Tomasz Tunguz: How I Raised $230M; ChatGPT vs. Google; How LLMs Work; Trump vs DeSantis | E1004

1. 0:00 – 2:50

   Meet Tom Tunguz

   1. TTTomasz Tunguz0:00

      Any time we talk about machine learning, there's always this question around, like, "What is the moat?" I think the answer is the one that it's always been, which is better execution is the moat.

   2. HSHarry Stebbings0:08

      Tom, it is such a joy to have you on the show. I just checked, and it is, uh, 2016 when you were last on the show, so seven years ago. I've missed you dearly, my friend. But thank you so much for joining me today.

   3. TTTomasz Tunguz0:20

      Uh, thanks for having me back, Harry. I can't believe it's been seven years. Time flies. Look at you, huge audience, new font. I mean, look how far you've come. It's incredible.

   4. HSHarry Stebbings0:29

      I mean, that is so, so kind. But I wanna start with, uh, you, bluntly, which is obviously we recently founded Theory, such an exciting time. I wanna dive in and say, first, why did you decide to leave Redpoint and why did you decide to start on your own?

   5. TTTomasz Tunguz0:44

      Yeah, I had, I had a great time at Redpoint. I was there for 15 years, learned from many wonderful people. And, uh, after, after that amount of time, I decided that after seeing so many founders start companies, that I really wanted to start one of my own. When I was, when I was about 17, I started a little, little company. And, uh, over the last 15 years, maybe more, 20 years, I've watched all these startups grow and, uh, I really, I, I wanted to have that feeling for myself, and I also wanted to experiment a bit more and try a little bit... You know, everybody has an idea about how they wanna create their own business, and, uh, as you know, I've been a startup, uh, student of startups for a long time, and so I really wanted to, to build a venture firm in, in a slightly different way. Uh, and so in September of last year, jumped in and, uh, and we were off to the races.

   6. HSHarry Stebbings1:33

      Man, uh, there's nothing more special than having your own firm. I couldn't agree with you more. You mentioned that, like, the learnings in the 15 years. If there are one or two big takeaways for you from your time at Redpoint, I'm asking you to distill 15 years of lessons-

   7. TTTomasz Tunguz1:46

      (laughs) Yeah.

   8. HSHarry Stebbings1:47

      ... into, you know (laughs) , a short sound bite, but what would they be and how does that influence how you think about building Theory moving forward?

   9. TTTomasz Tunguz1:55

      Yeah, so I really believe in thesis-driven investing, and what that means is going deep into space and spending six, nine, 12 months researching it and really understanding it. Uh, as a board member, I will never know about, uh, as much about a space as a founder, but if, if I can deeply understand a space, then I think I can be a very helpful board member, and that's one of the reasons why Theory is called Theory. The other reason or the other sort of difference is I, I really believe in concentration. Like, uh, venture capital, the industry is governed by a power law, and the more dollars you can have closer to the Y axis, so to speak, on the power law, the better your returns will be. And so I wanted to set up a firm that was set up for thesis-driven concentration. That was the whole idea.

   10. HSHarry Stebbings2:41

       I mean, listen, portfolio construction is what gets me out of bed in the morning.

   11. TTTomasz Tunguz2:45

       (laughs)

   12. HSHarry Stebbings2:46

       So when you were speaking my language, um, I wanna start there. This is like a process

2. 2:50 – 6:51

   Closing $230M Fund: Inside Look

   1. HSHarry Stebbings2:50

      that's shrouded in much opacity and we just see fundraises announced, so I wanna talk about the fundraise. How many meetings did it take to close out the fund, my friend?

   2. TTTomasz Tunguz3:00

      Uh, so it took, uh, a bit more than you. Uh, it took about 150, uh, LP meetings. So the, the fundraising market was a very challenging one over the last couple of months, I'd say, but it took about 150, meeting about 150 LPs, and the way I thought about it was, I mean, you know me, I thought about it just like a regular software sales process, right? Where sales-assisted, 15% close rate, and so built a, built a, uh, funnel and a pipeline that was large enough with 15% close probability that we could hit our target, and we were very lucky where we, we exceeded the target. We raised the hard cap and, uh, ended up at two, about 230.

   3. HSHarry Stebbings3:38

      Okay, so 150 meetings. I'm fascinated. How many of those did you know before the raise itself?

   4. TTTomasz Tunguz3:44

      I probably knew about 40 of them, 40 to 50, ahead of time.

   5. HSHarry Stebbings3:49

      That's super.

   6. TTTomasz Tunguz3:50

      Yeah.

   7. HSHarry Stebbings3:50

      That's, so how many of them committed having not known you before? 'Cause I always say invest in lines, not dots, and the importance of building that relationship outside of the fundraise. How many committed having not known you?

   8. TTTomasz Tunguz4:03

      Uh, so about 50% of the capital was, w- were from new relationships, about half of the capital.

   9. HSHarry Stebbings4:09

      That's pretty incredible 'cause that's quite rare. So, I mean, well done to you there. Can I ask, uh, for the starting checks, they're the hardest to get. When you think about, like, the strategy there, did you go for large institutional anchors first or did you go for the friendlies who were much more likely to say yes?

   10. TTTomasz Tunguz4:27

       I went for the large institutional anchors. Uh, I, I had some relationships there and, uh, you know, I figured if I could get... So the LPAC has five members, and if I could fill two members of the LPAC right out of the gate, then that would assuage a lot of concern from newer LPs because they were institutional backers, uh, from the beginning, and, uh, they were wonderful. Uh, they took many reference calls on my behalf and gave me a lot of advice through the process. It's kind of like, uh, if you think about finding a lead for a series A or a series B. If you can find that lead who then, um, does the diligence and, and then will talk to everybody else who comes in and sort of guides you, uh, it set, it sets up the process really well. It's a little bit different because LP f- (laughs) ... Most, most venture firms are basically large party rounds. It's just the number of investors, you're talking about 15 to 25, 30 sometimes, and so it doesn't really have that dynamic of a lead, but it's clo- Like, the LPAC is basically, the Limited Partner Advisory Council, is like the board, is the closest thing that, that I think you could get to, unless you have a super concentrated LP base.

   11. HSHarry Stebbings5:36

       Did you have a limit on check size? Often, when I was raising, people were like, "Oh, don't let them invest more than 20% of the fund." Did you have a limit on how much they could invest as a percent of the fund?

   12. TTTomasz Tunguz5:47

       I did, yeah. The, I think the largest LP is no more than 12% of the fund, and so there's, there's different, there's different theories here. So one of the wonderful things in, in raising this fund was the number and the s- the...I got to see startup land in all its beauty and glory. The number of people who reached out, who, you know, I wouldn't have thought to help, uh, and talk about their journey and, and their experience and their fund construction was amazing. And so I learned about funds where, you know, a hundred million or $200 million dollar fund with, like, 80% of the capital from two LPs was super concentrated. And then there are other funds that are the opposite, where it's just lots of small checks. And, and then you have people in the middle or one significant anchor. And having that breadth was eyeopen- You know, it was just absolutely eyeopening to understand that there are many different ways of creating a venture firm, uh, at least on the capital formation side.

   13. HSHarry Stebbings6:40

       Totally many different, many different ways. I'm glad you went with the 12%. I think diversification-

   14. TTTomasz Tunguz6:46

       (laughs)

   15. HSHarry Stebbings6:46

       ... is important. (laughs) So I'm thrilled that you went for the latter. Can

3. 6:51 – 8:57

   Secrets of a Successful Pitch Deck

   1. HSHarry Stebbings6:51

      I ask, on the materials side, what did you go for on the materials side? Like, did you send pitch decks to everyone beforehand? Did you have data rooms? How did you think about getting the right materials in place for the raise?

   2. TTTomasz Tunguz7:03

      I did. So I prepared a data room. There was a track record in there, a pitch deck, a bio, some of the blog posts that I had written, some metrics that I, I put together. And what I would do... So I, I set up a, like, a, a brief and a bio. And that was sort of, like, theory one. Um, that was the outbound email. So if, if I had been introduced, I would send them my bio and then the deck. And the idea was to use those materials as pre-qualification. So some LPs prefer not to invest in solo GPs. Different LPs have different mandates. They might only invest in the US. They might prefer early stage, later stage. And the idea was to qualify just like an SDR would. And so those briefing materials, that's, that was the entire purpose. And I, I used a DocSend, and I didn't allow downloading because I wanted to understand where people were in the pitch deck, where they were stopping. And that informed the, the way that I would pitch them if they decided that they wanted to meet.

   3. HSHarry Stebbings8:03

      Tom, what was the biggest reason that people said no? You mentioned the solo GP element there. What was the biggest reason people were like, "Ah, not for us"?

   4. TTTomasz Tunguz8:10

      Yeah. So solo GP is one. There's key person risk, and so some LPs are just not comfortable having a single person, um, be the general partner. The other challenge was timing. So, you know, I was raising during a time when the public markets had been down, but the private market valuations had remained elevated. And so the combination of those two put a lot of LPs in a place where they didn't really understand the nature of their portfolios. In other words, if you thought you were 50/50 public-private, and then the publics fell by half, all of a sudden you were three quarters private, one quarter public. But the VCs and the private market was going to be written down, but hadn't been written down. And so one of the biggest reasons was, "I just need more time in order to understand where my portfolio is so that I can figure out allocation in the future to, to this asset class

4. 8:57 – 14:23

   Mastering the Art of Closing Deals

   1. TTTomasz Tunguz8:57

      in light of that."

   2. HSHarry Stebbings8:57

      The hard thing is, um, first close, second close, final close. How did you approach the closing mechanisms?

   3. TTTomasz Tunguz9:05

      There's this sort of aura around that first and only close, and no one really explained it to me. I, I talked to one friend who is an executive at a publicly traded company, raised a venture fund, and he said he had 15 closes. And so every time an LP would commit, he would have a close. And his perspective was, "Well, if somebody signed up, it doesn't cost me any more to, to close them and have them wired, and we'll just keep going." And then there are, you know, other investors who said, "Ideally, having a single close means you raise from a position of strength." And we had a single close. The purpose of a close date is just to drive people in unison to a cadence that you're trying to set. There's nothing magical about it. There's nothing terrible about having multiple closes. It's just a way of organizing a particular process. And that close date is drawn out of thin air, and it's driven by how strong of an option you can develop and what your LP relationships look like. Maybe I put it... It's a vanity metric for VCs in the end.

   4. HSHarry Stebbings10:05

      I totally agree, and I think one of the biggest mistake managers make actually is they, they get a load of people who say yes, they don't close them, and they kind of leave them hanging. And then a month later, they come back and say, "Hey, Tom, you committed to my fund." And you're like, "Oh, I forgot about that. I actually kind of allocated the money elsewhere, and it was actually two months ago now." And, like, that happens a lot, which is, like, don't let it go stale. I always say, like, close as soon as you can, as fast as you can, to not let it go stale. So totally with you there.

   5. TTTomasz Tunguz10:35

      And it shows momentum, right? If somebody asks, like, "How, how much have you closed?" You wanna show a, a level of progress, and, and it gives you a reason to come back to people, or like you said before, a reason to email people. So I think this mystique around a single close is, is completely misplaced.

   6. HSHarry Stebbings10:49

      Did you do anything to drive urgency in the LP base? 'Cause as you said, the single close is one way to do it in terms of ensuring that people move faster down pipe. Did anything else work for you in terms of just ensuring there was efficiency and urgency in the process?

   7. TTTomasz Tunguz11:04

      The thing I did was that every time I had another verbal commit, I emailed the LP base, and I, I gave them the update. And the idea was just to, to set a cadence of... I, okay, I wrote this blog post a long time ago talking about when you're fundraising what you want to convince people of is inevitability, that the, that the company or the fundraising round, its conclusion, its positive conclusion, is inevitable. And so any data point that you can provide to investors that supports that is hugely helpful.

   8. HSHarry Stebbings11:38

      When you say, "Hey, this person just committed," or, "We just got another great institution," for me, as someone who's already committed, I'm like, "Oh, actually, Tom's gonna be a winner. I should introduce him to more people." Do you see what I mean?

   9. TTTomasz Tunguz11:51

      Totally. Absolutely. Exactly.

   10. HSHarry Stebbings11:53

       So I should-

   11. TTTomasz Tunguz11:53

       So that, that was what I did. I was very deliberate about that.

   12. HSHarry Stebbings11:56

       Can I ask, on review, what would you say worked with your raise that you'd do again? And what would you say did not work and you would change for next time?

   13. TTTomasz Tunguz12:06

       Yeah, I think, so after I raised capital, I went and I talked to, I should have done this before, but I went (laughs) and I talked to some of the most sophisticated capital formations people across different disciplines, and I just asked them, uh, how they do their job. It was really interesting, uh, because like, okay, look, we start here today and at some point, like we wanna get better and better, right, we wanna improve every day, and so I wanted to understand what is the best in class? And the really sophisticated fundraisers, they're always in market, they're referencing LPs, they are building pipeline, that's a full-time job. And so I think one of the things that's really important is building long-term relationships. That worked really well for me, so I was really happy to have a lot of long-term relationships that I could lean on. That was a really big deal. Things that didn't work so well, I think there are different geographies where LPs as a whole are more conservative, and it took me a while to appreciate that. And so I probably spent more time traveling than I should have. Next time, there'll just be longer lead times on, on some of those geographies.

   14. HSHarry Stebbings13:09

       Did you find in-person worked much more effectively than remote calls in terms of conversion and closing, or actually no?

   15. TTTomasz Tunguz13:16

       No. (laughs) No. No, there's no correlation. Ah, boy, I'd have to go back and look, but maybe, I wanna say like a quarter to a third of the LPs I only met after they committed in person, and, uh, that's probably an overhang from COVID where a lot of funds were raised entirely virtually and people are comfortable.

   16. HSHarry Stebbings13:36

       Yeah. (laughs)

   17. TTTomasz Tunguz13:37

       Yeah. (laughs) There you go.

   18. HSHarry Stebbings13:38

       Not entirely with me. I, I literally had an AGM recently and I was like, "Oh, it's so lovely to meet you." And they're like, "Great, you, you exist. You're not an AI, how wonderful." So I tot- I totally agree and get you there. I actually, the thing that I do is every week I meet two new LPs, and with each LP meeting I say, "Hey Tom, I loved this discussion. Are there one to two other great LPs that think like you and you think I'd have a great discussion with?" And they go, "Oh, you've gotta speak to Satish and Logan." "Oh, that'd be fantastic. Would you mind making the intro? I'll send you a blurb now that you can forward." "Oh, of course. Super happy to." And the flywheel is self-fulfilling.

   19. TTTomasz Tunguz14:15

       You're a machine, Hari. (laughs)

   20. HSHarry Stebbings14:18

       No, I'm just terrified about raising, and so I always make sure of (laughs) the relationships

5. 14:23 – 15:36

   Fundraising Advice for Managers

   1. HSHarry Stebbings14:23

      are there. Can I ask if, okay, so you now have this breadth of experience raising Theory in a pretty fricking hard market. Um, what would you advise managers going out to raise having had the experience you have done with Theory?

   2. TTTomasz Tunguz14:37

      The point of LP diligence that was new for me during the process was the business model. I think in the last 11, 12 years where we've had this incredible bull market, the portfolio construction hasn't mattered as much. The volume of questions that I received consistently from LPs about what are your assumptions on number of seeds, number of A's, the fatality rate, the multiples on those, uh, how does that compare to sort of the standard venture capital distribu-... Like, the number of questions that I got about that I think suggests that it's really important to have a business model in, in your deck. That, that, that I think is f- has changed just as a result of the cost of capital increase.

   3. HSHarry Stebbings15:14

      I think it's sad that it wasn't always there given the fact that there's always

   4. NANarrator15:18

      Yeah.

   5. TTTomasz Tunguz15:18

      (laughs) No, it, it's a reflection on the part of our venture capital ecosystem. I mean, in 2008, re- whenever we met, I mean, at the series A, we were looking at financial plans and we were putting together financing rounds that were a function of the capitaling to the business. And then, you know, whatever, in 2012, all of a sudden it was not about fundamentals anymore and it was

6. 15:36 – 23:24

   Building a Winning Investment Portfolio

   1. TTTomasz Tunguz15:36

      really just about access.

   2. HSHarry Stebbings15:37

      I love that discussion, and so I wanna just dive into it 'cause 230 million, okay, that's the fund size. Um, why did you decide 230 million was the right amount to raise?

   3. TTTomasz Tunguz15:47

      Yeah, it's... So I, a couple of different... Well, it was all about portfolio construction. I ran a lot of math in order to figure out, uh, using historical venture data. I ran Monte Carlo simulations for optimal portfolio construction for a solo GP and-

   4. HSHarry Stebbings16:03

      Uh, uh-

   5. TTTomasz Tunguz16:04

      This i- this is-

   6. HSHarry Stebbings16:04

      What does that look like then? How, help me understand. How many companies, like how much for initial, how much for reserve?

   7. TTTomasz Tunguz16:12

      Yeah, so it's about, uh, it's about 12 to 15 portfolio companies. You have significant concentrations, so you probably have like 40 to 50% of the fund in the top three holdings, uh, maybe more. And so that's a, it's an unusual portfolio construction. The analysis that I, there's basically, you know, there's a couple of different... The Monte Carlo simulation spits out a couple of different dominant strategies, and, uh, this is one of them. And this is the one that aligned with when we talked at the beginning about the way that I like to invest about being thesis driven and really understanding the space. If you go deep into a space and you can understand it, ideally you're in a place where you have a lot of conviction and you can keep investing and keep supporting a company. And I also think in, in a venture environment that's going to be significantly different this 10 years over the last 10 years, setting up a venture firm to be able to consistently invest behind its companies should provide founders a bit more comfort from their financial partner.

   8. HSHarry Stebbings17:07

      So I, I similarly did the math on Monte Carlos, and I found that at 23 companies you get something like 82 to 84% of the benefits of diversification. And so what you're saying is essentially with deep theses and deep thinking and a lot of time, you need less diversification because your ability to pick is significantly better. Correct?

   9. TTTomasz Tunguz17:28

      That's exactly right. That's the-

   10. HSHarry Stebbings17:30

       Totally, yeah. So, so we're doing series As I take it. Like, what's the check size per company estimate?

   11. TTTomasz Tunguz17:38

       Yeah, it's about eight to 12 initially.

   12. HSHarry Stebbings17:41

       Okay. Eight to 12 initially. Blunt question, is the fund big enough given how large and like AI machine learning rounds are today being 30 to 50 million on a pre-seed or seed as we're seeing quite often now?

   13. TTTomasz Tunguz18:00

       It, so it, we can flex. Uh, I don't, we're not in a position to be able to lead a $50 million series A.

   14. HSHarry Stebbings18:06

       Okay.

   15. TTTomasz Tunguz18:06

       Uh, we could co-lead.Uh, so that's one way of doing it. And if you're raising a $50 million Series A, you probably do want it from, I would say you probably do want it from two different VCs. But, uh, we've had a lot of flexibility, and that was by design. So, uh, because we're so concentrated, we can focus our resources where, where we have the most conviction.

   16. HSHarry Stebbings18:25

       Can I ask, how do you think about the decision on doubling down? You said there about kind of three companies could be, say, 50% of the capital base. What does that conviction building process look like to putting that much capital behind one of the three?

   17. TTTomasz Tunguz18:41

       Yeah, I think it's- it's a lot of diligence. I mean, it's, we'll spend six, nine, 12 months researching a space. Like, one of the themes that we have is the decade of data. So I've been investing in lots of different data companies for a long time, have a pretty strong network there. So one component is just really understanding the market, understanding the buyer base, the different segments, their- their needs. Another component is benchmarking companies, so I've been doing that for more than 10 years. I've got a- a pretty significant database of- of, uh, data points there. So just understanding on a relative basis what is the ultimate performance. A third part is understanding what the exit markets look like and the entry prices and what is a reasonable multiple expectation over what period of time. 3s There's ...

   18. HSHarry Stebbings19:18

       With the exit market analysis, I- I go back and forth on, is it worth doing because it's so variable? You could look back on the prior 24 months and say, "Well, it could be that, or it could be today, or it could be way, way worse." We can't project out seven, 10, 12 years. Is it valid doing it?

   19. TTTomasz Tunguz19:37

       I think it is because, okay, so the historical forward multiple is about 5X, let's just say, right? It's a little higher than that, it's about 5.5. And in the heyday of quantitative easing, the top quartile companies are trading at 40 times. And so you can't go... And- and today, it's about maybe six. And so you can't go into a company today and say, "Okay, I'm- I'm going to project a 20 times forward multiple on this company at the time of IPO if it's at $100 million growing at 70%." You just can't. That's irr- I mean, I would say that's irresponsible because it's just completely unrealistic. And if you've, you know, if you've spent the majority of your time in venture during a- a time when you've had those kinds of multiples, you need a s- check on what do you think your return expectations are going to be given that, you know, whatever it is, a 4% ESOP, uh, employee stock option pool dilution by year, and the dilution of created by other venture rounds and that. And so going through that discipline, I think, is as much just like a se- uh, particularly if you're working in a team or... It's just a really important discipline step. The- the other thing that- that I really love to do is, um, there's this awesome book called Superforecasters that a guy named Ted Lowe wrote. And he talked about Enrico Fermi, who created the atomic bomb and was one of the team for the Manhattan Project. And Fermi had this way of thinking which was all about conditional probabilities. It's called fermization. The idea is like, just enumerate the conditional probabilities. In order for this comp- in order for this generative AI company to succeed, the first thing he needs to do is hire a PhD team. Okay, what are the odds that they can do that? Then they need to raise a Series A. Okay, what is the- the base rate for hiring, for raising a Series A from seed is about 60%. Then they need to raise a Series B, base rate is 50%. Then they need to do this and this and this. And you put it all together, and then you tie it to your expected value, and you kinda come out with a big range of what you think the ultimate outcome can be. And each company, it's going to be different. A marketplace has to do supplier acquisition, uh, supply side acquisition and demand side acquisition. And so like that framework, at least for me, really helps me to think about what are the two or three key issues or questions that a company needs to answer over its lifetime? How do those odds change? And ideally, they improve, and the more that they improve, the more comfortable one ought to be in concentrating.

   20. HSHarry Stebbings21:48

       Can I ask you a weird question, which is like, I think-

   21. TTTomasz Tunguz21:50

       Yeah.

   22. HSHarry Stebbings21:50

       ... ƒLafont or one of the Lafonts says that if you know a market better than anyone else, you can pay a higher price than anyone else because you know more about it than anyone else (laughs) . My question is to you, do you agree with that statement? And how do you think about your own price sensitivity?

   23. TTTomasz Tunguz22:09

       I think it's true. I think it's true because, okay, so if you know more about a market, the range of expected outcomes is- is far more narrow, which means your certainty in making a bet is better. And so the result of that should be you- you should be will- be willing to pay a higher price, right? Because the, okay, so the, let's think about it a different way. The more you know that an option is in the money, the more valuable it is, or the lower the variance on an option, actually, that's a function of time. I'd have to go back through the option math. It's been a long time. But I think conceptually, I agree. Up to a point.

   24. HSHarry Stebbings22:44

       Yeah.

   25. TTTomasz Tunguz22:45

       Up to a point.

   26. HSHarry Stebbings22:46

       How do you think about your own price sensitivity and like ownership-wise, how- do you need 10%? Do you need 15%? How do you think about ownership sensitivity on a per company basis?

   27. TTTomasz Tunguz22:56

       Yeah, so I think the- the way I'd put it is it's important for us to have meaningful ownership because we're so concentrated. And, um, the idea is because we have such a small portfolio, we can spend a significant amount of time with each portfolio company. So ownership matters. Ownership also matters, I think, in the- the form of returns. Uh, so it's- we want to have significant ownership. And the idea with the firm is that you don't need to have significant ownership out of the gate, but you can build a position over time.

7. 23:24 – 26:50

   Lessons from Snowflake on Reserve Management

   1. TTTomasz Tunguz23:24

   2. HSHarry Stebbings23:24

      I- I totally agree. A- and in terms of like multi-round investing, how do you think about it? Like, can you do a 5% ownership on A and then get five more at the B, five more at the C? Like, how do you think about that cross-cycle investing is a bit ...

   3. TTTomasz Tunguz23:37

      Yeah. So that's a tough configuration just because the dollar amounts that you're talking about just go up so significantly, right? 5% of the A, that's hot A, so you're probably talking, you know, whatever it is, five at 100. And then the next round might be 200 or 300, and so to buy another 5%. You can kinda do the math doing ...

   4. HSHarry Stebbings23:54

      Mm-hmm.

   5. TTTomasz Tunguz23:54

      ... pro rata- pro rata. So, you know, I think about it a bit more as like, you know, can you get 10% at the seed? Maybe you can buy another 10 to 15% at the A and then buy another, say, 5% at the B, something that looks like that.

   6. HSHarry Stebbings24:07

      Can I ask, with the concentration on a per company basis-To have such concentration, you also have to not do pro rata or not concentrate capital in a lot of companies too, 'cause you have to preserve dollars for the best. How do you think about that aspect of bluntly being a little bit more disciplined around reserve dollars and not allocating to anything in the middle or underperforming?

   7. TTTomasz Tunguz24:32

      Yeah, so the business model of the firm affords both. So there's reserves for every company. Yeah. And the idea is with every business, you either, uh, you know, there's a very sort of blunt instrument, which is with every stock position that you have, if you're any kind of investor, you either, you should either be a buyer or a seller of that position, right? And if you're in the middle, you probably don't know enough about a business or... And so the, the idea behind like concentrating in reserves is we will run diligence processes on those existing portfolio companies in order to understand where to concentrate. Um, but we also have the capital to be able to support, like, companies go up and down. I mean, one of the stories that, um, I think hasn't been told enough is like the Snowflake series C and the series D almost didn't happen, because the company was burning so much, the gross margins were in a really rough place, and, uh, there was a flat round in there somewhere. And, you know, then it became the fastest growing software company in history. So again, like one of the reasons for this portfolio construction is if you're in a radically different capital markets environment, you want a financial partner who has the wherewithal to be able to support you across multiple rounds. And that Snowflake round, I mean, the, the multiple on that finance, that was a big lesson, uh, that I learned at Redpoint. The multiple on that financing is legendary.

   8. HSHarry Stebbings25:46

      So what, talk- talk to me about that lesson. I wanna dive deeper on that. What, what happened and what was the lesson for you?

   9. TTTomasz Tunguz25:52

      Snowflake at the time was, I mean, competing with giants, right? So there was Redshift and there was GCP, and the, the company was growing very quickly and the market was there. The company had a really tough time, and I can't remember exactly if it was a series B or the series C, but it was this middle round, maybe it was a series, maybe it was a series C, and the company was burning a ton of capital and couldn't raise money from the outside, and it was the insiders that stepped up and led that round, because they believed in the business. And so the ultimate result was if you have an accurate thesis and you can find the right company and you, you have the wherewithal to be able to support that business through good times and bad, you can be disproportionately rewarded for it.

   10. HSHarry Stebbings26:36

       I love that, and I, I agree with it. I also didn't know actually that in, in terms of the series C. I, I do have two just questions on like thesis-driven investing, which is, I always worry about confirmation bias, which is, you know, you develop the thesis

8. 26:50 – 29:20

   Avoiding Confirmation Bias in Investing

   1. HSHarry Stebbings26:50

      and then you find something that aligns to it and you're like, "This is it." And actually, theses can be wrong. How do you think about the dangers of confirmation bias and not just falling victim to your own predictions of the future model of the world?

   2. TTTomasz Tunguz27:03

      Yeah, totally. I mean, you could become enamored with a, a particular view of the world. And I think in order to mitigate confirmation bias, you need to have many conversations. You just need to keep testing and keep pushing and... And at the end of the day, the great part about investing in B2B software is there's a buyer that is buying a product and either they buy the software or they don't. And so (laughs) the greatest sort of foil to confirmation bias is a lack of customer demand. You know, if you're so enamored, like I have this view around the future of the marketing ecosystem being tied to the blockchain and this decentralized infrastructure, and I've been working on it for nine months, and the thing that I consistently look for is, okay, where's the pipeline? Who's the buyer? Who's willing to spend? Where are the experimental dollars? What are the advertising agencies saying? And so I can come up with that idea consistently, but if I can't find a buyer for it, I can have, I can still have this beautiful vision, you know, glass pyramid, so to speak. But if I can't find a buyer, then I need to abandon the thesis or at least set it aside for now.

   3. HSHarry Stebbings28:01

      Okay. So, uh, you get multiple different conversations from multiple different kind of parties within the ecosystem. The other question, and it kind of aligns to what you just said there about kind of blockchain applied to marketing, which is timing. A lot of these can be right, but can just be too early as you know (laughs) and as I know. Um, how do you think about bluntly market timing, especially on theses where you can know kind of too much ahead of market?

   4. TTTomasz Tunguz28:28

      You, you have to look for pipeline. Um, so it, it all comes down to customer need, right? So I mean, well, pets.com versus Instacart or Peapod versus Instacart, I think is kind of the canonical example. The market timing, I think as long as you have a strong pipeline, you can have a lot of confidence, as long as you can extrapolate the needs of one buyer to another. And so that's why spending time and trying to get as broad of a, an understanding of as broad a cross-section of the customer buyer population is absolutely essential in developing these theses, because somebody has to buy it at the end of the day.

   5. HSHarry Stebbings29:01

      No, I totally agree with you and I, I, I like the, the simplicity there. I do wanna move to the future and discuss kind of really what, A, where you're investing, but also some of the broader topics around it. Uh, a question that I have which I can't really find an answer to, but it's like when we think about the future, especially in terms of AI models,

9. 29:20 – 35:28

   Approaching AI Investing: Tips & Tricks

   1. HSHarry Stebbings29:20

      like does the rise of large AI models mean the future of AI is like as an ecosystem is dominated by a single general model or one or two single general models? Or will we have a decentralized kind of fragmented ecosystem?

   2. TTTomasz Tunguz29:35

      I think you have both. I think the analogy of Apple and Linux is really useful here, or Apple and Windows, where you'll have a, one system that is basically fully integrated and closed, and then you'll have another world where people are, are building little, I mean, open source models and they're... You know, some people believe that there's going to be a single dominant model. I'm of the mind that there's probably an interface that, I mean, if you look at Microsoft's Jarvis or you look at LangChain or Fixy or any of these companies where they take an input and then they're basically a mediator across a bunch of different models for different purposes, I think that's probably going to be the dominant model, at least in the consumer world. And then in the enterprise world-You have... You, you'll have, like, the Stripe, Twilios who are creating platforms where it's very, very simple for developers to get started with, uh, large language models. And then you'll have, like, full enterprise services firms where a big Fortune 500 just wants the problem solved. "I just... I... Pepsi needs a generative model." For whatever reason, they don't have the discipline, they want the whole thing in a box. And so you have this really nice spectrum, uh, of it. I think at the foundational model layer, that's a big boys game or big girls game just because of the capital intensity required both for training and the GPU access and all those kinds of things. So, maybe there's a startup or two that's able to raise a couple billion dollars in order to compete, um, but I think it's... You know, I think it's more at the application layer. I ran this analysis. So in Web2.0, if you take the top three clouds and you look at their market cap, so AWS, GCP, and, um, uh, Azure, it's about a $2.1 trillion market cap just for the cloud businesses. And then if you take the top 100 publicly traded cloud companies, both on B2C and B2B sides, so Netflix and ServiceNow, they have equivalent market cap of about 2.1 trillion for both. So, one's at the infrastructure layer, one's at the application layer. Market cap is basically equivalent. The difference is the infrastructure layer, there are three businesses, and at the application layer, there are 100. And so if the analogy holds, then you really... You know, as an investor, it's, it's... The odds of success are gonna be significantly higher at the application layer because the diversity of needs there is, is greater.

   3. HSHarry Stebbings31:44

      I, I, I love that. I, I also would love to, (laughs) to see that. That's fascinating. Uh, can, can I ask, in terms of, um... You mentioned kind of enterprise usage there. The thing I can't get my head around is like, bluntly, some of the biggest companies in the world will not allow the majority of their data to be put through a different solution stored in some cloud infrastructure they've got no idea about. Like, this is some of the most sensitive data they have. If they won't wanna run any form of queries or models on it, it will need to be on prem in their HQ under lock and key. How do we think about, like, enterprise access when data access is so core to their needs?

   4. TTTomasz Tunguz32:25

      Yeah, it's a great question. So there... Okay, so Salesforce was... Well, Siebel, the, the first generation of software, all the software was run on an enterprise's machines. And Salesforce said, "Let's move it to the cloud." And we convinced, as an ecosystem, everyone that the cloud was safe. And the cloud is also expensive, we're, we're starting to realize, and so now there's a bifurcation where data remains in the customer's account, but the application is being run by, by the software company. So you have the separation of the application from the da- the application plane from the control, from the data plane. And I think, you know, we will see sort of... I think we'll see a very similar architecture where the model actually goes to the data and then comes back out with the result, so the data is actually within the customer's account. There's some compute that's then put next to the data, the model is executed, and then it goes away. Um, and that way, whoever's managing the model-

   5. HSHarry Stebbings33:14

      Mm-hmm.

   6. TTTomasz Tunguz33:14

      ... can update the model, modify it, do whatever they need to, and then at the time the model is needed, it's then deployed and pulled back. So I, I think that's probably a dominant architecture. I think if you're in, uh, like finance or healthcare, you'll probably be completely on prem for the foreseeable future. There are other kinds... I mean, there are other kinds of issues, like if Copilot produces a bunch of code and you're a Global 2000 and that code is actually copyright- copyrighted by somebody else, what do you do? Right? Like, um, if a model produces a bunch of PII that's, like, quasi-related to somebody else. So there, there's, um... I put together this presentation on the opportunities for AI startups, and one of them is enter- is this whole bucket of enterprise readiness, like SOC 2 compliance, uh, legal shielding, um, uh, data security. There are all these kind of deployment models. There are all these kinds of challenges and issues that are associated with them, and there's a big business there, a really, really, really, or many multiple, many big businesses to be built there.

   7. HSHarry Stebbings34:06

      Can I ask a question? Do you, like... Do you think this is a bundled environment or not? I always think about Jim Bartsdale, unbundling or unbundling. Like, as you said that, there's many different big businesses to be built there, but they could also be bundled into an enterprise software suite. Do you think it's a bundled or an unbundled world in that envisioning?

   8. TTTomasz Tunguz34:24

      So, my learning has been that in early markets people want bundling. They want bundling because they don't yet understand the tech- The technology is moving so fast that most people don't really understand it end-to-end, but they want the technology to solve a problem. And so they'll get... If, if you're a Global 2000 and you want a generative model, you're not yet in the place where you can... Most people aren't. Say that these are the five different layers, these are the best of breed across the five different layers, and these are the, the sort of, the parameters upon which, like, I'm going to choose best of breed. So at the embeddings layer, the two most important things are, I don't know, X and Y, right? And at the model serving layer, the latency versus cost. Most people aren't there yet in their level of sophistication because they don't have enough experience with it. So my sense is in the beginning, people want an end-to-end solution. Just give me a thing that works, that's simple, and then as I learn what my needs are and what my customer needs are and what I need the software to do, I will break it in a particular way, then I will go and look for a best of breed in the market, and I will swap out that layer.

   9. HSHarry Stebbings35:21

      I think that's fascinating and, and, like, I'm learning here, so this is great. (laughs) Um, ƒ4. I get, I get, I get paid to

10. 35:28 – 39:13

    Are enterprise buyers ready for AI?

    1. HSHarry Stebbings35:28

       learn, so this is awesome.

    2. TTTomasz Tunguz35:29

       Yeah.

    3. HSHarry Stebbings35:29

       My, my question to you is, like, I'm worried about this asymmetry of knowledge, because, like, we mentioned kind of enterprise buyers there and then the providers. Tom, I'm European. I, I know how some of these large enterprises think, especially in Europe. Like, AI, it kind of, like, you need to remind them it's artificial intelligence. LLMs is you're gone, I... You, you've lost me already. Like, how ready do you think enterprise buyers actually are, and do you think the hype cycle is ahead of the enterprise propensity to buy?

    4. TTTomasz Tunguz36:02

       So, I think this is a technology that most buyers won't need to understand how it works. It's like a database. How does Snowflake work? I bet most people who buy Snowflake don't know. There's this awes-... I don't know if you ever started seeing the story, but there's this, there's this Italian artist, and he, he was exploring this idea, it's called the, um-... the illusion of explanatory depth. So, he found a hundred people in Milan, and he asked them to draw a bicycle, and then he 3D printed all those bicycles. And out of the hundred bicycles, how many do you think worked, Harry?

    5. HSHarry Stebbings36:36

       Ten?

    6. TTTomasz Tunguz36:37

       Two.

    7. HSHarry Stebbings36:38

       (laughs)

    8. TTTomasz Tunguz36:39

       (laughs) So-

    9. HSHarry Stebbings36:40

       What can we lessen from that? (laughs)

    10. TTTomasz Tunguz36:41

        (laughs)

    11. NANarrator36:43

        Yeah.

    12. TTTomasz Tunguz36:43

        So just because we're very familiar with a technology or, uh, an innovation doesn't really mean, doesn't necessarily mean that, uh, we understand how it works. And so, I think, I think in the case for most enterprise buyers, like I said before, I think they want an end-to-end solution that will just work and will work in 85 to 90% of the time, and that, that'll be good enough. And if in those 80 to 95% of the time, it can save you half, half your time just as Copilot does, then that's good enough. And as long as it checks all the boxes for my security team, my IT team, and my compliance team, then that's good enough.

    13. HSHarry Stebbings37:16

        Can I a- ... You me- you mentioned that Copilot, and you've mentioned it quite a few times. Um, today, G- I think it's like, you know, uh, code generation, 40% of new code generation is artificially intelligent code generation. What do you think that will be in 10 years time?

    14. TTTomasz Tunguz37:31

        I think it'll probably be 70 to 80%. And the, the reason I say that, I bet that 40%, a lot of it is what's called boilerplate code. It's just like standard. A lot of it is standard code or, um, code that's been slightly modified, right? Like, I'm creating an HTML page, I need the HTML them- the header, and the header and then the title. And, and so that's probably 40% of the content of an HTML page. It's probably the same for a Ruby file or a Python environment. And, and so we're prob- we're at 40% today, and I bet we're at 75%, 75 to 80%, because most of the code that's written is slight modifications of existing code. One, Pepsi's website is probably not that much, not that different to Coca-Cola except for the underlying assets in the text. And so we'll get there. So then the question is, okay, I mean, Goldman projects a 7% reduction in the labor population as a result of artificial intelligence, but overall a, uh, two and a half percent increase in GDP. And so that's massive, right? Like the US GDP is growing at about two and a half. Over the last 20 years, grew about two and a half percent a year. And so you have this impact where you could literally double the GDP growth of the US as a result of, of AI. And so the reason I think a lot of people are super excited about it, and the reason I'm so excited about it is like macroeconomically for the US, we're in a hole where we've printed way too many dollars for the GDP that we're producing, but now we have, we're f- you know, we're, we're faced with a technology that could actually create sort of like the, replicate the post-war surplus out of World War II that, that drove the next 40 to 60 years of, of, um, prosperity. But you've got a technology that's, it's not really a wartime technology, that it could do it. Uh, and that's the reason I think so many people are so excited about it, and, and,

11. 39:13 – 44:45

    AI & Wealth Inequality: A Discussion

    1. TTTomasz Tunguz39:13

       uh, that evaluations are as astronomical as they are.

    2. HSHarry Stebbings39:16

       S- I agree with you. The one concern that I have is like when you look at it, it does, like in my eyes, it does bring about a concern on like distribution of wealth and the concentration of, uh, income. Like, how do you think about like really, like wealth inequality over the next few years and the dangers of it actually concentrating wealth further into the hands of fewer?

    3. TTTomasz Tunguz39:39

       Yeah. Yeah. I mean, now we're sort of getting into politics. So I, you know, I think it's the role of the private markets in order to drive innovation forward, and it's the, it's the role of government in order to encode the values of a population into its laws. So, I think those are the forces that exist in tension. And, um, you know, uh, I think the, the network effects and the power laws that we are all chasing definitely create those dynamics when it comes to wealth. But it's not a new problem. You look at railroads or telecommunications or, uh, whaling, um, it's been around for forever.

    4. HSHarry Stebbings40:13

       Final one on this, but how do you think about like regulation? I, I'm concerned about the asymmetry of knowledge between private and public. Um, you know, w- we're very fortunate to spend time with some of the most brilliant entrepreneurs in the world. And then you go in to speak to, you know, regulatory bodies which, you know, bluntingly just don't have the same level of information and, and knowledge, and they're setting the regulation. It's concerning. How do you think about that chasm of knowledge between those two bodies and where it means will shake out from a regulatory standpoint?

    5. TTTomasz Tunguz40:44

       Yeah. I think, um, (laughs) this is a longer conversation. I think regulation on the whole as a... One, it benefits incumbents because the cost of adhering to regulations are significant. You take a look at, you know, in the mid-'90s you could have 25 million in revenue and go public. Today, if you have 100 million in revenue, it costs, and it c- it costs you $15 million in your first year to go public. And that's just a byproduct of regulation. So, regulation benefits the, the winners or the bigger companies. I think the second thing is a lot of the times when regulation is imposed, people don't anticipate the second order effects. You look at real estate prices in California as they're three to four times what they are in the rest of the country because of the law that was passed in the 1970s called Prop 13. You know, there are all these sort of like second and third order effects that a lot of regulation doesn't anticipate, and the legal process doesn't move fast enough in most of the cases. Uh, you look at crypto, right? It's taken the US government, I don't know, 10 years to kind of catch up to what's going on. And now with Operation Choke Point, they're starting to really regulate that ecosystem, and they've finally gotten around to it. So, I think, I think the system, and let me about indulge in another example. Like if you think about airplanes, okay, post World War II airplanes, people flew... We were just invented the jet engine, and the commercial airlining business was growing, but it was still really risky. And, but, you know, the FAA put a bunch of regulations around planes and we kept flying planes, and, and then one day, we realized that planes with square windows crash more because they create stress fractures along the points of the squares. And so we regulated that out. It's- And so I think the... Okay, what does that teach you? We don't know what we don't know. And so the best path to regulation is incremental when we identify that there's something wrong.You know, will bad things happen along the way? Yes, there's no doubt. Uh, but that's sort of the, and this might sound callous, but that's the path of, uh, the price of progress.

    6. HSHarry Stebbings42:36

       Final, final one. You mentioned kind of it sometimes benefiting larger companies and incumbents. This is my also big question, which is like startups versus incumbents. Alex Rampell of Andreessen says a brilliant one, which is like, "Will the incumbent acquired, uh, innovation before the startup acquires distribution?" When we look at kind of the two ends of the spectrum for the next generation of, you know, AI and LLMs, which is like, will existing incumbents integrate it well enough into their distribution channels to be highly effective and continue their dominance? Or actually are startups with agility, flexible code bases much better place to win in this next generation? How do you think about that kind of startup versus incumbent mega war? (laughs)

    7. TTTomasz Tunguz43:17

       Yeah. So my thinking's evolved here. In the beginning, I thought the incumbents were gonna win the whole thing. (laughs)

    8. HSHarry Stebbings43:21

       (laughs)

    9. TTTomasz Tunguz43:22

       And I, I thought that because, uh, the incumbents have far greater distribution. You know, Microsoft has an incredible channel. Microsoft has a special relationship with OpenAI. The pace with which Microsoft is injecting its, uh, products with, um, L- LLMs is astounding, right? And so startups are on, in this unusual position where they have negative time to launch. They're actually behind the market, which is unusual. I mean, think about mobile apps and the launch of the I- uh, Apple Store. Startups were the first ones to understand how to write mobile apps with Objective-C and ... But I think anytime we talk about machine learning, there's always this question around like, what is the moat? And the, I have this ini- this reaction, which is like, it's a data moat, it's a data moat. And I think the answer is the one that it's always been, which is better execution is the moat. If you can build a better CRM, it, and, and get it into market, you can win, right? You take a look at like what Notion has done with documents or what Snowflake did with databases facing two big incumbents. Like there, there are these stories, they're all over, they, they dot, they sort of, um, create this beautiful constellation within startup land of the David versus Goliath story. And I think if you're a venture capitalist or if you're a startup founder, you have to believe, uh, that. I think it's in your fabric that, uh, no matter how big the incumbent is or the advantages that they have, that if you have really great execution, you can, you can still win

12. 44:45 – 50:01

    Which company is losing the AI race?

    1. TTTomasz Tunguz44:45

       and you can win big.

    2. HSHarry Stebbings44:47

       I totally agree. I always say the speed of execution is the biggest determinant that I see in the differences between achieving and not achieving product market fit. Um, Tom, I think we both agree that Microsoft's absolutely killed it in terms of their embracing and approach to this next generation. Who's done really badly? (laughs) I'm just like looking on the flip side, incumbent-wise. Is it Apple?

    3. TTTomasz Tunguz45:08

       Oh.

    4. HSHarry Stebbings45:09

       Is it Amazon? Is it Facebook? Like, which one of them is like, "Oh, you really missed the beat on this one guys?"

    5. TTTomasz Tunguz45:15

       Oh, it's gotta be Google. Uh, it's my former employer, so it pains me to say it, but, and I, I didn't believe that chat would replace search. But I think it, for many use cases, it will. And I think Google had a rude awakening where, I don't know, for 20, 25 years they were uncontested. And now all of a sudden there's this disruptive technology that, to some extent they developed in-house but ignored. So it's a classic innovator's dilemma. And so this technology went to other places and now is, is challenging the hegemony, right? The monopoly power. And that is so exciting. I think if you are into the consumer ecosystem, if you think about like the ads ecosystem, like the B2C ecosystem has been relatively quiet over the last 10 years because of that dominance of Facebook and Google. And now all of a sudden you have a technology and a re-platforming where all that market share is conceivably up for grabs. Uh, you could create a new, uh, travel agency. You could create a new shopping experience. You could create a new stack overflow. You could create a new social experience based on chat. And, and so it's wide open.

    6. HSHarry Stebbings46:13

       (laughs)

    7. TTTomasz Tunguz46:15

       Yeah. Wide open.

    8. HSHarry Stebbings46:15

       Can I, I mean, blunt, what was DeepMind ... Was Deep ... Like, they, they, they did, they were so strategically ahead of the game acquiring DeepMind and an amazing team there. Like, what went wrong there? Help me out.

    9. TTTomasz Tunguz46:27

       I, I think it's a classic thing that it always, when you have a golden goose, when you have the, the, a gr- an incredible business model, you're always faced with the choice of disrupting yourself and destabilizing the ship or waiting until somebody destabilizes it for you. And it's, I think as a leadership team, it is so difficult to have the discipline to say, "We are going to destabilize this ourselves." And so, you know, uh, I, I just think that's what happened.

    10. HSHarry Stebbings46:54

        Do you think they knew? And what I mean by that is like Netflix did, did destabilize the golden goose. They took their mail order business and put it online 'cause it was obvious. It would lose revenue in the short term, but it was obvious. Like, the move from search to chat still isn't actually obvious. I mean, it's potential, but it's not obvious. Do you think that's why?

    11. TTTomasz Tunguz47:16

        I think it's part of it, but you know, the cost per query. So if you think about like the cost to produce a GPT-4 query versus the cost to produce a Google query, I bet it's like 100 or 1,000 or 10,000 times different. And so, you know, Chris Dixon had this post, "Every, every major innovation starts out looking like a toy." I bet the chat, the Google, uh, or anybody working in search is looking at those technolo- And then we, I had conversations with friends talking about like the cost per query on this stuff, you just can't get the economics. But you can't look at it at a point in time. You've gotta look at it on some geometric curve or some logarithmic curve where you've got effectively a Moore's Law happening, uh, for you. So I think that was definitely a mistake that I made in anticipating the technology. I think the other thing that we really didn't rev- I didn't really appreciate until some of the later models came out was just how s- sophisticated the emergent behavior can ... So there's this paper that talks about that how, how these LLMs learn. And it's like, it's, the, the analogy's like humans, right? So I can learn math by reading a book, right? I can learn addition, I can learn division. And in that way, like the next time I see four plus four, I know what the answer is. Or what are cube root of, of 27. I also learned how to swim. And in order to learn how to swim, I can read a book about the physics. I can, you know, fluid dynamics and I can understand like what's happening with the vortices and where my arms need to be and what my needs to do.But after reading that book, you throw me in the pool, I will drown. Like there's no f- there's no question, right? I can read all the theory in the world. And so there's two different ways that we learn, like we w- we learn by effectively memorization and we learn by doing. And what we thought at the beginning with these LLMs was that they're primarily memorization systems, and th- that's why th- there's improvements in the GMATs and the LSATs and the AP tests just because they have more and more exposure to those questions. What we're starting to realize is they, they learn also by doing. And so, uh, there are these sort of w- they are called emergent properties where the more questions that they're asked, th- the more they figure out how to answer those questions in a better way. There's this beautiful feedback loop that exists that only happens when they swim more. When they swim more, they learn how to swim m- faster. They're asked more questions, they learn how to answer questions better, just like a human would. And not to say that they're humans, um, set that whole thing aside, but that's, that, uh, that I think has a compounding benefit that is really difficult to appreciate. I mean, humans are very good at linear stuff and they're terrible at geometric stuff. And I think what happened is that the quality of the answers and the breadth of the knowledge and some of these emergent behaviors, like the models learning how to swim, all of a sudden snuck up on everybody. And now, I don't know about you, but like the pace of innovation in this space is so fast. You wake up every morning and there's a new model, there's a new way of putting it together, there's a new application. It's just, it's really hard to stay on top and that's because we're on the steep part of this geometric curve for the sophistication of these models. And at some point, it will make the, the shelf of an S.

13. 50:01 – 52:10

    Data Ownership: Who really owns your data?

    1. TTTomasz Tunguz50:01

       But we're not (laughs) it doesn't feel like we're anywhere close.

    2. HSHarry Stebbings50:05

       The final, final one, I promise. The one thing that as a European, I'm like shocked that no one else is th- thinking about or seems to be no one thinking about it is like the data or content ownership. And what I mean by that is like, you know, Google will redirect you to a newspaper website where the cool page is, where the original post is. ChatGPT will leverage, you know, the internet and the world's contact base and retain you on their website and simply scrape the information to theirs. That won't work. Like that, like content providers will not be able to build a business when ChatGPT just scrapes all of their content and they have no way to monetize in any way. Like how do we think about the future of data attribution and content attribution in that model?

    3. TTTomasz Tunguz50:49

       So Google's had this problem for forever with snippets, you know? Uh, you ask it like, "Uh, who is Hari Stevick?" And it puts three paragraphs about how amazing you are on the search results page, right? And that could come from the New York Times and, and the publishers in Google have been fighting back and forth in Europe and other geo- other geographies. I, I think it, it definitely exists here, right? Who owns that content? I mean, uh, what it, the, the notion of fair use, like if I take two music tracks and put them together, that's a new, that's a new product. And so I have that copyright. If I take the New York Times article about whatever, um, the events in Taiwan and I mix it with a CNN article and it produces a new article, is that a new thing where I should have copyright? And so basically, the internet becomes one huge walled garden that's just summarized by ChatGPT. Uh, I, I don't, I don't think any large language model operator wants to see that world because the reality is you need CNN and New York Times or any other content producers to have a viable business model in order to put into the system. And the large language model companies probably do not wanna get into that business. Um, and so what does the revenue share look like and what those arrangements and, and features, you know, I think that's all sort of TBD. I mean, you look at, I, I wonder if you can look at like the Amazon, sorry, the Mozilla Google deal or the Google Apple deal or some of the publisher contracts

14. 52:10 – 56:16

    Quick-Fire Round: Fast Q&A with Tom Tunguz

    1. TTTomasz Tunguz52:10

       or, um, even like distribution agreements across media companies today. I, we'll probably get to something like that, be my guess.

    2. HSHarry Stebbings52:16

       So will we be in a better or a worse place macro-wise by the end of 2023?

    3. TTTomasz Tunguz52:22

       I think we will probably be in a worse place by the end of '23.

    4. HSHarry Stebbings52:25

       Why?

    5. TTTomasz Tunguz52:26

       I think the Fed has overcorrected on rates. The rate of money production in 1M2 is decreasing faster than anyone expected. I think there's just like a human psychology to want to over-rotate on things and, and be slow, so. And then I think the risk of conflict in Taiwan is significant. So the combination of those four risk factors I think puts the odds of a US recession meaningfully higher than, than I think a lot of people appreciate.

    6. HSHarry Stebbings52:51

       We mentioned Microsoft is like the leader and Google is like, you know, bluntly behind. Who's like second to be chasing Microsoft? Who are you like they have a shot at chasing them?

    7. TTTomasz Tunguz53:02

       Adobe I think is, uh, doesn't have, um, doesn't have the recogni- recognition it deserves when it comes to using generative. I think about like the applications in Photoshop, they launched a product called Firefly. I think they're right there.

    8. HSHarry Stebbings53:14

       What trend in, in AI and the next generation of AI do you see that you don't think others are spending enough time on?

    9. TTTomasz Tunguz53:21

       Enterprise readiness. I think if there's one big market opportunity that people haven't focused on, it's how do you bring this to the global 2000 in a way that they will accept and buy, that's consistent with ways that they've bought software in the past.

    10. HSHarry Stebbings53:33

        You can invest in and you can short one multi-stage firm. Which firm do you invest in and which do you short?

    11. TTTomasz Tunguz53:41

        (laughs)

    12. HSHarry Stebbings53:42

        (laughs)

    13. TTTomasz Tunguz53:43

        I would invest in founders fund and I, I don't, I don't wanna say on the shorting side. (laughs)

    14. HSHarry Stebbings53:54

        (laughs) Okay. On the seed fund boutique side, um, like pure play seed fund, you can invest in one fund and I guess you don't want a short one, so we can just say invest in. Which one would you invest in on that side?

    15. TTTomasz Tunguz54:06

        On the seed stage, I'd invest in Goodwater. I'd invest in Goodwater because it's completely orthogonal to B2B software, and I really respect what Chi-Hua is building with, uh, his huge, I mean, it's pretty significant engineering team to identify B2C opportunities all over the world. And like I said before, the opportunity for LLMs to destabilize the existing B2C internet is really huge. And so I think he's got a nice market opportunity in front of him.

    16. HSHarry Stebbings54:34

        Tell me, what's your biggest investing miss and how did that impact your mindset?

    17. TTTomasz Tunguz54:40

        Yeah, I've missed, uh, so many companies, you know, like Datadog and Twilio, um, a- and many others. The thing that I've learned is that the startups are the ones who create the markets. And so, if you have a rabid user base in a really early market, it will f- it will, most of the time, surprise you on the upside.

    18. HSHarry Stebbings54:57

        I love that. Um, tell me, uh, what's the biggest investment mistake of the last 18 months for you, do you think?

    19. TTTomasz Tunguz55:04

        It's too early to tell. (laughs)

    20. HSHarry Stebbings55:07

        (laughs)

    21. TTTomasz Tunguz55:07

        Uh, I don't know. I mean, uh, I've been raising capital and building a firm, so I, I was lucky enough to invest in the Omni round and the Mother Duck round, and, uh, both of those companies are doing well. So I am, at least for now, I don't think I have a glaring mistake.

    22. HSHarry Stebbings55:22

        That- that- j- that- two good rounds to be in. Can I ask you, you mentioned now the fundraise, what would you most like to change about the world of LPs?

    23. TTTomasz Tunguz55:30

        I think the thing that needs, that I'd love to see happen in the LP base is LPs educating VCs on their goals more. This sort of happened in venture where venture capitalists explained their business models in a r- in really clearer ways about, like, fund construction. And I think the, the most impenetrable part about the LP, for in a lot of cases, is just understanding what drives them, what's their portfolio construction, and, and then figuring out how to map that to a fund. I think that's been the hardest part for me.

    24. HSHarry Stebbings56:00

        I think it's really hard also for me. I agree with you, and I asked many LPs to come on the show, but a lot of them really don't like to be public.

    25. TTTomasz Tunguz56:06

        (laughs)

    26. HSHarry Stebbings56:06

        And so they're willing to be... It- it's like, you know, it's like, you know, about 10 years ago, venture was not nearly as transparent, and I hope we bring a level of transparency

15. 56:16 – 59:23

    Election Prediction: Will Trump win in 2024?

    1. HSHarry Stebbings56:16

       to the LP market that we haven't had before. Um, will Trump win the election, Tom?

    2. TTTomasz Tunguz56:22

       I don't think so. I think, I bet DeSantis wins. I think it will be tough for him to circumnavigate all the legal troubles, and, uh, I wonder if the RNC doesn't get involved. And-

    3. HSHarry Stebbings56:33

       Would DeSantis be good f- would DeSantis be good for our business?

    4. TTTomasz Tunguz56:37

       He's a very complicated person. I think, on the whole, the Republican Party is still the party of business and capitalism. And so I would say yes. I would say yeah, and I, uh, maybe I'll put it a different way, which is the entitlement spending in the US over the next 10 years is projected to consume something like 95% of tax receipts. And so, that can't be. So we, we need, and I don't know who it will be, but we need some reform on the entitlements, just... And France is going, this, this is funny. France is going through this now, and it's, you can see it's extremely painful. And there's strikes in the US and there's strikes in France, and I think we're looking at a long period of time where the relationship between governments and, and people are, are going to change over the next 10 years pretty meaningfully. And so we need a leader who can guide us through all that.

    5. HSHarry Stebbings57:26

       Tom, penultimate one. Who's your favorite angel to work with, and why them?

    6. TTTomasz Tunguz57:30

       My favorite angel to work with is... Do you know? I don't work with that many angels.

    7. HSHarry Stebbings57:36

       So like, I love to work with Guy Podjarny, who's the founder of Snyk, um, the developer tool.

    8. TTTomasz Tunguz57:41

       Yeah.

    9. HSHarry Stebbings57:41

       Just fantastically insightful, really helpful, very granular advice. It can be a founder that you bring in.

    10. TTTomasz Tunguz57:49

        Oh. So one of the operator, one of the angels I really like to work with is a man named Alan Black, and Alan was the CFO at Zendesk, and he was on the board with me at Looker. And Alan has, um, he took, I mean, he took Zendesk public during the crash of '08. And so his experience going through financial carnage is just awesome. (laughs) Just to have that story, to have lived it, I, I really respect it, and it, it, um... Uh, I think he's got a really great world view as a result of that.

    11. HSHarry Stebbings58:20

        Tom, final one, my friend. What's been the biggest home run cash generative investment that you've made from a DPI perspective, and how did it come to be?

    12. TTTomasz Tunguz58:30

        It was Looker, and, uh, the, you know, the, the story there was in 2012, Redshift was the fast-growing product inside of AWS, and Tableau was the dominant BI product. And there was a thesis that there would be a new BI product that would be architected for the cloud, and, um, a friend of mine from Google introduced me to Lloyd, the founder, and we clicked, and I loved the technology that we had built, and there was a post that, I think it was Josh Kopelman or Ph- Finn Barnes wrote, and the question was, "Who took a bet on you when you were young in your career?" And Lloyd took a bet on me, and brought me in as DA, and forever grateful for it.

    13. HSHarry Stebbings59:06

        I love that. I- I actually haven't read that one, so I'm gonna read that. Tom, listen, I've loved doing this. I, I hope my interviewing style's changed a little bit over the years. (laughs)

    14. TTTomasz Tunguz59:13

        (laughs) It's, yeah, it's been-

    15. HSHarry Stebbings59:14

        Um, this has been so much fun. Thank you so much

    16. TTTomasz Tunguz59:18

        Yeah, the pleasure is all mine. Thank you so much, Hari. I really appreciate it. Congratulations on all your success, too.

Episode duration: 59:23