Loading summary
Transcript111 lines
- [00:01]
Marc Andreessen
Are ads purely destructive or negative to the user experience? Or are they actually, if done properly, are they actually either neutral or even positive? You know, if you're not Apple, do you really want to be a company that basically sits there and says, yeah, the world's moving and we're very deliberately not going to lean as hard as we can into it? I think there's a lot of survivorship bias in these kinds of strategy discussions where people look at the one company that's able to pull this off and they don't look at the 50 other companies that are in the graveyard because they didn't adapt.
- [00:27]
Podcast Host (Intro/Outro)
Marc Andreessen went live on TVP on this Week, and today we're dropping that full conversation here on the pod. Mark gets into it all. What's really happening in AI right now, how Apple is playing its hand, the return of open source and why perfect products can signal the end, not the peak. He also shares his take on how to break into venture capital in 2025 and what he's actually using AI for day to day. Let's get into it. This information is for educational purposes only and is not a recommendation to buy, hold or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies and individuals are not endorsed by AH Capital Management, LLC, A16Z or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.
- [01:22]
Chris Dixon
We have Marc Andreessen joining us. He's live from the TVPN Ultradrome. Welcome to the stream. How you doing, Marc?
- [01:28]
Marc Andreessen
Hey, what's happening?
- [01:30]
Jordi
Great to see you.
- [01:31]
Marc Andreessen
Yeah, you too. A lot.
- [01:33]
Chris Dixon
It's a little bit of a slow news day, but exciting stuff with GPT open source.
- [01:39]
Jordi
It's not a slow August.
- [01:40]
Chris Dixon
I will say it's not a slow August. We're glad. We were just reflecting that we've taken exactly one day off this summer. That was July 4th, and we're showing the Europeans how American companies work.
- [01:51]
Marc Andreessen
American work.
- [01:52]
Chris Dixon
We're setting an example and we have proof of work because we exist on the Internet and you can see us live every day. So we're setting an example. How are you doing? How's your summer going?
- [02:03]
Marc Andreessen
Fantastic. Going really well. So how long is it going to be until you guys put up avatars that make claims that you're working hard all through the summer, but it turns out you're on the beach?
- [02:12]
Jordi
You might have caught us.
- [02:13]
Chris Dixon
I think you'll know better than us as to when the technology gets there. We've been demoing some of the stuff people have been doing a lot of deep fakes of us, and unfortunately, all of them have been clockable. So it doesn't feel like a brand risk, but they're getting closer and closer. And I know that there's gonna be a moment when we have to say, hey, that's actually using our name and likeness to endorse something that we don't necessarily endorse. Can you please take that down? So we're approaching the touring test. The uncanny valley. We're escaping the uncanny valley.
- [02:41]
Jordi
I think a question looking back over the maybe 10 or 15 years was what moments did you feel like? There just was not a lot of action happening because this summer is just the pace from so many different teams has been absolutely insane. Everybody's like trying to keep up and it didn't used to feel that way, at least from my point of view.
- [03:04]
Marc Andreessen
So my view it always is, there's like these, there's this, these disconnected kind of patterns or trends. There's sort of the sort of day to day phenomenon where like engineers show up every day and they make things a little bit better. And then every once in a while, you know, you get a technical breakthrough or a new platform and that process kind of this, you know, kind of sawtooth kind of up to the right kind of process kind of plays out over time, kind of regardless of what else is happening in the world. And so it keeps happening through recessions and depressions and wars and like all kinds of crazy, crazy, crazy stuff that's happening. But basically, you know, the technology keeps getting better. So there's, there's kind of that curve and then there's the sort of enthusiasm curve and then the adoption curve, you know, which is basically like, when do these things actually show up in the world? And then, by the way, when are people actually ready for the new thing? Like, if you talk to the people who worked on language, I'm sure you guys have talked to people who work on language models, they will tell you that they were surprised. The ChatGPT was the breakthrough moment because they thought everybody already knew what these models could do for three years before that. And so they were shocked that it was the chatbot interface that made the thing go. And so there's somewhat of a sort of arbitrary disconnect between what's actually happening in the substance and then what people are seeing and feeling. And so it's just it's really hard to predict when these things pop, but also if you're in this day to day, it's really hard to tell when things are going to be hot or not because it doesn't necessarily map to how much the technology is improving.
- [04:24]
Chris Dixon
Yeah, we were just talking about that in the context of Google's new world model. It's this generative video game that you can kind of move around in and it feels like DeepMind is just absolutely crushing at the AI research frontier. They have the best world model simulator that you can walk around in. The question is like, if they let another lab do the ChatGPT thing and just get it out into the consumer three months earlier, they might wind up kind of chasing and trying to catch up if somebody actually figures out how to make it like a dominant consumer product. Now in the enterprise, it's more oligopolistic, but consumer seems to be winner take all. I guess the question is like, how much value do you place right now in the AI race to just like moving fast, breaking things, you know, dealing, having like the thick skin to deal with like the safety constraints and all of the different stuff. Obviously not being irresponsible, but just speeding up the organization as much as possible. It feels like now's the time to really push on that.
- [05:24]
Marc Andreessen
Yeah, well, first of all, I need to correct you. It's moving fast and making things.
- [05:28]
Chris Dixon
I don't know where that's right.
- [05:29]
Marc Andreessen
I don't even know where that came from. I have no idea what you're talking about.
- [05:34]
Jordi
Never heard of it.
- [05:35]
Chris Dixon
I mean, TRAGGPT didn't really break anything. I think that's a good point. It really did just move fast and make things. The first things it made were weird, but that was fine and it failed and it hallucinated a ton, but it didn't really break anything.
- [05:45]
Marc Andreessen
I don't know. Yeah, yeah. I believe in this case, total Deaths attributable to ChatGPT are still 0.
- [05:51]
Jordi
0.
- [05:52]
Marc Andreessen
So notwithstanding all of the, all the caterwauling. But yep, yeah. So look, I think the AI industry in particular has a very acute version of the sort of challenge that you identified with. And you know, and I don't say this negatively, just an observation which is that there, you know, like in sort of a normal technology company, you've kind of got engineers who make products and then you've got, you know, kind of salespeople or marketing people who sell them. You know, in the, in the AI companies you have this third tier of, you know, the quote unquote researchers. Yeah, right. And so, you know, which is, which has worked out incredibly well. I mean, the researchers have done, you know, they've just done like amazing breakthroughs at these companies. But, you know, the handoff, you know, there's not necessarily a clean handoff from the researchers to the market. And so it kind of raises this question of like, okay, like, is there are these companies therefore kind of three, you know, kind of three segment companies where they have research and then they have product development and then, and then they have go to market. And I think that's a really open issue. I mean, if you, you know, Google's kind of a case study of this. You know, you alluded to DeepMind, but even more broadly, Google, you know, Google developed a transformer in 2017 and then they basically let it sit on the shelf, right, because it was a research project, they didn't productize it. They were very worried about, you know, from people I've talked to, they were very worried about the, you know, brand issues and safety, you know, kind of all these, all these, they had all these reasons to not productize it. I talked to somebody senior who was there at the time, and I asked them, when could you have had ChatGPT with GPT4 level output if you had just gone flat out starting in 2017? And they said by 2019, they already knew how to do it and then they've now caught up, but it took an extra five years to catch up. And so I think a lot of these companies kind of have that challenge. Elon, as usual, of course, is provoking this question I'm sure you guys talked about. But he has now, within Xai, he's now collapsed. He's eliminated the distinction between research and product. And so, you know, of course he's pushing this as hard as he can. And I think it's a good question for a lot of these other companies, kind of how hard they want to push on actually getting these things in fully productized form out to the market.
- [07:47]
Chris Dixon
Yeah, yeah. On Elon's distinction, it feels like there is more research to be done, but it feels like we're entering like a new cycle of, you know, just focus on the engineering, focus on the deployment, the applications, let's get all this technology out into the world, let's reap all that benefit. And yes, there will be a different track of fundamental research that's happening somewhere, but it's really, really hard to predict. And so if you have something that's working, just double down and just go really aggressive. On it, I'm wondering more on. On that, but also on Apple strategy. It feels like Apple's been kind of like, you know, people have been maligning them for not for missing the AI opportunity. And Tim Cook's just there on the earnings call being like, look, we acquired a couple small companies and seven this year. Seven companies. But then it seems like they're taking more of like an American dol dynamism approach. Like, there was news today in the Journal that they, that they're investing $100 million in American manufacturing. They're certainly doing stuff. They're just not chasing the, you know, the, the shiny tennis ball headline, $100 billion capex. So I'm wondering about your thoughts on. On when you have a, you know, when, when you have a platform, how hard is it to resist chasing the new shiny object? Is that the right move? Or are there any other things that you think Apple should be, you know, strategy on?
- [09:12]
Marc Andreessen
Yeah. So look, Apple's always had this very clearly defined strategy that Steve and Tim working together figured out a long time ago, which is. I forget the exact term, but it's something like, basically, they invest deeply into the core of what they do. They'll basically work internally on things for many years. They only actually release things when they feel like they're kind of fully baked. And so as a consequence, they have this thing where, and Jim says this, right? They're rarely first to market with new technologies. They, they're more often in the category of what, you know, Peter, Peter Thiel calls last to market. You know, they're, you know, they'll, they'll, they'll come out whatever, three years later, whatever five years later. You know, they're, you know, there were tablets for years before the iPad. There were, you know, smartphones for years before the iPhone. Folding phone.
- [09:52]
Chris Dixon
They're about to do a folding phone. It's like 10 years into that technology. But I'm sure if they do it.
- [09:56]
Jordi
They'Ll see the last mover. The last mover.
- [09:58]
Chris Dixon
Yeah, yeah, yeah, sorry.
- [10:00]
Marc Andreessen
The last mover. And I guess. Yeah, well, what I was saying is, like, look, that clearly works if you're Apple, right? And so it clearly works if you're Apple. But I would say there's a fine line between that strategy and just. And simply becoming obsolete. Right? And so the problem is, like, if you're not Apple and you don't have all the other kind of super strengths and kind of now the market position that Apple has, do you really want to be a company if you're not Apple. Do you really want to be a company that basically sits there and says, yeah, the world's moving and we're very deliberately not going to lean as hard as we can into it. And so I think there's a lot of survivorship bias in these kinds of strategy discussions where people look at the one company that's able to pull this off and they don't look at the 50 other companies that are in the graveyard, you know, because they, you know, because, because they didn't adapt. I mean, you know, all the other smartphone companies when the iPhone came out, they were like, oh yeah, well we could do touch too, right. You know, we'll just, you know, we'll get to it. Right. And you know, you know, they're gone. Yeah.
- [10:48]
Chris Dixon
What was it?
- [10:48]
Marc Andreessen
What are you BlackBerry Bold?
- [10:50]
Chris Dixon
I remember it was like an iPhone knockoff.
- [10:52]
Jordi
What do you think?
- [10:53]
Marc Andreessen
Yeah.
- [10:53]
Jordi
You know, right now people are variety of, you know, shareholders are annoyed at Apple around their reaction to AI LLMs. John's annoyed around just like transcription generally super basic stuff. But it doesn't feel like the, the core business is immediately threatened today. It feels like it's still on the horizon around these sort of like, you know, eyewear based computing, you know, potentially net new devices that we're, that, that we'll see from you know, companies like OpenAI over time. But where do you like, how real is the threat this year versus 10 years from today and kind of what's your framework?
- [11:36]
Marc Andreessen
Yeah, well, look, I mean, I think the biggest ultimate danger, I mean the biggest ultimate danger is very clear, which is just like at what point do you not carry around a pan of glass in your hand, call the phone because other things have superseded it and everything becomes obsolete at that point. So there will come sometime when we're not carrying phones around and we'll watch movies where people have phones and we'll be like, yeah, look at how primitive they were. Right, because, because we'll have moved on to other things. And whether those things are I based or you know, you know, other kinds of wearables or whether it's just kind of, you know, computing happening in the environment or just, you know, entirely voice based, you know, who knows what it is. But you know, there will come a time when that happens. You know, is that time three years from now because there's like some, you know, huge breakthrough, you know, from, from some company that figures out the product that obsoletes the phone right away or is that 20 years from now because the phone is just, you know, such a Standard platform for everything that we do in our lives and everything else, you know, kind of remains a peripheral to the phone. I mean that, you know, that's, you know, that that's the game of elephants that's playing out there. You know, obviously, I think, you know, I think it's highly likely that we'll have a phone for a very long time. Having said that, it is, it is exciting that there are companies that are going directly at that challenge and you know, whoever cracks the code on that will be, that will be the next Apple. And by the way, that, that may in the fullness of time be Apple itself. You know, they may be the company that figures that out.
- [12:50]
Chris Dixon
Yeah, I remember being in a board meeting at Andreessen Horowitz maybe a decade ago or something and Chris Dixon showed me the HoloLens and I was like, okay, we're one year away from this being everywhere and I feel like today I'm still in the like, yeah, VR, it's definitely one year away. The next quest I'm going to be wearing daily. And it feels like we're always there. But it does feel like Apple did a lot of work on the fundamental pixel density of the resolution of the display. And then Meta has been doing a ton of work on just getting it light and affordable. It feels closer than ever. But you always got to wait until you see the churn numbers, until you really call the game.
- [13:29]
Marc Andreessen
Right. Well, you say the same, but you know, I think that's true. But you'd also say, you know, I'm on the, on the metaboloard, so I'm kind of a dog hunt on this one. But like the Meta Ray Ban glasses are a big hit.
- [13:39]
Chris Dixon
Oh, totally.
- [13:40]
Marc Andreessen
Right. Like they're a big, you know, so I think we now have a form factor that we know works, you know, for I based wearables, you know, there's not VR and then VR, you know, on top of that, but you know, just the, you know, the glasses and you know, and then the glasses of camera, you know, sort of integrated camera, integrated microphone, integrated speaker.
- [13:54]
Chris Dixon
Yep.
- [13:55]
Marc Andreessen
You know, that's a very interesting platform. You know, the watch clearly works, by the way, which Apple of course, you know, has played a significant role in making happen. You know, that now sells in huge volume. So that's the second data point. And then, you know, like, I think these, you know, these, these. I think some form of AI pin is going to work. I also think, you know, headphones are going to get a lot more sophisticated, which is already happening. And so, you know, you do have these, you know, kind of data points coming out. And then. Yeah, look, the trillion dollar question ultimately is, are these. Are these peripherals to the phone, you know, which is what they are today, or are these replacements for the phone? And, you know, we. Yeah, I would say we. You know, we have. We allowed. We. I think we have a lot of invention coming both from new companies and from the incumbents who are going to try to figure that out.
- [14:32]
Chris Dixon
Y. I always think about the value of like narrowing the aperture on these new technologies. Like with the meta Ray Bans, I feel like the fact that they aren't also trying to be a screen is actually a feature, not a bug. And I always go back to the iPhone. Like, it was first and foremost a phone and people bought it because it could make calls and then it could make text messages and then it was an ipod. But do you disagree with that, please?
- [14:56]
Marc Andreessen
Well, I don't. You guys might be too young. The first iPhone actually was a bad phone.
- [15:02]
Chris Dixon
How so?
- [15:02]
Marc Andreessen
Don't you guys. For the first two years, I couldn't reliably make phone calls.
- [15:07]
Chris Dixon
I had like the third one and a friend had one, but I feel like it was still like people were carrying cell phones and that was the. At least of the expectation. But, yeah, I mean, I guess you're right.
- [15:17]
Marc Andreessen
So for the first two, it was a classic Apple strike because for the first two years, the thing couldn't make phone reliably make phone calls. And then it turned out there was an issue with the antenna and with how you held it. And there was a. Yeah, you would.
- [15:27]
Chris Dixon
Record it and you would disconnect it.
- [15:29]
Jordi
You could basically brick the device.
- [15:31]
Chris Dixon
Yeah.
- [15:32]
Marc Andreessen
Based on how you held it. And somebody emailed. This is when Steve would respond to emails from random people. And somebody emailed Steve saying, if I hold the phone this way, it doesn't make phone calls. And he's like, well, don't hold it that way. Right. So even there it was like, yeah. And people forget. It took like five years for the iPhone to find its footing. It took like two years to get the. Remember also the original iPhone, it didn't have broadband data. It was on the old 2G. It was called the AT&T edge network. So it didn't have broadband data. And then of course, it didn't have an app store store. Right. It was completely locked down. Right.
- [16:01]
Jordi
So the challenges. The challenges for Apple now is that people are so used to perfection with the device that launching a product that isn't perfect, like, is Embarrassing, right? Like you look at the Vision Pro and it's like, well, the battery's big. Steve would have hated this, right? Like how he never would have shipped this and that. Being constrained and not being able to innovate because you're tied to this like impossible standard of being on whatever generation 17 of the iPhone and perfecting every element is a real challenge.
- [16:32]
Marc Andreessen
So I would say there's a corollary to that. One of the things I've observed over the years is I think technology products become obsolete at the precise moment they become perfect. What I mean by perfect basically is, yeah, it's like the perfect idealized, complete product. Like it does everything you could possibly ever imagine, everything a customer could imagine, everything you, as the technology developer can imagine. It's absolutely perfect. And there's been tons of examples of this over the last year, 50 years, where it's like the absolute perfect, what seems to be the permanent version of that product. And then it just turns out that's actually the point of obsolescence because it means creativity is no longer being applied right into that platform. You're just like, there's just nothing else to do. You're just like, you're done. The product has been realized and then the cycle is what happens? To your point, the cycle is other people come in with completely different approaches, completely different kinds of products that are broken and weird in all kinds of ways, but are fundamentally different. And so that is one of the time honored traditions. And one of the things you could say about Tim is, you know, his willingness to kind of break the mold of Apple only shifts perfect products. But, you know, being willing to ship the, you know, the Vision Pro, you know, you know, shows a level of determination to kind of stay in the innovation game.
- [17:39]
Chris Dixon
I like that.
- [17:39]
Marc Andreessen
Which I think is very positive.
- [17:40]
Chris Dixon
Yeah, yeah, yeah, yeah. That's great.
- [17:43]
Jordi
Updated thinking on open source since we last talked. There's, there's a lot that's been.
- [17:49]
Chris Dixon
OpenAI is an open source company.
- [17:52]
Jordi
Yes. OpenAI is open again.
- [17:53]
Chris Dixon
Yes.
- [17:54]
Marc Andreessen
Yeah, yeah, look, very encouraging. You know, a year ago I was very, you know, I was getting very distressed about open, you know, whether open source AI was going to be allowed, it was even going to be legal. And so, and I think we're basically through that at this point. We're through that in the US we'll see about the rest of the world. And then look, the US China thing is obviously a big deal, but I think it's been net positive for the world that China has been so enthusiastic about Open Source AI coming out of China, which has been great. And then, yeah, look, OpenAI leaning hard into this and releasing what they did is I think fantastic, both because of what they released, which is great, but also just the fact that they are now willing to do that. And then Elon reconfirmed overnight that he's going to start open sourcing previous versions of grok. And so, yeah, so we seem to be in the timeline where open source AI is going to happen right now. I think what you would say is it kind of lags the leading edge proprietary implementations by six months or something like that. But I think that's a good, if that's the status quo that continues, I think that would be a very good status quo.
- [18:55]
Chris Dixon
What are the rough edges that we need to kind of sand down when we're thinking about Chinese open source models specifically? Uh, is it we need to do some fine tuning on top of them to add back free speech or do we need to watch for backdoors? It's phone and home. If it runs into this specific thing, like the Chinese open source thing, it was remarkable because I feel like it really does accelerate the pace of innovation because everyone gets to see, oh, this is how reasoning works. I think that's great. Uh, at the same time it made me very, it made me much more appreciative of AI safety research and capability research and actually being able to interpret what's going on and say definitively this model is going to behave weird in this weird way. Like the Manchurian Candidate problem. We haven't found any of that, but it certainly seems like something we'd want to keep an eye on. But from your perspective, like what are the, what are the risks that we need to be aware of going into a world where China is really pushing hard into open source?
- [19:46]
Marc Andreessen
Yeah, there's two, there's two and you identify them. But let's, let's, let's talk about both of them. So the phone, so the phone home thing is the easy one, which is you can put up, you know, you can packet sniff, you know, a network and you can tell when the thing is doing that. And plus you can go in the code and you can see what it's doing that and so you can validate that that's either happening or not happening. And I think that, you know, that's important. But, you know, I think people are going to, people are going to, are going to figure that out. You can kind of gate that problem practically. The bigger issue is we have this term in the field right now called open weights. And open weights is a loaded term. It uses the open term from open source. But of course, with open source, the thing is you can actually read the code. You know, with open weight you have just a giant file full of numbers, as you said, that you can't really interpret. And then what you don't have, what most of the open weights models don't have, including deep SEQ specifically, what they don't have is they don't have open data or open corpus. So you can't actually see the training data that went into them. And of course, most of the people building models are kind of obscuring what that training data is in various ways. And so when you get an open weight model, the good news is the software source is open, the good news is you can run it on your machine, you can verify that it doesn't phone home, but you don't actually know what's happening inside the weights. And so I think that that is going to be a bigger and bigger issue, which is like, okay, how the thing behaves, like, yeah, what has it actually been trained to do and what restrictions or directives has it been given in the training that are embedded in the weights that you need to be able to see? I would say this is coming up as sort of, I would say a global issue which we worry about when these models come from China. Other countries worry when these models come from the U.S. so one of the phrases you'll hear when you talk to people kind of outside the US is kind of this phrase people are kicking around, which is not my weights, not my culture, or by the way, for that matter, not my weights, not my laws, which is like, okay, what actually is this thing going to do? Right. And to your point, the Chinese models, for example, might never criticize communism or something. Tell you, the American models have all kinds of constraints also implemented usually by a very specific kind of person in a very specific location in the US. And so I think that this is a general issue and we're going to have to see basically people's tolerance levels, being willing to run open weights models where they don't fundamentally have access to the data. And then correspondingly, I think what we'll see is more open source developers also doing open corpus, open data, so you can see what's actually in them.
- [22:17]
Chris Dixon
Yeah, obviously open source is very important in terms of just distributing intelligence broadly, giving people the ability to run their own models and really fine tune them, have control. There's also a big push just to make frontier models and High capability models, free. One model is you charge for the premium, you give the free away. It's a freemium model. That's what we're seeing at most of the labs right now. There's also this kind of specter on the horizon of potentially putting ads in LLMs and what that would do to the world. Jordi got in a little dust up with Mark Cuban on the timeline, deciding whether or not it would be net good to put advertising in LLMs. What might happen that might be bad there? What?
- [23:02]
Jordi
Yeah, my, my point broadly was that ads have been an incredible way to make a variety of products and services online free. And just saying like default, just no ads would potentially, you know, be incredibly destructive. But yeah, curious, your framework.
- [23:22]
Marc Andreessen
Yeah. So I should start by saying like, whenever I personally use an Internet service, I always try to buy the premium version of it that doesn't have ads. Right. And so if I can like live personally inside an ad free universe and pay for it, like, that's great. And I'll freely admit, you know, whatever level of, you know, hypocrisy or incongruence, you know, kind of, kind of results from that.
- [23:40]
Jordi
But no, the point is choice. The point is choice.
- [23:42]
Marc Andreessen
Well, the point is exactly what you said. It's affordability. So the problem is if you really want to get to five, if you want to get to a billion and then 5 billion people, you can't do that with a paid offering like it just at any sort of reasonable price point. It's just not possible. The, you know, global per capita GDP is not high enough for that. People don't have enough income for that, at least today. And so if you want to get to, you know, if you want the, if you want the Google search engine or the Facebook social app or the whatever AI, you know, frontier AI model to be available to 5 billion people for free, you need to have a business model, you need to have an indirect business model. And as is the obvious one. And so I do think if you take some principle stand against ads, I think you unfortunately are also taking a stand against broad access just in the way the world works today. And then look, the other really salient question is the same question that the companies like Google and Facebook have been dealing with for a long time, which is, are ads purely destructive or negative to the user experience, or are they actually, if done properly, are they actually either neutral or even positive? And this was something that Google, I think, to their credit, figured out very early, which is a well targeted ad at a specifically relevant Point in time is actually content. It actually enhances the experience. It's the obvious case. You're searching on a product, there's an ad, you can buy the product. You click, you buy the product. That was actually a useful piece of functionality. And so can you have ads or other things that are like ads or look like ads, different kinds of referrals, mechanisms or whatever? Can you have them in such a way that they're actually additive to the product experience? And you can imagine, just like with search and with social networking, you could imagine lots of examples of that. People will, you know, people will, you know, they'll whiner in lots of different ways. But I think, you know, I think that hasn't been a bad outcome overall. And I think that, I think it's entirely possible that that's what happens with these models as well.
- [25:31]
Chris Dixon
Yeah.
- [25:31]
Jordi
So kind of similar kind of question, what should be legal? Kind of trying to create legal frameworks on a number of issues with AI. There's been a number of IP cases that have been working their way through the courts. What can labs use to train models, etc. There's been some good outcomes recently. Sam also was talking about how a lot of people are using AI as like a confidant, like a, you know, a friend, things like that. And he mentioned that currently your chats are not privileged. They can be used in a, in a lawsuit or, or other situations. How, how optimistic are you that our sort of legal system in the US can get some of these issues right, where maybe it can't just be, you know, total free markets, kind of lawless, whatever goes.
- [26:24]
Marc Andreessen
Yeah, so in the case of training data, I think that there, I mean, there's a bunch of these copyright, you know, kind of lawsuits happening right now. There's, you know, the big New York Times OpenAI1, and there's, you know, been a bunch of others, I think, in that, for that particular problem. My guess is that problem ultimately has to be solved through legislation. It's, it's ultimately a legislative question. The reason is because it goes to the nature of copyright law itself, you know, which, which is legislation. And of course, you know, the content industry is already claiming that, of course, you know, using, using copyrighted data to train, you know, without permission and without paying is, is, is sort of, you know, they, they believe illegal on its face, you know, due to violation copyright law. The counterargument to that, which, you know, which we believe is, well, it's not copying. Right. There's a distinction between training and copying. Just like in the real world, there's a distinction between reading a book and copying the book, you know, as a person. And so there, there, there's going to need. I, I think, you know, the courts are trying to grapple with that. There's a whole bunch of cases. There's juris. Know, probably ultimately Congress is going to have to figure out a, you know, figure out an answer on that. And by the way, the president has kind of, you know, thrown down that gauntlet in his, I think the speech he gave last week or two weeks ago, you know, where he said that, you know, Washington probably needs to deal, deal with that as an issue. So that's one. On the, on the, on the, on the, on the privacy thing, I think that that one feels like it's a Supreme Court thing to me. It feels like that's the kind of issue Supreme Court. And that in other words, like whether, for example, your transcripts are considered your property and whether they're protected against warrantless search and seizure and the observation I would make there is if you look at the march of technology over time, so the Constitution has, like, very clear, you know, fourth, fifth Amendments, you know, very specific rights around, you know, the things that are yours, you know, such as, you know, your home, you know, being in your home, you know, by the way, the thoughts in your head, right, you know, that the government can't just like, come in and take, they can't, you know, they can't just come in and search your house without a warrant. You know, they can't like, you know, put you in a jail cell and beat you until you fess up. Like, you know, there are, there are, you know, we have constitutional protections against the government being able to basically take information fundamentally as well as possessions. And then basically what happens is every time there's a new technology that creates a new kind of sort of thing that you own thing that's yours, thing that you would consider to be private, thing that you wouldn't want the government to be able to take without a warrant out of the gate. Law enforcement agencies just naturally go try to get those things because there are ways to solve crimes. And it feels like that's a legal thing to do. And then basically the courts come in later and they rule one way or the other and basically say no. That actually is also a thing that is protectioned against, for example, warrantless search, warrantless wiretapping. And so I feel like this is the latest of probably, I don't know, 20 of those over the last hundred years. And I Don't know which way it'll go. But I think it's going to be a key thing because as you know, people are already telling these models lots of things that are very personal.
- [29:13]
Chris Dixon
Okay, lightning round, quick questions. We're letting you get out of here in a couple minutes. We're in this age of spiky intelligence. Models are great at some things and then terrible at others. Where are you actually getting value out of AI right now? Where is it falling down for you? Where are you, how are you using AI day to day?
- [29:30]
Marc Andreessen
Yeah, so I, I, I have two kind of, I don't know, bar, barbell approach. One is for, for serious stuff. I love the deep research capabilities. Yeah. And so, and I'm doing this in a bunch of models but like the ability to basically say I'm interested in this topic. And then I just, I just felt like write me a book. And I, you know, I'm kind of hoping for the longest book I can get. I always tell it like go longer, go longer, more sophisticated, you know. But the leading Edge models now, they' 30 page PDFs that are completely well formulated, basically long form essays. It's just incredible richness and depth. And if it's 30 pages today, I'm sort of crossing my fingers that'll get to 300 pages coming up here in the next few years. And so I'm able to basically have the thing generate enormous amounts of reading material with just, I think incredible richness and depth and complexity. And then on the other side of the barbell is humor. And I've posted some of these to my X feed over the last couple of years. But I think these models are already much funnier than people give them credit for really. I think they're actually quite highly entertaining. A while ago I posted specific formats.
- [30:33]
Jordi
Like chatting back and forth, be Mark Andreessen, that format take a dip in.
- [30:39]
Chris Dixon
My pool in my office.
- [30:41]
Marc Andreessen
They're really good, so they're really good at green text that works really well. But for some reason the ones I find hysterical are the, I haven't write screenplays for like TV shows or plays or movies. And I posted I had it right new season of the HBO Silicon Valley, you know, set 10 years later. And I had it right like an entire, I had it right like 10, 10 scripts for a complete season. And of course I just said, you know, make it like Silicon Valley, except you know, it's happening in 2021. It kind of peak woke. And I thought it was just, I think it's just, you know, I'll sit there at two in the morning just like laughing my ass off at how funny this thing is. And so I think these things are actually, are actually already like extremely funny. They're extremely entertaining when they're, when they're, you know, when they're used in that way. And I do, I do en that a lot and I generate a lot of those that I don't post.
- [31:28]
Chris Dixon
Stay in the group chats is probably a good idea.
- [31:31]
Marc Andreessen
They're your property.
- [31:32]
Chris Dixon
Yeah. Hopefully the Fourth Amendment holds on these.
- [31:34]
Marc Andreessen
Yeah, it's great.
- [31:35]
Chris Dixon
I have one last question.
- [31:36]
Jordi
Go for it.
- [31:37]
Marc Andreessen
And then I've got one more.
- [31:38]
Chris Dixon
How do you get a job as a venture capitalist in 2025?
- [31:42]
Marc Andreessen
So, I mean, look, the, the best way, the best way to do it is to have a track record early as somebody who is like in the loop specifically on your product development. And so somebody who, you know, be, be like deeply in the trenches at one of these new companies, in one of these spaces, you know, participate in the creation of a great new product and a great new company and, you know, really demonstrate that you know, how to do that. You know, there's, there, you know, There are great VCs who have not done that. But, you know, I think that is sort of a foundational skill set, you know, for working with the kinds of founders that you want to work with who are going to, who, you know, are going to want you to have, you know, kind of very interesting things to say on that, as I think, you know, still the best way to do it.
- [32:17]
Chris Dixon
Yeah, like feel the growth. Be immerse yourself in the growth, the aggressive growth environment and then you'll be able. They identify it when you see it from afar.
- [32:26]
Marc Andreessen
Yeah, that's right.
- [32:27]
Jordi
Last question for me. State of M and A in your mind, how are you advising, you know, companies where you're on the board or just the portfolio broadly around what they should expect now and, and in the near future.
- [32:41]
Marc Andreessen
You mean in terms of whether you can get things approved or.
- [32:44]
Jordi
Basically, yeah, yeah, yes.
- [32:46]
Marc Andreessen
So look, approval, still, approval is not a slam, doc. There was, there was a, you know, there was a buy, as I just saw, there was a medical device company this morning, you know, where the acquisition was not allowed by the ftc. So, you know, look, there is still scrutiny. You know, it's obviously a very different political regime in Washington. But you know, this is, this is not by their own statements. This is not an administration that believes in total laissez faire and then it definitely wants to, you know, in their view, maintain a very healthy level of market competition.
- [33:12]
Jordi
Do you expect certain companies to be negatively impacted by the FIGMA story? Right. You have this deal gets blocked, successful, you know, ipo. Lina Khan is taking a victory lap. You know, many people were responding and joking, saying, you know, someone. Lena cuts off the arm of a pianist, and they endure and can create a masterpiece. And then, and so I expect. And then you look at the example with, you know, Roomba, I think it was where Roomba had a deal with Amazon, it was blocked and fell apart, and the company is in shambles ever since. So my concern is that people look at FIGMA and say, you should be independent. You just figure it out.
- [33:53]
Chris Dixon
Nothing can go wrong.
- [33:55]
Marc Andreessen
Yes. Yeah. Miscon. Taking a victory lap was very disconcerting and for exactly the reason you said, which is survivorship bias, which is you pick the one that worked out and then it's the airplane, the red dots, the airplane. You ignore the 50 that are in the ground that you've never heard of. And so that was very disconcerting because it's sort of the central planning fallacy, which is like, we make centrally planned economic decisions. We have one example, you know, it's like in Europe, it's like, yeah, well, the bottle caps actually don't fall off the bottle, right? Like, you know, it works, right? It's like, okay, but do you want to live, do you want to live in an economic regime in which the government has dictated bottle cap design? The answer is clearly no, because the.
- [34:37]
Jordi
Downside consequences, or even looking at the Chinese model, which is, you know, people can say they're picking winners, but to get, get to maybe picking a winner, you have this intense bloodbath of competition where, you know, teams need to rise to the top and sort of prove themselves before they get any of that real, like, you know, meaningful state benefit.
- [34:59]
Marc Andreessen
Yeah, yeah, that's right. And so you just, you just, yeah, you just, you just have this adverse selection survivorship bias thing where you just, you don't pay attention to all the collateral damage. So I do think that mentality is like, super dangerous. And so, yeah, look, I, I think companies just have to be very thoughtful about this, both acquirers and the acquirees. And the big thing is if you're selling a company, you just need to anticipate that you might not get it through. And if you don't, they're sort of like, okay, number one, is there like a big enough breakup fee? Are you going to get paid for the damage that you're going through? And how is that structured on the one hand and then two is, yeah, look, do you have the kind of company culture that's going to be able to withstand that and is your business strong enough to be able to get through that? And it is a real risk and something worth taking very seriously personally. Yeah.
- [35:43]
Jordi
And that's why it felt emotion. We were at Nice last week. It felt emotional that the Figma team was able to effectively just restart the business and say, we're taking this all the way.
- [35:59]
Marc Andreessen
To think about it, if you talk to any really successful company, what they'll tell you is, yeah, over the years we had these crucible moments in which we almost died. But we pulled together and we pulled it off and then that became one of these central kind of mythical events in the history of the company that we always refer to. And like, my God, we got through that and we're so strong and tough and we've been forging fire and now we can do anything. And it's like, yeah, that's great. And then there's 50 other companies that hit those crystal moments, blew up and died. So, yeah, like it's, it's all of the lessons learned on this stuff, they're all conditional on life survival and so they, they. These things need to be taken incredibly seriously, you know, which, which the great CEOs do.
- [36:36]
Chris Dixon
Yeah. Well, thanks so much for joining. We'll let you get back to your day. We are busy five minutes over. Next time we have to book five hours because this is fantastic. I got 10% of the way.
- [36:44]
Jordi
First 24 hour TVP.
- [36:46]
Chris Dixon
Yeah, we would love to have you again.
- [36:47]
Marc Andreessen
Marathon.
- [36:48]
Chris Dixon
Enjoy the rest of your day. We'll talk to you soon.
- [36:50]
Marc Andreessen
Mark. Have a great day.
- [36:51]
Chris Dixon
Bye.
- [36:51]
Marc Andreessen
Sounds good. Thank you guys.
- [36:52]
Chris Dixon
Thank you.
- [36:56]
Podcast Host (Intro/Outro)
Thanks for listening to the A16Z podcast. If you enjoyed the episode, let us know by leaving a review@ratethispodcast.com a16z. We've got more great conversations coming your way. See you next time.