E166: Mind-blowing AI Video: OpenAI launches Sora + Is Biden too old? Tucker/Putin interview & more

Jason Calacanis 0:00 Zacks, how's tequila?

David Sacks 0:01 I tasted 10 samples the other day.

Jason Calacanis 0:05 Yeah.

David Sacks 0:05 One was heads and shoulders above the rest, which was a good sign.

Chamath Palihapitiya 0:09 Oh.

David Sacks 0:09 So now we're taking that and making it the new baseline and I'm gonna try 10 more samples and we're gonna go from there. I think we're very close to having something excellent.

Jason Calacanis 0:17 Great.

David Sacks 0:18 And then I did a, a side by side taste test of the sample I liked the best compared to Clase Azul Ultra and it was already better than Clase Azul Ultra-

Jason Calacanis 0:27 Hm.

David Sacks 0:27 ... which-

Jason Calacanis 0:28 Really?

David Sacks 0:28 ... retails for, like, $3,000 a bottle.

Jason Calacanis 0:30 Really?

David Sacks 0:31 Yeah.

Jason Calacanis 0:31 Holy (beep) .

David Sacks 0:32 So it's gonna be better.

Chamath Palihapitiya 0:33 V1's better than their final product is what you're saying.

David Sacks 0:36 Yes.

Jason Calacanis 0:36 And that was g- not just your taste test, but, like, you and several people?

David Sacks 0:40 It was me and two other guys who I trust, so we, we did it together.

Jason Calacanis 0:43 Now, when you say trust, do you mean underlings or do you mean actually, like-

David Sacks 0:47 (laughs)

Chamath Palihapitiya 0:47 No, he means alcoholics. (laughs)

David Sacks 0:48 (laughs)

Jason Calacanis 0:48 Yeah. Where in your pay hierarchy are they?

Chamath Palihapitiya 0:51 He recruited them at Alcoholics Anonymous in, (laughs) in West Hollywood.

Jason Calacanis 0:55 Are they level two employees or level four employees?

David Sacks 0:56 No, one, one works for me and one's a consultant.

David Friedberg 1:00 What you're willing to give.

Chamath Palihapitiya 1:00 Oh, so they both work for you. (laughs) So they're both on the payroll.

David Friedberg 1:04 So let your winners ride.

Chamath Palihapitiya 1:07 Rain Man David Zack.

David Friedberg 1:09 I'm going all in.

David Sacks 1:11 As I said, we are open source to the fans and they've just gone crazy with it.

Jason Calacanis 1:15 Love you guys.

Chamath Palihapitiya 1:16 Queen of Quinoa.

David Friedberg 1:17 I'm going all

David Friedberg 1:19 in.

Chamath Palihapitiya 1:19 Okay. Are we over or underestimating the impact of the AI boom? NVIDIA has been on an all-time heater. If you don't know, its market cap is now 1.8 trillion, Friedberg. This week it surpassed Google and Amazon as the fourth most valuable company in the world. There it is, folks. Right behind Saudi Aramco. 16 months ago, stock was trading about $112 a share. Since then, share price has gone up 6.5X on a big number. Shares went higher this week after it was reported that NVIDIA plans to build custom AI chips with Amazon, Meta, Google, and OpenAI. In other news, Financial Times reported OpenAI hit a $2 billion run rate for their reoccurring revenue. It's about $165 million a month and they think they can double that number in 2025. Tons of competition coming out. If you didn't see, Google rebranded their generative AI suite to Gemini and they're charging $20 a month for it. And then there was this bizarre Wall Street Journal headline that SAM was gonna raise $7 trillion. That makes no sense. Maybe that was the TAM. We'll get into that. Friedberg, your thoughts on this ascension. Are we underestimating or over-hyping AI?

Jason Calacanis 2:35 I mean, it seems like the compute buildout is underway, so this is gonna continue for some time. I think the real question that keeps getting asked is, you know, when do you really see the tides recede and what's, what's there? The compute buildout is underway, but are we seeing application and productivity gains associated with the capability of that compute? Some would say yes, some would say no. But you've got to build out the infrastructure if you're gonna assume that there's gonna be these productivity gains in the application layer and everyone's assuming they'll be there, so we need to meet the demand, which is almost like building out an entirely new internet. And so the core of AI is this matrix multiplication problem (laughs) and you need these chips that do that really efficiently. They're not traditional CPUs, which are general purpose chips, but they're chips that are very special purpose for a particular set of applications that render the output of whatever model is being run. It's sort of like the internet. Like, you could pretty much create a bogey of any number on how big this can get. At the same time, you're seeing some of the other players with different architectures and potentially that are gonna be beneficiaries of this buildout because they are participants in other parts of the data center for alternative solutions like ARM. I don't know if you saw this, but in just the last couple of weeks or last week or two, ARM's valuation doubled. It's come down a little bit since that big doubling. Today ARM is sitting at $140 billion valuation. Last week it got as high as $156. And remember, it was only last August... So if you guys remember, ARM was bought by SoftBank and the Vision Fund and then I think 25% of ARM was held by the Vision Fund and 75% was owned by SoftBank. Last August, SoftBank bought, from the Vision Fund, the 25% that the Vision Fund owned at a valuation of $64 billion. So now SoftBank owns 90% of ARM, the last 10% trades publicly, so at $140 billion valuation, SoftBank's position in ARM is $125 billion. And I think it really begs the question, did Masa have a masterful move here yet again? Because everyone said for years with all the bets this guy was making in the Vision Fund and some of the misguided bets he was making with SoftBank and the leverage he was applying on himself and SoftBank to deploy all this capital during the ZIRP era, everyone was saying, you know, this guy had a one hit, it was Alibaba and that was it, and clearly he doesn't know what he's doing. But now all of a sudden his bet on AI was fairly prescient and it's paying off pretty powerfully. He bought the whole business for $32 billion and now he has a, a stake that's worth $125 billion seven years later.

Chamath Palihapitiya 5:27 Yeah. I mean, chip and a chair, right? I mean-

Jason Calacanis 5:28 Well, it's really begging the question on, like-

Chamath Palihapitiya 5:29 ... you only need one, the power law, right?

Jason Calacanis 5:31 ... when you're making a long term investment, you can very easily look like a moron over the short term. And to your point, the power law, all it takes is one of them to work out and it can wipe out-

Chamath Palihapitiya 5:41 Yep.

Jason Calacanis 5:41 ... the losses on the rest of the portfolio and we may be seeing Masa with his whole AI vision story that he's been telling for seven years, we may see Masa clawing his way out of the hole. You only look like a moron to losers. Meaning when people were throwing shade at Masa, it wasn't anybody that's any good. I didn't see headlines that said, "John Malone thinks Masa is a fraud." "Warren Buffet thinks Masa is a fraud." Nobody was saying that.

David Friedberg 6:08 ... because those guys know what the game is. And the game is exactly what you said, Jacob. You must always survive and you need to keep your wits about you. And over time, if you're prepared and do enough good work, you'll get waves of good luck that amplify your hard work. And Masa is a, an example of an incredibly talented person that has found a way to survive, more than anything else. And so waves of success will hit him. The people that complain are the ones that are sort of a little bit stuck on the sidelines and are more frustrated with their own impotence and inability to do anything. Now on Arm specifically, the thing that I would keep in mind is that those guys floated I think about 15 million shares. So there's about a 145, 150 million shares outstanding, 90% owned by them, so there's about 10 or 15 million shares outstanding. And there was about 12 and a half million shares short going into the earnings report. And so the thing to remember is that there was almost an entire turn of the float that was short. So I don't know how much of this reaction was just an enormous short squeeze of people that were largely long other chip companies and so they thought they could short Arm essentially as a spread trade to minimize volatility. But this is an example of where if you don't do this analysis properly, GameStop was another where you misjudge or you sub-analytically look at the actual share count and what's actually free floating, you can get caught on the wrong side and these things can be very violent.

Chamath Palihapitiya 7:47 And to explain the float to folks, you can make a certain number of shares available to the public to trade. When you have a small number of those, correct Chamath, available to the public, weird things can happen because there's not enough available to short.

David Friedberg 8:02 That's exactly right.

Chamath Palihapitiya 8:03 And so that could've been the situation here with this run up, but I think, I think the overall point of placing bets and being a capital allocator is the important one to discuss here. Sacks, what are your thoughts on all the crazy bets that Masa made? We watched a lot of our companies or companies adjacent to our companies get massively overfunded. Some of them literally got drowned in capital, but then you only need to hit one and, and maybe win it all back. What are your thoughts on this high alpha, you know, betting style of his?

David Sacks 8:36 Look, I mean, he is a gambler and he is playing a power law so he only needs one really big return to basically make up for all the mistakes and it seems like he's hit that.

Chamath Palihapitiya 8:46 Once again, he did it with Alibaba too, right? I mean this is like-

David Sacks 8:49 Yeah, I mean he's a major gambler. Most people would be a little bit more risk-averse. I don't think it was a great strategy to be pumping $500 million checks into what were effectively seed companies like they were doing with the Vision Fund. However, obviously he made enough good bets that they're gonna pay off that fund.

Jason Calacanis 9:06 Well, it's a good point, Sacks. I mean, none of the bi- I, I don't, I, I, I probably need to be careful 'cause I'm not super familiar with the portfolio, but it doesn't seem like any of the big winners were those seed bets. Like, he doesn't have any returns coming out of that kind of investment strategy, but invest-

Chamath Palihapitiya 9:24 DoorDash. DoorDash.

Jason Calacanis 9:25 Yeah. DoorDash was not... Like, what point did they invest?

Chamath Palihapitiya 9:28 Billion or $2 billion round was it?

David Friedberg 9:29 Yeah, pretty important moment. We've heard at dinner, Jason and I at dinner, we've heard the, the waterfall from the SoftBank side and it sounds like it was an enormous number.

David Sacks 9:41 Yeah.

Chamath Palihapitiya 9:41 That was definitely their, their investment.

Jason Calacanis 9:42 But it was a de-risked business, it was a scaling business. Seems a little different than some of these companies that were fairly early on that got monster valuations and large chunks of capital and then struggled to find ways to deploy that capital, which led to unhealthy behavior.

David Friedberg 9:59 I think the thing you have to give him a lot of credit for is it's so easy to give up when the peanut gallery is chirping. And so the thing that you have to give him a lot of credit for is he has the ability to tune those people out. Now he probably isn't even living in a world where any of those people, that chirping even gets to him, and I think that that's a very important thing. But kudos to him. I mean-

Chamath Palihapitiya 10:25 Yeah.

David Friedberg 10:26 ... to be able to tag something for 100 billion here, 70 billion there, I mean, keeps swinging. And by the way, at the end of it all, what is it gonna change for him? Absolutely nothing except he'll have some incredible stories to tell about the vol that he took and the rollercoaster ride he was on. And I think that's gonna be the most valuable thing. He's gonna be able to tell his grandkids these stories and his, their, their eyes will just be like, "What?" And his friends, you know, I mean, I would love to have a dinner with him. I don't know him-

Chamath Palihapitiya 10:54 Yeah, let's do it.

David Friedberg 10:55 I don't know him.

Chamath Palihapitiya 10:55 We should have him on the pod. Come on the pod, Masa. Let's talk about it.

David Friedberg 10:57 I've only met him once, but I, I have a great story to tell about Masa when I was at Facebook, I met him once. But anyways-

Chamath Palihapitiya 11:01 Go ahead and tell it. I mean, can you tell it?

David Friedberg 11:04 All I'm saying is I would love to have dinner with him 'cause I just want to hear these stories of vol, like up and down and up and down and e- amazing.

Chamath Palihapitiya 11:11 I had two lunches with him. We had lunch one day, we had such a good time. He said, "Can we have lunch again tomorrow? How long are you in Tokyo?" I said, "Yeah." And I came back the next day and I had lunch with him twice.

David Friedberg 11:20 Oh, that's amazing.

Chamath Palihapitiya 11:21 And it was great, like talking two, three hour lunches where we're just talking about every web 2.0 company. This is in, you know, the 2009 Uber kind of Robin Hood period and he was just fascinated by what was going on and, and all the disruption happening. DoorDash Series D, $1.4 billion post-2018 he did.

Jason Calacanis 11:39 Well, the question I was asking is like DoorDash was a lot like an Arm, there's a point of view that the business has proven that there's a market that it can grow into, that there's a management team that knows how to deploy capital versus-

Chamath Palihapitiya 11:52 Oh, yeah.

Jason Calacanis 11:52 ... a lot of these early stage bets where when they got too much money, they ambled about try- well, they didn't amble about, they kind of-

David Friedberg 12:02 Incinerated it.

Jason Calacanis 12:03 ... incinerated because what do I do with this money? I don't yet have a machine or an engine that can grow with this money. And until the engine can handle the capital-

David Friedberg 12:10 Like the pizza thing. Like the pizza thing.

Chamath Palihapitiya 12:12 ... Zoom, Pizza you're talking about?

Jason Calacanis 12:14 Yeah.

Chamath Palihapitiya 12:14 Yeah, I mean if you get-

Jason Calacanis 12:15 Well, there's dozens of them in that portfolio, right?

Chamath Palihapitiya 12:17 Yes.

Jason Calacanis 12:17 If you don't have a business that can handle the gas you're putting in the tank, the engine's gonna blow up. The engine has to grow, it has to build, the volume has to build so that it can actually handle the fuel. And I think that's been part of the challenge. And I will just say, I've said this in the past, but there was an era in Silicon Valley that started around 2010, 2009, kind of right after, right as, right as Web 2.0 was starting to venture into more traditional markets, meaning tech companies were becoming other companies, kind of the mobile local era was, was kicking off. And there was a strategic concept that got popular in Silicon Valley on take the market, which is when you feel like you could be the winner, give the person all the chips and say, "Now go over bet and take the whole market." Because these are gonna be winner take all or these could be winner take most markets. So get first mover advantage, here's more capital than you need. And I remember being pitched this when I was raising my Series B, that I should raise more than I was trying to raise because I should use the extra capital to take the market. And then that mantra just became the de facto standard, which is, oh, you have something that has something. It has some edge, it has some advantage. It started to move in the market. Here's a (censored) ton of capital you don't really need. Now go take the market. And the reality is most of those businesses had absolutely no way of using that capital to quote, take any market-

Chamath Palihapitiya 13:40 To your point-

Jason Calacanis 13:40 ... because they didn't even know what their market was yet.

Chamath Palihapitiya 13:42 Yeah, they, when you're in the product market fit, product market fit phase, and you don't have the engine as you called it, how do you make the engine go faster? You're literally trying to take a bunch of car parts and then run a Formula One race. (laughs) The car's not even trackworthy yet and Katerra, uh, you know, not, not to pick on any companies, but I remember Katerra, which was doing construction and they just thought they're gonna take over the whole construction inju- you know, industry with this company. It raised like $900 million and I, I believe it's like a zero the whole way. You cannot pu- you can't put the money to use, capital as a weapon maybe works if you're Uber, Lyft, and DoorDash and you have customers and you can reduce the price, but...

David Friedberg 14:20 Well, you know, what's interesting is when you now look at all a bunch of these AI companies every day, what you see is every round is a minimum of $100 million. But the, the dirty little secret is not that these companies have found product market fit or are scaling revenue massively. I'm not talking about OpenAI, I'm talking about these other long-tail businesses. The $100 million is largely going to compute. But as we know, and I think we'll talk about this next week when all these announcements are done, but you're about to see a one-tenthing of the compute cost. And so what's interesting for those companies is they're gonna be left with a lot more money, but back to Freeberg's point, they may be ambling about with too much money at that point. So today you raise $100 million because you need to spend 70 or 80 on compute. Tomorrow when you only need to spend seven or eight for the same amount of throughput, on one hand you could say, "Well this is great. I have so much more money I can spend." But again, we've seen example after example where if you raise too much money when you don't need it, you will just run yourself into a brick wall. You'll over-hire, you'll mis-hire, you'll misallocate, and it just goes back to the same thing that we've said before, which is there's just not enough people that actually know how to run companies.

Chamath Palihapitiya 15:33 Yeah, and it just becomes a huge distraction.

Jason Calacanis 15:34 Huge distraction.

Chamath Palihapitiya 15:34 Uh, did you see Sax today?

Chamath Palihapitiya 15:36 Sora? Did you see Sora? Could try OpenAI.

David Sacks 15:38 Yeah, I was just looking at this, this is breaking news right now.

Chamath Palihapitiya 15:40 Yeah, this is unbelievable. If you haven't seen it, Sora is the text video model from OpenAI. They've been talking about this, I guess this, uh, would compete with... Is Stable Diffusion do video as well? Anyway, you give a prompt, I'll just read this one out loud. "Beautiful snowy Tokyo City is bustling. The camera moves through the bustling city street following several people enjoying the beautiful snowy weather and shopping at nearby stall, stalls. Gorgeous sakura petals are flying through the wind along with snowflakes." And this, if you weren't studying it would look like a major feature film or TV show. Yeah, Sax?

David Sacks 16:22 Yeah, I mean it's unbelievable. I mean, it looks to me like 4K video. I mean, I'm not sure it actually is 4K, but it looks pretty darn close.

Jason Calacanis 16:30 Yeah, the rendering resolution can be whatever you want it to be.

David Sacks 16:34 Right, the fact that you can describe the, the way that you want the camera to move.

Jason Calacanis 16:38 So I'm gonna speculate on this. You guys know I talked about this quite a while ago on how...

David Sacks 16:44 Yeah.

Jason Calacanis 16:44 I think we're gonna be a prompt to video, meaning people are gonna have personalized video games and personalized movie experiences and music eventually, where rather than everyone watch the same product made centrally by a producer, we'll all make our own product locally. I spent some time digging into this over the last year, and the big challenge was that the traditional approach for rendering video is you create three-dimensional objects and then you have a rendering engine that renders those objects and then you have a system that defines where the camera goes and that's how you get the visual that you use to generate a 2D movie like this. And the best tool out there is Unreal Engine 5, which is made by Epic Games. I don't know if you guys have ever seen the demos of it, but it's really unbelievable.

David Friedberg 17:25 Yeah.

Chamath Palihapitiya 17:26 Is that what Star Wars uses with their, like, virtual sets now when they did The Mandalorian? I think they're using the Unreal Engine or Unity.

Jason Calacanis 17:32 They probably are and then there's other tools-

Chamath Palihapitiya 17:34 Yeah.

Jason Calacanis 17:34 W- if you're in movie production you'll use different stuff, but Unreal Engine is just so incredible, but it requires the placement and the rendering of 3D objects. So where do those objects go? And then the physics of the light moving off those 3D objects-

David Friedberg 17:46 Yeah.

Jason Calacanis 17:46 ... to generate a 2D video is pretty straightforward. That's been done for 20, 30 years. And so the, the density, the, the rate of improvement, the color contrast, et cetera has all improved, made it look incredible, now it's super photorealistic, you can create photorealistic video. But it still requires this concept of where are the 3D objects in 3D space. This doesn't do that. This was tr- a trained model, so we don't know what's going on in the model. So how would you train a model to do this without having a 3D space?... because clearly there's thousands of objects in that rendering. If you look at that video, thousands of little objects. The compute necessary to define each of those objects, place them in 3D space is practically impossible today. So this is being done through some model that starts to render all of these things together. So it is very likely, my guess is that OpenAI used a tool like Unreal Engine 5-

Chamath Palihapitiya 18:43 Yeah.

Jason Calacanis 18:43 ... and generated tons and tons of, uh, video content, tagged it, labeled it-

Chamath Palihapitiya 18:49 Hmm.

Jason Calacanis 18:50 ... and were then able to use that to train this, this model that can, for whatever reason that we don't understand, do this.

Chamath Palihapitiya 18:57 Make the output.

Jason Calacanis 18:58 But what's interesting-

Chamath Palihapitiya 19:00 Yeah.

Jason Calacanis 19:00 ... is that this output, underneath it-

Chamath Palihapitiya 19:02 Synthetic- you're referring to synthetic training data-

Jason Calacanis 19:04 E- e- yeah.

Chamath Palihapitiya 19:05 ... is your thesis here. Yeah.

Jason Calacanis 19:05 Exactly, exactly. And, and, and, but what's amazing about this model is that underneath this, this model on its own has rendered physics. So all of the rules of physics, all of the rules of motion, all of the rules of perspective, all of the rules of how objects sit upon one another next to each other, how fluids move, if you look at some of the other demos on this tool, you can actually see fluid moving. It's rendering-

Chamath Palihapitiya 19:29 Or hair. Yeah.

Jason Calacanis 19:29 ... the movement of fluid or rendering the movement of hair. Those are, like, physics models that would typically be deterministically written, meaning that a programmer or an engineer writes the physics, and then the, the Unreal Engine renders the physics. In this case, the model figured out the physics and the model probably did it in such a reduced way that you can use a reduced amount of compute to generate something that would otherwise take a ton of compute time. I mean, look at this. It's truly amazing. And I think it-

David Friedberg 19:55 Nick, find the one of golden retrievers playing in the snow. This is way better than this one.

Jason Calacanis 19:58 But what's amazing is that the... And by the way, if you're listening to this, I really recommend going on Twitter and typing in OpenAI Sora, S-O-R-A, and looking at some of these-

Chamath Palihapitiya 20:07 It's bonkers.

Jason Calacanis 20:07 ... demo videos.

Chamath Palihapitiya 20:08 Yeah.

Jason Calacanis 20:08 It's bonkers, and I think it really speaks to what's so incredible right now, which is that software is writing itself, and it's replacing a lot of human determinism. The way that humans think about, uh, constructing software is being completely redone with these, uh, uh, learned models. This model learned physics on its own-

Chamath Palihapitiya 20:26 Yes.

Jason Calacanis 20:26 ... and it renders this thing that looks like the only way we know how to do this today is to create 3D object models in 3D space and use tons of compute to render the image, so, like-

Chamath Palihapitiya 20:33 And just to, to translate that from geek (laughs) to-

Jason Calacanis 20:36 Yeah.

Chamath Palihapitiya 20:37 ... you know, people listening, this is incredible because Pixar spent years, maybe a decade, trying to figure out hair. I remember talking to Ed Catmull, and he told me hair was just so hard to get right with wind and lighting. They literally built the physics engine for years, like, maybe even a decade to make hair cross the uncanny valley. What you're saying here, Freeberg, is somehow this language model, this image model just outputs it, and it doesn't actually have a physics model. It just outputs it, correct?

Jason Calacanis 21:07 Yeah, it, it... Well, we don't know, but the model, uh, th- there may be some determinism built into it. So, so I don't wanna assume anything. But the general approach, determinism means you write the physics and you write the formula of the physics-

Chamath Palihapitiya 21:22 Yep.

Jason Calacanis 21:22 ... so that the software can execute the physics and render the thing. But in this case, it may very well be the case, like a lot of the other models that are being trained today, when you create a model, they're trained, that it's being trained and on its own, it resolves to figure out how to do the physics.

David Friedberg 21:41 Is there a limit on the length of these Sora videos that you can generate right now?

David Sacks 21:46 It's one minute right now.

David Friedberg 21:47 60 seconds today.

David Sacks 21:48 Obviously, that'll come off at some point.

Jason Calacanis 21:49 But by the way, if you can do 60 seconds, you can do two hours.

David Sacks 21:52 Right. The problem that you have is, I mean, if we're talking about creating a movie, is making edits. So I'm obviously haven't used Sora yet because it just came out today, but I've used DALL-E, which is just their prompt, uh, image service, and I was designing the label for, or, or the bottle for the All-In Tequila for you guys.

Chamath Palihapitiya 22:11 Hmm. Yeah.

David Sacks 22:12 And what I noticed is that it would spit out a bunch of ideas, and then I would take one and then wanna just make little changes, and I would just say, "Okay, keep this, but just change this one detail." It would give me a completely new image. So, the AI doesn't realize the layers inherent in what it's producing.

Chamath Palihapitiya 22:33 Hmm.

David Sacks 22:33 Somehow it takes this prompt and generates a response, and if you give it that same prompt 100 times, it'll give you 100 different answers. And if you'd say, "Just modify this, this, or this," it can't do it.

Chamath Palihapitiya 22:45 Can't reason. It can't take criticism.

David Sacks 22:47 Well, it seems like on some level it's unaware of the output that it's created. Again-

Chamath Palihapitiya 22:52 Yeah.

David Sacks 22:52 ... uh, uh, a human creating this in Photoshop would have a lot of layers, and every object in that image would have its own layer, and you could have a layer for text and just quickly modify the text. When I wanna do something as simple as edit the text or change the font, again, it would give me a completely different image. And I just wonder about these videos. If you wanna create a movie, I think you're probably gonna be stuck with, like, the f- what they give you, and then maybe you can import that at some point into an Unreal Engine and then do work from there.

Chamath Palihapitiya 23:24 The editing part is, it doesn't exist yet. Y- you, when you do text, it's easy 'cause you just ex- you cut and paste it into your Grammarly or your Notion or whatever you use, and you start editing. Here, there's no exporting, like you're saying, to Final Cut Pro or something.

David Sacks 23:38 Yeah. I bet that when they turn this into an API that when you're working in Unreal Engine, you know, if you're, like, a VR designer or somebody like that, or you're, like, a, um, filmmaker, you could probably just get started with a prompt, and that could be hugely time-saving if they can make that integration work. I think that's pretty interesting.

Jason Calacanis 24:00 I looked at this a lot over the last year. This was one of the biggest challenges is in order to have an interactive approach to video production or video game production, you have to be able to say, "Well, take that out, put that in." But like, like you said, the model doesn't distinguish the objects. It doesn't distinguish each of those little flowers or each of those little cherry blossoms, or it doesn't even distinguish that there's two people walking down the road. The model doesn't have this concept of objects when it renders.... and so, um, the traditional model where you have objects in a 3D model that then get rendered into a video allows you to do those sorts of things. So, what may end up happening is a fusion, and this is something we talked about with Stephen Wolfram and others, that there are these models that are called compute models that do deterministic compute, uh, tasks. Then there are these AI train models where we don't really have a sense of how they work. But the fusion of the two will unlock a lot of opportunities, and that's very likely the next generation of models, is not just purely, um, non-human intervened train models, but the integration of those models with these deterministic models, with these compute models. And then you can start to do things, like instead of saying, "Hey, render this whole scene," you're saying, "Render all the objects that would go in the scene," they get placed in an unreal engine-

David Sacks 25:19 Right. Right.

Jason Calacanis 25:19 ... and then the unreal engine outputs the, the video.

David Sacks 25:22 Right.

Jason Calacanis 25:22 And so that integration work, I think, is the next big phase in AI model development and that'll unlock, like, insane (censored) .

David Sacks 25:29 Yeah, 'cause right now you can't use this to create dialogue, like actors in a movie scene talking to each other and acting, especially, like, close-ups of those actors with their facial expressions. I don't... at least, I don't see any examples like that.

Jason Calacanis 25:44 The way humans wanna have a creative process, it doesn't work. That's right.

David Sacks 25:47 Yeah.

Chamath Palihapitiya 25:47 Yeah. It'll get there though, and if you put something like this out there to the public and said it's 20 bucks a month to do this, or 50 bucks a month, you're gonna have a lot of consumers who wanna buy this product. So when I see this, I think people are gonna pay for this, and then going back to NVIDIA or ARM, okay, more infrastructure to your point, Freeberg, is like, (laughs) what, what infrastructure do we need for everybody to have Pixar on their laptop or on their, you know, phone? And people, people are willing to pay for these products.

Jason Calacanis 26:17 I don't know if you guys saw this, but there was another big breakthrough announced today with the, um, the new Gemini Pro 1.5. Can you just pull up Sundar's announcement on it?

Chamath Palihapitiya 26:25 Yeah, Sundar now has found the password to his Twitter handle. (laughs) He's, he's tweeting.

Jason Calacanis 26:29 Yeah. The reason I wanna talk about this, if you just scroll down, so when you prompt an AI model, you put in a certain number of words into a, like, an LLM model today and it spits words back out. The more words you put into the model, the more compute intensive or challenging it is to do the compute work needed to render the output. So, OpenAI, their GPT-4 Turbo model has 128,000 tokens, which means about 128,000 words can be fed into the model to get an answer back out. So you could put, like, a whole essay or 100 pages of text and it will read a document for you.

Chamath Palihapitiya 27:07 You could put a book. You could put a book-

Jason Calacanis 27:08 You could get, get a whole book and, and it could give you a summary in a couple of seconds. What Google just announced with Gemini 1.5 Pro is a one million token context window, up to 10 million token context window in total. I don't know what's going on here because the problem is the way these models are kind of traditionally built is that that creates a non-linear scaling and compute needs to be able to increase that, that context window like that. So there is some architectural work that's gone on to allow these models to scale up in a way that we did not think possible when these models first came out, which is crazy to think was only, like, 24 (laughs) months ago that they started to get kind of looked at like this.

Chamath Palihapitiya 27:45 Hm. 18 months maybe, yeah.

Jason Calacanis 27:47 Yeah, and so there's two kind of things that are happening. One is that these models are getting chopped up in a way that you can run small models in a highly trained way locally, and you don't need a massive general-purpose model, and so that's gonna make compute needs very small for very specific tasks. And the other one is that there's architecture that's happening in deploying the models that's allowing massive context windows like this, so it totally changes how scalable they are with the same number of chips, or the same amount of compute, or the same amount of power. So, there are all these amazing Moore's Law effects happening right now in AI technology development on the chip level, on the architectural level, on the, how they're being built and how they're being deployed that completely changes the concept that I think most folks had when these first came out two years ago, and the potential. So I think sky's still the limit. We don't really know how big this is gonna get, but all of these efficiency gains mean that we can do so much more with so much less. And by the way, when you can do more with less, it doesn't mean you need less chips. It means you need a lot more chips, because the fact that you could do a lot more with much fewer chips means everyone's gonna wanna do it now, e- more people can afford to do it. So, it unlocks the opportunity for, uh, for compute, and so it really starts to rationalize the NVIDIA thesis that it could be worth $10 trillion potentially.

Chamath Palihapitiya 29:01 Um, for people in the audience to understand what this context window means-

Jason Calacanis 29:04 I'm sorry if I was too technical. (laughs)

Chamath Palihapitiya 29:05 No, no, no, no.

Jason Calacanis 29:06 (laughs) Yeah.

Chamath Palihapitiya 29:06 I, I'm just... I always like to just give an example or two after you explain it. The context window, for example, you could upload the transcript of this podcast, right? And there were only certain services, like Claude maybe, where you could put a PDF attachment. Boom, you put it in there and say, "Hey, summarize this using your language model and using this content." Now you could be able to say, "Hey, take me every MrBeast video. Take me, y- you know, take the top 10,000 videos on YouTube, take the transcripts, summarize them for me, give some analysis on it," and that was only available to developers prior to these large context windows being open, and that's gonna just open up a massive amount of application. Right, Shoma?

David Friedberg 29:47 The results are actually not that much better. There's these third-party service benchmarks. Lambada is probably the most often referred to, but what you see if you just look at, for example, all the GPTs, even as they increase the number of parameters in a model, the quality of that model doesn't necessarily increase with the rate of parameter increase. So, you have this thing of a little bit of diminishing return, which is more a problem of what we said before, which is eventually when you're scouring the open internet, doesn't matter how fast or how, how big you scour the open internet. The open internet is a bounded space, and so you get the same answer. So, all of these things are... they're interesting marketing war, but I would just encourage us all to keep in mind that the shift is not necessarily these kinds of things.It's kind of like saying, you know, in a chip, like, "Oh, wow, we built this ship on three nanometer." Okay, what does it mean? Nothing.

Chamath Palihapitiya 30:42 Well, no, I think the piece, respectfully, you might be missing here is, like, your proprietary data might not be in that language model. So by having this context window and you can add something that you have on your hard drive...

David Friedberg 30:54 But that's not, that's not how it works. So my point is, the context window is not correlated to the quality of the model outputs.

Chamath Palihapitiya 31:00 Correct. But consumers, uh, and business users haven't been able to put their data into the models because they're not indexed to-

David Friedberg 31:06 But that has nothing to do with the context window. That existed when the context window was smaller but that's what you d- thing you do things like RAG with.

Chamath Palihapitiya 31:13 Right. But most people don't have access to that. I- this is like consumers can now just take something off their hard drive and put it up there.

David Friedberg 31:18 I understand what you're saying.

Chamath Palihapitiya 31:18 Yeah.

David Friedberg 31:19 RAG is different than what this is.

Chamath Palihapitiya 31:21 Okay.

David Friedberg 31:21 Like this is ... y- y- l- this, what, what you guys just spent all this time talking about, is equivalent to saying, "I built this chip with ... on a seven nanometer process or a three nanometer process." And what I'm saying is, it is technically more burdensome, as Friedberg said. It's technically much more sophisticated and complicated, but the, the translation to quality in the consumer is not what that means. So, a three nanometer chip is not two times better than a seven nanometer chip for the consumer. Similarly, a context window that's five times bigger than the one before it doesn't make a model five times better to the consumer. These are orthogonal ideas. It's not to say that it isn't an important thing, but I, I, I would say that it's a reminder for me that we're in the early phases where there's a lot of hay that people can make. But it's important to look at these third-party benchmarks that try to sort through and sift through what matters and what doesn't matter. And one of the interesting things that third-party benchmark services have figured out is that irrespective of the exponential growth of these things, like parameter size, quality is only growing linearly and now sublinearly. So, it's just an interesting point.

Chamath Palihapitiya 32:40 I understand your point now. My point is, man, when we s- when you are a consumer of this, and I've been talking to people trying to get their data into, you know, the context window to be able to ask a question, they're always like, "How do I put more in there? How do I put more in there?" And then these language models, because of the copyright issues, they're not able to upload Blade Runner and then say, "Hey, make me something." But I was able to. A- and just to give a very real example, I did, um, a podcast around the Ray Kroc movie, The Founder, that we were referencing earlier. I uploaded the screenplay, I found a PDF of the book, and I found two other, like, essays about Ray Kroc, like HBS studies, and I said, "Give me, like, a timeline of it." And I used something called Google Notebook, which has a big context window and you can upload all this stuff. And that's where I'm seeing consumers and what these things are capable of vis-a-vis the context window. It- it just unleashes in people's minds so many applications that you can do.

David Friedberg 33:38 There are two examples today that caught my eye that I think are worth bringing up. The first one is from Meta. Meta published this paper today, which was, like, pretty crazy. And essentially what they said is that for the realm of the software worker, the software engineer, what they basically built is this LLM tool that runs over LLaMA called TestGen. And it essentially does automated unit testing to such a high degree of accuracy. So, you know, in the software development, if you wanna spit out a product, you have to go through this thing where you have to functionally run unit tests and make sure the thing works as advertised. And this paper essentially now allows you to traverse a code base, understand it, generate these unit tests with high accuracy. So, where is this valuable? Think of, remember the software bugs in, like, the Boeing 777-37 MAX issue, J-Cal?

Chamath Palihapitiya 34:29 Yeah.

David Friedberg 34:29 Where it turned out that it was just bad software. Tools like this now will able, will be able to run unit tests where you will not be able to get out- get that plane out the door unless all of these functional unit tests not only have been written, but run and have been passed successfully. And it'll all be done by bots, which means that you can't game that system. So, that's, that's something today that Meta released-

Chamath Palihapitiya 34:53 And it's the last thing developers get to, is writing the unit test, right? It's always like-

David Friedberg 34:56 Which is why there's all kinds of bugs and security holes in code. Now, I thought that that was a pretty large breakthrough and I think we're gonna go through a bunch of fine tunes on this that make it interesting. And then I saw this other thing, which caught my eye even more, which was this thing that Nat Friedman tweeted out, which is this thing called magic.dev. And what essentially it is, is like a coworker. And what they are claiming, and I don't know if it's true, is that now you can basically have this thing understand your style of code and actually be able to write code that is just as good as the way you write code. So, now it's not a copilot, but effectively a coworker. And this is the first real example of what I've seen of a claim that basically says, "Hey, you know, we can now be an add-on to your workforce." And this, to me, is just so gargantuan because, like, if you look at the amount of OpEx inside of any company that sits inside of R&D, if you look at the amount of stock-based compensation, right? If you replace humans with bots, which is what this is, you're gonna crush those things. SBC, at the limit, goes to zero because you don't have to give stock to a bot, right? (laughs) You don't have to give stock to people.

Chamath Palihapitiya 36:14 Yeah.

David Friedberg 36:14 And OpEx ge- gets shrunk because now you just pay a license to magic.dev, or I don't know how they're gonna make money on this, but really, really, really fascinating stuff. It is a scary world.

Chamath Palihapitiya 36:25 Well, there was another open source project, uh, I think it's called Maggie, M-G-I-E. I don't know if you saw that, Friedberg, but Apple has been working on an open source image project. And so you're kind of gonna see, I think, in iOS 18 or the next iPhone 16-... they're gonna have an image model on-

David Friedberg 36:43 Jakeo, where-

Chamath Palihapitiya 36:43 ... baked onto that phone.

David Friedberg 36:44 Where is, where is Apple in all of this? It's unbelievable.

Chamath Palihapitiya 36:48 No, they, they're-

David Friedberg 36:48 Where are they?

Chamath Palihapitiya 36:50 They are building an image model and you're going to be able to talk to your phone and say-

David Friedberg 36:54 They haven't announced anything.

Jason Calacanis 36:55 I think you're gonna see a chip on-

Chamath Palihapitiya 36:57 Yep.

Jason Calacanis 36:58 ... the next set of mobile devices-

Chamath Palihapitiya 37:00 100%.

Jason Calacanis 37:01 ... by Apple that will run models locally.

Chamath Palihapitiya 37:03 Yep.

Jason Calacanis 37:03 And so this goes back to the point about how the hell is the government going to regulate AI models, which is the stupidest thing I've ever heard. Um, well, not the stupidest. It's probably, you know, probably top 20. But-

Chamath Palihapitiya 37:14 (laughs)

Jason Calacanis 37:15 ... the fact that they're gonna, like, be able to be fully vertically designed, that they can have the hardware and the software and integrate all the way through to the application layer, they're gonna have, I would imagine, incredible, intuitive AI features built into all of their products that can run locally rather than all the other stuff that's going on where you've got to make a request on the internet and get the results back over the internet. It'll be pretty powerful. But obviously no data, right? We don't know what they're doing. We're just speculating here.

Chamath Palihapitiya 37:43 Well, no, I mean, you have breadcrumbs, and the breadcrumbs are they have an image model, and I just dropped it in the hugging face of it. You can basically upload an image and just give it a text prompt of how to change the image. So, you know, if Nick were to use this, he could upload a picture of you, Chamath, and say, you know, "Give him five o'clock shadow, give him a mustache," whatever, and actually-

Jason Calacanis 38:01 So, Nick, Nick, will you do that real quick while we're talking? (laughs)

David Sacks 38:04 While you're doing that, can you bring up the video of the robot's life in a cyberpunk setting?

David Friedberg 38:10 Yeah, it's beautiful actually.

Jason Calacanis 38:10 Is this a Sora rendering?

David Sacks 38:12 Yeah, this is Sora.

David Friedberg 38:13 So beautiful. So detailed.

David Sacks 38:14 It's really cool. So but-

David Friedberg 38:16 Wow.

David Sacks 38:16 ... here's the, the limitation, and this goes back to what me and Friedberg were talking about earlier. Okay, so you've got these people in the background. Let's say-

Jason Calacanis 38:24 Look at their- look at their feet-

David Sacks 38:24 ... you're trying to make a movie-

Jason Calacanis 38:25 Their feet are backwards, by the way. You can see the human feet aren't correct.

David Sacks 38:27 Okay.

Jason Calacanis 38:27 Sort of like a DALL-E problem.

David Sacks 38:29 Yeah.

Jason Calacanis 38:29 But anyway, sorry, go ahead.

David Sacks 38:30 What's cool is that, okay, so those cars aren't moving. They're static. You got some movement of these background people, right? At least so they're kinda... they're the extras in a movie. But if you wanted to tell Sora, this is just my guess, that, "Okay, I want the people in the background to be doing this," it would render a whole different image. Like, your robot might change too, right?

Jason Calacanis 38:53 Well, Sax, this is where your context window matters. So if you-

David Sacks 38:56 Mm-hmm.

Jason Calacanis 38:56 ... repeat the statement, if y- instead of writing and then iterating, if you just re-prompt it and your prompt got longer and longer and longer with the degrees of specificity you needed, that's the way you use these models today. And to your point, what humans are looking for from a UX perspective is an iterative approach to rendering. It's an iterative-

David Sacks 39:16 Yeah.

Jason Calacanis 39:16 ... approach which isn't feasible in a world where the... this i- this is all a matrix. It's a three-dimensional matrix. There's two dimensions of pixels, and then the third dimension is the next image and the next image and the next image. And so that matrix is re-rendered from scratch. It doesn't have a concept of what are the objects that I'm rendering.

David Sacks 39:36 Right.

Jason Calacanis 39:36 It has a concept of a matrix of pixels.

David Sacks 39:38 Right.

Jason Calacanis 39:38 And so it will re-render a matrix of pixels based on a redefined prompt, but what you're describing would d- require a very different model architecture than the way all of these trained models operate today.

David Friedberg 39:48 This is gonna revolutionize pornography.

David Sacks 39:51 (laughs)

David Friedberg 39:51 I think it's the first pla- no, no, no, I'm not saying it as a joke. I think that's-

Jason Calacanis 39:54 Yeah.

David Friedberg 39:54 ... where you're gonna see this first because then all these issues, potential issues of underage people or exploited folks, it goes completely away and-

Jason Calacanis 40:02 (laughs) Your PR guy's gonna cut this out. (laughs)

David Friedberg 40:05 No, and, and, you know, it, it'll kill OnlyFans.

Jason Calacanis 40:08 Yeah, I was about to say, end of OnlyFans.

David Sacks 40:09 Well, look, porno gambling is where every revolution starts. That's where-

David Friedberg 40:12 Exactly.

David Sacks 40:12 ... that's where-

David Friedberg 40:12 Technological, that's where payments-

David Sacks 40:14 ... worldwide web started.

David Friedberg 40:14 That's where the web started. That's where payments got refined. But I, I think that in pornography, you're gonna see this basically destroy pornography and OnlyFans.

Chamath Palihapitiya 40:23 Well, I don't, I don't wanna bring up something, uh, that's super offensive, but there was a certain pop music star who had a bunch of images just a couple of weeks ago that were AI generated.

David Friedberg 40:34 Yeah, Taylor Swift.

Chamath Palihapitiya 40:34 Okay, well-

David Friedberg 40:35 AI fakes, deepfakes, right?

Chamath Palihapitiya 40:36 Yeah, there were a bunch of deepfakes of a pop star. I don't tell, like, actually highlight the person's name because then it just, I think it sends more of it. But X did a really good job of blocking all of them, watermarking them, and saying, "You can't do this." And so this is like a, a whole new world. You can take any person on the planet and make porn out of them that is realistic.

David Friedberg 40:55 Well, I don't, I don't mean that. I just mean that you'll be able to make pornography in a way that is totally-

Chamath Palihapitiya 41:01 Yeah.

David Friedberg 41:01 ... uh, non-human involved, which is generally better, I would say, for all humans involved. (laughs)

David Sacks 41:07 What's great about the deepfakes is that if you have a sex tape come out, you can just say it's a deepfake.

Chamath Palihapitiya 41:12 Totally. I was about to ask Chamath, what is the prompt you're gonna use to make love to yourself tonight- (laughs)

David Friedberg 41:17 (laughs)

Chamath Palihapitiya 41:17 ... with the new Sora model? (laughs) Chamath embraces Chamath. (laughs)

David Friedberg 41:22 All, all I hope is... I, I may just need to publish some data so that whatever deepfake gets some parts accurate.

Jason Calacanis 41:27 This is Apple. Oh, short Apple stock. This is terrible. (laughs)

Chamath Palihapitiya 41:30 It's not very good.

David Sacks 41:31 That's not very good.

Jason Calacanis 41:32 What is this model? This is an Apple model?

Chamath Palihapitiya 41:34 It's called Maggie, M-G-I-E. This is Apple's open source image model. It's way behind.

David Friedberg 41:40 This looks like GPT 0.1.

Chamath Palihapitiya 41:42 They got work to do, but it's open source. It's gonna go faster.

David Sacks 41:44 It's very hard to make fine tune changes in these models or to get the result that you want, but that's fine. This is just a... this is v0.9, so.

David Friedberg 41:52 If you guys want to see a great fine tune, the JuggernautXL fine tune-

Chamath Palihapitiya 41:56 Uh-oh.

David Friedberg 41:56 ... I would just say try it. It is a fine tune that does not allow bad-looking people. (laughs) The whole, the whole concept is only beautiful people. So it doesn't matter what your prompt is, but these people come out-

Chamath Palihapitiya 42:10 It's gonna be gorgeous, yes.

David Friedberg 42:12 ... looking the most incredibly good-looking people I've ever seen.

Chamath Palihapitiya 42:14 As if the standard wasn't, uh, already high enough with filters and everything else, but...

David Sacks 42:19 This is gonna be revolutionary for independent film. I mean, this is obviously just tip of the iceberg. But, you know, the problem in independent film right now is that the economic model literally doesn't work. It's too expensive to make an independent movie relative to the return. And so the independent movie business has almost dried up, almost completely. This is gonna revolutionize it because it'll bring the cost down close to zero. Now, I, I don't, I don't agree with Friedberg that everyone's gonna design their own entertainment because...... most people are lazy, but-

Jason Calacanis 42:48 No, I'm not saying design. Sax, the way you think about creative production processes is what's gonna change. It's all gonna reduce down to sitting down, and it's like when you ask your spouse or partner or whatever, "What do you wanna watch tonight?" And they answer something, that something will then be delivered to you instantly. That's what's gonna be so powerful.

David Sacks 43:05 Well, y- you mean the AI's gonna create it on the spot for you?

Jason Calacanis 43:08 It certainly will, but I, I think that there's a certain element of culture that gets integrated into personal AI. Uh, I, I've, I mean, I, I don't think we have time for this, but I have like a whole philosophy.

David Sacks 43:19 People still want curation. I mean, if you look at something like YouTube, um, or really any user-generated content site, there are roughly 100 times more people who are passive than people who are active and actually create stuff, whatever the ratio is. There's always a lot more passive viewers than there are creators. So, there's, I'd say most of the content will be created by people who actually care enough to do it. Again, let's call it 1 or 2% of the population. And then of those, 1% might actually be really good at it.

Chamath Palihapitiya 43:48 Yeah, but here, here's where Freeburg, I think, has a really good point 'cause I have been looking into this. Imagine if you took The Sopranos, something you and I love, Sax, and you just said, "Hey, for each season, give me 10 more minutes per episode and give me, like, two-"

Jason Calacanis 44:03 Exactly.

Chamath Palihapitiya 44:03 ... "more characters or whatever." Now, you, when you watch it the next time, you can say, "Just fill it in. I want every episode to be 90 minutes instead of 50 minutes."

Jason Calacanis 44:11 Right.

Chamath Palihapitiya 44:11 Or y- y- I've been doing this thing, uh, with my daughters. We're into Star Wars and The Clone Wars, et cetera. People have been cutting shows like Andor or Rogue One, Clone Wars, the prequels, et cetera, and The Bad Batch, and they're putting in together, like, timelines, and they're making their new versions of it by super-cutting the different series that are filling things in. So, you're gonna be able to say, "Hey, tell me the whole Star Wars saga with a little more perspective from Obi-Wan," or, "Give me a little more perspective from this person." And that, that's gonna just make these things, like, these franchises so much more watchable and deep and rich.

Jason Calacanis 44:49 Yeah, so this is my point, Sax, is I don't think it's about ground-up personal media creation. I think that it's about the curation model changes. The creator or the central creative force comes up with a genre with a bunch of characters, like a Star Wars, and general, like, arcs of stories, and then you can go in and you can view it from any character's point of view. Imagine watching the Star Wars movie from Darth Vader's point of view. Imagine watching the Star Wars movie where you just stayed on Endor the whole time, and you watched the news on the TV (laughs) like, while you were hanging out at the bar.

Chamath Palihapitiya 45:25 Wow. That's your fantasy?

Jason Calacanis 45:26 Yeah.

Chamath Palihapitiya 45:27 You with the w- you with the Ewoks, Freeburg? (laughs)

David Sacks 45:28 Yeah. Uh, so-

Jason Calacanis 45:30 Yeah.

David Sacks 45:30 So you're right that there's a great opportunity for some of these cinematic universes to kind of open up the creative assets and let the fan base do whatever they want with them or create, you know, these alternatives. I still think that it's that 1% of the viewership that actually cares enough to do that. And the reason is because that, I think w- what you see with all these, uh, demos is that it's really easy to get the first 90%. It's really hard to get the next 9% and then the final 1%. You know, to get to the level of-

Jason Calacanis 46:01 Of course, yeah.

David Sacks 46:02 ... t- to get it dialed in to the point w- again, the, the first thing it gives you-

Jason Calacanis 46:06 But the, the benefits could outweigh those costs. And this is true in 3D video game playing. I don't know if you play Fortnite, but there's all sorts of stuff that you feel could be better, but the overall value of that experience far outweighs the missing elements of it not being truly photorealistic. And I think that's where this stuff goes.

David Friedberg 46:23 It's ski week next week.

Jason Calacanis 46:25 Mm-hmm.

David Friedberg 46:26 And I've done this for all, both of my older boys, and I'm doing it now for my youngest boy who is four and a half, almost five. We are going to watch, this entire week, all of the Star Wars movies.

Chamath Palihapitiya 46:40 Oh, fantastic.

David Friedberg 46:41 And then I got a downloaded history-

Jason Calacanis 46:43 The feature movies.

David Friedberg 46:44 ... of all the movies-

Jason Calacanis 46:44 Yeah.

David Friedberg 46:44 ... yeah, from... And I got an order. The order was not what I thought. It's not as released, obviously.

Chamath Palihapitiya 46:50 No, you wanna do timelines, so you wanna have the prequels.

Jason Calacanis 46:54 I did that once, and I will tell you, it was-

Chamath Palihapitiya 46:56 And Andor-

Jason Calacanis 46:56 ... super disappointing.

Chamath Palihapitiya 46:57 ... Rogue One-

Jason Calacanis 46:58 Yeah.

Chamath Palihapitiya 46:58 ... uh, the Star Wars, the original sequel.

Jason Calacanis 47:00 Yeah.

David Friedberg 47:00 What, you're saying this order is not good? Here's what I was told. This is what was, the internet was the most popular. You start with The Phantom Menace.

Jason Calacanis 47:08 Yeah.

David Friedberg 47:08 Then Attack of the Clones.

Jason Calacanis 47:10 Okay.

David Friedberg 47:10 Then Revenge of the Sith.

Chamath Palihapitiya 47:12 Revenge of the Sith. Yeah.

David Friedberg 47:12 Then Solo is optional.

Chamath Palihapitiya 47:14 Optional, yeah.

Jason Calacanis 47:15 Yeah.

David Friedberg 47:15 Rogue One, optional.

Jason Calacanis 47:16 Yeah.

Chamath Palihapitiya 47:17 Rogue One, not optional. Mm-hmm.

David Friedberg 47:18 Then IV, V, and VI, obviously. A New Hope.

Jason Calacanis 47:20 That's when your kids'll lose their mind. They'll be like, "What is this?" (laughs) It's like-

David Friedberg 47:24 Empire Strikes Back, Return of the Jedi.

Chamath Palihapitiya 47:25 IV, V, and VI are too slow for kids.

Jason Calacanis 47:27 Yeah, that's the problem.

Chamath Palihapitiya 47:27 They like the prequels better.

David Friedberg 47:28 VII, The Force Awakens, and VIII, The Last Jedi, then IX, The Rise of Skywalker.

Chamath Palihapitiya 47:33 Skip the last three.

David Sacks 47:34 Yeah, the last three are garbage.

Jason Calacanis 47:35 They fall asleep-

Chamath Palihapitiya 47:36 Garbage.

Jason Calacanis 47:36 ... during IV, V, VI.

Chamath Palihapitiya 47:37 Oh, God. Yeah, IV, V, VI are the best.

David Sacks 47:39 The prequels are probably the most watchable.

Chamath Palihapitiya 47:41 They're very, they, in, in retrospect, they're great, but there's also The Clone Wars, which is absolutely fantastic. I would sprinkle in some Clone Wars episodes.

David Friedberg 47:48 Nat and my daughters, I'm forcing them to sit with us for this entire-

Jason Calacanis 47:53 Nat's gonna watch all this?

David Friedberg 47:54 No, she's ready to shoot herself in the face.

Chamath Palihapitiya 47:56 Yeah, thi- this is her worst nightmare.

David Friedberg 47:57 She and my daughters are like-

Jason Calacanis 47:58 I put the over-under on Nat sticking through maybe 100 minutes of this.

Chamath Palihapitiya 48:02 (laughs)

Jason Calacanis 48:02 And she's done. She's done.

Chamath Palihapitiya 48:04 Amori, why do you poison the kids with this nerd stuff? They're never going to get girlfriends.

David Friedberg 48:10 You told me it was Frederico Fellini. What, this is not Fellini. This is, uh, George Lucas.

Chamath Palihapitiya 48:14 How about that they watch something where the people make love instead of the light saber cut people hand off?

David Friedberg 48:18 Why are you watching this? Exactly.

Chamath Palihapitiya 48:20 Eh, no, there's no love and joy, uh, there's, uh, people with, uh-

Jason Calacanis 48:24 (laughs)

Chamath Palihapitiya 48:24 ... sword, like, light swords. It's no fun. But to bridge the gap between your two visions, Sax and, uh, Freeburg's, (laughs) sorry.

Jason Calacanis 48:32 (laughs)

Chamath Palihapitiya 48:33 Just, Nat's point is so great. Chamath and I will talk, when we talk to each other, for literally 20 minutes as Nat. We'll just have a conversation-

Jason Calacanis 48:43 (laughs)

Chamath Palihapitiya 48:43 But it's a conversation we're having. "I don't know why, Chamath, this deal has...... uh, such a high interest rate on the convertible note. So why would you do that to me? I'm your friend.

Guest 48:53 Who's going to be on the board? This person looks like a dum-dum. What's going on?

Jason Calacanis 48:58 (laughs)

Chamath Palihapitiya 48:58 I don't understand. Well, it's an independency, Chemo. You don't need to have to put a super smart person, just a different perspective.

Guest 49:05 This product is the opposite of a white truffle. The white truffle is delicious. This product smells like poo.

Jason Calacanis 49:11 (laughs) But anyway, between Sacks and Freeberg (laughs) , we're literally just talking, uh, for hours.

Guest 49:18 Hey, when are you coming tonight, (beep) ? Are you making it for dinner or not? We need to-

Chamath Palihapitiya 49:21 I have to... I'm not... I'm gonna try to make it by dinner. I just have, uh, I have to give a keynote at a podcasting conference.

Guest 49:26 Oh, my God. Are you... Wait, sorry, are you kidding me?

Chamath Palihapitiya 49:29 No, I'm not kid... You guys banned me from having advertising here. I have to make cheddar. I got bills to pay. Just because there's an eight in front of the Uber stock price doesn't mean I can retire, but it does. (laughs)

Guest 49:39 Will you be there by 7:00 PM or not?

Chamath Palihapitiya 49:42 I'm going to make it by 7:00. Yes. I... Keep me in for whatever hap-

Guest 49:46 So we are going to start dinner when you get there.

Chamath Palihapitiya 49:49 Perfect. I will not do David Sacks time. Okay?

Guest 49:52 Thank you.

Chamath Palihapitiya 49:53 But between Freeberg and Dave, you know, Sacks, you said it's like one or 2% of the audience. For, you know, show with a million viewers or 10 million viewers, that could be tens of thousands of lunatics. And I don't know if you knew this, but you remember when Luke Skywalker showed up, spoiler alert, in The Mandalorian, everybody was like, "Oh, his young face does not skip the uncanny valley." This is why you got to watch The Mandalorian. So then some kid did his own AI re-rendering of Luke Skywalker, you know, after Return of the Jedi, but, you know, after he defeats the emperor, and they then hired that kid to work at Disney and LucasArts. So what's happening is all these people who are making content for the Star Wars universe-

Jason Calacanis 50:35 That's awesome.

Chamath Palihapitiya 50:35 ... they are going to be the next generation of creators. So between a MrBeast way over here and LucasArt, there's going to be a new fill in, and it's going to be all these great one percenters making interesting (beep) with IP that's owned by other organizations.

Jason Calacanis 50:50 I think it's a whole new paradigm for media and content. It's going to be so different.

Chamath Palihapitiya 50:54 The automated stuff that's customized for you. Yeah.

Jason Calacanis 50:56 It's almost like what YouTube and Instagram did for user-generated content. There's going to be this whole new model of AI-generated content. The, the real winners are going to be the technology platforms that bring these tools in a simple, intuitive way, that, to Sacks' point, doesn't challenge the user to do work, but makes their life feel easier and better, and brings them a more rich experience than they're getting today from any other media source. And that's going to be a game changer.

Chamath Palihapitiya 51:22 All right, lots going on in AI.

Chamath Palihapitiya 51:24 I thought I'd bring up a really interesting story that happened just around stock options and employees and advice you're going to get, generally speaking. Just tons of chaos at the startup Bolt. If you don't remember, that's a one-click checkout startup. It was founded by a guy named Ryan Breslow. He's been on my other podcast, This Week In Startups. He went viral in 2022 after he called YC and Stripe, like, mob bosses and had this whole conspiracy theory of how they were trying to kill his company. Don't know if it's true or not. But at its peak, Bolt was valued at $11 billion, Chamath, and they're buying back shares at a $300 million valuation, down 97%, yada yada. What makes this interesting was that in 2022, Breslow announced a new stock option program for his employees on Twitter and he did this in a very bombastic, lecturing kind of way to the entire industry saying that Bolt would offer loans to any employee to exercise their options early. We all know about these type of loans.

Guest 52:22 Meaning, from a, from their balance sheet?

Chamath Palihapitiya 52:24 That is the idea here, yes. And he called it the most employee-friendly plan possible and said that half of Bolt's 600-plus employees took the deal. Uh, he since deleted the entire thread, but at the time, uh, a lot of VCs who'd been through this before, you know, explained to Ryan this is a terrible idea. Here's GGV's Jeff Richards. For what it's worth, almost every private company in Silicon Valley did this in the '90s, was an absolute disaster. Employees spent years paying back loans for worthless stock, tax bills from merely exercising, et cetera, yada yada. And his now-deleted tweet was, "Over half our employees chose this, plus I would strongly encourage my family and friends to choose this, but VCs say it didn't work in the '90s, so it's a disaster." VC Twitter pumps the tweet, "This is why VC-run companies are never able to make strides for employees." Obviously, all those options, at 11 billion, are worthless now.

Guest 53:15 But sorry, they did this when they were worth 300 billion?

Chamath Palihapitiya 53:17 11 billion.

Guest 53:19 Whoa.

Chamath Palihapitiya 53:19 And it was reported that at the same time, in a, a new industry publication, that Breslow was selling his shares in secondary. Breslow confirmed with that newsletter that he had sold 10 million and, uh, sources, uh, anonymous sources, yada yada, take that for what it's worth, said he didn't tell the team he was doing that. Two investors also accused the company of misleading shareholders and raising capital on inflated metrics. SEC investigated, but did not recommend any action. Sacks, Chamath, I don't know if either of you want to take this. This is really mechanical, but it does speak to what happens when things implode. Should employees be taking loans to buy these kind of stocks?

Guest 54:01 Yeah, at a minimum, hopefully these loans are non-recourse. And at the maximum, if they were recourse, but it looked like these guys were doing these kinds of shady things, they have to kind of forgive the loans and eat it.

David Sacks 54:14 There's a reason why almost no one does these types of exercise loans anymore. I mean, whoever the VC was who pushed back was right. I mean, these things were common 20-something years ago. People stopped doing it. And the reason why is that, well, there are a couple of reasons. So the mechanics of the loan are that the company loans you the money to pay down the exercise price, the strike price for the options, and then you're supposed to pay the company back at some future date, basically when you can sell the shares. And in effect, from the point of view of the balance sheet of the company, this was cashless, because it would loan the exercise price to the employee who would then exercise and pay that money back to the company. So everyone's like, "Well, what's wrong with this? This just seems like...... simple mechanics where no one loses. The employee gets to start the clock on their shares for long-term capital gains. That's why they would do it. The problem is that what people discovered is that if the company didn't work out and those shares became worthless, then the employee is still on the hook for the loan. And if they don't pay the loan back, then there's a loan forgiveness issue. In other words, the forgiven amount of the loan, or the defaulted amount of the loan, becomes ordinary income. And so what happens is you end up having to recognize as income the loan that you never pay back. So you can lose. Furthermore, if there's spread at the time that you exercise between the strike price and the fair market value of the shares, then although if, if it's the right kind of option, if it's an ISO, you can defer paying tax on that spread. However, people got whammied by the AMT, that that spread still counted for alternative minimum tax. So there were a lot of people during the first dot-com bubble who exercised thinking they can't lose because the market was always up into the right, and then the company would go bust, and they would get hit with a big AMT and then they would get hit with a loan forgiveness issue. So that's why everyone stopped doing these things.

Chamath Palihapitiya 56:11 Yeah.

Jason Calacanis 56:11 Yeah. I mean, really importantly, getting taxed on that spread, you owe that money no matter what.

Chamath Palihapitiya 56:16 No matter what.

Jason Calacanis 56:17 No matter what.

Chamath Palihapitiya 56:17 No matter what.

Jason Calacanis 56:18 The IRS demands that money, and the Cal- California Franchise Tax Board demands that money. So l- and, and I just wanna be very specific so people understand this. If the value of your company's gone up to $2 a share as a private company and your options are 10 cents a share, you pay 10 cents a share to get those shares, you owe tax on the $1.90 difference. And ultimately, if the company's now worth less than a dollar when you're finally able to sell or it shuts down, you actually have money you have to pay to the IRS and Franchise Tax Board, and you did not make enough money to cover the tax bill. The way a lot of these companies do this is they do a cashless early exercise or cashless exercise loan, meaning you don't even have to put money up to, uh, to pay the 10 cents a share. The company is just covering it for you. To Sax's point, you end up getting taxed for that later if you don't pay it back. So the whole thing creates a lot of tax obligations, and it's a very complicated structure and system. And the reason options exist is so that people could buy their options and not have to pay taxes at the time that they receive them. Because if you just get shares in a company rather than options, the value of those shares is your tax bill even if you can't sell the shares.

Chamath Palihapitiya 57:30 Well, it's income.

Jason Calacanis 57:30 So... It's income. And so options were created as a way that you would have a strike, and the strike was the fair market value. So you're actually not getting any value today, and so you can avoid the tax bill. But now it leads to all these other tax loops, all these other tax consequences. Yeah.

Chamath Palihapitiya 57:44 This is way too complicated for rank-and-file employees, is what it comes down to. Like, I-

Jason Calacanis 57:47 Totally.

Chamath Palihapitiya 57:47 ... I saw this with, like, senior executives trying to get, like, the top five positions, you know, a CFO or something, or CTO might do this.

Jason Calacanis 57:53 And the whole thing is to, is to reduce your tax bi- bill-

Chamath Palihapitiya 57:55 Yeah.

Jason Calacanis 57:55 ... from 50% to 20%.

Chamath Palihapitiya 57:58 Yeah.

Jason Calacanis 57:58 And at the end of the day, you open up all of these other problems if you try and do that.

Chamath Palihapitiya 58:01 Completely.

David Sacks 58:02 One of the stupidest things about the way this was used is that they would give the exercise loan to employees who were leaving the company because they said that, you know, normally when you're a, a departing employee from a company, you have three months to exercise your options or you, or you lose them. So they were saying that this would be a solution to that, right? "We'll loan you the money to exercise." All they were doing is creating a huge tax problem for that former employee. And you're right. The average employee is not sophisticated enough to navigate all of these issues. It's not clear also why you'd want to create this complicated benefit for departing employees. It seems like if you're gonna do this, it- you would wanna do it for a key exec who's joining the company. That's the only context in which I've seen it, by the way.

Chamath Palihapitiya 58:47 Yeah, same. Yeah.

David Sacks 58:48 So they were trying to solve a problem using this complicated device that, that really backfired, but it's not even a problem that you should try and solve in that way. I mean, there's a much simpler solution, which is extend the period of time that employees have to exercise. If you don't want them to lose their option if they don't exercise within three months, give them 10 years. And there are a bunch of companies that have done that, where they give you a lot longer to exercise the option.

Chamath Palihapitiya 59:12 The trend I'm seeing with startups, like early stage startups, seed stage startups, which are really indicative of what startups will do later on, hiring people, cash basis, paying them hourly, outside the United States, not having to worry about stock options, not having to worry about healthcare, just finding a global workforce and using these third-party tools. There's a lot of third-party tools to manage employees in Portugal or Manila or Canada or whatever, South America, and they're just saying, "You know what? Stock options are gonna be for, like, an elite group, the top 10% of the company, and then we're gonna automate the rest of the jobs with AI, and we're gonna outsource and have hourly workers or full-time contractors outside the United States." It's the undeniable trend.

David Sacks 59:55 I think that's terrible.

Chamath Palihapitiya 59:57 Why?

David Sacks 59:57 That's throwing the baby out with the bath water. One of the most magical things about startups in Silicon Valley is that when it worked, everyone got rich. I mean, you have the story of the Google chef who made $100 million or the secretary at Microsoft who got to retire. It, you know... And those are extreme examples, but-

Jason Calacanis 1:00:12 The graffiti artist at Facebook.

Chamath Palihapitiya 1:00:14 Yeah.

David Sacks 1:00:14 (laughs) The graffiti artist at Facebook. Yeah.

Chamath Palihapitiya 1:00:15 He did, really? What was his name? David something?

Jason Calacanis 1:00:16 David Cho. David Cho, yeah.

David Sacks 1:00:17 But the fact of the matter is that Silicon Valley had a share-the-wealth mentality where the rank-and-file employees did participate in the ownership of the company, and they got equity. And it was highly motivating, or it is highly motivating-

Chamath Palihapitiya 1:00:29 Sure.

David Sacks 1:00:30 ... for not just, like, lower-level employees, but mid-level employees and senior employees to basically get equity in the company. That's one of the things that makes Silicon Valley companies special or companies created outside Silicon Valley but using this-

Chamath Palihapitiya 1:00:44 The methodology.

David Sacks 1:00:45 ... Silicon Valley template.

Chamath Palihapitiya 1:00:45 Yeah, the pho- philosophy. Yeah.

David Sacks 1:00:46 So I think, you know, getting rid of options altogether is a terrible overreaction.

Chamath Palihapitiya 1:00:49 No, what they're doing is more, like, for the core team that's in Silicon Valley, options, but then, like, for work that is not considered as essential-... you know, just hiring people offshore who don't have, come with that expectation.

David Sacks 1:01:01 Well, trying to get those people then to work nights and weekends, the way that you might be if you're the owner of the company, they're not gonna have an ownership mentality.

Chamath Palihapitiya 1:01:08 Yeah, I agree. I mean, I, I guess-

David Sacks 1:01:10 You're never gonna get nine to five workers that way.

Chamath Palihapitiya 1:01:12 And I think what people are starting to realize is, like, maybe there's, with these remote workers, a lot of people are doing nine-to-five work and maybe that, like, hardworking culture just doesn't exist in Silicon Valley like it used to.

David Friedberg 1:01:23 You may be able to find hardworking entrepreneurial-like folks that don't necessarily speak English, that now with these AI tools basically-

Chamath Palihapitiya 1:01:35 Mm.

David Friedberg 1:01:35 ... become some of your best folks. But there are many countries that don't value equity, don't have an understanding of it.

Chamath Palihapitiya 1:01:40 Yep.

David Friedberg 1:01:40 And so to Jason's point, it's almost more of a complication to try to give it. So, that may be the balancing thing, Sax, that happens, which is that all these tools actually allow somebody in some country that speaks some language that's not English, but who can be completely committed, who becomes this linchpin of a business and gets paid an extremely great salary in American terms, that makes them rich in their country, get to the same place maybe. Maybe.

Chamath Palihapitiya 1:02:08 Yeah.

David Friedberg 1:02:08 I don't know.

Chamath Palihapitiya 1:02:09 Their infrastructure needs to be built for stock options in other countries 'cause I don't know that that exists as-

David Friedberg 1:02:13 Or just pay people more cash-

Chamath Palihapitiya 1:02:14 Yeah.

David Friedberg 1:02:14 ... because a lot of folks have different... Again, stock options make sense when you have a, a differential in taxation between stock equity and income.

Chamath Palihapitiya 1:02:23 Capital gains, yes, yeah.

David Friedberg 1:02:24 But in other places where you have a flat tax or no income tax whatsoever, what's the difference?

Chamath Palihapitiya 1:02:32 Yeah, if you're in Dubai paying 1% (laughs) like, you could just ship it, right? Yeah.

David Friedberg 1:02:36 You can ship it. You can ship it.

Jason Calacanis 1:02:37 I mean, you can always g- uh, go back to a model where people vest shares rather than vest options, and those shares vest on a trigger of time and a liquidity event. And that way they can't exercise or rea- or they won't get taxed prior to the liquidity event happening, but then they will have to pay ordinary income tax rates on their shares.

David Sacks 1:02:58 Oh, hold on. We're making it sound like exercising options is a bad thing. It can be a very good thing.

Chamath Palihapitiya 1:03:02 Yeah.

David Sacks 1:03:02 I mean, if you're an early employee of a startup and the options are worth pennies and you can exercise for, I don't know, a few hundred dollars, then that's a great tax benefit because you start the clock ticking on long-term capital gains treatment. And when you sell those shares, eventually if there's a liquidity event, you get tur- taxed at a capital gains rate, not an ordinary income rate. That's the tax benefit. That's why people want to be able to exercise their options in the first place. The problem is that as you join the company later and later, the strike price of the options goes up. And so; therefore, if you want to exercise, you're gonna have to write a pretty big check and so you're gonna take a meaningful risk. Some people do that, and that's fine. The trouble comes in when you try to get cute and go around taking that risk by saying, "Okay, well, the employee doesn't have to go out of pocket for their exercise price. We're gonna loan them the money." Well, they thought that they had figured out a way to avoid taking that risk. But as it turns out, all you end up doing is potentially incurring AMT and a loan forgiveness issue. So there's no way around the fact that if you want to exercise your options to get the potential tax benefit of a lower tax rate, you have to also take the risk of losing the strike price that, th- that check that you write into the company. That's a fair trade, right? If you want the potential benefit, you take the potential risk. But that doesn't, that doesn't mean that you should never exercise. There's a lot of cases, again, I think if you're early at a company, you should exercise.

Chamath Palihapitiya 1:04:33 Yeah, low thousands of dollars-

David Sacks 1:04:34 Yeah.

Chamath Palihapitiya 1:04:34 ... low hundreds of dollars, go for it, yeah.

David Friedberg 1:04:36 There's all kinds of ways where when the basis is extremely low, it's nominally not a lot of money, so you can take the risk. But when you start talking about tens of thousands or hundreds of thousands of dollars or millions of dollars, and then using loans and leverage, and-

Chamath Palihapitiya 1:04:52 You get the risk of ruin coming out.

David Friedberg 1:04:53 Bad news bears all the way around.

Chamath Palihapitiya 1:04:56 Yeah, no bueno.

Chamath Palihapitiya 1:04:57 All right, Sax, here's your readme. Did you see the poll about Biden's age and h- the unfavorable percentage of people who think he is too old?

David Sacks 1:05:07 How can I not see that?

David Friedberg 1:05:09 I think the most important thing that we should talk about is that President Biden has decided to not go through a cognitive test as part of his annual physical.

Chamath Palihapitiya 1:05:18 All right.

David Friedberg 1:05:18 It's the first time it's ever happened.

Chamath Palihapitiya 1:05:20 Joe Biden now 81 years old, 82 at the start of his second term if he wins. He'll be 86 at the end, making him by far the oldest president ever. Trump is 77. He would be 78 and a half when he, uh, gets his second term if he won. 82.5 at the end of his term. Here's a histogram of age of presidents. These are the oldest, just to give (laughs) you a... Everybody obviously refers to Reagan as the oldest president prior to this. He was a spry 69 at the start of his presidency, and by the end he was 77, which is younger, obviously, than Biden and Trump today. So, Sax, thoughts? Or Shamath, thoughts?

David Friedberg 1:06:02 Let's just say that we've had a bunch of these issues over the last few years. President Trump didn't disclose his tax returns. Is that a huge issue? I think people made it to be a huge issue, but it's more of an issue of judgment. But I don't think it fundamentally compromises his ability to lead the country. You had a different issue, not with the president, but with the current se- defense secretary who had prostate cancer. The disclosures were, let's just say, somewhat suboptimal. He had a UTI. He got emergency admitted to the hospital. He just had another acute issue, and those disclosures weren't good. But you're talking about an acute kind of a thing where if you're incapacitated, there's a chain of command that immediately kicks in. Those are a class of issue that I think-... people can find things that are wrong with, probably correlated to one's political affiliation. The other side probably cares more than the side that the- that the leader's on. But I think this issue is a really big one. And I posted this clip of KJP basically saying, "Nothing to see here, move on." And the first comment was Elon, which I think nailed it. He said, "A basic cognitive test should not be something that someone who controls nuclear strikes is allowed to skip." And I think that that is part of the fundamental issue that I have with what they decided to do here, which is that it just massively increases the risk that you could have somebody that was fundamentally unelected running and making decisions, and I think that is a huge issue. And it's an enormous reason why, for me, this was sort of the straw that broke the camel's back, and why I just absolutely cannot vote for Biden anymore. I was trying-

Chamath Palihapitiya 1:08:01 Hmm.

David Friedberg 1:08:02 ... to find every-

Chamath Palihapitiya 1:08:02 There's a new story? Okay.

David Friedberg 1:08:04 I was trying to find a lot of good reasons to support him. Stability, normalcy, all of this stuff. But this is not what stability looks like. And I go, again, back to this very simple thing. A cognitive test for somebody that can control the ability for us to go to war, or for us to completely eviscerate entire populations through weaponry that he and he alone controls, should be something that is just mandatory. You can't skip it. This is not a thing where it's a nice-to-have. I think it's a must-have. And in the absence of it, I think it compromises his ability to be a legitimate choice.

Chamath Palihapitiya 1:08:44 And we do have a minimum age, Sax. And what Chamath is saying is what the majority of Americans are saying. ABC News ran a poll, 86% of Americans across both parties think Biden is too old to be president. 59% both- said both Biden and Trump were too old. You said on a previous episode, Sax, that you thought the public should decide not to, and that there doesn't need to be a cognitive test. Are you sticking with that, or do you think you're going to maybe move to Chamath's position and mine? Cognitive tests should be mandatory to be president.

David Sacks 1:09:15 Well, I think it's a tell that they won't take the cognitive test, because they're worried that he might fail. Just like it was a tell that they refused to do the Super Bowl interview, which the president always does. That's the most softball interview that Biden could possibly have, and they refused to do it, obviously because his aides and advisors are nervous about how that might go. We had the Heuer report come out where the, uh, special counsel who investigated the classified documents case for Biden said that we can't prosecute this guy. He'll be too sympathetic a defendant because he's manifestly senile and he'll come across as an old man who can't remember what happened. So, that's why they decided not to prosecute him. The Babylon Bee had a hilarious headline or take on this, which was, "Man Deemed Too Senile To Stand Trial Still Fit To Run Country." That's the situation we're in. He can't stand for trial, but he's still running the country.

Chamath Palihapitiya 1:10:07 Yeah.

David Sacks 1:10:08 And the thing that was most damning about the Heuer report was not just what the special counsel wrote, but it was Biden's reaction to it. Apparently, he demanded that he wanted to go out and do a press conference that very day to refute the idea that his memory was bad or that he was senile. And he then did this press conference, and it completely backfired. He came across as kind of an old man shaking his fists, yelling, "Get off my lawn!" I mean, that was basically the- the vibe that he gave to the press. And then, the thing that he got wrong was he confused Egypt for Mexico in regards to the whole crisis in Gaza. So, it would have been one thing if it was just Heuer saying that Biden was senile, but then Biden himself came out and seemed to confirm it by making these mistakes and gaffes. So, they know they have a problem.

Chamath Palihapitiya 1:11:04 So, to my original question-

David Sacks 1:11:05 Yeah.

Chamath Palihapitiya 1:11:05 ... would you be in favor, and will you change your position here, that we should have cognitive tests for the presidency of the United States? Yes or no?

David Sacks 1:11:12 I think that the real issue is that Biden would fail such a test. I mean, that's the problem, is that he is cognitively-

Chamath Palihapitiya 1:11:18 Okay.

David Sacks 1:11:18 ... impaired. And the thing to realize about these sorts of conditions is they're not static. There's a trajectory to them. So, in the same way that Biden is manifestly worse than he was a year ago, or three years ago-

Chamath Palihapitiya 1:11:33 Yeah, when he took office-

David Sacks 1:11:33 ... or 10 years ago-

Chamath Palihapitiya 1:11:33 ... he was obviously sharper, yeah.

David Sacks 1:11:35 Whatever this condition is, it's going to be worse-

Chamath Palihapitiya 1:11:37 Okay.

David Sacks 1:11:38 ... in a year.

Chamath Palihapitiya 1:11:38 I- I wanted to ask you a third time to answer the question, but you're choosing not to answer the question. So, Friedberg, what do you think?

David Sacks 1:11:43 Well, look, when you say- whenever you say that, should the president take a cognitive test? Sure, he should take a cognitive test.

Chamath Palihapitiya 1:11:48 Should it be the law-

David Sacks 1:11:49 But- but at the end of the day-

Chamath Palihapitiya 1:11:50 ... of the land-

David Sacks 1:11:50 No, because-

Chamath Palihapitiya 1:11:50 ... just like we have a 35-year minimum age. I'm just asking you to be specific.

David Sacks 1:11:54 You could make him take a cognitive test, but the point is that, let's say, the doctor says, "Yeah, he's cognitively impaired." Do- is that gonna bar someone from running for office constitutionally? No, that's not going to be part of the- the constitutional requirements. Y- you have to pass a constitutional amendment.

Chamath Palihapitiya 1:12:10 Yes.

David Sacks 1:12:11 Would it be a good idea for someone to take a cognitive test as part of their physical to be president? Sure.

Chamath Palihapitiya 1:12:16 Okay.

David Sacks 1:12:16 But is that going to be a constitutional requirement? No. You can't have some doctor say-

David Friedberg 1:12:19 No, no. And by the way, that- that's my whole point. Like, you- you could take a- a physical. Bill Clinton did it, and people used to joke about his cholesterol levels and he would go for a jog and stop at a McDonald's. So, I'm not saying by any means that the results should bar you from the ability to run. It should still be up to the voters. But just like we would make sure that acute or chronic health conditions were well understood, right? The idea that this entire body of judgment which is around your brain, around the judgment of the President of the United States, to basically say, "No, we're good here, we can skip it," I think that that's really dangerous. And it's part and parcel of a systematic set of behaving now by the Democratic Party...... which I think is really not where they want to be. So the only person that's really stood up to all of this is Dean Phillips-

Chamath Palihapitiya 1:13:09 Yep.

David Friedberg 1:13:10 ... but he is being prevented and railroaded from even being able to put that issue on the table so that the voters can decide. My fundamental view is the voters should decide, but all of these tests should be done, and skipping them is a form of a tell, as David said. And I think that this is the most important tell that we've been given about the amount of cognitive decline that he's suffered. And Jon Stewart, by the way, had the funniest version of this. Jon Stewart's, essentially, monologue was like, "Hold on, everybody is out here giving sound bites that say he is a chess grandmaster in private, but then he shows up in public and it looks like he can't even play checkers."

Chamath Palihapitiya 1:13:48 Yeah. Friedberg, let me, let me jump to you there. You shared the clip with us. What, what are your thoughts, what are your thoughts on this whole issue? Cognitive tests?

Jason Calacanis 1:13:54 I don't ... I mean, I don't think a cognitive test should be the law, uh, 'cause you're starting to get into edge cases where you'll create lots of different requirements. Like, what's the strategic test and, you know, what's the under pressure test? And there's obviously, uh, a lot of ways that folks can game and challenge and create new systems for, um, new criteria. The system of democracy, however, is meant to solve these problems. The system of democracy is meant to allow individual voters to say, "We judge that someone is not equipped or prove us wrong," which is what the voters are saying and doing right now, based on the polling data. The majority of voters in the United States are saying, "We do not think