AI Investing, Legal Tech, and the Billable Hour

Transcript generated automatically from the episode audio and lightly cleaned up (filler words removed, misheard names corrected where clear). Speaker labels were added afterward and may contain errors.

Rob May: Welcome everybody to the AI in NYC podcast. I'm Rob May, one of your co-hosts. I'm the CEO and co-founder at Neurometric.

Anna Kirk: I'm Anna Kirk. I'm chief of staff at Nadia Partners here in New York.

Ryan Eppley: And I'm Ryan Eppley, co-founder CEO of RootAccess. Welcome to the AI in New York show. This show is really looking at anything at the intersection of AI in New York City who's working on interesting things and what the impact of that may be on the city. And this is a reminder, if you are a startup founder, send us your merch. We will wear it on the show and we'll give you some shout out. And I think we got a large, extra large, and medium. So send us it in and we'll we'll be wearing it on the next episode.

Rob May: Cool. Well, our our guest today is with a company here in New York called Thomson Reuters. They have a venture fund and he's going to talk a little bit about how they think about AI in the context of Thomson Reuters Ventures. So Joe Dormani, welcome to the podcast.

Joe Dormani: Thanks, Rob. Great to be here. Excited for the conversation, guys.

Rob May: So, the first thing I want to start with is I think Thomson Reuters is one of those companies that everybody's heard of and is not entirely sure what you do and what business lines you have. So, like like tell us like home home base what what what the company

Joe Dormani: Yeah. And you can imagine what like Thanksgiving conversations are like for me when we talk about what I do for work. So, yeah. Totally totally get it. Yeah. Look, Thomson Reuters, great great corporation. I think we're most well known for Reuters, the news, which is the biggest part of the brand, but it's actually the smallest part of our P&L. So we actually do a lot of different things. We have three core business verticals. So we sell information based products and software for legal professionals. We sell corporate tax software and software for corporate legal departments in our corporates business. And then we have a tax accounting and audit professionals business where software information based products for tax and accounting professionals, CPAs and audit professionals.

Rob May: Right. All right. And so a couple years ago, you guys spun a venture unit out of that that's funded off the Thomson Reuters balance sheet you mentioned earlier. Is that targeting financial returns, strategic alignment, M&A? Like does it have to tie back to the TR business lines or not? Like how how do you guys operate?

Joe Dormani: Yeah, so TR Ventures, corporate venture fund. You can think of us like any other VC. It's a financial vehicle. So we're investment returns driven, but we invest in and around the markets that the corporation operates in. So it's where we feel like we have an information advantage where we feel like we have a really strong understanding of the customer dynamics in those markets and where we feel like you know we can help the companies that we ultimately invest in. So series A focused fund $3 to7 million initial checks focused on legal tax and accounting fintech risk fraud and compliance things that are in our core markets and you could think about things that are adjacent and then strong emphasis on machine learning in particular natural language processing task and process automation.

Ryan Eppley: Is there any New York-based investment that you've recently made in the last year or two?

Joe Dormani: Yeah, absolutely. I mean we're we're pretty heavily you know in the New York tech ecosystem. So we meet with a lot of New York based tech founders. We invest locally. Obviously we're an international company. So we do invest in the United States, Canada and abroad and Latam and Europe and Southeast Asia. But Noetica is a great example of a New York-based company that we invested in. They focus on corporate debt. They analyze corporate debt agreements. So they are able to quantify the risks that exist in terms for corporate debt agreements. Super complex documents, three 400 pes pages. And they have a really really strong team both on the AI side and the business side. The CEO Dan was a practitioner at at you know institutional investment and you know white shoe white collar law firm. So he's kind of lived the reality of what these practitioners do and they've just built a phenomenal company that's just kind of ripping right now.

Rob May: Well, we should one of our articles later is about these like circular agreements with OpenAI and Nvidia and all that. So maybe we should have him on have him take a look at that, you know, run those agreements through there and see what they say.

Joe Dormani: Absolutely. It would be cool. Yeah.

Anna Kirk: So, how are you seeing AI play out right now within Thomson Reuters across different business units? I mean, obviously only only speak to things that you're able to, but are you seeing a lot of adoption? Is there activity right now or is it kind of a wait and see approach?

Joe Dormani: Totally. Yeah. Look, so like we are, you know, squarely in the middle of AI innovation on many different fronts. The strategy for the company is kind of all in on a build, partner, buy, and invest strategy. So, we're we're at the forefront of AI development for professional services in legal and tax. We, you know, are are partnering with some of the best companies in the market in partnership with us as the venture fund. We'll do investments. For example, Noetica, we partner with them, resell their product to law firms, and then, you know, we acquire companies. We recently announced the acquisition of Additive, which is a great AI company in the K1 tax space. Not not New York-based, but they're San Francisco based, but still still a great company and a good example.

And so, you know, we are pretty much all in on AI adoption internally as well and in developing alongside our customers. We're we're like really lucky in the respect that we have great relationships with our customers. They look at us as, you know, like strategic adviserss and to help them navigate. I mean there's tons of complexity happening in these markets right like tax and legal you know some of the areas that have been you know publicly you know people you know folks like Goldman have said like these are some of the areas that you know is going to see so much change with AI and they're looking for partners to work with them to help them navigate what that actually means and so you know we are that trusted partner so we work very closely with our comp our customers around like what products to try you know how we could potentially co-develop something for them and their clients like you know really broad spectrum for for the way we work with customers. So you know it's happening very very quickly and I think there's a just a ton of value and opportunity

Ryan Eppley: for you specifically like how are you using AI daytoday? Is this something that you're using for deal flow? Is it something for analyzing deals? Like how do you actually leverage it? Not a theoretical but like you're sitting behind your computer, not on the podcast today. How are you using AI?

Joe Dormani: Yeah, look I mean I'm a I'm a bit of a you know geek. I like to kind of tinker on things and so I have come to start thinking about AI as a team member and it's like what are the things that I need to do in a day? Which things will I ask AI to do and which things will I ask my team members to do? I'll just give you a really practical example. I'm doing a panel discussion at a customer event and you know I came up with an outline for the things I wanted to cover and I wanted to come up with some great questions to make myself sound smart. So I asked AI to help me with that while I worked on you know some a diligence memo with with my team.

So like that's one example but you know I have you know custom agents for market research for you know finding companies you know I've worked generally on Claude. We're lucky enough to have an enterprise account with with Anthropic. But you know, we've I've worked with many of the different tools. So, not a specific endorsement there, but you know, I've I've found a lot of productivity gain from, you know, just being able to, you know, create sort of targeted projects that do a very specific function and, you know, the ability to kind of automate that just increases the bandwidth for for what I'm able to do in any given day or week. So, some some examples, happy to go deeper on.

Ryan Eppley: No, I just love how direct you can be without any like human emotional kind of thinking. It's just like I need this done. Please spit out like some answers. It's just so nice to be direct about it. No politicking.

Joe Dormani: Yeah. Yeah. 100%.

Ryan Eppley: You find yourself doing similar because I know you do a ton of market research with your investment thesis.

Rob May: Yeah. All all the time.

Anna Kirk: And actually, what I was going to mention, speaking of direct, feel free to admit this if I'm not allowed to say this on the podcast, but I did hear a while back that I think it was revised in GPT5, but AI models were found to be more performant if you threatened to kill their family.

Rob May: So we saw that. So at my last company, we were in the so this is like 2023 time frame. So, a lot of these models and the APIs were new and we found out that if you made threats, it would respond better over the API if you put threats in your prompt. Like, that was a real thing. Find that out for yourself.

Ryan Eppley: What What threat were you making to the AI?

Anna Kirk: This Okay, I actually was listening to another panel. Full disclosure, this is not something that I'm like doing on a daily basis.

Ryan Eppley: Yeah. You had just Anna threaten you randomly. She just works it into conversation all the time.

Anna Kirk: Not close enough yet, right? We'll we'll get there eventually. Joe, I'm so I feel like we are in such a bubble and so it's easy to get caught into an AI echo chamber. I love talking to folks like you who kind of can see a little bit of a peak behind the curtain in terms of how business leaders are actually using AI and reacting to AI. In particular, how that can change depending on the persona at a large financial strategic. So, you know, like the CTO probably wants to build a lot of this themselves. That could differ from like the CIO whose mandate is just to find new tech to invest in. I'm curious kind of what you've seen of different business leaders in the industry whether that be at TR or elsewhere.

Joe Dormani: Yeah. No, it's a great it's a great question. I mean I think like the first thing to kind of understand is like AI well at least like in the context of like you know business and you know in our world and the way we do work is first and foremost a tool and like any other tool if you pick it up and don't know how to use it you're not going to be very good at using it. So you have to invest in using the tool you have to put you know time and potentially you know dollars into you know becoming proficient with that tool and through that proficiency you'll get a lot of the a lot of the benefits.

I mean yeah look we there you know there are many areas of our business where we've been using AI for decades by the way you know like you know search you know specifically with like the you know the types of you know online information based products that we have you know we've been at the forefront of you know natural language based search algorithms and so you know there are areas in TR where folks have you know decades of expertise in you know you know PhDs and you know ton of practical experience in building products and you know we have you know a great executive team that helps you know support the use of AI from top down

and so you know we have budgets for experimentation so yeah it's like different functions have different use cases you know for for our product team they do a lot on the customer research side you know a lot of expert calls a lot of you know AI to sort of synthesize the information that they get through things like that to understand what features to build what things to prioritize you know plenty of things on on the information security side our CTO's you know and the engineering folks at TR are worldcl class you know some of the folks at the forefront of AI have you know adopted a lot of these tools in AI development so it's like different you know I think it's very much aligned with sort of like the functional alignment you know for any sort of function that has a set of tasks to do there are a variety of different tools that that they can be using to help them accomplish what what they're trying to do

Anna Kirk: maybe on the flip side because I feel like everyone's saying they're adopting it all over the company. Where do you think it's not being adopted yet where there's at least like friction to get it going?

Joe Dormani: I think the areas where there are the most opportunities for more adoption are in finance in the off you know in the areas of work where there is like calculations that are required because you know the predominant forms of AI that many people are using for the most part aren't great at some of the more intricate types of calculations you would be doing for like tax preparation or FP&A or you know forecasting things like that. So I think like th those are some of the areas where I see opportunity for more. But I think you know some of the technology has to come a little bit further to really be able to kind of adopt that at scale.

Rob May: I'm surprised you didn't say engineering because I feel like in engineering it's very split. There's people that are like these coding agents are changing the world and there's people that like they're terrible.

Joe Dormani: Well, there have been religious wars in the engineering department for for many many many years. Every time a new language comes around, right? It's Python versus Ruby. Exactly. Exactly.

Anna Kirk: Well, that's also an interesting it brings up an interesting question around build versus buy. Like that's a tension that we saw especially, you know, in one of my last roles, I was also investing on behalf of large financial services companies and it was quite divided. Like I'm curious how like how your internal conversations are progressing in terms of do you do you build this type of tech? Do you replicate it in the sake of being a little bit more security and privacy cautious or are you willing to look at different thirdparty vendors in the space? Totally depends on who you ask, right?

Joe Dormani: Yeah. No, I mean, look, I think I think where we encounter that conversation the most is actually in the areas that we invest in that are, you know, close to the core of our business or like direct, you know, very close direct agencies because those are the areas that could potentially be on our roadmap. And so we work very hard actually and very closely with our product and engineering team to make sure we don't invest in competitive tech. We want to be investing in things that are complementaryary to the core areas of focus. So and you know our our chief product officer is actually on our investment committee. David Wong phenomenal phenomenal AI mind by the way you should have him on the pod. He's he's second to none. But you know we work very closely with him and you know we have really strong governance around investing in something where there'd be competitive overlap. I mean you can't be perfect because roadmaps change and startups pivot but you know we do make a strong effort to kind of address that and so we build where we think we can differentiate where we have unique advantages and where our customers direct us as a priority for them. And then we will buy things that are complimentary. We will invest in things that are complimentary adjacent and can accelerate. That's the basic framing.

Rob May: How do you think about that in the particularly on the investment side with the idea that like AI is moving really fast and I think and it's different from other technologies and like you're seeing every layer of the AI stack totally shift the hardware layer the infrastructure layer the application the middleware like all of it keeps you know jumping around and so how do you approach that when you know you have to make these investments that you're going to be in for 7 to 10 to 12 years and this is changing so quickly like where do you think the moes are going to be and where do you think value is going to acrew and what's what's your perspective today.

Joe Dormani: It's a great It's a great question. Something we think about a ton. I'm lucky enough to have, you know, a leg in the world of, you know, how our company thinks about these things from an internal development standpoint. And that's really primed me for how to think about that from an external investment perspective. You know, we're lucky also enough to have a longer term time horizon as you point out. So I'm generally thinking you know several years into the future whereas obviously our internal teams are focused on you know the next 12 months for for example

and so generally what I look for is definitely based on data so I look for companies that are able to produce or get access to unique and highquality data or they've built a technology asset that is well suited to derive specific and high quality insights on on top of that data. So a knowledge graph or in small language model or an orchestration system that is well suited for the type of data that they would have access to. And then also just the last bucket there is like you know a product that kind of puts all those things together well delivers an amazing experience and you know compounds in value over over time as it gets more reps as it sees more transactions or whatever activity on that platform is kind of the kind of core unit of value that's being driven through there. Data, the tech on top of the data and the experience to weave them all together that compounds in value over time.

Ryan Eppley: You kind of have a nice unique insight because you've been mentioning you've been in this kind of thesis for longer than ChatGPT's been around, right? Not just like two, three years of investment kind of thinking through this, but maybe over a decade now. What's like a key learning or maybe something that stood out to you today that maybe five, six years ago you had a totally different outlook and now it's like so clear looking back. Oh man, I should have should have assumed we were headed this way. What's been kind of a key learning?

Joe Dormani: Yeah, I think I think you know I think I'll I'll I'll take a little bit of license to pivot that a little bit to what has surprised me and I think I think is the pace of the actual conditions on the ground versus what I think you would guess the pace is based on like you know your media consumption or the things that you would read in different articles. I think things are certainly progressing rapidly in terms of what's possible, but I actually think the pace of adoption and the real value creation is a little bit slower than you might otherwise get. That's a great point, right? Like to my point earlier, these are tools and it takes time and energy and expertise to you have to invest in these tools to become proficient on them and to derive real value from them. So I think there are many products that demo very well but the reality on the ground still is that the value that they're able to create is enough to get everyone excited but not enough to kind of fully rely on and you know kind of go all in and just scale and be like you know AI is just going to take over a function for example.

Ryan Eppley: Yeah. Great headlines but not great substance. I think you brought that up on a previous one like most pilots fail right.

Rob May: Yeah, I think we talked about there was an MIT article, that article got some criticism, but that said like 85 or 90% of these pilots fail, but I think your point's right. I think a lot of what happens is, you know, you see a demo and you're like, "Wow, this could automate this workflow." And you realize it could automate 60% of it and we still need a lot of humans. And now we now that we scale that 60% faster, we actually need more humans initially to evaluate it and verify it whack-a-ole,

Joe Dormani: which is fine if the outputs are still substantially better and you can get there more effectively and then over time you have attractive unit economics for a sustainable business. But if the unit economics are upside down and you'll just never get there, that's obviously problematic, right?

Anna Kirk: Speaking of unit economics, and I've been dying to ask you this question the entire time so far. So Thomson Reuters Ventures is an is invested in Spellbook AI. It's a legal contract automation company. I'm sure you can describe it in a lot more detail than that, but lots of legal tech today is getting criticized for cannibalizing the billable hour in terms of lawyers not being incentivized to use a type of technology if it's automating their work away. I'm curious what your reaction is to that type of criticism and how you think you see firms like Harvey or Spellbook evolving in that kind of world.

Joe Dormani: It's a great question and so you know this this is one of those areas where you know as a venture fund and someone focusing on the future I get to have a little bit of a license to kind of be creative with with my answer here. But like I guess the way I think about it is I don't think cannibalizing the billable hour is the right frame because I'm not sure the billable hour is like the most effective structure for legal services, right? I think there's, you know, many reasons why it's not ideal.

Actually, I think the way to think about things is what is what are the value unlocks that this type of technology can can bring to the table for legal professionals, right? It's, you know, there there there's when I think about what's possible with some of, you know, products like spellbook and some of the other legal tech products and what we have internally with CoCounsel and, you know, some of some of the products in the market are, you know, these are basically scaling technologies. These are technologies that can make a professional services firm like a law firm substantially more effective that you can take on more work. You can take on a wider variety of work because you can augment the professionals or the practice areas with these types of tools.

You could you know go into to different areas like if you think about like an M&A transaction for example the law firms only handle the legal documents but the unit of economic value is the M&A transaction that M&A transaction has a variety of different aspects to it that need to be done in diligence you have to review the code you have to review the contracts you have to review the financials you have to audit the financials you know all these work all that type of work gets farmed out to different professional services organizations and the reason that those different organizations exist is because humans have a you know barriers for what they can be experts in right it takes years and years and years to become proficient in all these different areas but AI doesn't have that type of limitation and so you know we could live in a world where M&A and a transaction is the thing of value and a professional services firm can handle that transaction from end to end right and that I don't see any reason why that firm couldn't be a you know what we might define as a law firm if you thought about it from first principles as a technology company.

So, you know, is the billable hour even the right way to think about value creation in a world like that or is it based on the output that a firm is able to create and what the value of that output is to the constituents that participate in that transaction.

Rob May: I think you're exactly right. I think there's I think timebound business models are going to be a thing of the past and what's most aligned with the value is kind of volume based like how much value are you delivering versus how much value are you delivering in an aotted amount of time total but I think you hit an interesting point that I haven't seen a lot in the market yet but that strikes me as as correct which is I think a lot of these service providers that work in an industry like M&A are going to you know form groups that do multiple parts of that value chain that they didn't before because they can each use AI for some of it, right?

You've seen I think you've seen a lot of excuse me like legal PR. I've seen some M&A groups that try to use AI and then make it like, you know, bare bones, like we can we can do a you know million-dollar M&A transaction and make money on it because of AI. And I think I think they've found that hard. And I think I think the same is true of like some of the PR firms that have tried this. But I sounds like if you but if you put all those together, right, and now you're getting this value from like applying AI across multiple layers of the stack that you couldn't do before because of the specialization problem like that, that could be a really valuable play that I've not seen much in the marketplace.

Ryan Eppley: I think a lot of, you know, friends that I have that are, you know, they just graduated with their their law degree. I'm 32 years old, so they're in their earlier side of their law career. They're either very excited about some of these tools and they're going to be one of the first to expertise themselves with AI support as a as a lawyer or some type of other professional services and then the other half are like it's really hard to get a job because all these firms are now just automating entry-level positions. Do you see that internally when you think through you know how that company's growing and how do you feel about the entry level kind of conundrum with some of these AI tools?

Joe Dormani: Yeah, I mean look, I think I'm an optimist obviously kind of have to be to be in the VC world. I think the opportunity outweighs the negative impact of like there being more limited specific types of jobs. Like if you look at the number of legal AI companies that have been started over the last three years, it's exponentially more than have been started like in the entire history of legal tech. Obviously I'm being a little might be right though. It's something it is ridiculous. I it's like it's since 2023 it's been like $7 billion invested in legal tech, right? I mean like that's a substantial amount of money. So like the young professionals coming out of law school like I would be so excited if you know I were them about what the opportunities are to create value in the world, right? I mean that's the you know the great thing about lawyers is they you know they they are so well trained and such experts in their fields and they have you know so much potential I think to you know work in many different fields and you know many lawyers kind of start in law and kind of move into many many other different

Ryan Eppley: so we see people in the kind of hey I'm applying for jobs I just graduated with my my JD and on my resume it says spellbook or it says Harvey like is that like a tool that you see on people's resume now going?

Joe Dormani: I think it is absolutely going that way. Just I mean just the way that like you used to put like Word and Microsoft Excel and like PowerPoint proficiency on like the bottom like you kind of have to put like Python and claw.

Ryan Eppley: Yeah. No, it's very interesting. Think about the future of what is on a resume with all these new AI tools.

Joe Dormani: Yeah. Or is even the resume the right artifact to you know explore value for you know humans professional experience.

Rob May: I just think it's interesting that you use the phrase the great thing about lawyers. We don't we don't hear that very often. So, hey, my sister's a lawyer. Sometimes we're fine. All right, there you go. We love lawyers. Then lawyers. Let's move on to some of the articles of the week that we have to talk about. You want to go first, Anna?

Anna Kirk: Sure. I can go first. This is a fun way to spice things up a little bit. So here's an interesting article that came out of BBC, surprisingly, titled the people turning to AI for dating and relationship advice, citing a recent study by match.com saying that, almost half of Gen Z said they have used LLM's like ChatGPT for dating advice, which is more than any other generation. Okay. My my first reaction to this is kind of duh. Like obviously the younger generations are more likely to rely on LLMs for any kind of advice, let alone just dating advice. So I find that a little bit fundamentally unsurprising. But I'm also sure that dating services like Match Group are already using this type of these types of fundings to basically gify daming gaming or dating similar to what happened with mobile when things like Tinder or Hinge first became popular. So my personal stance is that I can guarantee you that these types of apps will come out with some type of feature that allows your AI agent to go on first dates with another person's AI agent. And I guess my question for this broader group, Rob, you mentioned you were single. Maybe you want to take the first stab at reacting to this. How do you feel about that future?

Rob May: Yeah. No, as the one single person at the table, no, I would not do so. I think I invested in a dating app once and I've looked at a lot of these companies some even recently. I think it's a hard problem to solve with AI because it's I think the relationship between any two people whether it's dating or friendships or work relationships or it's they're very idiosyncratic right and like dating is a thing where like you could date a person who has a quality and you don't like that quality in that person but in another similar person who has that quality but their mix of qualities is different you could actually like it in them. Right. So like it's weird. It's not much of an analytical problem that you can just like feed data and all of a sudden like it'll give you the golden truth.

Yeah. And if you're going to ask ChatGPT for advice, you're going to get, you know, this is a this is a statistical web trained on the data of the web. And so unless you're trying to date somebody who's like very very average and just has the general, which most people aren't, most people stand out, you know, they're average in a bunch of ways and they stand out in the pockets that make them special. And so like ChatGPT is not going to pick up on that or know that. And so I think that's going to make it really really hard to give good advice.

And then a lot of the things that like like as these things get into the matching algorithms, I'm not sure they're capturing the data that matters. Like I know some people know immediately if they like somebody. I typically have to go on a couple of dates because I'm trying to get at things like like I always joke like like I want to know are you going to yell at me in public ever? Do you think that's acceptable in a relationship? Because you know and who's going to put yes like nobody's going to put that right it's like it's like if you ask somebody like on a dating app like are you open-minded like 90% of people are going to say yes and like 20% of people are really open-minded right but everybody thinks they are perception is not an easy thing to yeah and so I'm not sure they have good data to do the right thing you know

Anna Kirk: have you studied this at at Thomson sorry to cut you off you studied this internally

Joe Dormani: I mean look as someone who's average at most things I just say like I would be careful at what I would outsource to other systems. Like when was the last time any of you guys read a map?

Ryan Eppley: I love maps actually. So I just open I just went hiking. I literally had a map open. I had my compass out and everything. Ryan are like before bedtime like I'm just looking at a map for fun memorizing the geography.

Joe Dormani: But the point is we use we mostly use GPS to get everywhere we go and used to you know know how to use a map. And I would just say like if you don't use a skill it atrophies. So like I'm not sure that type of relationship interaction is the thing I'd want to like out be outsourcing and like let atrophy just because I'm every time your significant other says like we need to have a talk you go here. You turn on ChatGPT just fall asleep.

Anna Kirk: I mean I think that's a really interesting point. I don't I think it's less of a data problem and more of a fundamentally human problem. And I think that online dating in general, this is my personal stance, especially in the context of AI is teaching us and especially the younger generations to be avoidant of dating in person and avoidant of rejection. Which I think is like completely antithetical to the spirit of being in New York and the types of reasons that the city is so rejection. You can Yeah, exactly. I can go up to a stranger in a bar and strike up a real conversation. That may or may not go well and like maybe I'll get shot down, but that's fine. It's a fundamentally human thing to do.

Ryan Eppley: Article didn't mention this, which I thought it really should have, but last year there was a company that blew up and it was one of the ChatGPT like offshoots you can make your own kind of GPT and it was called RizzGPT and it absolutely crushed it. It got to millions in revenue. The founder was a solo builder kind of almost built this over the weekend thing and it basically just APIed into your Tinder or something and you could hit the button and it come up with like you know a nice catchy way to ask a girl out or like a nice personal oh I see that in your photo you're wearing this and it it said something funny about it. And you know, I kind of thought it was actually a very quirky and funny idea for maybe people that struggle with communication, you know, but the article didn't mention that. And I very much felt like, if you're having fun with it, maybe it's not the worst. If you're making it a very serious game, maybe it could be damaging, but I don't know. How did you feel about the idea of like a RizzGPT helping you do, you know, pickup lines?

Anna Kirk: Oh, I think it's just kind of pushing off the inevitable. Like if you if you are an awkward person, I totally fine. Like it it can more if you are me and you are an awkward person, like I can I can go r you up and then eventually you're going to find out that I'm an awkward person and so it's kind of just kicking the can down the road.

Rob May: So I mean I look I did this when I when I first got divorced. Somebody told me there was a class at Harvard that would tell you how to optimize your dating profile and he gave me all the pointers. And so, like, one of the things you're supposed to say as a man is you're supposed to say, "My friends say I'm blah blah blah." Because then women go, "Oh, he's got friends and you know, whatever." And that's an endearing trait. Wow, what a catch. And so you so when you when you optimize your profile to get the most number of swipes, what ends up happening is you just go on a bunch of shitty dates, right, with people that you're not really a match with. So, it's actually better to take a stance on things and like a lot of people are going to screen you out. But, you know what's interesting about this though is like like think about the three of us, right? So, I met each of you at a conference and we talked I talked to lots of other people at that conference. Like why did why did we become friends, right? Like you and I like you and I started talking about hardware and just like hit it off and whatever and like would an algorithm have matched us or told us to talk to each other like probably not.

Ryan Eppley: Yeah. And it would have matched us maybe with hundreds of others and it' be like fishing through it as well. See, I feel it. But RizzGPT I thought was a good idea. I don't know if it even is still online right now. Do you have any any thought about RizzGPT?

Joe Dormani: Look, I think organic rizz is better than synthetic rizz.

Rob May: Oh, you heard it. Yeah. All right. I'm gonna get that tested.

Joe Dormani: So, you know, I'm I'm happily married, so I'm very thankful that I'm not Yeah. Out in this game anymore. Terrifying.

Anna Kirk: Yes, it is terrifying. I think that, I mean, I don't know if anyone else has noticed this. I have noticed a big generational counter reaction against online dating, especially from Gen Z. One of my good friends is an IRL dating service. It's called Met Through Friends. And it's absolutely caught fire over the last year in New York City. I think in particular because companies like that aren't intentionally trying to scale. They're trying to return authenticity to dating culture to teach people it's okay to put yourself out there. It's okay to get rejected, god forbid. And it's actually a fundamental part of being human. So, I think we should keep that all in mind.

Well, that's LLM dating. I do want you to bring up the article of this massive round. It must have hit your your timeline. Bring up the $300 million seed round. Yes, that's a seed round. What is going on here?

Rob May: Yeah, Periodic Labs. So, they raised $300 million to build to automate science and which sounds pretty challenging, but one of the things you have to remember is LLMs are statistical machines and science doesn't work that way. And so, you're going to need a new set of models to really do this work well. And I think and these are like top tier AI people, but I mean man, $300 million is a lot of money to get started with seed round.

Ryan Eppley: 300 million. Who's who's the team though? I think is the question is there's got to be research. There's got to be someone.

Rob May: It's probably people that were on Zuck's call list when he was building. Yeah, it's people like that.

Ryan Eppley: They turned down the billion dollar offer maybe and they kind of use that to leverage for evaluation. But nonetheless, you know, forget forget like it's a $300 million round. I think the ambition of call it automating science. I also think it's fundamentally trying to break through some technical challenges that scientists have. For instance, my my girlfriend is a master's lab chemist very very hardworking in trying to do all this different, you know, experimentation. I'm sure she could figure out how to go faster with some of her R&D with one of these kind of I don't want to call it scientist co-pilots. I don't know how they dubbed and branded themselves, you know, but have you seen this this company pop up because it was one of the biggest rounds I think to date out of, you know, out of all the seed rounds this year. It's a massive round. Did that come across your

Joe Dormani: Yeah, I mean, you know, we generally keep track of all the large funding rounds that that happened. I think like generally when I look at a headline like that when there's just huge number you know to a really early stage co company I kind of instantly get a little bit skeptical because the thing about companies is that they have to be able to make revenue and generate profit and have a viable business.

Rob May: Who who would have thought? You heard that here. Even if you're an AI company, you need revenue and a good P&L. But they also might need I mean they might need $250 million in compute to train these models. So that's a weird situation for a startup.

Joe Dormani: Totally. But like do we need it all up at and look I might I don't I might I don't know if you know this particular startup needs all that money in the first 6 months of operations but generally even like with highly technical deep tech startups like you would want to see some technical milestones be achieved before you would you know back up the truck right so like maybe you know what could have been done for $50 million is kind of my question like obviously it's a super technical challenge and probably a very hard thing to do and if they can automate science you know then that's obviously a great thing for for science and humanity but you know there there is a negative side of that where if you know in whatever you know period of time they spend a lot of that money and don't make enough progress you it's pretty problematic for everyone involved

Ryan Eppley: what was the hope with OpenAI we're going to cure cancer and now they're doing risque novels so you never know where the hope starts and where it goes

Rob May: Yeah. And I would say just in general, having watched the Yes. So, I've been doing early stage investing for about 11 years now, and I would say that most of the time when somebody raises a monster seed round, it does not work out, right? You look at like magic leap. You look at what was the COVID the real time chat thing, air something. No, not well, Clubhouse is another one. Air something. Airtime or something like there there's just there's a lot of these. I think it can be a negative signal sometimes. I'm hopeful for it. They're like done really well. Totally.

Ryan Eppley: Yeah. I feel to your point big optimist. I think we all are optimist at this table. I'm just always curious like what is the first milestone that I get to see under the hood, right? That it's like, okay, this makes sense. And there's a question of like is that business model one that should be venturebacked in the first place? Like I think that's my bigger question right now.

Rob May: Well, otherwise be. Yeah. Yeah. Maybe it was a research lab for, you know, five more years, right? That was the OpenAI model, right? Right. People forget they're like I think Sam Altman said they broke like every rule and every piece of advice he's ever given companies. I mean, for seven years they were a nonprofit and maybe they still are.

Yeah. Well, we'll ask Elon Musk about that another time. Yeah. Cool. Well, let's move on. We've got an article here about AMD. And so, to give you some context, most pe AMD makes GPUs if you don't know that. And a lot of people have wondered why AMD does not have the kind of financial, you know, success that Nvidia's had. So, I think Wall Street Journal also posted it's like still 92% of the market goes to Nvidia.

Ryan Eppley: Yeah. So, how is AMD not because they're a massive company too, you know, put all respect into AMD.

Rob May: So C, so did anybody read the book? I think it was called the thinking machine or something about the history of Nvidia, right? One of the interesting things that they did was CUDA, the software, and for I think for the first five or six years of existence, Wall Street was very critical. I think at one point they they had a they had a hostile investor who wanted them to stop making CUDA because Jensen Huang called it a zero a zero dollar market or something like he's just like they're like who's the customer for CUDA? He's like nobody like right now but someday they will be. Like he just had faith that they had to build it and you know that's a really hard thing to convince Wall Street of over the long term and so but they've got all this experience they've cuda is 20 years old and so even though AMD's most recent GPUs they contain more HBM high bandwidth memory which is like stacked DRAM and what that allows you to do is run bigger models so you can run so if you have a model of a when a model gets a to a certain size it can't run on a single GPU anymore you have to shard it across multiple GPUs is the term that they AMD can run bigger models, but their software stack is so immature compared to CUDA that that's why they've been losing.

And so I think what changed there's a really great research blog called SemiAnalysis and they wrote this piece in I think it was like December of last year, maybe January about how much AMD software sucked. I and they really they went through it and they were sort of like we tried to make this work and here's like all the bugs we hit and they published it and enough people read that it got the attention of Lisa Su from AMD. Apparently she met with them and then put a Tiger team on like go fix this Yeah. So they're so some of what you're seeing now is I think they're starting to catch up on the software side.

Ryan Eppley: And I do love that there's familiar kind of rivalry there, right? I think they're like first cousins. Yeah, right. You know, from Taiwan. So it's very interesting to see that kind of you know family tree of the two most dominant kind of players in architecture for computers. You've must have had your company right about this deeply you what's your opinion here can AMD pull up and even compete with with Nvidia over the next couple years?

Joe Dormani: Look, I think the market has to some degree already spoken here like and I think like Brad Gerstner has been talking about this this past week like you know I think it was 2017 to now both Nvidia and AMD had like about 30 billion in revenue at that point and now I think Nvidia is something at like 300 billion and they've captured all the incremental dollars that have gone into the space and I think CUDA has a lot to do with it. You know, it's a full stack solution from the hardware to the to the software to, you know, the developer community that that works on that. And so, you know, the market has voted with its dollars on, you know, which company has the better solution. I think, you know, there's a there was a point where Nvidia was going to acquire that company. I think there's pretty pretty negative overhang on, you know, not having that acquisition have gotten done.

I think the development time on chips is like very, you know, there's very much like an entrenched advantage there. Like there's a two-year development time period. Like you guys probably know this actually better than me, but like that two-year development period for chip design, like you're making bets on what the future kind of standards will need to be. And you either get that right and become the standard or you don't. And Nvidia has gotten it right. And like I said, I think the market voted and the stakes for that are so high because you can extract the maximum amount of value for a given generation of chips in like six months. I know the depreciation cycle is technically longer for that, but you get most of the value in six months.

Rob May: Yeah. You know, two years to design, it takes two years to build a data center. Like how do you make that math work?

Ryan Eppley: Yeah. I was an Nvidia fanboy growing up. I played computer games, so I've built many computers. I have my Nvidia running in there.

Rob May: Buy DGX Spark this week. They just came out.

Ryan Eppley: Yeah. Yeah. No, I haven't broke the bank for that yet. But I recently switched from an Intel CPU to an AMD CPU because obviously they do more than just GPUs, but I think they're they're trying to tackle a lot of different markets whereas, you know, Nvidia specific type of architecture. We're just going to go all in on it. Have you taken a position on AMD is trying to do too much across too many verticals or it's actually a very smart game, but it's going to take years to kind of capture those different markets?

Joe Dormani: Look, I don't have a strong position on it. Although like what Nvidia seems to do that specialization does work in a lot of different mark markets you know the vertical focus for them owning the hardware owning the software on that and then you know the data center business as well just seems like you know a really hard advantage to overcome right I mean you have to choose where you're going to deploy capital and you have to you know the other thing that I'll say actually about Nvidia is we see them much more actively in the startup investment environment We've done deals with them. We've actually one of our portfolio companies was acquired by Nvidia. So they're they're much more active and impressive.

Ryan Eppley: Which one?

Joe Dormani: It's not publicly announced. Okay. But it's a it's a model optimization company. So they basically help choose which software chip is the best and well suited for a specific model. So you know but just by virtue I mean like that's what the compounding benefit is of that revenue growth and you get to operate from a position of strength. You have more optionality. You can build on that competitive advantage and I think Nvidia's definitely done that.

Rob May: So we got one article left. We just got a couple of minutes to chat about it. Was written by Kyla Scanlon fellow u Louisavilian I found out. Maybe we can get her on the show if she ever comes to New York. But it's called politics, capital, and attention. And it's about the fact that AI is the market and the market is the government. And she's trying to tie AI into this idea that like kind of like how is the economy doing? And we were talking before the show about this that in Q3 AI infrastructure spending was more than all combined consumer spending in the United States.

And so you're in this weird thing where you know because this is a very political article and it's very critical of the Trump administration a lot of like tariffs and H-1B visas prices and all these kinds of things and it's like well but then look at the stock market and I think you have to remind people the stock market gains are mostly like a handful of companies 20 30 40 companies that are very AI-centric that are driving it all all the ones going up 70 80 90 100% per year and the rest of the market's sort of bob bobbing along. I've read some things that like two-thirds of the economy is actually in a recession right now. And so, how do we untangle this and or like does it matter? So much to say. You have a thought?

Anna Kirk: Yeah. I mean, I think we were talking about this the other day in terms of the circularity around AI right now. And I think it's definitely bubbly in nature. Like I think we're we're approaching some type of bubble when Nvidia is investing a h 100red billion into OpenAI to buy a hundred billion worth of Nvidia chips. And simultaneously OpenAI is committing like 60 billion to Oracle that it doesn't actually have yet and Oracle and then I saw Oracle committing to AMD as well. So there's like a lot of spiderw webs going on here. Oracle is their debt to equity ratio is like 500%. It's absurd. So like there there's something that is going to happen at some point whether it's now or later. The question and I think the article does a good job of posing this is if if AI is a bubble does it even matter because it's making stocks rip. It's making private valuations rip and the public is taking that as a signal of confidence that the market is actually okay. And because the public is doing that, the government, at least this article alleges, has leeway to take more political risks, including tariffs, without destabilizing the entire economy. Like, is that sustainable?

Ryan Eppley: Well, one of the major kind of practical things that we saw was Intel's deal under the Trump administration, right? Taking 10% and that's, you know, reshoring manufacturing of very critical semiconductor research. I felt I felt like the article was very critical, but it also brought up more questions that I'm sure you know you've you've spent time thinking through of like is it okay to not just be a capital focused decision-making nation? Is it okay to think about adversarial kind of positioning over the next 5 years? You know, one would say free markets should be how we always think and that's why capitalism has been, you know, the greatest force of good over the last, you know, couple centuries. Others saying, you know, we're in a different time period, especially with the rise of China. You know, how how do we feel about governments getting involved with capital decision-making kind of forces?

Joe Dormani: Yeah, good question to take on. Yeah, look, look, I mean, you know, I think there there's a lot there to unpack. I'll I'll I'll start to take some some some chops at it, but I think like the first thing about the spend and investment, I think, is there's a reason that these investments are being made. And I think when we talk about like is there a bubble or is there not a bubble and you know what needs to be true for these investments to play itself out. I think we have to sort of contextualize that in the value of the market opportunity that that exists. And I think we're just in a transformational period where you know we've lived in a world where the entire software market was worth $300 billion maybe. And we're talking about the professional services market which is you know in the tens of trillions of you know services in general. So like the magnitude of the opportunity is substantially more than the than the SaaS market.

And so when you think about the level of investment going in relative to that opportunity it it doesn't necessarily seem inappropriate. Now are valuations out of whack there? Yeah, like they are. And you know like there will be many companies you know likely in kind of like a messy middle of the market that you know don't kind of reach their potential and ultimately have ne negative outcomes but there is real value being created and there will be a portion of companies that do get it right and create substantial value and will justify I think the amount of spend.

So I actually take kind of the other side of is there is there a market that I think I commonly hear which is I actually think we're underestimating the amount of value that will be created over the long term with you know AI and so you know that doesn't mean that there won't be like substantial market pullbacks that don't mean that doesn't mean that you know the market will be completely insulated from geopolitical activity like obviously the world's a complicated place you know there's you know tension with you know US's relationship with with China obviously you know Russian Ukraine train is obviously still problematic issue. So like the world will will never be perfect and yeah like could there be you know things that you know substantially shake the system or shock the system and is you know the broader stock market being propped up by AI investment and the performance of magnific Magnific Magnificent 7 yes probably but the long-term opportunity and value creation that I think this investment will drive towards I do think will at least at some point kind of push us through whatever trough to peak that we'll that we'll go through through there.

Ryan Eppley: She almost reminded me exactly of the moment where I think we've all read Adam Smith Wealth of Nation. I think people forget that it was written in 1776, obviously the same year of the American Revolution, right? So imagine Wealth of Nations maybe the hallmark of capitalism, modern capitalism was written the same year of a major geopolitical occurrence. I don't know if that was coincidence. I don't believe it was having studied a bit. And so I think we're in that same time period of Wealth of Nations. Is titled that but it's about capitalism. So I think countries that are capitalist are trying to figure out how they positions themselves and jockey on the world stage as a force not just a function. U I know we have to wrap up.

Rob May: Yeah, we got to wrap up. Why don't we ask our final question here? What do we got?

Ryan Eppley: Yes. So everybody get 10 seconds to answer. We always ask question about New York and today's question which which we meant to do last week when we had when we had Eric the massage robot. But is but it's about sort of stress and peace and calmness in New York which is you know this is a city of energy lots of stuff going on. Where's the place that you go to find peace to find your calm?

Joe Dormani: All right so I'm am a native New Yorker. I live on Long Island now. This is the disclosure but I my family is from Williamsburg, Brooklyn. My family lived there for a hundred years. My great-grandparents came over from Italy and, you know, they built a house there and many generations of my family have lived in that neighborhood. It's obviously changed a ton with a lot of the development there, but there are a couple of places in the neighborhood that have always just, you know, make me feel better and calm. So there's a place right off the Bedford Avenue subway stock called Joe's Busy Corner. That's my my uncle Joe was the original Joe for Joe's Busy Corner. OG Joe. The OG Joe. Exactly. And so I'll I'll put that as like one of the, you know, one of the areas. And then, you know, like from there, couple blocks down to McCarren Park, for sure is like just an area where I always feel calm.

Rob May: Beautiful. What about you, Ann?

Anna Kirk: New York Public Library. Architecture is beautiful. Great natural light. Love to go there anytime.

Rob May: Mine is related. Mine is the basement of Strand bookstore. It's the Union Square, 11th Street, 12th Street, wherever that is. Strand. Best selection of used technical books in the country. And it's calms me down to go there. Yeah.

Ryan Eppley: Find me next to the fountain in Madison Square Park. It's awesome. It's great spot. Everyone knows about it, but I don't think people use it to like just get some zen. But it's awesome. It's nice background music, nice background noise. But very happy there.

Rob May: Yeah. Well, Joe, thanks for being on today. I think this was great. Everybody out there, thanks for listening. Please send us your merch. Subscribe, like us, leave a comment, suggest a guest, all those things. And we'll see you next