Brian Bell (00:00.859) Hey everybody, welcome back to the Ignite podcast. Today we are delighted to have Emmanuel Veloed on the mic. He is a partner and head of venture research at Hivemind Capital, where he backs frontier builders across AI, crypto and payments. And he's also spent 14 years teaching at Berkeley, go Bears. Today, we're talking about Dark Matter Lab, his new bet on funding researchers before the company even exists. Thanks for coming on, Emmanuel.
Emmanuel (00:25.1) Brian, thanks for having me. I'm very excited. Go Bears.
Brian Bell (00:28.187) Or should I call you Professor of Load? Well, I'd love to take us back to your origin story. What's your background?
Emmanuel (00:35.98) I'm a mass nerd, always been for as long as I can talk. And somehow I became a venture capitalist. So, I mean, it can happen, I think, in two ways. Like most mass nerds, I ended up popping up in the US to go work on Wall Street. Got the right place, the right time, and a lot of that I actually owe it to my thesis advisor. But effectively, after like the first part of life, couple of years, know, across banks, asset managers, BlackRock notably. I ended up jumping into startups, deciding to go build on the work I had developed during my grad school. That was in AI infra. That was actually, I think, a bit too early versus where market appetite was. And so after three painful years of being a founder, my lead investor basically suggested that I come on board to become a venture capitalist. And to be honest, I had never thought I would make that transition. I ended up loving it. Never left.
Brian Bell (01:43.717) Yeah, it's pretty, it's pretty addicting. You know, I have a pretty similar origin story. Wall Street, kind of a math nerd, finance, CFA. Hated Wall Street, washed out after a year, taught math in the Bronx, actually, in the teaching fellows program in New York. And then I did a bunch of random jobs all over the world, all over the country, backpacking, basically unemployed for the back half of my 20s. You know, off and on, just trying different things and... I ended up in the tech accidentally in Silicon Valley. I wanted to work in solar when I came to San Francisco 15 years ago and I took a sales job at a solar startup because I would take any job. I'd mop the floor at a solar company if I could because I really believed in clean energy and I happened to be in the old Twitter building at an NEA backed startup sharing an office with Instagram and that was my first
Emmanuel (02:32.526) Let's see.
Brian Bell (02:40.922) tech job in San Francisco. So I saw Instagram get acquired for a billion dollars and we kind of went out of business. You we sold for, you know, probably money raised, right? Wasn't a good exit for NEA and the rest is history. But tell us about, you, how has, you know, being a professor at Haas sort of informed what you do today?
Emmanuel (02:44.795) huh, yup.
Emmanuel (02:49.655) Yeah.
Emmanuel (03:04.366) So it started really as me trying to pay back what I owed to my thesis advisor. So she had asked me to come and teach within one of her main classes to share the way the industry puts into practice what she's teaching, which is basically securitization, structured products. That's how it started. And then it evolved into first me realizing I like teaching on its own as a kind of knowledge sharing thing. But I also realized that it was a fantastic conduit to appreciate what was really motivating students. Like what technical knowledge were they excited about learning and why? Like what did they want to do with it? It started surfacing very interesting research opportunities. It started surfacing effectively students that I really enjoyed the way they were thinking about problems. Out of the box thinking or going into what we it, less traversed paths of the world, trying to transpose. And I was like, okay, this is fantastic to basically spot people I would hire to work with me, or I would hire to go help portfolio companies, or would hire actually as founders, backing them investing in And so really like the last few years, what it's been for me is that really good source of what we call talent identification, talent validation, but also problem solving, business opportunity validation. The other part of that was I've always liked doing research. I've been doing research ever since I stepped into university. And so that has allowed me to keep a significant footing in research while actually having the flexibility of no publication requirements.
Brian Bell (05:09.274) Yeah. So you founded an AI infrastructure startup back in 2018. I think that was before it was cool. I was at AWS at the time. was leading AI globally. launched SageMaker Marketplace that year. So walk me through what you built there and kind of what experience you still draw on now as an investor.
Emmanuel (05:30.575) Absolutely. So it started from an observation I made at BlackRock that concurrently one of the other co-founders was a partner at Goldman Sachs, made at Goldman Sachs. And that's one of my thesis advisor at Berkeley had made through consulting, which is way too many people, whether it's Wall Street, whether it's tech. were spending time reading information, ingesting information versus prioritizing what part of that information should actually have more implications for their respective business lines. And so we said, okay, all of that, could we basically build infrastructure that ingest much faster, very accurately, in a very flexible way and ends up pre-tuing for these knowledge workers and speed up the time to information summarization and then prioritization, deep dives into whatever corner of the world you want to take this. That was the genesis for the startup. Back in 2018, still through this day, when you start going rubber to the road and figuring out, what are the friction points? Most friction points are actually technological friction points. How do you ingest fast enough in a robust enough manner, in an accurate enough manner? in a flexible enough manner, whatever content people are pushing at you. And then how do you make it an intuitive back and forth between the tech, the product, and the user that you don't force them into an answer, but you let them in a way drive their own path. And that was a big chunk of work I did during my first grad school. So some part of that was how do you accelerate AI workloads? Some part of that was... Notably, when it comes to natural language, how do you make it essentially agnostic, highly robust, super parallelizable, yada yada. Along the way of going from an idea to a business and eventually your business smashing into the wall, there were a few things that still mark me in how I try to act as a VC today. The first one was this very big understanding gap and language gap
Emmanuel (07:53.677) between basically the nerds and the VCs. So I remember a number of conversations where we would come into pitching to VCs, we would be explaining technical challenges that corporations are facing with regard to latency and speed and infrastructure costs. And we would be met with a blank stare. Or people trying to force you into what they've decided would be the right product. And you're like, that's literally not what clients are asking. So there was this first angle that was interesting and that still marked me. The second thing that marked me actually a lot was everything that could go wrong went wrong. Everything. Things that take much longer to build and are a lot more expensive to build. Sales cycle, that drag-on, you you thought they would take three to six months, they take a year, year and a half because along the way two executives on the other side of the table are leaving the company. and you need to reset your partnership negotiations. You're set by that. That was there. Issues, tensions among co-founders, bad stuff. The whole thing. I think it brought me a very good understanding and a very good empathy now sitting on the other side of the table for what you need to budget when you go raise and what may be in the back of your mind that you may not want to share with your investors, and actually you should absolutely be sharing with your investors so they can help.
Brian Bell (09:30.746) Yeah, it turns out you should tell your investors what's going on and you should give them if there's nobody else you tell the absolute truth to it should be your investors. Right? We don't know what we're doing and we're losing, you know, like we are losing sales in this in this area. You have to be brutally honest because investors have seen it all before. And we're the only ones that can like guide you because we have money on the line, right?
Emmanuel (09:39.744) I agree. I agree versus this. So back then, yeah.
Emmanuel (09:57.881) Correct, we have money on the line. We have the flexibility of speaking to a lot of founders and seeing enough of what's going well, what's going wrong that at the end of the day, it is in everybody's best interest to have that information. I remember I forgot when, but you have this like whatever content slop on social media with a number of... I call them influencer VCs or influencer startup founders, where it's like, your investors, they will take advantage of you. Why would we do that? Literally, nobody benefits from it. I'm going to benefit from it for like half an hour and then I'm going to be slapped back in the face for a decade. And that's, think, really like a big learning. And I keep telling founders in the portfolio, I'm not here to hear the good news. There is zero usefulness for me to only hear the good news. I only want to hear what's not going well so that then I can actually be helpful somewhere. It's just like patting you on the shoulder because you've got some good numbers.
Brian Bell (11:01.722) Hmm.
Brian Bell (11:05.688) So eventually it sounds like the startup didn't work out and then kind of tell me what happened next and kind of the origin of Hivemind and what that is.
Emmanuel (11:11.628) Yeah, so we had &A offers that got withdrawn from some very unfortunate stories. We had a bunch of things that didn't go well. But basically, my lead investor was like, OK, what do you want to do next? And I was like, honestly, I don't know. I might go into startups again, because I actually really enjoyed it despite everything went wrong. I might go back to Wall Street, because I know I can. Maybe I go into big tech. It was really like blank slate. And then my wife was basically like out of question, you're going back to startups. Because now you have a kid, so you're going to be a dad and you're not going to be a dad if you never sleep and I never dare, right? Okay, fair point. I didn't really want to go back to Wall Street. I kind of wanted to stay on the builder operator or more entrepreneurial side of things. And so my lead investor said, well, Why don't you become a venture capitalist? And I was like, well, you know, I've never really thought about that because I'm literally just a mass nerd. Like I love doing math, I love coding. And he was like, well, you know, that may actually help you identify interesting deals that others cannot. That may help you actually be value adds to some of these founders. And then on top of that, you understand finance, right? That should be helpful there. And so we started. with me just taking a look at the portfolio where I could be helpful and then just me digging so where would I want to spend time. So we did that kind of semi-informally for three months. I found a spot of FinTech where I really wanted to go spend time because I was like, okay, there are big opportunities there. So he brought me on board as a partner to formalize the global FinTech practice because prior to that, they had done it on the more... ad hoc and opportunistic basis. And as started doing that, in parallel, there were some portfolio companies that had to go through &A or had to go through restructure. And I was like, you know what, why don't you let me help with that so you can concretely see. And I can concretely see sitting on the other side of the table, not just the exciting stuff of who I'm looking at deals and doing deals, but like I'm solving problems. And I really liked both sides. So was like, here we go.
Emmanuel (13:36.788) That was really like the genesis of me doing venture capital. While all of that was happening, to be honest, I'm an awkward dude. I don't like speaking in events. I don't like being social. I'm not even on social media. And so there were parts too that were very helpful in being like, okay, what am I realistically bad at that I shouldn't touch? Or just no differentiation. And I was like, know, early stage is probably what I should do because you have a lot more tech, a lot more product. Yes, there is good to market, but in a way it's probably a better fit of person, a better fit of background. That was how the porridge came to be.
Brian Bell (14:26.874) And so how has the thesis evolved? Because it sounds like it might have started in crypto maybe and then now you're kind of more deep checker.
Emmanuel (14:34.211) No, no, it actually started in FinTech, not touching crypto. I was notably convinced, I remained convinced, and it's mostly been true, that finance would move to embed itself in non-financial platforms, whether it's enterprise SMBs or retail clients. I also had a very strong opinion that the way
Brian Bell (14:38.18) Vintec, okay.
Emmanuel (15:03.817) lending, debt capital market services work would completely get redesigned through this embedding. So that was really like the starting point. And then along the way, I started reading about blockchain in deep detail, like the papers, not the crypto projects, the Ethereum play paper, the Solana paper, some of the DeFi papers, like Morpho being an example, Aave being another one. There, got me excited wasn't so much the crypto angle as much as it was the blockchain angle. Because I was like, hold on, this is basically open source, trustless automation of financial services. mean, sure, you could do it in a fully DGEN manner in crypto, but to me, I was more taken to what you could do on Wall Street and inside payment institutions and inside, for example, small merchant banking type of services. So that was the second very big push into going really deep into blockchain. The third push, actually was of our first thesis that was me in a way going back to my roots, was you could call it the chat GPT moment, but to me there was a little bit of a different chat GPT moment. To me the chat GPT moment is the realization that the next five, 10 years of startups, notably AI startups, their mode is going to be technology first. And therefore, distribution will be very strong. While prior to that, distribution was the mode. Tech really was secondary consideration, mostly around scalability of robustness. And so when I saw that, was like, I want to go back to my roots. I want to go do really technical deals really early where the technology is a breakthrough and because of that breakthrough, distribution at scale will be possible.
Brian Bell (17:09.058) Right, love that. And you recently published a report on space data centers. Maybe you can talk about that.
Emmanuel (17:18.607) I had a lot of fun doing that one. Basically, what I started looking into was... Is a space data center an exciting narrative or is that a legitimate possibility? And if it is a legitimate possibility, how big, what gets in the way and the stuff that gets in the way, do we even understand how to tackle it? The background to that was in part, course, Elon Musk and SpaceX talking heavily. about space computing and space exploration and space mining. And obviously the guy is a brilliant engineer, brilliant product person, brilliant engineer with a lot of other downsides, but that's just what it is. But there was another angle to that, which is a meaningful amount of... research in mechanical engineering, in robotics, in AI happening at Berkeley, happening at MIT, happening at Carnegie Mellon, where a lot of it actually has relevance in the context of not just limiting ourselves to Earth. And I was like, okay, let's try to connect the two ends. so the third part was a friend of mine in France sharing a blog by a French engineer. who was like, hey, how do you call down the data center in space? And so I started taking from his piece, being like, okay, he looked into how to call things down in space, do I agree with the numbers, I agree with the numbers. And I was like, okay, great, we do know how to call down a data center in space, fantastic, but can we go a little bit further? Can we start figuring out how many of these data centers could we put in space? Where would we put them in space?
Emmanuel (19:16.151) And based on that, big of an addressable market expansion are we talking about? Because I was like, know, if it's small in a way, nobody cares. It doesn't matter whether you can do it or not. The reality is just a story. And so I went looking through how I would do it, basically concluding that the best way to place a data center in space so that it's in a way always capable of running and always able to cover some significant area on the surface of Earth was to put the data centers at the exact separation between the half of Earth that's in daylight and the half of Earth that's in nighttime. So one side always gets energy. So great, you have your solar panels, will have electricity getting produced. The other half is always in the interstellar darkness and so super good to cool things down. And I was like, okay, well, based on the size of that area, and the size of a realistic data center in space, how many can we put? Well, the answer is you could go, there is of course some uncertainty, but at least six times larger than existing data center addressable market on earth, all the way to 24 times larger. I'm like, this is a fairly big number, right? In terms of revenue, in terms of compute, right? In terms of how many more use cases you could serve. And so I decided to go then look into, okay, what prevents us from shuffling data centers in space. Cost is obviously one aspect. It's actually very expensive to go ship a data center in space. But cost, a way, and shipping, could say, well, that's the other part of what SpaceX is solving. Cheaper and cheaper, bigger and bigger spaceships to move stuff into orbit. So I'm like, okay, cost on its own is only a temporary block. there will always be enough money at some point to curb the cost down. I discovered the surfacing technology blockers that are first, very interesting, but second, are solvable. It basically boils down to, can we make communications fast enough?
Emmanuel (21:41.016) Right now, we don't fully know how to do that, but we have good directions on how to make it happen. And we've solved that on Earth already. And second problem was, how do we make it easier? How do we make it even possible to basically maintain, update, upgrade, replace these data centers? These are things that we have very good understanding how to go about. It's allowing smarter robots to go operate in space, do all this maintenance. It's allowing a mix and match of different GPUs, different models from different manufacturers, something called generalized heterogeneous computing, which is actually some of the work I did in grad school. And that was a core part of the info of my startup, just for like the cute story around. But at end of the day, I got really excited as I was going through the research because I landed on saying, it's not for tomorrow, it's probably not for in two years, but there is no way you can cross it off as not happening. No way. And when it happens, it's trillions in revenue. Potentially tens of trillions in revenue. Then you ask yourself, Would you be willing to spend five years for sure? Would you be willing to spend 10 years for sure? Would you be willing to spend hundreds of billions of dollars? Guaranteed the answer is yes. And so I ended up with that, I don't know, I found it really interesting. wasn't just like, I'm happy with the outcome. As I was going through answering the questions from my own understanding, just really interesting progression, really interesting elements I learned.
Brian Bell (23:36.909) I shameless plug here because what I think what you're getting into is this question of timing. Like and this is one of the hardest things I think as a VC to get right. And I just I wrote a new book called LP how to you know the insider's guide to investing in venture capital funds so shameless plug for the book I think it might be the first time I've mentioned on the podcast but this book's coming out in September. But the next book I want to write. is that question, right? When is the right time? And looking at like historical stories around timing, right? I think about this a lot as a VC. You know, if I was sitting in the VC chair in 1997 and Larry and Sergey got introduced to me, right? And they were like, okay, I got a new search engine, right? And you're thinking like, okay, well, that's like the 17th search engine to come along and like, how are you going to beat Yahoo and all these other...
Emmanuel (24:13.859) Yeah, I agree with you.
Brian Bell (24:30.904) Well-funded competitors that are way ahead of you and I think about that a lot right because that's an example of timing The other way around which is like you're coming in late to the market Versus you know investing Very very early like data centers in space, right? I looked at a YC company That was trying to do that and I passed I was like this is way too early and now they're like a two billion dollar company, right? And you know what I was thinking is like one it's too early And then two, know, SpaceX is vertically integrated and they'll be able to like eat your lunch on that. How do you kind of think about all that as an investor? How do you think about timing of data center investment, right?
Emmanuel (25:09.679) Yes, that's the eternal question. Last week, I had a discussion related to that with a friend at MIT who has been a repeat founder. And he said one sentence, Mark me, he said, you know, when you look back the last 20, 30 years of startups and you look at any point in time, the startups that are whatever unicorns, decacorns, where were they a decade before? They were never the cool kids. Never. And to me, that was a very interesting way of summarizing it, which is timing, the right timing, I think requires a bit of contrarian thinking. Not going for for whatever, the trendy things. There is also luck involved. I don't think we can pretend it's just skills.
Brian Bell (26:11.992) Yeah. Maybe you could repeat what your friend said. It's never what?
Emmanuel (26:17.579) It said the unicorns or decacons at a point in time were never the cool kids when they started.
Brian Bell (26:28.1) They were never what? The what? The cool kids. yeah. So they looked weird. They look too early or too late or, you know, impossible like renting your room out on Airbnb or like driving with strangers on Uber, you know, and name your favorite thing or like being the 17th search engine at Google or AWS with cloud or loud cloud before that, right? Yeah, that's really interesting.
Emmanuel (26:29.431) The cool kids.
Emmanuel (26:33.603) They looked weird. Yeah.
Emmanuel (26:55.151) Yes. So that's one point. The second one, too, is I think it's a blend of conviction but also process. If we try to think really far out what's going to be transformative, I don't think we can have 20 genius ideas and for each of these actually be helpful because we just don't have enough brain capacity. and time capacity. I think we also need in a way to pick a bit of our battles. So for example, I could tell you, I can totally see certain health care businesses being revolutionary in the next decade. I can also tell you I know zitch about health care. So I'm not even going to pretend that I'm going to try to go find the next weird project that is going to turn out to be a trillion dollar company. Then there is another part in timing. which is, I will call it influencing your luck. And I think Vinod Kosselaar had a comment on that, which is, at some point, you also need to accept that you are not alone in the world betting on some founders doing their stuff. There is also a market dynamics. markets may be wrong longer than you can afford to be right.
Emmanuel (28:33.167) And so at some point, think also recognizing that you may be right, but it may still be too early or in way it might be too late until the next phase of adoption of a better solution arrives. think we also need to accept that that's true. I am very stubborn. I had strong convictions in 2020, 2021 that directionally are moving in the right direction. But basically, I deployed capital in those four years too soon. So sure, I got in, but I had gotten in four years later, I would have entered at whatever 20 % markup. But they're risking significantly more. So yeah, timing is really... We'll call it art and science, think, right?
Brian Bell (29:32.81) Right. So tell us about the dark matter lab story, the funding gap you saw and what you built to close that funding gap.
Emmanuel (29:41.423) So. Really, what Dark Matter Lab is meant to do is act as a precede funding source before incorporation while researchers still sit inside labs. And bringing not just cash, but bringing compute, bringing legal expertise, bringing operating resources that will be relevant at that stage, which is technology there is key. Do you actually have a breakthrough? or do you have a feature or do you have a great people? The idea came from in part me sitting on the other side of the table with my technology breakthroughs and being unable to in part explain them in a way that VCs would actually get, that many clients would actually get and in part from me being a professor. still doing research, still spending a lot of time with masters, PhDs, post-docs, and realizing in many cases that they would be able to go one, two, three years faster on their research. And these are folks who are geared toward becoming entrepreneurs, right? They could save one, two, three years by effectively bringing, being willing, being able to bring these VC resources at the point where it's... typically not possible. So that's really the genesis of Dark Matter Lab. What made it possible at Berkeley to start with is luck and trust. It's the luck that...
Emmanuel (31:34.436) the university chancellor is a former professor of mine. People running venture services, people running entrepreneurship and innovation, people running industry alliances are former colleagues of mine, are former professors of mine. And so when I came to the table with that crazy idea, they were willing to listen and they were willing to try because we go back 10, 14, 15 years. And along the way, they've seen other things I've done. So that's luck and trust, right? I think there is also luck and trust in that Berkeley, and we both come from there, so we've seen it as students, Berkeley is very much of a startup campus. And it's a startup campus that's actually really willing to innovate and reinvent itself, not far from how startups themselves do it, right? So we are now at just... the onset of it, right? It formalized itself basically in May. But we now have that conduit where a researcher can get a million bucks, half of that capital, half of that compute legal operating resources to continue on their research. There is no IP encumbrance. We are not claiming a right to the IP. We're not restricting usage on that IP. They want to take it open source. They can take it open source. They want to do a patent. They can do a patent. The goal is to save time to market and your risk. Some examples mark me. So there is a research team who told me that it's going to take them three years to do what they actually need to do. Because it's going to take them three years to get a federal grant large enough that they can actually pay for the data in the compute that they need to pay for.
Emmanuel (33:34.966) I don't need three years. All of that can be brought up front. I'm like, dude, if you're saving three years on this, this is amazing. There is another researcher, I asked him, what's your biggest bottleneck? And he was like, I need training data that does not exist. So I need to create the training data set. And he was like, to create a training data set is half a million dollars, just to start with something acceptable. to start iterating. So it's very tangible technological hurdles that those folks are facing. Now, to be fair, I don't think those hurdles are new in the last few years. I think what is new is on one hand, federal funding for research has been collapsed. So what used to take time but was significant funding, now takes time and is no funding. The second thing that changes is the pace at which that research, that innovation is happening. And so here, I'll just relate from when I was in grad school versus now. When I was in grad school, a good lab, we would be producing maybe half a dozen papers a year. Now, a good lab is... two to four dozen papers a year. The size of the lab in terms of how many researchers has at most doubled. And so there is that pressure both in corporate and in academic labs to just iterate faster on research and to go bigger and to go bolder. And that of course requires more compute, more data, more hardware. I think the realization by the investment community and the realization by university ecosystems that startups are going to continue to drive a staggering amount of innovation and value creation has become front of mind for everyone. And so more people want to become founders, want to join startups and therefore want to go also undertake those more challenging projects. Therefore again,
Emmanuel (36:00.58) more resources needed.
Brian Bell (36:03.002) How does that differ from what University Tech Transfer Office or even DARPA or especially Accelerators already do at this stage? And you might even call this more like Angel Capital, right? Because this is, yeah.
Emmanuel (36:11.087) Yeah, that's a fair question.
Emmanuel (36:16.559) Yeah, in a way, it's more like Angel Capital. So accelerators will typically focus on the application layer because then they don't need to interface with tech transfer. It's very complex to interface with tech transfers. And then on top of that, we'll typically start engaging with students as they are about to graduate and really deploy resources when they graduate and are out of... the strict framework of university status. DARPA is one of the routes you can get federal funding. That route notably has concentrated on what we them frontier technologies. DARPA works for certain kinds of research, doesn't work for others. DARPA in terms of magnitude of resources could be significant amounts of dollars. but it tends to be distributed over time based on milestones that are effectively approved by your committee, hard to change, and actually are not necessarily milestones that are supportive of there is king technology for the sake of startup commercialization. And then tech transfer offices, the primary modus operandi has been through fellowships. or has been through corporate-sponsored research. Corporate-sponsored research comes with IP and income rights. The corporate sponsor has a right of first refusal on the IP, and that has its own complexities for a range of startups. The fellowship route tends to be much smaller in nature. It tends essentially to cover your grad school students' stipend with a little bit of buffer. To give you a sense, buffer is about 20 % of the stipend, and a stipend is 125k. If you are working in robotics, if you are working in AI efficiency, if you are working on large language models, on world models,
Emmanuel (38:40.58) that is not going to get you very far from a training, from a, you call it, know, vertical specialization standpoint, from the standpoint even paying for the hardware or creating your own robot prototypes. And so it's effectively the second gap of, you need to have bigger amounts. You need to interface tech transfer offices in a repeatable way. you need to make sure there is no IP encumbrance that's created. And you need to do that without waiting for folks to graduate. Because it's actually easier for them to do it within the vicinity of the lab versus after they are done with it.
Brian Bell (39:28.494) Very challenging, but you kind of you're kind of investing at the hardest juncture, I'd say, of a company's life cycle, which is, you know, pre formation, right? They haven't gone to market, right? You have no traction. You're kind of evaluating. OK. And maybe you can speak to how you evaluate, but the way I would look at a deep tech startup is, OK, if this works, how big is it? Is this the right team at the right time to do it? And is it, you know, protectable, patentable, or can they move faster, or is open source an advantage here? How do you, how are you, walk us through some of the ways that you're kind of, what's your weighted scorecard look like when you're kind of looking at these kind of companies?
Emmanuel (40:07.792) of high level, these are exactly the two main questions. Then there is a third one, which is if indeed, this would be a big opportunity, and if indeed we think that this is kind of the right time and kind of the right team, there is a corollary question, which is is that the right direction of research? Because it could be the right folks and the right, because when you talk about right time at that stage, right time is plus minus five years. I think that that's just being honest. Plus minus five years if you're on the right direction of research is probably minus five years. If you're on the wrong direction of research is plus five years. And so there you literally have a delta of a decade. And so that's actually the key starting point to me, which is I read a huge amount of research papers. I mean, just out of Berkeley since January, I read every single paper out of the AI labs, the blockchain labs, the finance labs, the robotics labs, and the hardware labs. it's probably, I don't know, 500 to 600 papers, something like that. And that's just Berkeley, right? But that's where I start because I'm like, okay, what are they researching? Do I actually understand? At the technical level, how they are setting up their experiment, how they are testing for what they need to test, what are... limitations to their approach that are data limitations versus modeling limitations versus simulation or testing limitations. Because that to me is step one. Is it a direction of research that I think is going in the right direction? If I think the answer is more likely yes than no, then I'm like okay so that right direction if it concludes positively.
Emmanuel (42:27.124) What problems can it solve and how much would we be making in revenue doing that? If my answer is, sounds big, then it's all centered around that research team. And it's trying to understand their ability to think out of the box. It's trying to understand, do they know what they don't know? because in research you're going to fail 90 % of the time. And that's you doing your job, failing 90 % of the time. And so you need to know what you don't know. So you don't have that illusion of it's a breakthrough when it's not. And then of course you need to make sure that they aren't just
Emmanuel (43:21.082) good researchers, they are good researchers with founder DNA, with entrepreneurial ambition. And so how do they think about their next steps? That's very key. And this type of discussion can take a really long time to really understand the way a brain thinks about the problems they are tackling and adapts to feedback, to new input. It could take... multiple months could take a year, a year and a half. But once you're there, you're like, okay, yes, that's the right team and it would be a big opportunity and it feels like a reasonable direction of research, then to me, there is a key question, which is actually a question back to them. What are your bottlenecks? And if I were to plug the resources that you need, what do you do with it? And if you get to a good answer, what's your next move?
Brian Bell (44:22.158) I'd love to spend the last five minutes doing a rapid fire. And we'll have to have you come back on the show at some point.
Emmanuel (44:27.024) Yeah, and I need to go get a plug-in battery. I need to plug in my charger. I'm so sorry. I already got a battery.
Brian Bell (44:32.237) Okay, yeah, go for it. Yeah, go for it. Yeah.
Brian Bell (45:01.38) Well, I'd love to do the rapid fire question segment. What's something you believe with total conviction five years ago that you'd argue against now?
Emmanuel (45:37.072) I think the main example there would be the true blend between decentralized finance and institutional finance. I really thought five years ago that we would have a perfect blend where we would solve on-chain identities and on-chain security and on-chain wallets. and on-chain transactions, on-chain products, we would solve everything in such a way that large payment institutions, large asset managers, insurance companies would be comfortable handling a significant percentage of their business in that completely disintermediated, anonymous environment, compliant, anonymous environment. I have, over the last five years, concluded that it's actually much more likely to be a co-existence with a sharing of technology versus a joint existence in the same environment.
Brian Bell (46:52.644) What's a popular idea in AI that you think is wrong?
Emmanuel (47:04.049) So I would say there are two to me that are wrong but popular today. One is the idea, and I think we have many startups like that, that really are wrappers around LLMs and agents of behemoths of the sector and thinking that they actually have moat. And in specific verticals where the data is really hard to obtain or the workflow requires a lot of expertise, I believe that to be true. But for most of them, I do not believe that there is lasting defensible mode. And then the second area that I think right now is popular but wrong is this recognition that we can tremendously innovate on AI hardware and that I agree with it. What I disagree on is the idea that there are founders with very little background relevance that would be capable of producing something of tangible value. To me in AI hardware, the right founders are people who have worked with AI hardware as product managers, as engineers, who understand distribution and absorb it versus dropouts from undergrads and masters.
Brian Bell (48:38.574) What's the best advice you ever got? You ever received?
Emmanuel (48:43.537) Stop being an asshole. And this was told to me by my mentor at BlackRock, Ryan Lafont. So Ryan, also a quant, at the time got assigned as a mentor to me by HR because I was regularly getting in trouble for being very straightforward in...
Brian Bell (48:47.162) That's good advice.
Emmanuel (49:12.421) what I thought was a waste of time, what I disagreed with and people didn't like. And he told me, look, stop being an asshole. And he was like, I'm not saying you're wrong in being an asshole. I'm saying there is no upside. So think of it as an arbitrage situation.
Brian Bell (49:34.171) There's no upside to get. Yeah, my wife says this a lot because, you know, I kind of have a temper, right? And I'll just get angry. You know, it's because of my background, you know, alcoholics yelling at each other growing up. But she's like, you know, anger does not help anything. It doesn't help the situation. Actually, my father in law told me that he goes, if if I thought it would help to get angry, I'd get angry. But oftentimes it doesn't help. And same with being an asshole. It doesn't really help anything.
Emmanuel (50:02.065) Yeah, I saw that after the facts. So going through startup and of course, even going through VC, I was like, look, you could strongly disagree with someone. You could strongly disagree with a situation. Find the most idiotic thing you've ever heard. There are constructive ways to go back at it. There are times where being right actually does not matter. Moving the situation forward for everybody is what matters. so it seems like the sharing, the advice back then was helpful point in time for addressing a behavioral issue. But I've kept whatever you could call it, recycling or maturing on it. Now in like, how do you manage conflict? How do you manage negotiations? How do you manage disagreement? it turned out to have been way more helpful than Ryan at the time probably anticipated.
Brian Bell (51:06.97) Well, I really enjoyed the conversation. But I definitely have to have you back on in the future. How can people get in touch with you? There might be founders listening that are doing research or fellow VCs like me that are looking at a team like that that want to get your opinion because you're reading 500 research papers every year or two, sounds like. How can folks get in touch?
Emmanuel (51:22.939) Yeah, well, happy to. Well, the easiest is my email, my hivemind email. That's the easiest since I'm not on Twitter. Now, X, that's you, that's me, that I say that. But that would be the easiest. The second one is, could always ping me on LinkedIn, but better than that, we have a dedicated dark matter page. which is darkmatter at hivemind.capital. And on that webpage, people can literally drop their research papers, their research plans, things like that. And of course, you can just ping me directly because now we have our contact team.
Brian Bell (52:08.61) Yeah. Well, thanks so much for coming on. enjoyed it.
Emmanuel (52:11.397) Brian, thank you so much. Great conversation.
