Guest: David Marra, Founder and Portfolio Manager at Markin Asset Management
Hosts: Seth Earley, CEO at Earley Information Science
Chris Featherstone, Sr. Director of AI/Data Product/Program Management at Salesforce
Published on: August 22, 2024
In this episode of the Earley AI Podcast we are joined by guest David Marra a seasoned expert in AI-driven investment strategies and founder of Marken Asset Management. David brings invaluable insights into the transformative potential of AI, particularly in complex payroll workflows, ERP systems, and financial services.
David joins our hosts, Seth Earley and Chris Featherstone, as they explore the current state and future of AI, including the integration of large language models (LLMs) and the importance of robust data management.
Key takeaways:
Insightful Quotes:
"They're not oracles, they're things that are great in the hands of people that know how to use them. But just like any artisan, it's really much more about the artisan doing things with the tools that they have rather than the tools themselves." - David Marra
"The low hanging fruit is simply let's do existing things better, and we can get probably a 30% bump on that. But the 300% bump from one trillion to three trillion is going to come from doing new things that we didn't even know we could do today." - David Marra
"Jensen made the example showing the Infiniband wiring—there is more throughput in just these cables than there is data passing over the entire global Internet in the same point of time. That's what you need because you've got a lot of data coming in being spread across so many chips." - David Marra
Tune in to discover how investment professionals evaluate AI beneficiaries and technology companies—and learn which data infrastructure decisions separate successful AI implementations from failed pilots.
Links:
LinkedIn: https://www.linkedin.com/in/david-marra-8693b44/
Website: https://markinfunds.com
Ways to Tune In:
Earley AI Podcast: https://www.earley.com/earley-ai-podcast-home
Apple Podcast: https://podcasts.apple.com/podcast/id1586654770
Spotify: https://open.spotify.com/show/5nkcZvVYjHHj6wtBABqLbE?si=73cd5d5fc89f4781
iHeart Radio: https://www.iheart.com/podcast/269-earley-ai-podcast-87108370/
Stitcher: https://www.stitcher.com/show/earley-ai-podcast
Amazon Music: https://music.amazon.com/podcasts/18524b67-09cf-433f-82db-07b6213ad3ba/earley-ai-podcast
Buzzsprout: https://earleyai.buzzsprout.com/
Podcast Transcript: AI Investment Strategy, Financial Services Transformation, and the Evolution from Machine Learning to LLMs
Transcript introduction
This transcript captures a comprehensive conversation between Seth Earley, Chris Featherstone, and David Marra about AI's transformation of financial services and investment strategies, exploring how organizations evaluate AI beneficiaries, the critical role of data infrastructure, the evolution from machine learning to large language models, and the emerging multi-trillion-dollar application layer market.
Transcript
Seth Earley: Welcome to the AI Early AI podcast. My name is Seth early and I'm Chris Featherstone. And today we're really excited to introduce our guest to discuss how AI is revolutionizing the landscape of investment strategy as well as business performance evaluation. We'll talk a little bit about some of the common misconceptions around AI tools such as LLMs, the importance of data hygiene and purpose built systems, the use of AI for extracting data from business databases, and accurate information retrieval in corporate settings. And we'll talk about utilizing algorithmic investment strategies to improve portfolio performance. So our guest today is a seasoned expert who has been in the AI space since the late 90s. He sold his first venture in 1999 which focused on Internet search. He spent time implementing quantitative prediction systems at large enterprises with Boston Consulting Group BCG. His academic background is in finance and in 2010 he focused, he founded an investment research firm building quantitative investment strategies for institutions. He brings a unique perspective with a background in coding systems implementation in large enterprises and the use of AI in financial modeling and in the investment space. His current venture is focused on algorithmic based investment strategies for institutions, wealth managers and retirement advisors. David Mara, welcome to the show.
David Marra: Well, thanks very much Seth. It's nice to join you.
Seth Earley: Terrific to have you. So tell me, give me your, your thoughts, tell me your perspective on the state of the IND industry. How well are organizers, organizations in your sector leveraging AI and what is their level of understanding the people that are doing the stuff you're doing in terms of, of quantitative analysis, qualitative analysis in the investment space?
David Marra: Yeah, sure. So just, you know, for the benefit of your listeners and your viewers, my day job as today is as a portfolio manager, an investment portfolio manager. So in this role where we have long only strategies, equity growth strategies and a hedge fund, those are the things, those are the investment funds and strategies that I, that I manage. And my job in this role is to, you know, obviously find opportunities to invest in that we think are going to perform well, you know, in excess of the risk that we have to take with them to build diversified portfolios of AI companies and AI related companies from the tech sector and, and then to manage, to manage risks. Right. So on the one hand, one part of my job, there's really two aspects that are to it that are relevant for our discussion today. On the one hand, we're looking at AI as an investment opportunity. That's one aspect of it. And the other aspect of it is that we're using the tools as a way to invest as well. So we're both kind of consumer of IT and a producer of AI. So I just wanted to set the stage back to your question about artificial intelligence and how, you know, I, I think your question gets at kind of how, how we view it from how it's affecting our industry. For our industry, I would say that our industry is probably one of the financial services is one of the slower industries to adopt AI and this gets at the importance of data for implementing really enterprise wide robust AI systems. The more data you have, combination of publicly available data that's relevant to your business and private, the more quickly and more broadly you can embrace AI and start to generate either new business models or efficiencies off of existing business models. Financial services, a weird thing in that a lot of the data is proprietary, that doesn't always lend. That's not the fastest way to take off in, in this, in this industry. So I don't really view again as a quantitative investor, we're a little bit different than financial services and industry because you know, we deal with more diversified data sets and more of them. But outside of our niche of quantitative investing, just more broadly speaking to the industry, which is what your question was about. Yeah, financial services is definitely embracing AI, particularly on the customer service side and the banking side on the investing side. I think it will come later. It's really other industries frankly that, that you know, I'm more excited about and see a lot more opportunities particularly to invest in.
Seth Earley: And so just for our audience because there are companies that are either early stage, mid stage, some startups. What kinds of companies are you looking at? Are you looking at large established companies that are doing a good job with AI to change their business models or evolve their business models? Or are you looking at earlier stages mid tier? What are you, what are you looking, what's your investment profile for?
David Marra: Yeah, so our investment universe is anything that's listed or anything that's listed that's heavily investing in unlisted things. So the, the Googles of the world, for example, is, is an example of a company that's listed but certainly a certain percentage of their market cap and their growth in their market cap can be attributed to a lot of their investments and things that are unlisted. Microsoft's probably the best example, having a stake in, in OpenAI, for example. Right. But, but primarily our, our universe is anything that's, that's, that's listed. So for your listeners, any company that's looking to go IPO or on timeline, as well as anybody who, who's listed would be the kinds of things that, that we're looking at.
Chris Featherstone: So yeah, I was just going to ask you. So Dave, so clarify the quantitative investment strategies. What does that mean? Is that just in terms of, of how you just described the companies you're investing in or is that something specific?
David Marra: No, no, that, that actually yeah. Just so there's no confusion. That quote, that's what we do every day. We use a lot of quant and compute to evaluate all these companies in the AI ecosystem that we're looking at for investment purposes.
Seth Earley: So one of the things you, you discussed and I think you're alluding to this now are AI beneficiaries. Right. So these are the companies that are doing a good job using AI. They're either achieving growth or reduction of costs using AI moving forward. So give us a sense of. And you, and you said, you mentioned that, that finance is a little bit behind in the quantitative investing side of things, but they are using it for certain for banking and analysis and portfolio analysis and predictive analytics in all sorts of ways. Customer service, but not necessarily in the quantitative investing which, which seems the more specialized area. But talk a little bit more about the AI beneficiaries that, that you're seeing. You know, how well are the organizations that you're evaluating for investment doing in their execution and operationalization of AI and how do you really separate what they say they're doing from what they're actually doing? I think we talked a little bit about in our prep call.
David Marra: Yeah. So AI beneficiaries is something, you know, I'm particularly, you know, excited about because it's really where the rubber hits the road with AI and it's certainly where all, you know, the, the, the devaluation gains is going to come from if companies don't adopt it. Everything that the hyperscalers and the data center people of the world are building is all going to be for naught. Right. There has to be efficienc generated by, from all industries for this, this, this to work out from, from an investment perspective. So, and we're very bullish on it. I mean we think this is a secular opportunity, secular very long term opportunity to invest in a revolution that's going to play out over multiple decades. So I'm glad you, you kind of mentioned beneficiaries. We also, you know, for the sake of your, your, your listeners at Markin Asset Management where, where you know, you know, we have an AI beneficiary strategy. So we're invested in 50 firms across industries that we think are well positioned to benefit from artificial intelligence in terms of how it's going to drive its earnings and drive its growth. I mean, what I see in the beneficiary space is first big picture that we are at this tipping point in time, machine learning came along. And again you, you alluded to my, my life at, at BCG and other management consulting firms that I used to be with. Like you, Seth, I was in the trenches there implementing these types of predictive modeling systems using machine learning. And up until large language models came along, it was a slog. It was very, very hard to find companies that had everything you need, which is you had to had pristine, clean data sets that are constantly being updated. You had to have the data pipes that, that combine those data sets. Because you've got all these data sets sitting in all parts of your organization, geographically, in different places, all with different permission structures and you know, the maps that have to go into making that a single easy way to query, etc, all of that infrastructure needs to be in place. Then once you had that, which most companies didn't, you then had to build these very kind of highly specialized prediction models that could be very difficult to maintain, very difficult to create. A lot of PhDs involved, a lot of data scientists involved, and, and then the, the payoff. So the, so the number of use cases, every time you got something like this, it's filtering down to a smaller and smaller number of use cases that delivers less and less return on investment. Right? That was up until the large language models came along. So we're really talking up until about a year ago, that was the world of machine learning. And then really large language models comes along and it's a much easier implementation use case where, because a lot of firms did put in the effort for machine learning type of algorithms to build all those data pipes and those database assets that they needed, they're kind of now ready for the LLM revolution. And then LLM itself being sort of a thing that again, to oversimplify, because I think you have a very sophisticated audience, everybody knows, but you can just take a mass of data and sort of throw it at it and you'll start to get reasonable results. We're gonna, we're gonna have to go beyond reasonable results to drive, you know, big ROIs. But LLM becomes something that's much, much easier to implement and start to drive return because it's so generalizable to many functions, many different types of queries. It doesn't require so much kind of quirky, very narrow use case type of fiddling that you had to do with the machine learning stuff. So that's, you know, a very colloquial way of explaining why we think we're at the beginning of a very generalized revolution that's probably got a multi decade run to it across many industries.
Seth Earley: So one of the things that you mentioned was that you have about 50 firms across industries that you were investing in. And obviously I'm not asking for your secret sauce, but what are some of the commonalities that you're seeing generally in terms of organizations that are doing a good job? Do the, are they the ones that have already made that investment in their data cleanliness, integration, the data pipes and so on, as kind of a precursor? Because we still need new good data even though LLMs will help us deal with bad data. And you know, we can use LLMs to cleanse data. Right, but what are some of the characteristics you're seeing in companies that you believe are kind of at the cusp or going down this path? Is it that they've already deployed this for certain applications? Is it that they're, they have new business models that they are enabling? What are some of those ways of looking at those companies?
David Marra: So on the one hand, I think, and I'll just speak, I'm not going to speak to our portfolio specifically, but speak to the kinds of firms, you know, that I think are the characteristics of the kind of firms that really kind of understand this revolution and then are starting to show evidence that they get it and they're going to be able to derive something from it. I think on the one hand they've, they've done that basic data asset work that we just discussed. That's something that, you know, that's not something you can do quickly. Right. That type of work to get your data assets in order often takes years to get right. In large global organizations, most of the firms who are going to be able to move quickly have that piece largely in place. There certainly may be work to do, but they've done that already. That's because they were already participating in kind of the machine learning decade or two leading up to the present moment. The second aspect of it is that they perform the kinds of activities, they're intensive in the kinds of activities that AI is good at. Because AI is not good at everything. Right. AI today is not good at physical activity. AI applied to robots is way behind just AI relative to query based activity. Right. We don't have C3POs walking around doing stuff yet today. So things that involve Physical work, construction, a lot of services, businesses like delivery and logistics, a lot of businesses like you know, hotels and things like that where it's about cleaning a room, thousands of rooms a day, you know, that kind of anything that involves physical work is not the type of thing that's ideally AI's position to do today. Earlier things that, where AI is very good at are things like accounting, tax, even legal. Right. It's, it's work that's often data retrieval oriented is not very good at reasoning. Reasoning is not, not a strength of it. Today human beings are much better at reasoning than AI is. That may change but the kinds of industries where data retrieval, you think something like tax, for example, a lot of it is just knowing the rules, right. It's a rules based industry. So where you have rules based industries that you can, you know, that lends itself to query and being able to. And in fact where there are simple queries is even better for AI. Today where AI is struggling and a lot of startups are starting to kind of focus in on trying to improve, is the multi step query. And in fact what's even more difficult is the multi step query that ha that has an answer that come from multiple domains or multiple databases. These are things that even today a struggles with. So things where there are simple queries from a single domain that really gives you, you can get you know, 30, 40, 50% productivity increases. And those are some of the industries, you know, I just mentioned accounting, tax, legal, those kinds of things as just some, some, some ready examples. There are others as well.
Chris Featherstone: So when you guys get into, you know, some of this from the perspective because you know, a lot of the investing used to go with like definitely a ton of education and also with your gut. Right. Like and I think that's where you're alluding to this where the AI models are really good at binary, you know, classification of yes and no. Right. Here's the outcome, here's the input, here's what we're driving towards. But we still need, I'm assuming that human in the loop to do some of the exception handling. Right. Can you talk a little bit about that and how you think about that from the investing side of things. But you know, and then also maybe what companies are maybe missing as they implement some of these because we have all the patterns now, we're using the reasoning behind it.
David Marra: Yeah. And I think what you're getting at there, and it's a great question is you know, when you're transforming an ERP system and an enterprise system the first thing you're gonna, you're always gonna look for low hanging fruit, right? You're in a new industry, you're in a new age, you're dealing with new technology to minimize the risks and maximize the return. What are you gonna do? You're gonna go for low hanging fruit? What's the low hanging fruit gonna be? It's gonna be doing existing things better, taking existing workflows and doing them better. An example would be like an ADP type of workflow where ADP is a payroll company, right? Payroll involves a lot of rules, right? Think of the kind of workflows that, that a payroll expert is involved with today on an ERP system like that, it's like somebody gets hired, a whole bunch of workflows are going to come from hiring. Somebody gets fired, somebody gets laid off, a whole bunch of workflows, somebody moves from location to location, tax implications, all kinds of pay, pay implications. The government changes its rules about how something is accounted for, whatever, a whole bunch of it. Each one of these workflows today involves a person who is skilled in that particular workflow working on an ERP system. Now if you can make that ERP system so that somebody with less skill in the hiring process, for example, and they've got to go through, you know, 40 steps in the ERP system, if somebody's skilled, they can go through all 40 steps and they know how to answer all the questions and plug everything into the model. If somebody's not skilled, you could have a query based system, say, I'm stuck here, this is what I'm trying to do. And have a chat bot that answers specific to your firm, exactly what you're looking for. So you can plug in the right data into the ERP system. This is going to reduce errors that you were talking about before, right? This is going to allow more people who with less training to do more work in a day and be able to process more hires, to be able to process more transfers to different locations. That's kind of an example of how the rubber hits the road. And like the ADP CEO, for example, has actually talked about this on conference, on conference calls, for example, quarterly conference calls. That's, I think to me that's a very illustrative way to show how, you know, you can take an existing ERP system and existing workflows and all of a sudden you can make it. And the average here with these kinds of types of innovation is about 30%. But the big gains will be from, you know, new workflows and Actually new, new things which we could also talk about.
Seth Earley: Yeah, we, we are, we're with one enterprise right now that is in the medical device manufacturing space and the number of knowledge sources, data sources, information sources that they need to deal with that field service tech needs to access. We did this applied materials years ago as well. You know, there's about 14 different information sources. And so just that, consolidating that, getting them the answers. It's the holy grail of, you know, knowledge management. The right information of the right person at the right time or personalization, the right information of the right person at the right time or recommendations. Right. These are all things that we've been striving for. Usability. Anytime we're trying to improve that user experience, we're trying to reduce the cognitive work, the cognitive load of the human right and be able to present information from people who need it. And instead of going to a colleague or not being able to find the answer or having a lot of other manual processes to be able to surface that information. So one of the things you talked about was the knowledge and, and the access to that knowledge. And of course LLMs are very good at understanding information. They're very good at understanding language and they're good at making things conversational. They do hallucinate. So you know, are these organizations that you're seeing be more successful? Are they paying attention to knowledge as a foundation? Knowledge management, knowledge engineering, those types of things that are a little bit in the weeds but are foundational and fundamental to some of the improved workflows that you're talking about.
David Marra: I think that the, I think we're in the early stages. I think that 2023 was the wake up call year. Wake up call year means we've got a new technology, it could be very useful. We need to start looking at this 2024, right? Yeah, exactly. 2024 is okay. We started running pilots in 2023 and the first half of 2024 and now we're starting to roll out production systems. 2025 is really the year when I think the, the results will start to show up for these more leading edge companies that are heavily investing in it because they've, they've the, the, the pilots. For example, a number of the firms with the early pilots were literally starting to transition from pilot to production, you know, three months ago. You know, this is a rough kind of, kind of rough kind of time frame. And so those results won't really start to hit bottom lines in an appreciable way until early next year. Right. That's when those results will start to start to show up. But this is the time when people are creating competitive advantage by being out there running more AI pilot projects to figure out what the priority priorities are for production. And you start to, you start to realize the benefits starting next year.
Seth Earley: Yeah. And part of it is also the data. I mean when content is data, knowledge is data as well. So when people are looking at remediation and getting their house, data house in order, it also has to be around content and knowledge, right. The unstructured information, which is what large language models are good at. And so when you think about these organizations that are making some progress, what, what are they missing? So they're getting some things right in terms, they're understanding the workflows that need, need to be enhanced. They're looking at the heavily rules based, unambiguous outcomes, right? Like you know, whether it works or not, right? By based on a use case, based on testing, you know, it's not ambiguous. So they understand the workflow, they understand the process, they understand what the outcome needs to be, they understand the rules. What are they not getting or what are they missing or what are they still struggling with? Would you say?
David Marra: So on this? I think, I think the total addressable market here for kind of the. I think this comes down to the application layer. And the application layer, you know, you get, you get the chips first, right? You get the LLM second. Well, I guess you get the data centers. The data centers and the chips first, you get the LLM second and then third, what do you get? You get the applications, right? And it's really the applications that explode the industry. I think the total addressable market right now for the application layer is just under a trillion dollars globally. Nine hundred and some billion. So just market at 1 trillion. I think that where this goes is probably to two or three times that. So we could be looking at a $3 trillion application layer market because that's where the action is. I mean that's, and that's huge, right? You know, triple, triple, triple. A very already very large industry. And that's the thing that's missing, Seth. Right. It's the thing that's going to come, but it's the big driver. You know, we talk about those kind of the low hanging fruit will never be the transformative fruit as you know. Right. The low hanging fruit is simply let's do existing things better. And we can get probably a 30% bump on that. Then we can get another 30% bump like kind of let's do more of everything we do better. Right, right. But the 300% bump that I'm talking at 1 trillion to 3 trillion, that's going to come from doing new things that we didn't even knew we could, knew we could do today. And there are certainly lots of, and that's where you know, you coming back to like startups and things that are in the, in the, in the, in the, in the research centers of the listed companies. Those are the really interesting things that are going to explode this market. And that's why I called it a secular growth opportunity. You know, over the next few years.
Chris Featherstone: I think of, you know, you went back to, you know, in terms of 2023 being a marquee year for a new introduction of technology. 2024 POC, you know, hell, right, 2024. Also I would argue part of this secular market is not only, you know, the advancement and innovation of the, of the LLM LLMs themselves, but all of that infrastructure that's sitting around it because you outlined it. You know, you have all the data centers and chips and then you've got, you know, the other infrastructure pieces. But we're also seeing a proliferation of how many different technologies now are there to help with architectures to help with, you know, you know, like ML Ops, but now LLM Ops, you know, and architecture centered around what's the most cost optimized way in order to put an LLM up and also facilitate the ability to switch out language models at will. Right. Things like that. So the part of that I'm excited about is to see this groundswell of all the supporting cast of characters that's also being raised and forced to be raised because we're just barely getting to optimal infrastructure or optimal architectures for these types of applications, which will then to your point, facilitate the two to three times growth that's happening. Right. So just anyway, go, go.
David Marra: Yeah, yeah, I mean, on that, Chris. So, so, so you know, one strategy that we, that, you know, Seth asked about is, you know, what are going to be the beneficiaries. And I talked about, you know, the 50 companies and then we had a chat about that. But the other is, you know, we have another investment strategy which is an AI technology strategy right here. You know, we're invested in like another 65 companies, completely different companies, which is purely on the technology side, which is what you're talking about. The, you know, who are going to be the beneficiaries within the technology space. And I mean, it's for, for a portfolio manager like me to have so many opportunities to look at many good opportunities where I want to keep a portfolio that really doesn't have more than 75 things in it. It's just, it's a fantastic time to be managing a portfolio like that. So let me give your, your, your listeners some examples. And you gave some examples. Like, you know, I was there at the Nvidia GTC conference where Jensen, you know, got on stage and introduced the new Blackwell chip. And, and one of the points he made during the introduction is that it's not a chip. The entire server, the, the entire data center is a chip is basically his point. The whole data center is a chip, right? So what you're to think about it, you know, in simple terms and the way they kind of express it is you can think that we had to. We had to start to put chips on top of each other, but don't think of, you know, thousand chips sitting on top of each other as a chip. As the thousand chips. It's a chip, right? So your whole rack becomes a single chip, and then your rack is now a row, and the row is really a single chip. And now you've got 30 rows. And really the 30 rows is the zinc, right? What enables. And there's, there's technical reasons why they have to go that way, right? There's physical limitations when you get down. So particles are so small and all this kind of thing. The, the, the. What enables making an entire data center work essentially as a single, extraordinarily fast chip with extraordinarily big throughput is networking, right? And I can. And one of the, one of the, one of the big networking companies is a company called Infiniband. For example, I remember buying my first Infiniband server at my investment research firm now a decade and a half ago, right? And Infiniband has just constantly improved the amount of throughput that can go through a cable, right? Because now if you've got 30,000 chips that are all acting as one chip, right? They're all communicating with each other, and they've got to communicate with each other very, very quickly. And Jensen made the example during the, during the conference. He was like, let me show you the cabling, the wiring, right? And he was showing, essentially what I think is, don't quote me on this, but I think it was Infiniband, all Infiniband wiring. And it was just a bunch of silver cables wrapped around each other within one rack. And he's like, there is more throughput in just these cables you're looking at here than there is data passing over the entire Internet in the same global Internet in the same point of time. That is so. And that's what you need because you've got a lot of data coming in. It's being spread across so many chips. All these things got to communicate with others, all got to be synced. And that's what that. So this was just one simple example to illustrate what Chris is talking about is that all of a sudden you had this kind of sleepy area of networking, which a lot of people didn't care about. Now people who, like me, who want to invest in this, don't really care about it. Right, right. Because it's essential now to this new.
Chris Featherstone: Revolution that's so essential.
Seth Earley: Fascinating. Especially when you start thinking of the physics and, and the, you know, all of those pieces behind it and the, the volume of communication. I mean, that, that boggles my mind what you just said. And I, I'll have to dig up that, that recording to, to hear that and see what that cable looks like. But it is, it never. I love the science behind. I'm a scientist at heart. You know, my degree is in science. I love to read about science. So that's really fascinating. Let's see. So I want, when you, when you start looking at the organization, yes, there is this big ecosystem, resources and technology firms and smaller technologies and boring sleepy technologies that was that. That are now exciting and leading edge. Right. Like networking, as you said. And, and there's lots and lots of different, like a billion flowers blooming. Right. There's so many organizations that are just building on top of foundation models that are building different applications that are integrating with other systems. You know, one of the companies I was talking to, and I get approached by a lot of these startups who, you know, want to be on our podcast or want to partner or whatever was talking about agentic technology flows and you know, and we've been using agents with LLMs. Even when you're doing a lookup, when you're doing retrieval, augmented generation, you're using an agent, you're using an API to call something else. Right. So that's getting more attention now, you know, as simply being a mechanism by which you can do other things. With an LLM, you know, it seems to be growing up as its own space, agentic workflows or agent powered LLMs or LLM powered agents, however you want to say it. Do you have a thought about that? Because again, what we're talking about is having an autonomous agent that can go off and do certain things, make certain decisions, be focused on a goal and then find ways to achieve that goal. And I just read or started to read a research paper, I didn't read all of it about LLMs learning what tools to use, you know, kind of autonomously, you know, and I again, I don't know how big that is or how valid that is, but it was talking about how these LLMs are able to start looking at research. And of course with APIs, you know, you, you can have a lot of descriptive information and what all those integrations should look like. And potentially an LLM could look at that and say, okay, this is something you need to do. What your thoughts about, you know, multi agent workflows and you know, this idea of orchestrating specialized agents that will go off and do things and work with other APIs either internally in the organization or externally. Do you have some thoughts about that?
David Marra: So I think that where we're going to see innovation first in that space is actually going to be in the consumer space. Think I'm thinking Google, I'm thinking Apple. So Apple demonstrated this agent activity, agent action type of use cases in their, in their know when they announced Apple Intelligence. And we'll see. It seems like some of the features are going to be delayed and you know, this is the usual stuff in tech, but this is, this is definitely coming. But I think it comes, and the reason I said I think it comes first in consumer is because you're not gonna let, at such an early stage of, of you're not gonna let it do take autonomous actions if they have big consequences. It's okay if it takes an autonomous action to remind you about you need to do something or buy something or go somewhere, whatever, whatever that you know, that's okay. It can kind of screw up and get it wrong. And you know, we're not, we're not, we're not, we're not gonna drop a bomb in the wrong location or whatever we're gonna do, you know. So I think the autonomous actions will start to come in places where the, the, the, the cost of getting it wrong is low. And that's why I say consumer. And certainly Google has announced they have a whole big set of a lot of projects working on autonomous agents. And Apple demonstrated it actually in their, in their, their launch videos. So I think that's that Seth, that's where we'll first see those actions. And as they become more reliable and dependable, those things will migrate toward ERP systems. And things, you know, enterprise systems and things of that nature.
Seth Earley: So David, when you, when you say that we want to be safe, obviously you need guardrails, right? You need to have governance and you need to understand what your processes are and what the limitations are and what triggers a workflow and what triggers an exception and when a human needs to decide. But it sounded to me like the, the whole ADP workflows could be agent driven. Right. So, so, or, or portions of them. Right. So you could speak. And I imagine this is what they're already doing. And then the question is when you do have that exception and that have human in the loop, then you're also augmenting that with that ready accessibility of that knowledge.
David Marra: Definitely on, on that, that, that, that journey from a 1 trillion dollar industry to a 3 trillion dollar industry. This happens along the way, right. You know, that's part of creating all that value. But I just, I just think it takes some time, Seth. Yeah, that's all I'm saying. It definitely comes in and it sort of has to come to go from one to three, so to speak. But I do think that, you know, people are going to be very cautious about making sure they implement that in a way that it doesn't cause bad things to happen in a costly way.
Seth Earley: Right. And as these things get more interconnected and as you have more dependencies and more APIs, it becomes more and more challenging for traceability, for audit trails, for troubleshooting, for optimization when something does go wrong with, you know, Joe's third party agents in your ecosystem. Right. You know, what's the liability? So there's a lot of governance issues around this as well. How are you seeing some of these leading edge companies doing on the government governance side? They must have to get that fairly locked down in terms of how they're using things, how they're making decisions, allocating resources, measuring results, you know, making, assigning accountabilities and so on. So that's a big piece. What are your observations there?
David Marra: I'll have to be honest with you there, Seth, as a, as a. More taking the lens of a financial analysis of these firms, that's probably at a layer below which I really look and get involved in. But I don't feel like I'm the right person to answer that question. Yeah, yeah.
Seth Earley: Well, you know, it's something that we're finding there's a lot of demand for an appetite in the industry because people are trying to figure this out. They're trying to understand, well, how do we make these decisions at the board level so that we can have guidance and policies and procedures on the ground in order to make this work.
David Marra: I will say that certainly I've seen analysts comment on that. And you know, the big issue there seems to be that, you know, obviously when you have a query, you have a query and it involves multiple databases and multiple layers of data within databases, all of which have their own sort of permission structures. Right? This becomes very, very complicated because I might be asking a question and if I'm the CEO, I can get all the databases involved to give me an answer. But if I'm somewhere else in the organization, only can answer from these databases and only from these sections of these data cases because of permissions. So obviously there are all kinds of, you know, there are startups working on these issues to kind of, you know, figure out in a very automated way, okay, for a particular query, depending on who you are. And that's part of the context. Right? Not only who's asking and what department are they in, but are they allowed to see, you know. Right. You know, so this becomes very, very complicated very quickly, as we all can imagine. Certainly that's part of the $1 to $3 trillion journey is companies need to be able to figure this stuff out because we can't be calling somebody up to say, hey, does he have permission? Right. It's all got to be built within, into the system.
Seth Earley: Yes. The permissioning, the provisioning, the respect of privacy and even confidential communications within the organization. There's one company we're working with that has a, an orchestration agent and you know, it's monitoring all these conversations that are going on to give some, provide some memory to the individuals having those conversations. But some of those conversations are privileged. So while the orchestration agent knows about it, it also knows that when Joe and Bob spoke about, you know, potential layoffs or terminations or whatever it is, that is a privileged communication and that cannot be surfaced to others. So it is, as you say, it's a set of problems and provisioning, it's, it's kind of analogous to an organization working with IBM and Apple on semiconductor fabrication. Certain things they can share, certain things they can't share. So all those security models have to be very well thought out and so on.
David Marra: I was just laughing a little bit because I find that there's always share point databases in my company that I should be able to access that are telling me I need to ask for permission. Theoretically the CEO has access to everything, but in reality, I don't know why that's Happening. I should be super admin.
Seth Earley: Hey, I know this is a little bit off topic from the We Podcast itself, but I'm really fascinated by the GLP1 ecosystem that we had discussed. Right. And just because I'm sure people are fascinated by that stuff, but we know that those are being used for controlling appetite and hunger. But you had also mentioned that this receptor and the drugs that target it have huge potential beyond weight loss. I, I would just love to hear you riff a little bit on that. Just, just because I think it's a fascinating topic. Sure.
David Marra: If there's an AI connection, great. But if not. Yeah, I mean, just again, for the benefit of your listeners, as I mentioned earlier, you know, I manage a, a portfolio that is, you know, technologies, stocks, but another portfolio brings in, keeps that portfolio. But for people who want to also invest in the healthcare innovation sector, so it's, it's AI driven innovation and healthcare driven innovation, we have a, another strategy which adds on the GLP1 GLP1 for the benefit of your listeners is the weight loss drugs. And the GLP1 refers to the actual mechanism by which it's at, its accomplished within your body. But broadly speaking, it's, it's, it's, it's weight loss drug innovation. But the interesting thing about weight loss drug innovation, it's not just about weight loss. And that's, that's what you were getting at. It's turned out that these, these drugs are delivering outcomes, health outcomes across a variety of, of diseases. So from liver disease to diabetes to heart disease and even to Alzheimer's. And it seems like not a quarter goes by where I don't see some study come out showing that there's promising applications in some other area of health. So we think that the ecosystem, just like as we've been discussing, is huge on the AI beneficiary side, the AI technology side, that there's a whole ecosystem there to invest in GLP1s. And we think that's another secular trend, meaning when you start to be able to show efficacy and promising outcomes, health outcomes in an area like heart disease, in an area like diabetes, in an area like Alzheimer's, I mean these are massive, massive markets on their own. Even just a hit in one area would be massive. And when you're talking about a drug that could potentially do it across these multiple areas, that's an innovation stream that a lot of investors, you don't want to be invested in.
Seth Earley: So, so, and that's interesting because it's not just about the weight loss it's about another mechanism that is impacted by that receptor and by that mechanism of action and those biochemical pathways, not just weight loss. And it's not just the consequence of weight loss, but it's a separate impact.
David Marra: That's right. Yeah, that's fast. That's right. And, and it seems to, again, we don't know because there, there are completely separate mechanisms going on and then there are overlapping related mechanisms. So just think about someone who loses weight, becomes just more healthy, right. They're closer to where their natural body weight should be. Part of it is, is benefits related to that. But as you're pointing out. But then there are these kind of specific mechanisms that can be more targeted and the drug can be modified to be more targeted that and get an even better outcome.
Seth Earley: Wow, that's great. That's awesome. I know we have just a couple minutes left and again, fascinating aside because I'm a chemistry geek, degree was chemistry. I love biochemistry, biology. I still read, you know, all sorts of publications and research just for my benefit. So I love hearing about that stuff. I just want to learn a little bit more about you for our audience. And you would tell us about your time in Japan. What was the catalyst and interest and how did those experiences impact your career and your personal growth? You want to talk a little bit about that?
David Marra: So for the benefit of your listeners, I lived, I lived on sort of multiple tours of duty. I lived in Japan for about 14 years. Two of those years were education, learn the language, study economics, things of this nature. And then I worked there for a large multinational for a few years. And then I started that, that Internet search company that we talked about. I started that in Japan as a VC backed, you know, startup back in 1999, as you alluded to earlier. And it was, I mean, that change which happened in Japan was, was, you know, a seminal event in my life because that was what, you know, having not really been a technology person, been more of a analytic finance kind of, you know, MBA type of person prior to that. That sort of woke me up to the power of how data. When you bring together data, when you bring together computer and when you bring together algorithms, these three things, man, magic can happen. Things that just weren't possible before. At that time it was searching the Internet, searching this global thing which I. Know. That just gave me a bug for the rest of my life was, you know, that we could do things with, at the nexus of those three things that would be exciting and bring people benefits and just would be exciting place to be and so that's where I spent the rest of my life, at the nexus of those, those things.
Seth Earley: That's wonderful. That's very exciting. Another question I like to ask people.
David Marra: And just last thing to add there. Sorry, interrupt Seth, but I will say Japan, obviously Japanese markets have done very well in the past year. And a large part of that is not just to do to the macroeconomic background, which is certainly part of it. The other part of it is Japan is. Japanese companies are right at the center of the AI ecosystem. I mean, there's a lot going on in Japan, particularly on the hardware side of AI that's absolutely essential to everything that goes into the data center and chips that, that go. That are powering the industry, but also that are going into your phone as well. Right. I mean, Japan has a very, very. Is well positioned on the hardware side, in particular of the industry to, to grow and benefit from AI.
Seth Earley: That's wonderful. That's very interesting. And you have some very unique insights from the language and cultural perspective. Absolutely. Another question I like to ask people when we have them on the shows, you know, if you could go back to the time when you were graduating from college and you could, you could give yourself some advice, what might you say to yourself? Is there anything you would tell yourself with this kind of perspective?
David Marra: Well, I think when you're my age and you look back on yourself in college, you think, man, geez, that guy knew nothing. Right? Absolutely nothing about the world. I mean, I, at least I, I hope we all say that about ourselves because, you know, if we didn't, we wouldn't have learned anything along the way. But yeah, I mean, question what I. Know about the world. Yeah, I mean, you know, on, on the one hand, maybe I would get, if I were to talk to myself at that age today, I would say, first of all, realize, you know, very little, so, you know, which is a good place to be. It's a good mind to have because then you're open to constantly learning, right? To trying to, you know, ingest as much as you can, learn as much as you can, you know, move along the learning curve as fast as you can, and you're inquisitive, you're curious. I think that's always a really good place to be in life and kind of stay that place and stay inquisitive, stay knowledge hungry, stay skill hungry. These are the kind of things that benefit you at, at every, at every age. But it's better to learn it younger and sooner.
Seth Earley: Get outside of your comfort zone and be comfortable, comfortable with that and that.
David Marra: I mean, I will say the last point on that is, you know, to any, any young people heading off to university or parents of people heading off to university, I did read a very interesting study recently on this. So there's, there's empirical academic work behind what I'm about to say. But it's been shown that, and in at least one study it's been shown that people who did dual majors have better lifetime earnings outcomes and actually happiness outcomes. And they speculate why is because, you know, there's a lot of volatility in the world these days. Right? The world is. Technology is making the world change at a, at a much more rapid rate. Not just technology, other factors. But that's one big one and a volatile world. You really don't know what the skills are going to be in demand over the work, your, your future working life. So it's better that you have a few pockets of skills to pull from. I personally recommend left brain, right brain, right analytic, liberal arts. Right. Do a dual major and, or you know, a major, minor, whatever is the case. But pick up something that's analytic, kind of oriented, scientific analytic, kind of oriented. Pick up something perhaps more liberal arts kind of oriented, which is what I call critique oriented. Critical thinking.
Seth Earley: Oriented thinking.
David Marra: Yeah, I think. Anyway, the study says do a double major. It's better than doing a single major. I would add to that if you're going to do a double major, do a left brain one and a right brain one.
Seth Earley: That's wonderful.
Chris Featherstone: I mean, from a portfolio manager, you know, it's, it seems very, very in line with diversify your education. Right? Diversify your portfolio. Diversify your education. Yeah, exactly, exactly. I like it as well. It's a good summary.
Seth Earley: Well, guys, this has been great. David, I've really enjoyed this. I wish we had more time, but maybe we can have you back. Participants conversation thank you so much for sharing your insights and joining us today on the early podcast. We really appreciate it.
David Marra: Well, thank you, Seth and thank you, Chris. It was a delight to join you. Both today and you will have your LinkedIn.
Seth Earley: Profile in the show. Notes. Your company website is Markin M A R k I n funds1word.com market asset management.
David Marra: Oh, I'm sorry, we had market funds. Markinfunds.com is the, the URL. But the name of the company is Markin Asset Management. Sorry, okay.
Seth Earley: Market Asset Management, but Markin Funds is the URL. Okay, great. All right, very good. Thank you. And again, thank you so much and thanks to all of our listeners and Carolyn, for doing all that work behind the scenes. And again, David, thank you so much for. Thanks. David, you thank thank you. All right, Take care and we'll see you all next time in the next Early AI podcast.