is just how much other technology market cap is currently underneath Apple's banner that might really belong somewhere else if Apple took less of a tax? I bring up this clip not just because it's interesting, but I think it just shows how the market has changed. It felt like a couple years ago people just didn't want to criticize Apple because they might get dinged in their App store, renewal permissions or things along those lines. But I think the market's changed. I think everyone's pretty willing to be blunt about this, and so Apple might be time for another act now that the Vision Pro has yet to catalyze revenue growth. All right, second clip here. We've talked a lot about AI coding assistance with the CEO of Codium on the show recently, and we have a couple more interviews in that vein coming in the next couple of months. But here is Mark Zuckerberg, who employs lots and lots of engineers, talking about the pace of AI development in the coding space and how that will impact Meta's own operations in 2025.

We at Meta, as well as the other companies that are basically working on this, are going to have an AI that can effectively be a sort of mid level engineer that you have at your company that can write code. And once you have that, then in the beginning it'll be really expensive to run. Then you can get it to be more efficient. And then over time we'll get to the point where a lot of the code in our apps and including the AI that we generate is actually going to be built by AI engineers instead of people engineers. But, but I don't know, I think that that'll augment the people working on it. So I mean, my, my view on this is like the future people are just going to be so much more creative and are going to be freed up to do kind of crazy things.

Couple of things I really like in that clip. First of all, the note that AI systems that are being built today may help actually build new and better AI systems because that to me sounds like a system that is self improving to the point at which we can really step back and let the computerized systems run their own show. That to me sounds like a much faster development cycle and the potential for upgrades and improvements that we humans might not have seen. However, if we do end up automating away a lot of jobs, especially jobs that have been historically relatively high value, there is the risk that our economy could go through some disruptions. Now, Zuck said there at the end that he expects people to be freed up to do more creative things. I think that's probably true, but I do think that we're talking a lot about long term solutions, I. E. The labor market, sorting out the impact of AI and we're kind of diminishing the impact in the short term. If you're a developer who lost their job recently and can't find a new one, you probably don't care. If in 10 years you can do more poetry for money, you probably care a little bit more right now about the economic stability of your family. One more thing to keep in mind. This is not the first time we've gone through a technological period of disruption. There have been several revolutions, the Steam revolution, thinking about, I don't know, cars, the Internet, and yet through all of those we have wound up with roughly full employment. So my general view is that this will be duff in the meat in the short term, but overall long term positive. So depending on where you sit in the economic chart today, it might hurt you. But I am still very optimistic about the future. And Mark made me ever so slightly more bullish about AI. All right, now we have a couple of folks in the wings. I want to start talking to a couple of Twist 500 founders. We have been greatly expanding the Twist 500 list. If you haven't seen it, twist500.com we are out there hunting for the top 500 companies in the world of technology today. Startups, if you will, that we think are going to have absolutely outsized economic impacts on both their backers and their founders, but also the world at large. Our first company today is called Profound, and here's my thesis about why this company really matters. SEO Search engine Optimization has been a big industry for as long as I can recall. However, that was kind of predicated on search staying mostly the same. Now you could argue that Google got worse over time, but essentially everyone was still chasing that Google love. Now AI is changing things. AI search now does a lot more answers and summaries than just big lists. Of Blue Links. And so in my view that makes traditional SEO just a little bit obsolete. So we're going to need new tools for the equivalent of SEO in the AI search era. And Profound, a company that recently raised three and a half million dollars, is building just that. It's backed by Khosla and a couple of other big names to the tune of about three and a half million dollars. So it's one of the most early stage companies we have in the twist 500. But I want to talk about what they're building and how it's working today. So please welcome to the show, it's James Cadwalader, the co founder and CEO. James, how are you?

Hey Alex, thanks for having me. How are we doing?

I'm doing really good. I'm also very curious about how Profound works. Let's start though by talking about why this matters today. So I'm kind of curious, how popular are AI search tools today that Profound interfaces with compared to traditional Google 10 blue links style search?

You know, before going into numbers, this is clearly a platform shift. If you're a techno optimist and you believe that the better technology wins, which we certainly do. You know, this feels like a sort of streaming to CDs type moment where instead of thumbing through pages of blue links, you just ask a question and get an answer. It feels like we've entered this new era where the world is talking to the Internet and it now talks back.

Yeah.

And that creates this huge second order problem for every company on the planet of, you know, how do I show up in the answers?

Hey founders, you want to build the next great billion dollar business, right? Well, you're going to need two things. A killer idea and a properly set up company. I can't help you pick the idea, but I can help you get your business started the right way. With Northwest Registered Agent. With Northwest Registered Agent, you can form your business for just $39 plus state fees. One of the most affordable options out there. No hidden fees, no upsells, just a simple, straightforward process. And here's how it works. In 10 clicks and 10 minutes, your LLC is officially formed and ready to operate. Northwest handles all the paperwork fast, accurate and hassle free. So you can focus on building your business, navigating all those crazy legal forms. Thousands of entrepreneurs trust Northwest Registered Agent because they make business formation affordable, efficient and stress free. And with their expert team standing by, you'll always have the support you need. So here's your call to action. Don't let paperwork hold you back from Your entrepreneurial dreams get started today@northwestregisteredagent.com twist. That's northwestregisteredagent.com twist. For just $39 plus state fees, your business can be up and running in no time at all. Get more value, more convenience, and more peace of mind. Only with Northwest Registered agent.

And this is day zero. This is very, very early innings and I think, you know, it's SEO, it's the closest thing we can compare to. But I think as with most things in life, the past isn't always a reflection of the future. And I think this, these, these AI systems are far more probabilistic for folks.

Who don't know what that means. James, can you break down how probabilism impacts search?

Well, it means you can ask the same question lots of different times and get different results. There's a probability of showing up. It's not predetermined. So it's not, there's no longer, hey, I rank position one on Google for marathon sneakers. It's just, it's a probability of showing up, which is due to the way these LLMs work and they're a black box. No one really understands them. It's far more opaque and yeah, lastly, I'd say it's like more condensed. So, you know, if you search for running sneakers on Google, there's pages of links you can go through. If you ask ChatGPT, you'll just get three or four options.

Exactly. It's much more constrained. And my take on this is that the reason why Perplexity is doing well, the reason why Google has rolled out its AI overviews to more and more markets is that people are responding to this. But it also sounds like we don't have good data yet on how popular this type of search is compared to the old school list of blue links.

Well, to give you some kind of number, we, you know, and these are fuzzy numbers, I wouldn't sort of live or die them, but there's roughly about 3 billion unique monthly users of Google search. And now to put that into perspective, we estimate the AI answer engines are somewhere in the region of 600 million unique monthly actives.

So that's 20%.

Well, ChatGPT, to give you some sort of rubric there, ChatGPT has 300 million weekly actives. That's not including, you know, CoPilot, perplexity. Google AI overviews, you could probably throw in to that to an extent, even though that's less input based. Right, because you don't change your behavior when you Use a Google search Is just a Google search still?

Sure, but I think that goes to show the scale of the market that you guys are going after. It's not like there's eight people out there currently using these products. In fact, hundreds of millions of people are each day, each week. Call it whatever you will, but that opens up a big market problem, which is if you're only going to show three or four possible brands, websites or sources, people are going to be competing like crazy to get onto that list. And this is where I'm curious about Profound's technology. How much can you guys influence those models and their probabilistic outputs to better surface something from a particular website, customer or client?

Yeah, sure, I guess. You know, just the last point on the first question is that, you know, we really can see clearly now that this is going to be AI visibility. He's going to be a hair on fire priority for every company on the planet over the next year or two.

Big or small, because everyone's had an SEO strategy, I presume, up until this point, and so they're going to absolutely have to have something to fit into this new market moment.

Yeah, I mean, we're working with dozens of enterprise brands now and everybody is scratching their heads because at least before talking to us, everybody is scratching their heads because everyone's using these AI platforms for search now because they're better, but no brand has any understanding or control over how they show up. So, yeah, answering your question, profound. The way that our system works is we send millions of prompts to these systems and paint a picture of how they're responding. So If I ask 2 million questions in lots of different variations, semantic variations about sneakers and running sneakers, which brands does ChatGPT suggest or which, which brands does copilot suggest? And there's huge delta in the suggestions for the record as well. So it's not, it's not like you get the same answer. The crazy thing is that there's not a huge correlation between traditional SEO and the answers that we're getting from these answer engines.

Ah, okay. I want to double click on that because that's very curious. So the old strategies, the things that worked in traditional SEO, do not automatically map over to the new AI search world.

Absolutely not. Yeah, I mean, that's interesting. That's. And to answer your question of how you control, I mean, I can say unequivocally, 100% it is possible to control your AI visibility to improve it, because the way these answer engines work is, you know, back in 2022 when ChatGPT launched, it was obviously, you know, for all intensive purposes disconnected from the Internet. So you'd have that response of as of my last training data. But today there's retrieval. Internet retrieval systems are baked into all of the major AI models. So they're getting their information from the Internet. So as a company or a brand, you just need to understand which sources these AI models are going to for their information and what's the shape of the content that they prefer and how.

Different is that AI friendly shape of content to the things that I don't like on the Internet today? Kind of like forced listicles and websites that are clearly designed frankly for the Google crawl bot instead of me. Alex, the person who needs something.

Well, that's it. Yes. Now it's become a game of telephone, right? Because as a company you're creating content for crawler agents rather than human beings. So your job is to create content for a retrieval agent to prefer and pick up because then it becomes a game of telephone. Because if the retrieval agent picks up that information, then that's what it might spit out in the answer. The answer is, yeah, very different. I think it's only going to get more different as well. I was using Google Deep, Google Deep Research the other day, which is Gemini's latest model. Obviously Jason and co have spoken about it on the all in podcast. I asked it a question and it used, I think it was, 73 sources for its answer. And when you grok that, when you really think about that, that's the direction that this is going in. Human beings will click. We will typically click somewhere between three and five links when we're researching a product. But that's just because we lack patience and because of our fleshy human brains. When you think about a retrieval agent, why wouldn't it look through 200 links to find me the best answer and then can then synthesize the best parts from each of those 200 links to give me the best answer.

But what's the value of that? So if I can end up being the top three organic results for marathon running shoes today on traditional Google search, I think you can put a dollar value to that pretty easily. But if you're going to be one of 200 sources in a deep research report, that just seems like so much less valuable to me in terms of what impact that could have for brands. So are we moving towards a future in which, yes, companies still want to make sure they show up in these AI results, but they might not have the same. I don't know, impact or commercial lift that traditional search had.

I imagine my vision here and my vision of the future is that it seems obvious to me that a huge brand like a Walmart or a Coca Cola, they're going to have entire departments dedicated just to AI visibility. Got it. And there's going to be teams of marketers working on ensuring that their brand shows up more visibly and AI answers. I just don't see a future where we're as consumers, where we're going to be clicking through blue links.

Yeah, I think that's going away. And ironically it's going away after Google effectively neutered it by turning it into essentially just an advertising canvas. So I'm not actually that sad that it's going away, but I bet seven years ago I would have been screaming and holding onto my blue links because it worked better back then. So, black box, you said earlier about how these AI models work. You can tell kind of what they're pulling for. So how big of an impact can Profounds tech today have for, I don't know, someone like me? I have a personal blog. I want it to be available in AI Answers so people can find my work and click on it. So if I worked with Profound, how quickly could I improve my overall AI visibility, to use your phrase?

Well, within about a week of working with us, you would have a clear understanding of how visible you are across all of the major answer engines such as ChatGPT, Copilot, Perplexity and Google AI overview. So immediately or within a week, that cloak is pulled back and you understand, okay, I'm completely invisible on this answer engine and I have a bit of a footprint here. And then from there we typically expect, you know, there's no silver bullet. It's the same as any traditional marketing pillar. There's no big red button in Profound where you click go and then your AI visibility rockets through the ceiling.

This is, I was hoping for that. I love, I love big red buttons that do quite a lot.

This is going to be something, as I said, that companies are going, this is going to be a new pillar of the marketing stack is the way that we're thinking about this.

Okay, that makes good sense to me. But this is not a services based business. This is still very much in the realm of software first enterprise.

Yeah. I think to make it clear, as we enter this generative era of the Internet, Profound will become a platform that's relied on by every marketer on the planet to navigate the generative era of the Internet. In a world where you speak to the Internet and it speaks back.

No one likes dealing with taxes, especially a busy startup. Did you know you're required to pay digital sales tax wherever your customers are? That's where Paddle steps in to put your billing on autopilot. When you launch your SaaS product, an app or a game, you're building something that can reach billions globally. That's the magic of the Internet of app stores and it's awesome. But scaling brings a lot of challenges like taking payments in local currenc and navigating all those regional tax laws plus managing fraud and refunds. That's a tremendous headache that will slow you down. Maintaining a messy payment stack eats up time, energy and money. And these big company problems compound with every new market you enter. Paddle has seen these challenges a thousand times or more. As you're a merchant of record, you're more m o R. They're going to handle payment localization, fraud prevention, tax compliance and customer billing support so that you can grow faster. You need to launch in a new country. You're done. You need to set local pricing. One click you want to add payment methods. Easy peasy lemon squeezy. So here's your call to action. Wherever you want to take your business, go there with paddle. Go to paddle.com twist to get started with your exclusive fee free period. That's paddle.com twist to unlock your no fee startup period.

I'm very optimistic about this feature, but I don't think that the SEO era ended up making better Internet for regular folks. And so I'm curious, do you think we're going to end up with AI visibility leading to the same sort of homogenization of content and content, you know, display that we saw with SEO. Because I will be very disappointed if we end up with the same problem we had before, even under a new search paradigm, which feels better. But if it ends up just being the same people who are just gaming the system, then I worry we won't actually get that much better.

Yeah, I don't think there'll be, I don't think they'll be gaming the system. But I do think, you know, we in a world of capitalism, I think companies are going to want their brand to be more visible and then I think eventually what we're going to see most likely is advertising open up in these answer engines as well. Like Perplexity is already playing with this right now. And I think the unlock of generative content in a conversational interface is going to be one of the biggest marketing unlocks in the history of the industry. I don't think the world is ready for just how interesting that marketing is going to become. And again, profound will help marketer's plan there.

So I'm trying to make sure that I understand the future you're describing to me. So one thing that people loved was wirecutter before the New York Times bought it. Because you could go there and you could get essentially trusted reviews from people that were just trying to make sure you got the right thing. I think people use AI today for what wirecutter used to be for. And I can definitely see If I'm asking ChatGPT tell me about the best marathon running. Should just stick to that example. The links it brings up will have enormous value. Okay, I can see that. It does seem to narrow the aperture though a little bit if ChatGPT kicks out a handful of sources. So it feels like it'll be more cutthroat because there's not an infinite number of pages to try to rank on. There's only a couple of spots.

But I think the sources will be broad. I think the sources will be in the hundreds. I think it will reference speaking to your question. This is the point where I was going to get to is I think it will be less controlled by one source and it will be more a spread of different sources that create an answer. Because the only reason why that isn't the case right now is because we don't have the patience to click through so many links. But if it's a retrieval agent, it's got the patience to do so.

Okay, now on the profound side, just a couple of business questions. You guys just raised some money. I'm curious, how long has the product been commercially available?

We've been working with big customers, predominantly mid market or enterprise or even large enterprise since like September, October of this year. Sorry, of last year of 2024.

Yeah.

So we're still relatively fresh out the gate. And yeah, we're predominantly an engineering team. We're based here in New York City. We're in office six or seven days a week.

Yeah.

Working late nights, real grind and having a lot of fun as we do it.

So just to be clear, James, you are a London born person with a Welsh name building in New York City. I feel like that touches on a couple of cultural and political things that we've been talking about here. But I'm a little surprised that you're not building in San Francisco, one of my favorite cities of all time. Why New York? For the company?

I think as a Business. You know, we think that the New York, the New York City tech ecosystem is strong. Like, we have Karim Atiya, who's one of the co founders of Ramp, is one of our investors as an example. And I think, you know, there's, there's, there's a lot going on here in New York City. Me and my co founder met at a, an accelerator here in New York City called South Park Commons, which ironically did actually begin in sf.

But I was about to be like, wait a minute, I've been to.

Yeah. But yeah, I think it's just, you know, it's where we lived, there wasn't a ton of strategic thought that went into being New York City based. It's just where we were based. And yeah, it doesn't, I don't think we lose anything from not being an SF right now. And a lot of the brands are based here as well.

As a fellow east coast transplant of the U.S. i mean, I'm sympathetic to where you've ended up now, business model, I'm presuming this is something akin to SaaS when you break it down.

Yeah. Right now it's a monthly subscription and the pricing is typically based on the amount of configurations. Right. So as a customer, you might just want to understand how often you show up in a general topic or a category or an industry. But you know, say, for instance, if you were a large consumer electronics brand like some that we're speaking to, they have hundreds of SKUs, hundreds of different topics, hundreds of different markets, different languages. So the size of that configuration increase the cost.

So earlier you mentioned running, let's say 2 million searches to figure out the distribution of, you know, what's being linked, how and when for different types of searches. Given that you're working with customers that have hundreds of SKUs, different languages, products, I mean, the number of possible queries you might need to run to figure out what's going on sounds enormous. What is your OpenAI API bill right now? It must be absolutely bonkers.

Yeah, I mean, we're definitely getting LLM bills, but, you know, it's cogs at the end of the day. Right. So we only do those. We're not just running those prompts for the fun of it. We're running those prompts when we're working with a customer. So it's cogs to an extent. I mean, another thing that I'd say just to sort of plug what we're working on in terms of innovation, this is super. This is our kind of like Edisonian moment, is we're working on a new feature you can see on our website. It's called Conversation Explorer. It basically pulls back the curtain to help people understand what the world is actually searching for inside these AI answer engines.

I'm curious today, with your current technology, how granular can you get?

At the moment, this is very news. We only just announced this. It's very much in the early stages, but the place that we're going to get to is you'll be able to come to profound and search for any topic. So if you were Apple, you'll be able to come in and search iPhone 16 and see monthly or weekly how many people are searching or having conversations in ChatGPT about that topic. And it's incredibly hard to do. It requires a ton of very hard to get pretty expensive proprietary data and then a very complex predictive model that we're building on top of that data set as well. So this pulls back the curtain and yeah, you'll be able to understand what people are searching for within these AI answer engines.

On the point about the LLM bills going into cogs, this is a slightly nerdy financial question for the eight people out there who care about it, but does that mean that profound gross margins hinge a little bit on OpenAI's pricing changes?

A little bit. I would, you know, I'd say it's relatively negligible though.

Really?

Price changes? Yeah, it's the amount of prompts we're sending is relative to the size of the contract. So yeah, it's okay. I mean, listen, if OpenAI cut their bill by orders of magnitude, it'd be good news for us.

Okay, so essentially your business gets better as the different search AI technologies you have to interface with become cheaper. So therefore, essentially, efficiency gains will trickle down into your overall business health. That's cool because it does seem like the industry is getting cheaper and better. So.

Yeah, and there's also a moat that forms. Right. The more, the more prompts we run, the more interrogations we run on these AI answer engines, the better we understand them, which as a net net, but gets better for all of the customers that we work with. So, you know, I would say, at the risk of sounding obnoxious, we understand the outputs of these answer engines better than most people on the planet right now.

And that's the show to be obnoxious on. So you're dead on with us. Okay, Are you going to end 2025 calendar with seven or eight figure ARR? And if you say six figures, you're fired. Get out.

I'VE learned my lesson. Putting flags in the sand. Alex, I'm going to forfeit the question. I'm happy to do a dare instead.

Okay. I dare you to break 5 million ARR this year.

All right. Dare accept it. We'll give that a go.

All right, I have one last real question and then we'll get you out of here. Do you think that you're going to end up with anything akin to an adversari relationship with the major AI search providers who might not want you to be helping people to better speak to their AI tools?

My view is that we are absolutely supplementary to their business. We are doing the Lord's work for the answer engines in that we are going around promoting the answer engines of the future to enterprise brands all over the planet. I feel like we should be getting a kickback right now. I think we've seen that with Google Search, there's been huge supplementary businesses built off the back of Google Search or the App Store App Lovin being built on the back of the Apple App Store, which is what you touched on at the start of your introduction. Yeah. So I think it's supplementary. I think brands are going to need to if chatgpt or OpenAI and Perplexity, if these guys want to win, they're going to need big enterprise brands to take a lot of interest in these new platforms and we are doing that for them.

Well, I think you just made a really great pitch for both Perplexity's founder and Sam Allman to both invest in your eventual Series A. But James, thank you so much. It's try profound.com correct?

Yeah.

All right, founders, let's kick off the new year by talking about your website. If you're launching something new in 2025 or you need to refresh your brand, and I bet you do, you need to use Squarespace. Squarespace makes building a stunning professional website ridiculously easy. Whether you're selling products, offering services, or just sharing your ideas, Squarespace gives you the tools you need to make it happen. And you know, Squarespace is at the cutting edge with AI tools that make building your site easier and faster than ever. For example, its design intelligence tool is just like having a world class designer at your fingertips. You ask it a few questions and their AI takes it from there. Building a fully customizable website that's perfectly tailored to your brand. In just minutes, we're talking professional layouts on brand visuals and premium content, all ready to go. Day one. So here's your call to action. Start the year strong, check out squarespace.com twist for a free trial. And when you're ready to Launch, go to squarespace.com twist to get 10% off your first website or domain purchase. That's squarespace.com twist.

It's the new year, friends, which means we have a lot of people to talk to. As a reminder, we're Talking about the Twist 500 bringing on founders to the show to talk about how they're building companies that we think are going to have outsized and market impacts. And to that point, next up, we are going to talk about Resistant AI. Now here is my general thesis here. I think that AI is going to become an absolutely critical method of detecting bad things on digital canvases. And that could be, you know, looking at scans to find tumors or just looking at transactions to find financial fraud. AI is going to automate a lot of repetitive work that humans do, but also it does some things better than we humans can and therefore it's going to help us tighten up our digital systems. So Resistant AI, Resistant AI, if you want to look it up online, is a startup that's applying AI to find and combat fraud in both documents and financial transactions, backed by about $30 million from Index, Google Ventures and others. So please welcome, it's Martin Rahak, the CEO and founder of Resistant AI. Martin.

Hey.

Hello.

I see you're stuck in a really, really sad call room. I hope you're at the office.

Yes, I'm in our office in Prague.

I wanted to start there. You guys are building over in, well, Eastern Europe, frankly. So I'm curious, what's it like building a cutting edge AI startup that far away from the South Park Commons and San Francisco itself?

Well, we like to think that it must be hard to be in San Francisco so far away from Prague.

Oh, fair enough, the old switcheroo. No, but I'm serious. You're the first person I've talked to in the last at least year, year and a half who's based there. So what is the AI vibe like on the ground? Lots of talent, lots of people, lots of companies. Is it quieter? What's going on?

It's quiet, wild. Actually, there is plenty of AI in Prague and what we like to do is to bring a little bit of the west coast vibe and a bit of San Francisco here so that we can enjoy that locally.

So eight $8 mochas and lots of little dogs. Like, what does that mean to bring the San Francisco vibes to Prague?

It means creativity. It means different way how to run a company. It means different Way for people to develop it means all of the good things about west coast that we love and that we always liked. We bring them to different cultures and for me this is the right way. How to essentially spread what I learned in my first startup.

Just for people who don't know, tell us about your first company and then I want to dive right into what Resistance is doing.

Look, our first company is something we stumbled on pretty much by accident. In 2006 we were doing AI research at the university. It was way before anyone else was interested in that thing that will never work. And we found out it actually worked perfectly well to catch cybercriminals. And I remember some first VC pitches when we spoke to VCs and the answer we were getting was no one will ever use AI for network security. That's impossible. It's so important. You always need people with hands on keyboard and they will never relinquish responsibility to anything automated. Well, we know how that ended up. Yeah, but we basically we built a company, we got funded by Credo Ventures who also fund our second company now. And we were their first investment ever. And in 2013 we became part of Cisco as a first acquisition in the region. And this is how we got to know the west coast culture.

So now you're working on resistant. And the thing that really caught my eye about your current company was how you seem to be working on what to me appears to be several different challenges at once. So I want to start there because to me building a bit of AI technology to look at documents, to scan them, to search for fraud is one task. And then to do what you guys call transaction forensics, to essentially spot fraud and potential money laundering in a flow of transactions is something entirely different. But I presume that I'm wrong here. So is there a shared anti fraud AI model that you guys are using that applies more broadly than I might have anticipated being possible?

You are perfectly right. In 2024 and 2025 you will be wrong in 2026, which is why we are doing this. Essentially what we see and what we have seen since starting to do this in 2019 was an increasing sophistication of criminals. So what worked as a technique in 2019? It was really primitive and most it was like script kiddies in the 90s, pretty easy hacks, easy ways how to get into the systems and like hacking and process in 2019 took about two minutes flat. It was really easy. It became much more sophisticated and the criminals became also quite sophisticated. So we are fighting an escalation warfare, which is something we love.

And what happens is that just because it's fun to be challenged.

So first, it's fun to be challenged. Second, for me it's a sound business principle because if you build a security company and if you can perfectly solve this problem immediately, essentially you don't have a business. I know this sounds harsh and it sounds bad, but we need to make a living. I need to feed my programmers and my kids. What we need to do is to find a place in the market where the attackers will be getting better and we will be fighting them. And for that, eventually it will be defense in depth. We look at the documents when we onboard someone, we look online sources, we look at the behavioral sources and we look at the money flow and we pull all of this information together, paint a complete picture and take a decision. This is the end game for us and this is what we are building progressively.

So later there's going to be one shared AI model, brain, whatever that can do everything. But today in 20, 24 and 25, it sounds like you're still using different models for different tasks.

Exactly. And these models will be connected seamlessly in the cloud, which is something we are building right now. And that's kind of the holy grail we are going to. The interesting thing is how much better are criminals getting? And you could remember what Mark Zuckerberg said about mid level programmers being automated. So you can literally say that about mid level criminals being automated. Criminal ganks.

I'm now trying to imagine the stack ranking for criminals. Sorry, you're in the bottom 10%, you're out of the gang. I don't know.

Yeah, and you don't want to see that HR interview because it typically is not as polite and nice as with can be company. That's a low bar. Arguably. Like you don't want to be a failure in a criminal gang.

No, no. I don't think HR has little bowls of candy or anything like that. I'm pretty sure it's mostly what baseball.

Bats and shouting probably or that candy is kind of very not sweet. I would say it's the one you eat once. But what happens is that they are actually automating quite dramatically. And we see this on the documents. Two years ago when you did ID fraud, you could still zoom in and you could see the modifications. You could use human eye to verify things. You can't do this anymore.

Is that because they're better than they used to be?

They are better.

Got it.

They are better. They are better techniques. They use AI in every place where it makes sense. It's still not a fully synthetic fraud. We don't see basically image of an ID card or something created by the AI from scratch, because that's still very hard to achieve despite the claims of some criminals. But they. Well, they are criminals, so they don't quite hesitate to make wrong marketing. And if you go to our website, there is all of the marketing stuff about our products, but you can also read about the criminals, which is actually funny stories about how they live and how they operate. So I recommend you that section, which is plenty of fun.

I want to make sure that I'm understanding this correctly, though. So essentially, in the old days, you could do a certain level of sophistication with human talent when it comes to trying to abuse transaction systems or documents or whatever. Now with AI, the bad guys have a expanded arsenal that is constantly improving because, I mean, look, open source AI is going to be a thing. I'm a big fan of it. I don't mean to sound negative, but the tools that are available are getting better. And so essentially you're in a market in which you get to use AI to combat the things that AI has done. So effectively, you're going to get paid to make AI take better care of itself, pretty much.

And we are on the side of the good AI against the bad AI.

Right, right. You are the good guys. But still, it's funny that you're a firefighting department that also accidentally sets fires. The AI industry as a whole, I just find that ironic.

It's a different kind of AI, but definitely there is an irony in that. But it's the same. Like if one side of the war gets fighter planes, you don't want to be the one caught without them.

No, you want to have 10 times. I mean, look, we all know our Cold War history.

Yeah, exactly. So you want to have enough on your side, and that's pretty important. What I want to say is that humans are getting lost in that game and they are realizing that increasingly, because the AI versus AI is getting so much faster by day, is that people who were saying, I can still see this with my eyes because I'm a professional, they are disappearing. And even if you could have people trained to do that, you don't want to, because criminals augmented by the AI can produce so much more.

Oh, so it's quality and quantity that's overflow.

Exactly. Because if you. There is. In a computer security, there was a moment where basically spam and malware was an exception, and then it flipped Completely within two years. Where most of the emails sent were basically spam marketing and malicious emails. And it flipped very abruptly, it was almost immediate. And we are facing the same things. When you hit the API of any financial institution for the onboarding, we still see mostly majority of legitimate customers, but that's going to change very quickly. Imagine a world where most of the things hitting your API will be criminal.

So the way that I my first thought about that just talking this through with you is that email has been so persistent in our digital lives because in a sense it's kind of an open system, you can send emails and I know there's a lot more nuance to that. I've talked to email marketers, I'm aware, but generally it's kind of open. Whereas the JP Morgan API, I presume, is not just an open endpoint out there that anyone can just latch onto and fire in information to.

So if you install an app of any kind of mobile application, any fintech application on the mobile, that's literally what you do. You hit an API, you use that app, it collects some information, but more often than not it probably runs on a phone, vicious virtual phone in a farm somewhere in China or southeastern Asia with people trained to use it and onboard profiles of people on financial services using fake documents, fake histories, fake transactions. And then they sell those profiles of companies or individuals to others who commit crimes with them. So it's a very lively and very active ecosystem. It's not people doing script, like script kiddies crimes at home, it's really professionals doing this at scale.

So a lot of companies are going to be faced with a build or buy situation here. And I think that what you're doing is super cool. But I'm curious about how good companies will be able to do on their own using their in house technology teams, open source AI models and not needing to go out to a vendor. I think a lot of people are asking today in the new AI era, how much software do we buy versus build? So I'm curious, for your company, when you're talking to potential customers, how often are they thinking about building on their own versus going with your solution or a competitor's?

So many people thought about that, many people tried and they eventually became our customers. The reason for that is because, and I don't believe that we are smarter than those people. It's just that if you are running a fintech company and you operate a fintech company, do you really want to employ 20 programmers or 40 programmers doing in depth research of image files of the ways, how cameras work, how mobile phones work. Do you want to know all of the arcane details of 1990s vintage PDF specification? Do you want to learn how every single bank on the planet looks like in terms of bank statements and documents? So it's really about the roi. We met with a bank and they are the ones who want to build this in house. And they said we also historically built our own operating system and we built our own AV engine and they were very proud of that decision. Which, well, they were a customer. I was very polite, but I didn't say that they were idiots because.

And which bank, which bank was this? Can you give me the full name please just so we can tie that comment to that customer?

No, I can't obviously. And they didn't become a customer. They are still trying to build on their own and they are obviously not succeeding. Like, there is a good reason why most companies don't build their own AV engines simply because you can build it, but to maintain it and to keep it on the edge of the development is a day to day task. And that's the difference between a good security business and a bad security business. It's not about what we give you today, but it's about the promise and the trust our customers give us for the future.

How can you be confident that you're going to be able to stay ahead of the bad guys? Because if this is an arms race between you on the side of the good guys, frankly, and the rising sophistication of threats that are impacting the industries you guys focus on, are you confident you can stay ahead or is it going to be kind of like they pull ahead and you catch up? Kind of like, I don't know, nuclear bomb size in the Cold War.

So so far we are ahead and we know that we have statistical numbers to show that to our customers. The reason why we do that is that we have a different twist on the AI and something we have been building for the 18 years since we started, which is called Ensemble systems, where we have plenty of small AI systems that we connect together. So it's not a huge LLM model like everyone is doing now, but it's a combination of small models looking at different angles of the and they check each other's outcomes and we combine them together in different ways, which gives us much more stability and gives us an edge in defiant. So this is our secret sauce.

That sounds awesome. And honestly, I think we're going to see more of that in different places in the technology world, in the next five years. But we're talking about documents and transactions, two things of which there is a lot. And the system you described to me sounds super cool, super powerful and expensive. So I'm curious, is it costly to run a constellation of models, even if they're smaller, to do the sheer amount of processing work that I think that the company has to do?

So it's costly to develop and it's the maintenance and it's the learning and it's the training that's costly. Running itself is something we try to keep cheap.

So the inference is cheap.

The inference is something we try to keep cheap because we have to do this many times per second. When you do a payment with your card, we are often in the loop for that. When you get a viral transfer into your account or ACH in the US we are in the loop for that. And we check those things and we verify. Are you a fraudster getting paid or are you a money mule getting money so that you can send them away for a grant? Or are you a legitimate person who merits and deserves to get paid? And we have to do these things in couple of milliseconds.

That seems to be incredibly hard. But it sounds like in the future you could you get down to the point to which you're verifying my credit card tap at the store to make sure that it's the right and non fraudulent person doing it.

There is other companies doing the credit card itself. What we do is the money laundering. We do find the fake merchants. So what we do find is fake is criminal gangs that do funny thing, they basically create an alter ego for a restaurant, a legitimate business. They open a new bank account for them. They pass all of the onboarding process using fake documents. They occasionally take insurance for that business. They start paying themselves with the stolen money into the business accounts. They pay taxes and this is the way how they launder the money. So essentially you have the sibling, a digital sibling, a digital twin of the legitimate business in a different bank with different bank details. And this is the way how many lots of money is being laundered today? It's fake merchants, fake merchant onboarding used to process payments from stolen cards. So this is an example of pretty sophisticated fraud we are going after.

I just don't get criminals. I know this is a failure of my own imagination, but that sounds like a ton of work. Why not just do the thing instead of doing the fake digital twin of the thing?

Well, because actually using AI you can do this so much more efficiently. Again, going back to the Efficiencies. You can actually now do this quite well from the comfort of your home in Ireland, Cambodia, Laos or other Russia, Ukraine, whatever.

Martin, I wanted to have you on because I was hoping you could have this uplifting positive conversation about how we're going to beat back cyber fraud. And it sounds like instead we're just taking on bigger and bigger waves and it's never going to be a solved something that we fix.

Well, the advice, if you want an optimistic take on world, don't do never invite a security person on this podcast.

Absolutely true.

That's not going to work.

You had a demo though. I want to get it on the screen to show people kind of what you're talking about because it's been a lot of words from us and I think pictures are worth several thousand of them.

So this is actually. So obviously all of these frauds are synthetic done by our team because we can't show the real fraud data. We have to protect defraudstress identity because they, they are humans, some of them. And this is kind of the historical legacy, low quality fraud that professionals can tell apart. And you can see that things are slightly off on this one. When I zoom in you can see that the fonts are not quite right, but honestly this would pass muster in 80% of processes by a human. Yeah, because simply if you see hundred of these per day, you don't notice a small thing like this kind of mistake on misalignment here. Oh yeah, you're right, you can see the subtle mistake here. But like professional that's well rested and has a good capacity and doesn't get too many of these can notice this. Well, honestly, not many professional SEO call centers do have that time to take a look and to zoom in. They have a couple of seconds per document and they have to make a decision in the in a snap. And this one is an easy one. So this is something you can buy online. So actually today when you need a face, when you need to get onboarded, you can get a template like this. This one is even like you know what to fill in. This is not even finished. And we can basically tell you this is kind of template form document and we can link this to other cases and other attempts that have been on your API. And we give you an information about this without naming anyone because we don't kind of want to tell the criminals what exactly do we know? So I'm hiding a bit of the information here in this kernel.

So you're saying we noticed a couple of things, but we're not going to give you the exact report card, so that way you know what to fix the next time.

Well, I'm not going to give this on the live demo that anyone can watch. I'm sorry, but we.

No, that's very reasonable.

There is a little bit of the secret sauce still we want to keep. This one is actually much better. Like if you look at this one, it's better quality. You can see that it's a little bit out of focus. That's normal. You can see that. You can see that. Well done. There is a fault created in the paper, which I think is done digitally. You can see that the geometry works. It's kind of. It looks okay, except it's completely fake.

I mean, this looks like exactly the type of thing I get from my gas company and I never open or look at or so I wouldn't be able to catch it because I don't even know.

This is something that people use so that you can get on board with the US Government systems to prove the identity. Because they typically ask you for an ID and then for one other, let's say utility bill or something else. And this could be that second piece that someone uses to steal your identity so that they can get, let's say government subsidy or something else. So again, we get the same warning synthetically created document and we get reference to similar template form documents that we are seeing in the same data set.

Okay, well, this convinces me that you guys definitely have a great place in the market. So how do you guys charge for this to your customers? Because I can see this going in two different ways. One, SaaS, very standard. Or every time you catch fraud you could charge them five bucks. So I'm kind of curious, how are you guys going out into the market and extracting value for your work?

Right now we do the SaaS pricing traditional. When we started with the very first customers, we did what you suggested, basically getting paid for success. Which actually is an incredibly tempting. We are a very honest team. But we were so tempted thinking about that because how many frauds can we create using our systems per second and how much money can we make per hour?

Did you hear the story about the tech company that just paid bills that were sent to them randomly for several years and someone racked up a couple hundred thousand dollars in free money because tech companies just paid too much?

You would be shocked how many companies using RPA system and some automation did just that. Because the first thing you do in a company where you want to make things efficient is that you have accounts payable. You get an email with an invoice, you find the account number and you do the wire. You have just created a huge ATM on the Internet for anyone to ask for money. And you would be surprised how many billions per year are disappearing criminals. It's about business. Email Compromise is about 10 billions per year in the US alone.

That's an insane amount of money. That's a lot of deadweight loss.

You know, it's like if you cannot get the basics right, it's fine. But it's so easy to commit. And there is basically no enforcement because there is no one going to go to Nigeria or southeastern Asia to catch those criminals. Right.

So we have to have digital walls here essentially, or Prague, basically.

If you live in the physical world, you have windows, you have houses, you have locks on the doors, you have alarm systems. If you look at what processes are used in financial systems or any other systems, we don't have any of that. That or we have very little.

It is kind of crazy that on the bottom of my checkbooks it has my routing number and my account information. All that you need to just take my money.

So every single European who sees the American check system is amazed by the fact that not all the money has already been stolen in the U.S. yes.

I've had the same question, but somehow it kind of works. I just want to focus on the business itself and growth. It sounds like your tam, your addressable market's growing day by day as criminals get more sophisticated thanks to AI. So how fast is the business growing? I guess maybe the right question is last year, in percentage terms, how fast did you grow?

I can't say the exact numbers, but it's always doubling to tripling every single year of the company life.

And since 2000, I think was 1919.

Yes. Like it was obviously faster at the beginning, but these were small numbers, so easy to make. I have to say that I'm grateful to criminals for creating such an amazing business for us. Essentially what we do is that we see our job as a mission because as people we love hunting criminals. We love protecting our customers from what they do. And we are coming to the office every day to play this game.

Yeah. Do you have little sheriff's badges and guns on your hip? Because I feel like you're essentially the sheriffs of the modern AI Wild west.

So at the last off site, the people who survived the first day actually got some badges in. Sheriff's team.

No. Are you serious?

Yes.

Dead on.

Yeah, it was dead on. It was an idea from our people department and it was amazing.

Well, I guess I know where I'm going to go work next is the people department. Let's get that question pulled up here and we'll put you to the test quite live. So the question is, I have been using social and cyberpsychology with linguistics to design a custom email security system, performed user training with the same. Are you going to use similar technology? And that's via a LinkedIn user as of about two minutes ago. So Martin, what do you think?

We are using similar approaches, but not quite directly because on documents what we do is that we don't actually look at the content too much because there is obvious GDPR implications and privacy implications.

You don't want to ingest my PII while you do this.

I don't want to process your PII too much. We obviously have to look at the way how is the PII written? But we don't want to look at what is actually written in there because that's something that's very confidential, very private and there is plenty of restrictions in different countries. So we want to be mindful of privacy restrictions. On the other side, when we do the transaction checks, for example, we know when you send money, we know what is the likelihood that someone named Martin was born in a given year in a given country.

Oh, so you can just. Oh, that's interesting.

So we can estimate and in the US you can quite easily estimate the likely age group for a person looking by the first name. With many cases, I know how true that is and so much more. So these things are actually quite subtle. And what we do as a team is that we connect plenty of weak signals, plenty of weak evidence into a very good decision by doing that. And so this is the power of the model is that we can take information like that, which is more sociology than linguistics, it's more social and structural. There is plenty of economic modeling, there is plenty of business modeling. There is understanding what does a transaction buy. One interesting finding in transaction system was a taxi driver in Colombia who was making $50,000 per day.

Seems slightly dicey. I'm surprised that someone could make so much money just driving a taxi.

Well, I wanted to apply for the job.

I'm getting paid. Not nearly enough. Martin, absolutely I have to let you go, but I am glad that you committed to 300% growth this year at Resistan and the website is Resistance AI. But I want to give you the last minute to shout out anything that you're working on or looking to hire Just before I let you go.

So look, we are always looking for people willing to try what we do and we are always looking for people who have trouble with criminals because that's what we love.

We love crime resistant AI.

That's pretty true. And don't tell our marketing department I said so.

They will not see this show. So don't worry about it. Martin, thank you so much. We really appreciate it. Thank you very much. And we'll see you in like maybe six hours, months. I would love to know how this year works out for you guys.

Thank you very much. Thanks for the invitation.

Thanks, man. I do want to hit on a couple things before we go because they're just too important not to get into, even if we don't have tons of time to really get into the nitty gritty. The first thing is this. The Biden administration today came out with essentially a new plan they're calling AI Diffusion to limit certain exports of chips to certain markets. A lot of names you might expect, Russia, China, etc. But the plan also creates a second tier of nations that would have restrictions or possible limitations on how much AI chip they can buy, usually from, let's be honest, American companies. Now, when I read through the overall plan, it struck me as technocratic and I would say more restrictive than I expected. But I didn't anticipate the sheer response that it got from the market. Now, for example, I can just pull up what Oracle wrote here. Oracle is vehemently opposed to, to the Biden administration's plan, saying that, quote, the diffusion framework misses the mark by a wide margin and chooses instead to disrupt US leadership in cloud chips and AI. It also says that this plan could shrink the global chip market for U.S. firms by 80% and, quote, hand it to the Chinese. That's pretty bad if you're a big bull on the United States. But it's not alone. Nvidia also came out with a really blistering statement that I think might even be sharper nosed than what we saw from Oracle. So Nvidia said in its own post that, quote, the new rules will control technology worldwide, including technology that is already widely available in mainstream gaming PCs and consumer hardware. And then this is the real boot to the chest. Rather than mitigate any threat, the new Biden rules would only weaken America's global competitiveness, undermining the innovation that has kept the US ahead. Now, there is a little bit of political posturing going on here. Both of these companies definitely want to make sure that they are in good stead with the incoming administration and sniping at a Biden era rule even as they exit stage right is not a bad way to do that. But as far as I can tell, everybody absolutely hates this new idea from the Biden administration. And you know, I'm being honest with you because Secretary of Commerce who's behind this to some degree is Gina Raimondo, and she is the former governor of my state and is a beloved Rhode island export. But here it really does seem that what she's putting together as she leaves is a flop. I don't know how long this is going to last into the next administration, but it's not going to help. I would say Democratic bundlers raise money in 2028 if this is the taste that the current administration is leaving in the mouth of technology. But that's not even the end of recent AI policy work. OpenAI today dropped what they're calling an economic plan. This is a, I would say medium detailed overview of what they think the US Government should and should not do to help bolster the domestic AI industry once again in a somewhat contra China perspective. So I'm going to pull up just a couple of screenshots here that go over what they think we need to do. So if you want to know their policy overview, this is what OpenAI has in mind for the US chips, data and energy and talent are the keys to winning on AI. And this is a race OpenAI thinks that America can and must win. Now they do talk a little bit about a thing called democratic AI or what you might call AI with democratic principles, essentially open government, open data and the free market. I don't think that any single thing that OpenAI put forth in this is going to be controversial, but there are some really good ideas. For example, they think that all models, AI models should aim to be politically unbiased by default, supplemented by users having the ability to personalize the tools. I think that's going to be generally popular today. I don't think anyone wants to have an AI model that is too much in any particular direction, be it, I don't know, Tanki or libertarian or, I don't know, Maga right or Woke Left or pick your brand. I think everyone's going to want something that's probably as unbiased as possible. Fair enough. OpenAI also says that in any future policies we should have the ability for individual users to choose their own preferences to personalize their AI tools, including controls on how their personal data is used. I think this is super important. We're going to have a lot more data ingested into AI models in the coming years. So being able to protect our own personal information, that just strikes me as pretty good policy. Of course, the proof will be in the pudding. Some folks are going to want basically no rules at all there. But I think some rules of the road just make good sense. There's also notes in this document about how the government should make more of its data available and classified computing systems available, so that way AI can learn and then better serve the government. I think that also makes a lot of sense, though, of course we do have security questions inside of that. But I presume that we can get past those with the right technologists in the right places. And certainly AI companies are hiring a lot of folks, so that should be possible. A couple of other things that really hit me here was OpenAI thinks that we should have the creation of AI economic zones, essentially allowing for significantly sped up permitting processes for building AI infrastructure like new solar arrays, wind farms and nuclear reactors. I think that makes a lot of sense. Two more things that OpenAI put forward. One is the idea that federal AI rules should definitely supplant and remove or supersede state level AI rules. We've seen states like California and Texas pursue their own AI regulatory vision, kind of because the federal government isn't. But OpenAI says that is not the way to go. Why? Well, you don't really want to have to comply with 50 different rule sets. What you like to have is one. So to make an EU example, do you want to have one unified set of rules or one for each nation state over there or just state here? Probably you want one system that's simpler, lowers regulatory overhead, et cetera. But we do have a lot of states that are very active in their own way. So I'm curious to see how that pitch ends up as policy, if at all. But I do think it would be faster for industry. And then finally, there is a note in here from OpenAI about copyright. And they make a general kind of hand wave towards protecting copyright while keeping data accessible by AI companies. There we will see some fight, but I don't think that's anything that we aren't already looking at and seeing in action. So to kind of summarize here, I think that OpenAI wants more government support for the AI industry and fewer government roadblocks overall. Not shockingly controversial given the state of AI development here, and also I would say the political climate of the day, but seeing this OpenAI document land on the same day we saw the Biden administration try to do good, but seemingly missed the mark. Based on what I've read from literally everybody, it certainly does say that we might be onto a new page, a new chapter, maybe even a new genre of interaction between the government and AI companies. All right, one last thing before I let you all go. Today, there's a company over in India called Grow G R O W W. If you're not familiar with it, it's essentially India's largest retail stockbroker. So you can kind of think of it as if you want the Robinhood of India. TechCrunch reports that this company is going to go public within the next 10 to 12 months, which means inside of calendar 2025 if timelines hold and the company could be seeking a valuation between 6 and 8 billion dollars. A lot of money, but critically much higher than its 2021 ERA $3 billion valuation. We've talked, I don't know, ad nauseum on the show about how companies that raised a lot of Money back in 2021 are struggling to either live up to those lofty valuations or are simply dying on the vine. Well, here's a company that's not doing either. Grow is growing and it's going to double to maybe even 3x roughly its prior valuation. That's a huge win. But there's something else brewing here that I think is actually even more important. So pulling from TechCrunch here, my old alma mater, here is a bit of a quote that I'm going to read for folks on the audio version of this India has emerged as a bright spot for tech listings globally, with seven technology startups going public in 2024. Food delivery platform Swiggy's $1.35 billion IPO was the largest tech listing in the world last year. More than 20 Indian startups are planning IPOs in 2025, TechCrunch writes, including B2B Marketplace, Zetwork, managed workspace provider TableSpace, process owned PayU, and pharma platform PharmEasy. Throw grow into the mix and a couple of other names, and we're looking at a very strong IPO cohort from India. To put this into context, we have seen venture investment in markets like India and China decline in recent years. The VC global market has become a bit more US Focused in the last couple of quarters, but I do think that this much liquidity from the Indian market is a mark in its favor. We have had a liquidity drought here in the US So I think it's good to see other markets put up points on the exit board because that will help money recycle and also attract more overall venture investment. All right, guys, that is the news for today. My name is Alex Twist. We'll be back on Wednesday with live show. Jason will be back with us. And then we have quite a lot planned for later this week and next week as well. So stick close to Twist. I'm over on X as out. I'll see you guys soon. Goodbye.