Watch directly on Youtube for best experience. Youtube Link

‍

Transcript by AI

Hello, welcome to Stack Genius, the podcast for data-driven investment professionals. Today, I have Ernest with me again, actually. So we already had you as a guest when you were telling us about your product, Starwatcher. Today, we have an hour planned to discuss technology and the state of investment and all of it in general. So thanks for joining today, Ernest. Happy to be here. Thank you, Silvan. I'm a bit of a fan, I have to say. You're not only the founder of Starwatcher, but I think you have a super high nutritional value podcast yourself where you interview people from the industry. Tell us a little bit about your podcast. Yeah, it's... So it's called Star Watcher Observatory. So Star Watcher is out there looking at stuff, observing startup universe. And actually, I started this as a Substack blog, but the first entries were podcast episodes. And I might change that in the future where I keep on writing about what I think about the world. In our blog, I have in past written some stuff about AI and that sort of topics. Just to have like a snapshot particularly. And the podcast this far has been... particularly around data-driven VC initiatives. And I focus on getting people who actually are doing the, walking the walk. In the VC firms, those are the people who actually build stuff. And sometimes, in some cases, those are partners, but in most cases, those are actually professionals. Yeah. and who were guests so that people get and get a feeling for, um, what to expect. Oh, I have, uh, had wide range of, uh, data-driven VC, uh, circle celebrities. So Andrea from early birds, uh, Mike from moonfire, Mike Arpaia, uh, And I have had Atomico, Sarah, Pietro from EQT. Yeah, so I have had a good bunch of people. Yeah, I like the discussions a lot that I listen to. So I highly recommend to anybody to listen in. But more about you. So maybe in 30 seconds again, tell us what Star Watcher is so that people can understand the following combination. Yeah, Starwatcher started out as a database, but it has evolved basically in mapping projects where we map startups and VCs. So essentially, you can grab any company name, type in the Starwatcher, and we will show what happens around the competition, target customers investors everything what happens around this particular company and you don't have to we don't have to have the company in our database to show you insights it's simply you add url and we show the the we do the company profiling and mapping in real time That's quite interesting and different to what many others do out there. What made you decide to build it like this? Let me tell you a story. Long story long. Back in the days, I was thinking about why VCs are not data-driven, and that evolved into talking and getting to know people in the industry what's happening and i thought well the main one of the main problems is a lack of data but then when i start to scratch the surface then it's not the data it's actually the it's culture and it's uh easiness so to say to use the data and those are those are two separate cases so one is culture where a lot of in the past industry, it has been very inbound driven. So some people have money, a lot of people have ideas. So tendency, how the information flows is to people who have money. And so people start to look at stuff. That kind of starts to change. There are more and more people with money. The information flows more freely. And so VCs start to compete on deals. And then we come to deal flow and everyone considers like, yeah, deal flow is super important. Quality deal flow is super important. The problem is that people don't, maybe they have feel, well, they definitely know what they want. We just want to have cool companies and look at the potential unicorns. Okay. But then when you again scratch next layer, it's what exactly that means, cool unicorn, in your context. And then it's about the investment strategy and everything. And so we started to unpack that. what are the building blocks of investment thesis and what are the building blocks of startups and how we could map both of these sides and find unifying aspects. So let's say there is no need to give you access to the full deal flow if you are investing in PropTech. It doesn't make sense. if you're investing in real estate, but you're focusing more on construction and B2B side and not on the B2C side and finding cool apartments. And we map out all these features and try to find the best match for startups and investors. So that's something we have been on idea level working for quite some time and on more practical level, for a bit more than a year. I mean, you touched on it a little bit that there's quite a few sourcing tools out there, maybe 30, 40, 50 different tools and similar number for matchmaking platforms. So why is there so many, what you think? Because the promise is huge. And We just have to be on every platform and then we will find the unicorn. And that applies for both sides. As a startup, we just have to be and fill in the profile and then we will be matched with hopefully Sequoia. And from the other side is that, yeah, this is like we should track people who are working on some stealth startup. And then we will just like, you know, out of the blue, get that unicorn material. It's, it's the promise and people have this feel that yet the only missing thing is data and matchmaking platform, but it's actually not the case because when You go there to these platforms, they are basically like a data dump and both sides are focusing not on finding the best. So there are two dimensions of data, data dimensions. One is who you are and the second is how you present yourself. And in all of these platforms, it's kind of mixed together. And so what happens is that everyone over-promises themselves. They over-push who they are and what they do. And we're looking for next unicorns who will disrupt industries and ambitious founders. Yeah! What exactly that means, because it appears that they invest in BHC, below million, and they are really, their ticket size is 500K. And on the other side, they are a startup. Yeah, we are about to just like getting break even and everything, but it's like, yeah, we got two customers for our MVP and we actually work on very narrow stuff and not like disrupting the whole industry. And those are you know what where that leads is that yeah it doesn't work because both sides have uh uh both sides essentially have uh i wouldn't say lied but uh overestimated who they are and so it doesn't work and you said the lack of data is not the problem can you unpack that a little bit more for us so i hear people speak about specifically in early stage companies not having enough signals so um what do you mean when when you say the lack of data is not an issue What does it mean, signals? People want that one single unicorn signal straight from the darkness in their face, like this is it. But it's not that simple. The data, I know people who are buying the... the data room and pitch book and whatnot and what they end up is that they have to unpack that data themselves because it's it's not the best it may be the only one but it's not the best data set uh and so people are trying to unpack it and it's really hard and then It could be easy to blame on lack of signals. I know, but then, yeah, we would like to see hard data. Oh, well, cool story, bro. We would like to have an email updates regularly. Well, that's also a cool story. What it ends up is that I spoke with one VC who said, well, I have access to more than 300 startup Google Analytics. Here you go. Have fun. You have hard data. But there is no context where that person would go like, ah, today I will go through 300 of my accessible Google Analytics. It doesn't work. The signals are there. It's just how you're going to capture it. The same applies for those email updates. Email updates, they suck. Maybe that's not the right term to say, but again, it's how valuable are there and is that the best channel? And do you actually know what you're looking for? And the right way, which doesn't scale, would be that after the conversation for VC with the startup, they agree, well, here's the next step and just let me know how you go towards that one. But then again, imagine startup having to report 10 different kind of goals for 10 different VCs. And so they end up the classical now classical joke that in startup life startup life is pretty boring so you just have three bullet points what happened in past couple of months and then you go to chat gpt and make it more juicy and then on the other end we see that reverses through the same chat gpt make me three bullet points what happened which again shows exactly like Both sides expect that they have to have this exaggerated information flow, like, yeah, yeah, everything. But in the end, they want reality. They just don't talk about that. If we speak some more about data, I mean, definitely access to data is an issue. It might exist already, but not everybody has access to those 300 Google Analytics or other sources of data. for the less technical people in the audience, you end up needing to normalize data from different sources and then also deciding what is the truth or how do you combine the data? Can you speak to us a little bit about architecture for such a data lake or how to build something like this? Yeah, that's quite a topic. I like the joke about data lakes that pretty fast they become data swamps. And that's a true story because you become data hoarder. oh, there is a really nice piece of data set and some overlap with that piece of data set. And then you start matching and it doesn't match. And then you keep on hoarding. And then your data lake becomes data swamp. And that's an engineering task. And for a lot of people, it hasn't been accessible before. They might have clear view about their investment thesis or companies they're looking for, but it wasn't available before because you had to get the data, then you had to structure them in a particular way. And then you had to have great UI, UX, and access, and then, yeah, make sense of it. And that has been a problem since, a major problem since forever till last year. Where last year, with the rise of AI technologies, the game is changing. The game is changing, but yet... Let's say AI enables different kind of pattern how to look at the data, but the technical people are not there yet in the most part. So yeah, that changes. I think we will go away from these data lakes and really a lot of relational database patterns where you have hard categories, you have hard structure, and it will be more fluid. And how do you manage, for example, entity matching or resolution or whatever you want to call it is a big issue. Yeah. It's the same whether it is in relational or in lake format. Right. So what's the issue with entity matching? Why is it so hard? Because startups are living organisms. It's, oh, we did a pivot and just in case we changed our domain name. And it's not actually only the startups. I recently saw a trend. I noticed the trend in my dataset, in the Startwatcher dataset that VCs are switching to .VC. And then, well, if we're switching to .VC, maybe we could just, because now we have access to this domain space, we could just drop fancy schmancy capital now could be fancy schmancy .VC. And that would be like pretty cool, eh? Yeah. And so that changes also. So entity matching is a big problem. And I spoke with, I forgot the name. From Red River West, we spoke about entity matching and he said that that also is the case with some datasets which are out there. He mentioned at the time it was crunch-based data where he said, There was a certain percentage, if I'm not wrong, like 1.5 or even more, where there was entity matching problems. And those entity matching, that's one of the things where with availability of data sources, that's what happens because you have no clue what people are doing on that end. Like you're not at the beginning of the data flow. you don't know how data are created, how they are matched and formatted. So you're missing out on control what's there. And so some companies have this identity matching, some don't, some data are outdated and now that thing has switched. So yeah, and you end up with data swamp. If you look at it from 30,000 feet in the air, right? So you need to decide which data you want to procure or get. You need to build access to the data set. That's the extract in the ETL, right? And now you have extracted the data. You need to transform it somehow. And we speak about entity matching right now. If this is such a big issue, why are there no good APIs for it? If everybody's solving it separately, or are there good APIs that you are aware of for entity matching? For entity matching, no. I think that's because that's a pretty hard problem to tackle. And VC industry in itself is not big, so that's not that big of a problem over here. Well, it's a big problem, but the market is not as big, so people don't do stuff. And the data analysis in other parts, I'm not sure they... the problem is as big uh right it is but there's this uh negative uh negative response rate or how was that so basically People don't care in other industries as much about that because it's used for cold outreach and then basically it's just spam and you don't notice that you have been spamming the same people with three different kinds of messages. I recently got a personalized email which had the company name I used to have 10 years ago. And I was like, wow. And that was exactly the case where there was some crazy entity matching where they got my latest email and company name for 10 years ago and my name, which hasn't changed. That's right. But it's amazing. Yeah. It is a problem, but for those people, they don't care about this spamming part. Yeah, well, whatever, just spamming. And in VC industry, where the entity matching is important, people deal on their own. That's my feel. Yeah. Yeah, thanks for sharing that. I've been shopping around and looking for good solutions myself, and I did not find any. That's why I was so curious. I've been shopping, for example, in the KYC space. So where you need to do creditor risk analysis and these type of things, but typically they don't work for small private companies. So they work for bigger private companies. So the solution space is definitely an issue. Let's switch gears maybe a little bit. So we spoke about the extract, the transform a little bit now, then the load. You need to put it somewhere exactly. And you already started to speak about AI or the changes that are currently happening. So what is the current state of AI, do you think? And what do you feel is... happening in LLMs and other technologies that helps do the transform or the load. Yeah. Yeah. You won't believe what LLMs and AI are going to do in the future. Here are the 10 tips and prompts you should nail. But I have a bit different kind of view. I think most of the people look at this as a next generation of kind of the database stuff. And in like next google instead of googling you will find you you will just get an answer but technology technology behind chat gpt and other uh models it's uh i like how andre carpathy says it's a lossy compression of information so in past 10, 15 years, we have generated gazillions of data, like whatever, the crazy amount of data. And we have been using it in form of search, which was pretty big leap when we moved from the catalogs of Yahoo, like, oh, I'm looking for entertainment sites and I will just go to catalog and then scroll stuff. Yeah. which was a big leap from the notebooks with addresses in 1995. Agreed. Yeah, right. And then, so we think of these terms with Google because we have been using it since 2010. 1998 and 2000. And that's the paradigm there is I'm coming up with a search query, which now is prompt. And then I have these pages and then I go and then I go through one, two. I don't like this one. It doesn't look like good one. Then I get this one. I skim over. I look into it and then I formulate my opinion and blah. And that worked pretty well, but now the information is, there's so much information that actually we can statistically build the representation of the world. There's one good part and one not that good part. The good part is that, wow, you can ask questions and it will give answer. The problem is that, and it's a problem and amazing stuff. it can give you back answers which you didn't expect but which are really cool creative ones based on the documents but it can also hallucinate but it's the same it's two sides of the same coin and that's where we are moving away from hard data and and and into more of these, yeah, basically the model, these models are, we expect that that's like a next generation leap in the search, but actually it's more of a reasoning engine where our goal as humans is to describe the information, gather that information and describe what exactly we want out of that information. We treat ChatGPT as a question-answer machine. And then you ask, and they go, oh, that's not true. Well, that's the point. It's statistical average answer what it has come up. The more details you give, the narrower statistical answer is there. And if you have very exact question and you, to have exact question, you facilitate with additional information, you give examples, you give materials, and then it comes back with a brilliant answer. Like how to build a wooden lodge. And it will come back with statistical answer. Oh, that's how you build. And you're like, yeah, but I have rocky ground. ground there well now it has an opinion that not not an opinion but there's a new dimension and then it says well this is the hut you can have but then it's like well but i have three kids and i have a car and so you add on more and more and then you're like here's the um here's from my um my local authorities the map of surroundings and the elevation and everything and then you feed in more information but that's totally different kind of approach than we used to use information where previously we were googling making our mind up and using our stuff like thinking about but now over here it's it's It's very important that we understand what we want to have, what we want to achieve. Going back to investors, it's like, what is your, exactly what is your investment thesis? Investing in unicorns, it's, well, that's, of course, that's the end goal. That's an exit. But what is your strategy? How are you going to pick? And it's like... And if you can formulate information and gather the right data, then you can ask the machine to make up the mind. The good part is that you can automate that. And that's where I think we are not there yet. We still think about these engines and everything as search machines and everyone is super hyped, overhyped. And in my view, OpenAI and some companies basically are just feeding the hype. Like there are no... there's so much to do with the underlying technology, but they keep on giving, you won't believe what happened next. This changed totally everything. Developers are out of the job because we will have something. And then we don't look back to auto GPT or GPT engineer or whatever, because there's new thing to be hyped about, but there's so much to do. Yeah. I don't know. This was a long rant. I appreciate it. Maybe sidestepping a tiny bit from there. So Ben Horowitz and Marc Andreessen, they also have their – they call it Ben and Marc show. And they recently discussed also AI and what's happening there. One of the thoughts they had in there I thought was quite interesting and relevant to what you said. They said – by nature of the technology how it's built currently it gives you average answers right but if you are looking for exceptional things then a technology that outputs average might not be the right thing so are we just trying to find a nail for the hammer that we now have and try to apply it, for example, in scanning startups, in the screening process. But the results of that screening will be quite average by nature. What do you think? Yeah, I heard that podcast is pretty nice. I think in a way we are missing the point with LLMs. It is average, but as I said, it's average when it comes to reasoning about generally. Let's say it's average when it's general, but when you feed in information, then it can make sense. If people have tried to add in the chat, like chat with PDF, that's a good example. You add PDF and then, wow, it actually extracted the information and figured out what to do because now it has structured information and the context. And I think the answer will be not on the LLM side to become smarter, Because the underlying technology, I'm not sure it's that possible to make it much more smarter. It is smart, but it works based on the human data. And you can, of course, fine tune it on a more high level information. But the interesting, the researchers have shown that if you start to like skew the model and just take out some parts, for example, there was some research where they had grounded models where those were without swear words and everything. And they performed not as good as No, so there are grounded models and those who are just like, yeah, they will go with swear words and everything. And those models which are not grounded and go with swear words, they actually are better. And why? Because they have broader perspective on the world. And Those models, there are thousands of dimensions and we don't understand what are those dimensions. We have kind of clue. And if you take out the swear words, you're basically taking out sarcasm and edgy thinking about stuff. And so the model in a way becomes more dull. And then if you fine-tune the model to look at the world from the super smart, arrogant person view, it again won't perform on understanding the information, the real world information, which you would like to have. That's my working hypothesis because I'm not the expert. the Oracle to answer this one, but my bet is not on the models, but on the information and the agents working with that information. Because the agents is that you can give very small task to the particular agent and it will perform because now you have like very specific task. come up with the so for example you want to have like i don't know some executive summary your task your single task for this this particular ai agent is you have to be the best one who is just making the description of solution which is made by this company That's it. Here are examples for good, bad, ideal. And here's the company description I have. Make the best one. And you can nail it. And then that together with other agents, basically they all together create this executive summary. But essentially what it is, it's... you divide your thinking process and then optimize each individually and dump your view on the world in algorithm or in prompt or whatever. So will AI be able to handle more of the investment decision process than currently visible? I mean, venture success forecasting is just super hard, right? I mean, just one dimension. What is the biggest problem? The cycle is ten years. I agree. And I recently watched Moneyball, and I think that in a way, some people have the expectations that there is this Moneyball moment Potentially, it's going to be there. But I don't think so. Because Moneyball and these kind of things are for games. What are the games? Games have clear results. They have a set of rules, set of items, players. And you can figure out, you can deconstruct that. But the nature of investment world is by itself, it always... You cannot step in the same river twice. And that's what the innovation is just going forward. Oh, we built the model based on how Intel came about back in the days. Well, good stuff. Now we have invented processors. The world is not going to be the same as back in the days. I think AI is going to become big assistant to VC world. And gradually, we will automate some of the decisions Bigger, smaller, I don't know. But I don't think that there's a Moneyball moment coming in, particularly to VC industry or this quant moment in trading where they basically built the algorithms which just were trading stuff. Because those are... You can make those decisions. But investing in startups, it's one quarter idea, one quarter the team, one quarter personal stuff, and one quarter money. And it's like, you're just like, I don't know. But getting to the short list, what we do at Starwatcher, we basically want to show you different kind of short lists based on your different kind of perspectives. The best shortlist. One of the elements, you spoke about team and market and the typical T's, the four T's that everybody's looking for, but even something like tam sam and some um you know getting current data for something is hard then many of the companies that are outliers they built their own market right so the song exactly have didn't even exist seven years ago right and you know how do you do that so they i would you know follow following your logic i would definitely argue that It's very unlikely that you can forecast things reliably. So you will end up with a little bit of an art more than a science in the end. And we look for this. There's a lot of silver bullet seeking. Yeah, which will be the case where I can. I need the solution which will estimate the sum to all my portfolio, potential portfolio, any company. Sometimes you can do that. Like, I'm building something for architects. Well, that's very, like, hardcore stuff. Like, that's it. But then there are solutions which just come in and they create the markets. They simply, there is no such metric. Sometimes it's, yeah, we are investing just in these kind of companies who have very good... very clear unit economics if you cannot tell about unit economics then yeah you're not for us again well sometimes it works sometimes it doesn't sometimes it doesn't make sense back in the days when the web 2 came about everyone was just like just grow fast and we will figure out the business model afterwards when you have millions of audience there is the business model somewhere there And I think it's not the answer as typical philosopher. It's not outside, it's inside yourself. The answer is inside yourself. you have to look inside what exactly you're looking for. It's like, oh, and the same applies for the startups. Like, here are the 10 things to put in a pitch deck. And then people are like, okay, one, two, three, four, five, six, seven. Yeah, that doesn't make sense, but that was said there. And so I'm putting in... it's it's it's really like start start from yourself when asking question what exactly i want to accomplish what what are my companies what is my view and then go from there what kind of information i need to make up my mind where can i get the information how can i analyze and then kind of evolves more naturally than any all of these you won't believe how these 10 prompts changed my investment life like getting inspired all day instead of going for a long walk and thinking i mean if i listen to you um Just pondering in my head right now. So should investors then build their own data lakes so that they learn over time? Some people like Correlation Ventures in the US, they started, what is it, 20 years ago or 16 years ago or something like that, right? um and they've been performing well i don't know if it's because of that but it's definitely part of their lp story and now recently many more have followed they're investing money in building their own lakes does it make sense to build your own swamp or lake again It depends. I think a very good answer was at some point from Mike Arpaia, who basically said, well, at Moonfire, it's our competitive advantage to have the platform. And they have worked on that since day one. And I think it works for them because it's not just technology, it's the culture. That's the way how they built the company. with technology in the mind since day one for all sorts of size um bc firms like we would yeah i think we need the data lake and then you get the data lake and then Yeah, no one is using that. And it's not the lack of data, it's the culture. And that's the question I have asked in every single podcast episode about the culture, how people change the behavior and do they use the data. And all of my guests have said, yeah, that's one of the things you have to work on to get people to work. And that changes the processes, that changes the thinking. And I think those are more important questions than should we need or don't need the data lake. That's the easy answer. Because do we have a budget or not? Yes, we have a budget. Yes, let's have a data lake. And then it's like in South Park. In South Park, there are underpants gnomes episode. I don't know what the episode was. There's underpants gnome. And underpants gnomes, they have a business model. Collect profit. No, they say, well, our business model has three phases. The first one is we collect underpants. Then we have the second phase. And the third phase is a profit. But what's the second phase? Well, the first phase is to collect underpants. And then the third one is the profit. And I say a lot of times to the startups, you have like underpants business model, underpants gnome business model. We are building the platform. We are collecting the data. And then we will have a profit. Okay. But what's the second step? We will get the data link and then the unicorn. And that's not the straight correlation. At my times with Cisco, we always had the running joke that in the OSI model, the layers, right? OSI model layer 10 is the typical problem, which is management, right? Yeah, and that's the challenge that they hold the money, but they are also the problem. And then how to get the money before they realized the problem. i mean what are the the reasons to invest maybe if we twist it around the profit is not clear but the risk is much clearer of not investing in technology and augmenting the performance of the humans. So maybe you can get away with that for a little bit more, for a few more years. But I would argue that in 10 or 15 or 20 years from now, it's going to be hard if you are not augmented with tech. Yeah. So the VC firms have an interesting lifecycle. And so there are the big ones. And even the big ones are sometimes they have a hard time to raise money and they live on a brand. And historically, they are like a big tanker just going. And then there are small ones where if you don't perform, you don't get healthy money and then you die. And then that's it. And then you move on. And I think that's a good part because then you can have, like, you can reborn and create new fun with new culture, new people. And in 10 years, there's going to be a new generation which will be fluent with ChatGPT and everything. And I see the Gen Z already, the up and coming VC. Now they are associates, but they are coming in also as partners. And they have a totally different kind of perspective on how we should work. and it's not around like in like it it was before so i think the change is coming and again not necessarily for from per because of technology per se but because of the different kind of culture coming in i mean there's quite a few funds that are changing generations now and i mean generation of managers right so um it's going to be interesting um if that generational change is going to run smoothly and maybe it is also a timing question that the departing guys currently they don't see a big need to invest anymore or learn this new trick this new technology But the following ones might want to. Let's change topic again a little bit and speak about the state of the industry. So Doug Leone in a podcast said that it's a commoditized industry now. So it's much different than it was 20 years ago. And emerging managers might have a tough time raising funds currently. What is your thought on the industry and what do you see with your customers maybe even? I think we're at the interesting time when the rates are again up. So we have lived in a world where for some 15 years, there has been very low rates. And so that means cheap money, and that means a lot of people had an opportunity to come into industry because they could get to the money easily. And so it was either you give to the borrowers or you're investing in higher risk, which is VC world. Now the game is changing and now it's a reality check where it's not that easy to get the money. And so LPs are evaluating more, more like what what's the performance and it could have been that like five years it was commoditized industry because oh now everyone is in in the industry and capital is easy available and And now I think it's reality check. Is this really commoditized industry? And is it that easy for the funds to raise money and then move on and just have it as a cycle? So we will see that part. We will see that part. Yeah, that's on the VC side, I think. What I'm more interested in is what I'm actually interested in is that because of the data-driven approach, there are emerging models where it's not about venture capital, but it's actually about borrowing or lending money to the startups based on their KPIs, where let's say you have 10K MRR, 20K MRR, and then the Gillian, for example, used to be our capital. They will give you money against your Google Analytics and Stripe, if I'm not mistaken. That was the case. And those are really interesting models in the time now with... higher rates, that actually could make more sense. And then the question is, is that the venture capital or not? Venture debt. Yeah, venture debt. That is growing. And that's a very interesting direction. And of course, founders are interested sometimes going for venture debt because you don't give out equity. So that's something I'm interested in. their liquidity cycles are also a little bit better with that for sure. Yeah. Okay. Sorry. Good. Yeah. Yeah. That, that, and that, yeah, the liquidity is also the big, big problem now in a way, in a way, because, um, If no one is buying companies, that means that you could buy, you could build a great portfolio of startups and then sell down the road. But apparently private equity is also in trouble, so they are not building startups. of startups anymore. So we'll see. So meanwhile, startups and VCs are stuck without exits and LPs are overexposed to the venture capital. And so we are doomed. Let's not make that the last word of this conversation. No, no, no. I think we, in general, industry is in interesting times that, All of these aspects are coming together. And also, there are funds who have a lot of dry powder. And the capital has to be deployed. True. There's pressure rising. I mean, Terry Pratchett, one of my favorite authors, he has this line. Actually, Douglas Adams it was. Apologies. And he said, it's interesting times we live in, and it's always true, right? Yeah, it's always true. Exactly. So having these four ghosts and bold statements, I'm always a bit cautious. Yeah. Yeah. Maybe gearing towards the end of this conversation. So what are people that you follow or books that you love? So maybe some inspirations for other people. The books. What are some books? I have a bunch of books. I like this one. So it's a good strategy, bad strategy from Richard Dermelt. That's very good. What's good about it? It's actually a good framework for thinking a bit bigger, but in a structural manner. Having an actual strategy. A lot of times, startups just come in and it's like, yeah, boom, we're going to just smash it and do stuff. Yeah. Well, underpants comes afterwards, but... that this one is really good in a way it helps. And if you look up good strategy, bad strategy, there's one guy wrote a really good summary blog post where you can get the gist. So Rumelt wrote the book, that guy wrote the blog post. So that's really good to have a strategic view. And then I have this one, which is Howard Marks, the most important thing. This is a really good one. What is it about? It has very bad reviews. I think it had really bad reviews on Amazon because it's so simple. But the point is that Howard Marks writes these very deep essays and sometimes... those things are kind of obvious, but he puts them in right perspective. And about investment thinking, Yeah, that I really like, how you operate with the risk, how you think about the portfolio, how to mitigate risk, what's the psychology of investors in the market, what are you actually seeking, is that return or value of an asset, and a lot of that applies also to startup investing. Whom am I following? from the data-driven VC space, maybe on the more technical side, who do you think? Yeah, of course, you have to follow Andrea, data-driven VC. That's like if you're interested in stuff. Great content. I look a lot for stuff on Lex Friedman podcast. that's that's like you have to go for a long work walks to listen to his podcast those are like sometimes goes up to like for a couple or a couple of hours sometimes i listen to his podcasts while i'm in the countryside doing some some getting my hands dirty and just listening the podcast about the future of ai like doing something really manual um there's a latent space podcast which is about ai practical stuff about ai knowledge project i'm listening to the knowledge project sometimes i listen to the 20 vc Harry has good guests and he's teasing and he's actually also asking some controversial questions to get interesting opinions. So I like that. What else? Yeah, it's about, I have to then look up, yeah. But I have a bunch of books I have to read. Yeah, Ben Horowitz and Mark Andreessen is something to follow there, of course, have deep conversations. I'm always fascinated by the depths of thinking where they, like, Mark Andreessen, he's like, yeah, and there's this book, and there's this idea, and then that comes from on top of that idea, and then it's like, what the heck? Where was the beginning of this one? Yeah, so those are things I'm following. That's cool. Thanks for sharing that as well. so we're almost at an hour so um definitely i recommend people to also follow your podcast because uh you know it's it's very uh unfiltered real accommodations about you know how to work with data and how to build data different products so um i think that's a big suggestion And I really hope to see you again somewhere soon and wish you a fabulous day. Sunny, it looks like, where you are right now. Yeah, in Latvia, we finally have got summer. Very good. A couple of weeks ago, it was snow. It was a surprise for everyone. But, well, yeah. Well, Ernest, thanks so much and have a fabulous day. Thank you. Thank you for having me. Bye.

‍