GPT-4o 发布、Glue 演示、Ohalo 突破、德鲁肯米勒押注阿根廷，谷歌是否杀死了 Perplexity？

Jason Calacanis (00:00:00): Alright, everybody. Welcome to your favorite podcast in the world's number one podcast, the All-In Podcast. It's episode 1790. Oh, wait. That's just how it feels. Welcome to episode 179. With me today, of course, is your Sultan of Science. I don't know if that's a movie background or just his favorite vegetables. What's going on there? What's the crop? David Friedberg (00:00:22): That's AI generative. Jason Calacanis (00:00:23): It's AI-generated crop. Okay. David Friedberg (00:00:25): I'm trying AI backgrounds. I'm gonna try that for a while. Different crops. Jason Calacanis (00:00:28): Your fans are gonna be crushed that you're not doing deep movie polls with us, of course, man about town, DC, new products being launched. David Sacks. The Rainman. Yeah. How you doing, buddy? David Sacks (00:00:41): Good. Good. Yeah. Good week. What's going on? Jason Calacanis (00:00:44): Yeah. Yeah. Definitely good week. Chamath Palihapitiya. Chairman Dictator. He puts the Chairman in dictator. Chamath Palihapitiya (00:00:52): I would like to take this opportunity to wish my child a happy birthday. I absolutely love you. David Sacks (00:01:04): Well, now the rest of us look Jason Calacanis (00:01:05): like Yeah. Great. David Sacks (00:01:07): I've never done that before. Jason Calacanis (00:01:08): Sacks, in your desk, in your desk is a piece of paper with your children's names and their birthdays. You wanna get David Sacks (00:01:14): 3 birthdays a year, and I've never done one. Jason Calacanis (00:01:22): David Sacks. Speaker 5 (00:01:26): And I said, we open source it to the fans, and they just call me recently. Love you. Chamath Palihapitiya (00:01:34): No. No. No. No. But I'm saying it rarely lands on the same day. Today is the day. Today is David Friedberg (00:01:39): the day. Today is the day? Okay. Chamath Palihapitiya (00:01:41): Today is the day. David Friedberg (00:01:42): Congratulations, child. Congratulations. Chamath Palihapitiya (00:01:45): Yeah. No gender name or any other folks. We can't we can't tip anybody off. No pronouns. Jason Calacanis (00:01:51): No pronouns. Yes. Absolutely. So how are they experiencing their birthday? Chamath Palihapitiya (00:01:57): This child has experienced a wonderful life, and this child is an incredible person for whom I have tremendous admiration and love and compassion and hope for the future. Jason Calacanis (00:02:08): Alright. And did you order them some chicken fingers? Chamath Palihapitiya (00:02:11): I cannot comment on who this person is. Jason Calacanis (00:02:14): Are you are you talking, of course, about Phil Hellmuth's? Your child, Phil Hellmuth. Chamath Palihapitiya (00:02:20): Can we please talk about last weekend's festivities Jason Calacanis (00:02:22): in What a disaster he Chamath Palihapitiya (00:02:24): is? Oh my god. You guys just so you guys know. David Friedberg (00:02:28): So actually, we missed you last weekend. Missed you. Jason Calacanis (00:02:30): So much fun to make. David Friedberg (00:02:30): Chamath, we missed you on Saturday night. Saturday night was really fun. I had Chamath Palihapitiya (00:02:35): such a lovely time coming home, to be totally honest with you. David Friedberg (00:02:37): We had a cabana set up. On Saturday, played blackjack. I missed you guys too. I had a FOMO, and I saw the videos. It was so fun. Well, you you don't have to Jason Calacanis (00:02:45): have too much FOMO because Phil sent the entire group chat to pokernews.com. They didn't normally look twice. Theflop.org Chamath Palihapitiya (00:02:56): poker dash update. David Friedberg (00:02:58): Oh my god. It was like he kept reading all weekend. Jason Calacanis (00:03:00): He leaked every single person who's there and the check and the check numbers David Friedberg (00:03:04): He's like, look, here's me and Elon came by for my dinner. No. No. No. Chamath Palihapitiya (00:03:08): It was worse than that. No. It's worse than that. He said, I got to hang out with our guy Elon for 10 minutes 14 seconds. David Sacks (00:03:16): Like what? Jason Calacanis (00:03:17): He died. He intercepted him at the valet. Speaker 6 (00:03:22): Like, what? Chamath Palihapitiya (00:03:24): How many minutes? Ten minutes, 14 seconds. He had the exact time down to the second. Oh my god. David Sacks (00:03:29): Well, it's not one Jason Calacanis (00:03:30): of the ways David Sacks (00:03:30): to go home with a happy birthday because I did miss his 60th party. Chamath Palihapitiya (00:03:34): Yeah. It's coming back to his birthday. Jason Calacanis (00:03:37): Actually, his birthday. It was Bill Gurley. So we just hijacked Bill Gurley's birthday. David Sacks (00:03:41): I also David Friedberg (00:03:41): got to enjoy for my first time ever, the experience of baccarat I've decided is the most degenerate game on earth. It's the most the most you just flip coin. Chamath Palihapitiya (00:03:52): Flipping coins. Well, correct. You make David Friedberg (00:03:55): betting decisions. All you do in baccarat is you say bank or player, and then you freak yourself out about how you flip your cards. And the smartest people I know on earth are all sitting around this table at 2 or 3 in the morning saying, turn this corner this way, no. No. No. No. No. Turn it this way. Turn it this way. There's two dots, and they're debating the right way to slip of. Chamath Palihapitiya (00:04:16): No. Fox is the most incredible, performative act images. Don't have to say. David Sacks (00:04:21): Yeah. You're right. Everyone's got their own little techniques about how they bend the card. So it's all destroyed by the end of the deck. They don't they David Friedberg (00:04:28): don't try. I go like Chamath Palihapitiya (00:04:30): this, and I try to see Jason Calacanis (00:04:32): like you're always holding your mustache like an evil villain. Chamath Palihapitiya (00:04:34): You gotta go like that. Right. And then you call it, oh Jason Calacanis (00:04:37): my god. No spotter. Chamath Palihapitiya (00:04:38): If you see a roast spotter, we're too across. David Friedberg (00:04:42): And then you get to decide whether the bank turns over their cards and it when they turn Chamath Palihapitiya (00:04:46): Then you lose a small house, and then you're like, let's David Friedberg (00:04:48): try it again. You're convincing yourself that you have all this control and ways to change the outcome. Right. Because literally flipping a card. It's high card. Well, it's it's a high card. That's all it is, Chamath Palihapitiya (00:04:56): it's high card. David Sacks (00:04:57): It's even worse than that. You're basically sitting down at the casino's table, and then they tell you whether you've won or lost. In order to convince yourself that that's not what's going on. You have to play with the card. Jason Calacanis (00:05:08): Yes, sir. But, really, they David Sacks (00:05:10): always tell you you either win or lose. David Friedberg (00:05:13): I'm not watching the smartest guys we know staring at the window at the little machine that tells you whether bank or player won, and they're studying it. It's been four wins. Jason Calacanis (00:05:24): It's gotta go black. David Friedberg (00:05:25): Healthy is like, I'm pulling it now. Bank bank player player player. All the guys are like, let's do it. And then everyone Jason Calacanis (00:05:30): headset sales. Chamath Palihapitiya (00:05:31): So tell me so tell me if asked us to play in the high stakes poker game on PokerGO. So it was me, Hellmuth, Stanley, Sammy, House, and then Jen Tilly, and Nick Airball, and Robl. So most of the guys from the house came plus Jennifer Tilly and Nick Airball. Jason Calacanis (00:05:48): Jennifer Tilly is amazing. We're not great human Chamath Palihapitiya (00:05:51): and actress. Well, listen to this hand. Literally, the second hand of the actual poker game. Jennifer Tilly is in the big blind. No. Sorry. She's under the gun. She raises. Hellmuth 3-bets. It comes all the way around to me on the button. I look and I have pocket kings. David Friedberg (00:06:16): Oh, if Chamath Palihapitiya (00:06:17): I ship the whole cheeseburger, comes back to Tilly, she ships, Hellmuth ships, listen to these hands. Jennifer Tilly has aces. Jeff Hellmuth has kings. I have kings. Oh my god. I've never seen a cooler hand like this in my life, and the second in the second hand of the game. Anyways, wow. Don't worry, guys. It's so back. And I want She triples up. She triples up. Jason Calacanis (00:06:43): And then into lockdown mode, I the first time ever played with her. No. David Friedberg (00:06:46): Then I stacked Chamath Palihapitiya (00:06:47): and I stacked her right. And I stuck. Anyways, well, I I don't want to reveal the game, but David Sacks (00:06:50): it was Chamath Palihapitiya (00:06:50): it was wonderful. This one, Jason Calacanis (00:06:52): I show up at a mutual friend of ours' game, and there's a, like, beautiful Porsche something in the driveway. It's a really notable car. And the I I noticed on the license plate, it says, but it's spelled with a j. And I'm like, oh, degenerate. What a great license plate. I wonder whose that is. I go in. It's Jennifer Tilly. Oh, she's just so cool. Chamath Palihapitiya (00:07:10): She's very charming. Great. Very turn. Great actress. Jason Calacanis (00:07:13): Great. That's what she was in. David Friedberg (00:07:15): Bam. Bam. Chamath Palihapitiya (00:07:16): That's what it was. Yeah. David Friedberg (00:07:19): Yeah. Exactly. Towards the course. Jason Calacanis (00:07:21): Exactly. Gangster film. David Friedberg (00:07:23): Yeah. With, Gina that's the old g. My gosh. Jason Calacanis (00:07:28): That that David Friedberg (00:07:28): film that film. Oh my god. Jason Calacanis (00:07:29): Well, let's not get canceled here. Okay. Yeah. It is quite a film. Alright. Speaking of action, big week, the AI industrial complex is dominating our docket here. Apologies to Biden, Ukraine, and Nikki Haley, but we gotta go AI right now. OpenAI launched ChatGPT-4.0. Monday, three days after Sam Wise came on All-In as a programming note, and we'll go to Friedberg about this. We probably made a bit of a strategic or tactical error in not postponing his appearance. In fairness, Friedberg, Sam Altman, did tell us originally he was coming on to talk about those things, but then it got pushed back. And then he wanted to add to that. It was a programming note because people are wondering. David Friedberg (00:08:16): What happened? I've been talking with Sam for a while, a year, about coming on the show. And every time I see him, we're like, "Hey, you should come on." He's like, "I wanna come on the show." Okay. Let's find a date. We never got a date that worked. I saw him in March, and he said, "Hey. I wanna come on the show." I said, "Okay. Well, come on. Let me know when works." And then a couple of weeks later, he's like, "What about this date in May?" And I'm like, "Yeah. That's that's fine. We can make that work." He's like, "Well, I've got a big announcement we're gonna be doing." And I was like, "Perfect. Come on the show. That that that sounds great." And then the night before, he asked me he told me he texted me, like, "Hey. We're actually not gonna have this announcement happen tomorrow." "It's gonna be delayed." He didn't tell me how long. And I'm like, "Well, is it ChatGPT-5?" He's like, "No, it's not GPT-5." And I was like, "Okay. Well, you know, come on the show anyway," because he didn't tell me when he's doing the announcement or when it's being pushed to. So it didn't seem like that big a deal. Thought we were just gonna be able to have a good chat anyway. So it's really unfortunate. I think the fact that the announcement happened two days after, and he had to stay quiet about it during our interview, but, that's the story. I think in the future, if someone says they've got a big announcement to do, we should probably push them, if they, if they have to wait or something Jason Calacanis (00:09:22): like that, but don't beat yourself. But I David Friedberg (00:09:24): don't think we're gonna do a lot of these interviews anyway. I think people clearly don't love them, and it's better for us to just kinda hang out and talk. David Sacks (00:09:31): I think if we had just gotten Sam on the day after the launch of GPT-4, as opposed to, what is it, three days before? Yeah. He could have talked much more freely about it, and it would have Jason Calacanis (00:09:43): been interesting. David Friedberg (00:09:44): Yeah. It was supposed to happen same day, so it's unfortunate this all worked out this way. Jason Calacanis (00:09:47): The other little trick is to say you can tell us under embargo, but my understanding is they were still doing the videos over the weekend. So I think those videos and stuff, they were still figuring them out. And so, yeah, lesson learned. In terms of the interviews on the show, just to recap for people, we've done a dozen. Half of them have been presidential candidates. Sometimes they break out. Sometimes they don't. But we follow our interest and our passion here on the pod, it's gonna be interesting for us too. So if we think this person's gonna be interesting, we do it. And, yeah, we understand you miss a news subject, but, yeah, it is what it is. Chamath Palihapitiya (00:10:21): And to your point, a lot of the people that come on and increasingly a lot of people ask to come on because they know we're not journalists. And so for all of those folks that expect us to be journalists, that's not what we are. We're four entrepreneurs, four business people, four friends, four technologists, four curious people, four poker players, but we're not four journalists. And so we're gonna ask whatever we feel like asking. Sometimes those things will touch a chord because it's what you wanted to have asked. And sometimes we won't go to a place whether we didn't have time to or whether we forgot or whether we chose not to. And I think it's important to have that disclaimer. Like, we have day jobs, and this is what we do to coalesce a bunch of information in the way that we're thinking about the world. So we are not journalists. So we don't David Sacks (00:11:13): have that expectation. I think what that means is that if the guest doesn't wanna talk about something, we're not gonna start peppering him with gotcha questions and things like that. I appeared at a conference a couple of days ago to promote Glue, which we'll get to. And the first half of the conversation was, like, a normal conversation about what we were launching. And then the second half was basically the reporter peppering me with fastball questions. Which is fine. I knew what I was signing up for. It's a totally different style. It's a totally different style than coming on the pod and just having a normal conversation, but it's not really our job to make somebody open up if they don't wanna talk. Jason Calacanis (00:11:48): What was the smartest question, Sacks? That was the fast ball. Anything come close to your head? David Sacks (00:11:53): No. I mean, it's not worth really getting into. You can watch it if you want. Jason Calacanis (00:11:56): I was just curious. David Sacks (00:11:57): Like, look, I kinda like sometimes when reporters pitch me fastballs because you can strike out or you can hit it out of the park when they do that. Jason Calacanis (00:12:04): That's an important part here. I think, you know, as a former editor-in-chief journalist myself, I sometimes like to ask, I would say, a challenging question in a respectful way. I did that, for example, vague, you know, just clarifying his thoughts on trans and gay rights. Wasn't disrespectful? Was thoughtful? Oh, would you consider it spicy or hardcore? I don't think it was hardcore. He liked us because No. Chamath Palihapitiya (00:12:28): But that's because you asked it from a position of curiosity. You weren't trying to catch the guy. You see the difference? Jason Calacanis (00:12:33): I'm actually interested in his opinion. That's Chamath Palihapitiya (00:12:35): this is my point. That's why it comes out differently, and that's why I think people enjoy these conversations. And sometimes we don't get to the other kind of answer because I'm not interested in trying to gotcha somebody that's working hard. Jason Calacanis (00:12:46): I always have the same conditions when I do interviews, which is I don't clear questions, and I don't let people edit it, but you know, everybody's got a different view on how to do interviews and feel like deferrals. If you like it, you like it. If you like Lex Friedman's version or Tim Ferriss's version, or you prefer, you know, Fox or CNN, go watch those interviews there. You can have a whole range of different interviews and interview styles available to you in the media landscape. We are about 1. Sam Altman mentioned on the pod last week that the next big model might not be called GPT-5. So on Monday, they launched GPT-4O. The O stands for Omni. It's everything you love about tech. It's faster. It's cheaper. It's better. But, from my perspective, the real show is the massive amount of progress they made on the UI/UX. The O stands for Omni as in omnivore. It takes in audio, text, images, even your desktop, and video from your camera to inform what it's doing. You can consider it like 360-degree AI. Producer Nick will show a couple of videos I described them here before we go to the besties for the reaction to the announcement. First, they made great progress in solving the CB problem we mentioned last week. Where, like when you use Siri or any of these tools, you say, you know, hey, Jack, GPT, what's 2+2 over, and you have to wait. And then if you talk over each other, it breaks. They now have that working much smoother. They did an example of counting where they said speed up, slow down. They did a translator that worked really well. Speaker 7 (00:14:13): I would like you to function as a translator. I have a friend here who only speaks Italian, and I only speak English. And, every time you hear English, I want you to translate it to Italian. And if you hear Italian, I want you to translate it back to English. Is that good? Jason Calacanis (00:14:30): Mike, you Speaker 8 (00:14:35): do. Mike, she wonders if whales could talk, what would they tell us? Speaker 7 (00:14:42): They might ask, how do we solve linear equations? Speaker 8 (00:14:56): Certainly. Yes. Speaker 7 (00:14:58): Great. Looks like it works. Jason Calacanis (00:15:00): I think Duolingo's stock took a hit during that. Most impressive to me at least and also I I had Sundeep Madra on my other podcast and we talked about it was their desktop and iOS app. What this app does is fascinating. It watches your desktop on your Macintosh or on your iPad or iPhone, and it does things like adaptive learning. Here's a clip of Sal Khan from Khan Academy and his son basically using a drawing app to do some arithmetic or geometry. And it does adaptive learning and basically makes a personal coach because the app is watching his son try to solve a problem. Speaker 6 (00:15:39): I'm here with my son, and I'd love you to tutor him on this math problem but don't give him the answer. You can ask questions and nudge him in the right direction, but I really want to make sure he understands it himself. And he's here in the room, so you can talk to him directly. Speaker 8 (00:15:54): Can you first identify which sides of the triangle are the opposite, adjacent, and hypotenuse, relative to angle alpha? Speaker 9 (00:16:05): Alright. So I'm pretty sure this is the angle alpha right here. Right? Speaker 8 (00:16:10): Correct. Okay. That's correct. Now looking at the triangle, which side do you think is the hypotenuse? Speaker 9 (00:16:19): Remember any hypotenuse? I'm not totally sure. I think it might be this one, but I really am not sure. This side AC. Speaker 8 (00:16:29): You're close. Actually, side AC is called the adjacent side to the angle alpha. The hypotenuse is the longest side of a right triangle and is directly opposite the right angle. Can you find which one is the hypotenuse? Speaker 9 (00:16:45): Oh, okay. I see. So, I think the hypotenuse is this really long side from A to B. Would that be correct? Speaker 8 (00:16:53): Exactly. Well done. Jason Calacanis (00:16:55): It can also participate in Zoom calls, explain charts, all that great stuff. And so it's gonna be your guide on the side. It's gonna be a present, you know, personality while you're using your apps. It's really impressive, I have to say. So I guess let's start, Friedberg, with your takeaways on all of these innovations that we saw. David Friedberg (00:17:14): I think it's become quite apparent that there's an evolution underway in model architecture. We've and I think you may remember we talked about this briefly with Sam last week, but we're moving away from these very big, a lot of money to rebuild every time they get re-released towards a system of models. So this multimodal system basically leverages several models at once that work together or that are linked together to respond to the inputs and to provide some generative output and that those individual models themselves can be continuously queued and or continuously updated. So rather than have, you know, hey, there's this big new release that just happened, this new model just got trained, cost $10 million to train it. It's been pushed. These models can be upgraded with tuning, with upgrade features, and then linked together with other new smaller models that are perhaps specialized for specific tasks, like doing mathematics or rendering an image or rendering a movie. And so I think what we're gonna see is soon more of an obfuscation of the individual models and more of this general service type approach where the updates are happening in a more continuous fashion. I think this is the first step of OpenAI taking that architectural approach with GPT-4o and what's behind the curtains we don't know. We don't know how many models are there. We don't know how frequently they're being changed, whether they're being changed through actually upgrading the parameters or whether they're being And so this seems to be pretty obvious. If you look at this link, one of the criticisms that initially came out when they released GPT-4o was that there was some performance degradation, and Stanford actually runs this massive multitask language understanding assessment, and they publish it, I think, daily or pretty frequently on how all the models perform. And you can see the scorecard here that GPT-4o actually outperforms GPT-4. And so this goes counter to some of the narrative that in order to get some of the performance improvements and speed improvements they got in 4o that they actually made the model worse, and it seems actually the opposite is true that the model's gotten slightly better. It still underperforms Claude 3. David Friedberg (00:19:35): Claude, which, you can see here it ranks top of these charts, but there's lots of different charts. All the companies publish it on charts. They all claim that they're better than everyone else. But I like Stanford because it's independent. Jason Calacanis (00:19:44): Chamath, any thoughts after seeing it in combination with our interview? Do you think chat is running away with the consumer experience? Or do you think this is neck and neck with some of the other players? Chamath Palihapitiya (00:19:57): Not to tell tales out of school, but somebody that we all know in our group chat posted something about the fact that the consumer growth had stalled. I don't know how they knew that. Maybe they got some data or maybe they're an investor. You guys know what I'm talking about, and they said that they're trying to reinvigorate growth into the consumer app at OpenAI. Jason Calacanis (00:20:19): Any insights as to why it might be plateauing from your perspective? Chamath Palihapitiya (00:20:23): I wrote this in my annual letter. But there are these phases of growth. And when you look at social networks as a perfect example, Friendster was magical when it was first created. Right? And then you had MySpace that just ran circles around them because Friendster didn't really invest the money and the quality that it took to create a moat. And then MySpace really wasn't able to compete. So we were, you know, Facebook. We were the eighth or ninth when we showed up on the scene, and we ran circles around everybody. I think what it means is that there are these phases of product development, which exist in many markets. This market, I think, is going through the same thing. And right now, we're in the first, what I would call, primordial ooze phase. Which is everybody's kind of like running around like a chicken with their heads cut off. There's all these core basic capabilities that are still so magical when you see them. But we all know that five to ten years from now, these things will be table stakes. Right? And what Friedberg just showed is a table of many companies and many trillions of market cap, all effectively running to the same destination. So I think where we are is probably within two years of where the basic building blocks are standardized. And then I think the real businesses get built. So I will maintain my perspective here, which is the quote unquote Facebook of AI has yet to be created. Jason Calacanis (00:21:50): Okay. And here it is ChatGPT web visits, as you can see, have plateaued. This data is SimilarWeb. I would agree with you, Chamath. It seems like the use cases and the lookie-loos who were just trying this software because they heard about it. They've gone away, and now we have to find actual use cases. Sacks, I'm wondering, Chamath Palihapitiya (00:22:11): but our friend, Jason, just to kinda complete the thought, said something about the premium conversion. Right? That's what he I don't know how he David Friedberg (00:22:16): knows paid. Paid versus paid. Jason Calacanis (00:22:18): So to be clear, paid versus free. And then what Sam said on the podcast last week was it seems like whenever they come out with something new, the old stuff becomes free. In my talk with Sonny this week, he mentioned that these new models are so much more efficient that you actually can throw the old model into the garbage because it's so inefficient. And these are now becoming about 90% cheaper every year, which means every two years, these things are gonna be 99% cheaper and better. Yep. David Friedberg (00:22:46): Yep. Jason Calacanis (00:22:47): And it might be that OpenAI, Sacks, on a strategic level, is going to make all this free or close to free and maybe just charge for a multiplayer version. That seems to be where it's heading. You don't have to log in to use 3.5. You don't have to log in to use Google services, no. You do have to log in still on Google services, but I think these are gonna just be free. So on a product basis, what are your thoughts? And then maybe you could talk about free to paid. Do you think everybody in the world is gonna pay 20, 30, 40 bucks, 500 a year, 200 a year to have one of these, or are they just gonna all be free? David Sacks (00:23:21): Well, I think you're assuming there that the long-term business model of OpenAI is in B2C subscriptions, and I think that's probably the least attractive model they have available to them. It's sort of the first one and the most obvious one because they put out ChatGPT, and then it's pretty easy just to roll out a premium version. But in my experience, B2C subscriptions aren't a very attractive model because consumers aren't willing to pay a lot, and they have high churn rates. And there's no possibility of expansion, really. So I suspect they're gonna move in more of a B2B direction over time because that's where the real money is. And probably the way they do that is by monetizing all the apps that are built on top of it. And I think that in that sense, GPT-4o is a really important innovation. By the way, the "o" stands for Omni, which I think stands for Omnichannel. I think you may have said omnivore. David Friedberg (00:24:16): I'm just gonna tell Jason Calacanis (00:24:17): you, yes, it's Omni. Yeah. Which means all the different media types currently coming in, right? Like, that's the difference. It's not like you just give it an image or give it a video. It's absorbing all those at the same time in parallel, I believe. David Sacks (00:24:29): That's right. There's three big innovations with this model. Right? So one is omnichannel, which means text, audio, video, and images. Second, it's more conversational. Like, it understands the tone of people talking and understands sort of sentiment in a way it didn't before. And then the third thing, which is really important, is that it's just much faster and more performant than the previous version, GPT-4 Turbo. In the speed test, they say it's twice as fast. We've played with it at Glue. We can talk about that in a minute, and it feels ten times as fast. It is much faster. But it's the combination of all three of these things that really makes some magical experiences possible because when you increase the speed of processing, you can now actually have conversations in a much more natural way. Before, the models were just too slow, so there'd be a long delay after every prompt. Yeah. So now, like you showed, it can do things like you point the camera at a blackboard or something with math equations on it. And it can walk you through how to solve that problem. Or two people can be talking, and it does real-time translation. You know, there's that old saying that every Star Trek technology eventually becomes true. They've just basically invented the whole natural language, real-time universal translator. Jason Calacanis (00:25:45): Yeah. Anyway, so those David Sacks (00:25:47): are some interesting use cases, but I just think they're gonna be able to unleash a whole lot of new applications. And if they're metering the usage of the models and providing the best dev tools, I think there is a business model there. Jason Calacanis (00:26:00): This thing is moving so fast. We're in, like, Leonardo DiCaprio mode. Every two years, they throw the old model away. Okay. Let's keep—thank you, Sacks. David Sacks (00:26:10): bing. Jason Calacanis (00:26:11): Well, is this still on? Did you David Friedberg (00:26:12): write that ahead of time, or Jason Calacanis (00:26:14): did you guys say it's David Friedberg (00:26:15): in the moment? Yeah. That was good. David Sacks (00:26:17): Just one point on that. Yes, there are a whole bunch of startups out there that were creating virtual customer support agents. And they've been spending the last couple of years working on trying to make those agents more conversational, quicker, more responsive. I think their product roadmaps just became obsolete. Now that's not to say there isn't more work for them to do in workflow in terms of integrating the AI with customer support tools and doing that last mile of customizing the model for the vertical-specific problems of customer support, but my guess is that hundreds of millions of dollars of R&D just went out the window. And probably this is the best time to be creating a customer support agent company. If you were doing it 2 years ago, 5 years ago, your work is just like been Jason Calacanis (00:27:09): Well, I mean, that is the thing of this pace. Like, you know, you used to have to throw away client-server stuff or, you know, whatever. You had a web thing. You get an app out. You throw away some of the old code. But this is like every 18 months your work has been replaced. David Sacks (00:27:22): If you're an app developer, the key thing to understand is where does model innovation end and your innovation begin? Because if you get that wrong, you'll end up doing a bunch of stuff that the model will just obsolete in a few months. David Friedberg (00:27:36): I think Chamath Palihapitiya (00:27:36): you're totally right. I think that's such a really important observation. That's why I think the incentive for these folks is going to be to push this stuff into the open source. Because if you solve a problem that's operationally necessary for your business, but it isn't the core part of your business, what incentive do you have to really keep investing in this for the next 5-10 years to improve it? You're much better off, like Klarna, for example. Right? We talked about the amazing improvement and savings that Klarna had by improving customer support. Release it in the open source, guys. Let the rest of the community take it so that it's available to everybody else. Otherwise, you're gonna be stuck supporting it. And then if and when you ever wanted to switch out a model, you know, GPT-4, 4.0, to Claude, to LLaMA, it's gonna be near impossible, and it's gonna be costly. So I also think, Sacks, the incentive to just push towards open source in this market, if you will, is so much more meaningful than any other market. Jason Calacanis (00:28:35): Yeah. I mean, listen. You were there when I think you were there at Facebook when they did the Open Compute Project, and they just were like, oh my gosh. Chamath Palihapitiya (00:28:43): Talk about torching an entire market. Explain what it is. So there was this moment where when you were trying to build data centers, you'd have these, like, 1U rack-mounted kind of machines that you used. And what Facebook observed was there was only a handful of companies that provided it, and so it was unnecessarily expensive. So Facebook just designed their own and then released the specs online. Just kinda said, here it is, and they went to these Taiwanese manufacturers and other folks and said, please make these for your cost plus a few bucks. And it was revolutionary in that market because it allowed this open platform to sort of embrace this very critical element that everybody needs. And I think there's gonna be a lot of these examples inside of AI because the costs are so extreme, so much more than just building a data center for a traditional web app. The incentives to do it are just so meaningful. Yeah. And I Jason Calacanis (00:29:39): just showed it on the screen. Sacks, you've actually been dancing along this line. Last night, I was using your new Slack killer or code just around not sure. It feels like a Slack killer to me because I'm moving my company to it over the weekend. We're moving to Glue. David Sacks (00:29:52): And you Jason Calacanis (00:29:53): and I were doing some very I think I may need to wet my beak on this one. Speaker 6 (00:29:58): Well, you wanna wet your beak. Jason Calacanis (00:29:59): It feels like a 100 bagger to me. I need to slide it. David Sacks (00:30:04): That's the way we're thinking Jason Calacanis (00:30:05): about it. Chamath Palihapitiya (00:30:09): Voice, please. Jason Calacanis (00:30:10): I need to wet my beak. That's like a 100x. Sliding 500. Wow. Sacks. Tell me about product decisions. Where does the AI end and your product begin. David Sacks (00:30:25): Yeah. Well, it's a good point. I mean, I think where the AI ends, we want to use the most powerful AI models possible, and we wanted to focus on enterprise chat. So you could think of us as, for sure, a Slack killer or Slack competitor. It says that Slack wasn't built for the AI era. The glue is AI native. What does that mean? No channels? You know, I showed this to Chamath. The first thing he says, you had me at no channels. Right? The people are so sick of channels. You have to keep up with all these hundreds and hundreds of channels. And the real problem with channels is there's one thread in a channel that you want to see. In order to see it, you have to join the whole channel, and now you're getting all this noise. People just want the threads. So if you look at what's the chat model inside of ChatGPT, it's just threads. Right? You create a topic-based thread in ChatGPT, the AI comes up with a name for it, puts it in the sidebar. And then if you want to talk about somebody else, create a new chat, that's exactly the way that Glue works. It's just multiplayer. You just put the groups and individuals you want on the thread. Let me just show you real quick. Here's my Glue here. And you could see that in the sidebar, I've got all threads that I have been involved in. And like I said, you can address them to multiple people or groups. And then you've got the chat here. Now we've also fully integrated AI. And so Nick, who's our producer, at this and this thread, said, at Glue AI, what countries does Sacks talk about most in episodes? Episodes is a group we created to be the repository of all of the transcripts of our episodes. And so Glue did a search, and it said David Sacks frequently discusses Ukraine the most. Yeah. Really? So then Nick said, be more specific about Sacks' stance on Ukraine Russia war. Jason Calacanis (00:32:04): Oh, boy. And it's gonna overload the server. David Sacks (00:32:08): Well, here, it said here, David Sacks has a nuanced and critical perspective on the Ukraine Russia war across various episodes, the All-In Pod. Here's some key points encapsulating his stance. And it, like, nailed it. It's talk about prevention through diplomacy, opposition to NATO expansion, humanitarian concerns, skepticism of military intervention, peace deal proposal, you know, I'll copy and paste this onto Twitter X later today, but the point is it, like, nailed it across all these different episodes. And then this is a feature of Glue. It provided sources. So it cites where it got all the information from. So imagine now, we're doing this for the All-In Pod, but you could imagine that instead of it being transcripts of a podcast, it could be your work documents. You now have in your main chat, the ability just to ask, hey, at Glue AI, remind me where we left that project or tell me who the expert is on this subject matter, or who's contributed the most to this project. I've actually figured out using Glue AI, who's contributed the most deal flow at Craft is pretty amazing. Now let me show you, like, some David Friedberg (00:33:11): of the stuff. What yeah. What's the answer? David Sacks (00:33:14): I'm not gonna say it here. David Friedberg (00:33:15): I think it's Brian. Okay. Jason Calacanis (00:33:16): Brian's my guy. David Sacks (00:33:17): Those 2. Jason Calacanis (00:33:18): Ryan, right? It's my Chamath Palihapitiya (00:33:18): guy. I David Sacks (00:33:19): wanna hang on Jason Calacanis (00:33:20): on that. If I gotta get sacked, I go Daniel and Brian. Boom. That's how I get some Chamath Palihapitiya (00:33:24): of them. Okay. You take Daniel. I'll hire Brian. I would, you know, reconstitute Kraft. We'll call it Kraft. David Sacks (00:33:34): Alright. Look. May make sure you're a couple Chamath Palihapitiya (00:33:35): Hold on Jason Calacanis (00:33:35): a second. Sacks, I just looked at the ChatGPT-4 Omni server. And when you did that query, it actually rolled its eyes. Keep going. David Sacks (00:33:44): Okay. Jason Calacanis (00:33:45): Hello. David Sacks (00:33:45): So we talked about how ChatGPT-4 understands sentiment and conversations. And I think this is a really good example. So Evan, who's my co-founder at Glue, was just testing this for us. And he asked, what is each bestie's personality type based on conversations and episodes? And first of all, it figured out who the besties are. Like, we didn't program that. It just figured it out on its own, which is pretty impressive. And here's what it said about each bestie's personality. So it says Jason Calacanis' role: the moderator and entertainer. Personality traits: charismatic, humorous, lively, and the glue of the group. Often praised for his ability to make the podcast entertaining and keep the conversation flowing. Jason's energy and ability to bring humor to discussions are key elements that define his role in the podcast. Comments he's acknowledged for his role in bringing the group together. Chamath Palihapitiya, the visionary and hustler; personality traits: insightful, driven, persistent. Chamath's relentless drive and ability to hustle inspires peers, making him a core part of the podcast and electric discussions. I'll skip me, but David Friedberg, the pragmatic scientist and realist; pragmatic, methodical, and a bit reserved. Friedberg is often the voice of reason, bringing a scientific and realistic perspective to discussions. He focuses on analysis and logical reasoning. And then it cites where it got this from. And it says here, overall, the dynamic between the four besties creates a well-rounded and engaging podcast with each member bringing their unique strengths and personality traits to the table. I think that's pretty incredible. Jason Calacanis (00:35:08): How woke is this? Have you, have you put any rails on, or is this just pure GPT-4.0 compliant with the data? David Sacks (00:35:17): Yeah. Yeah. So what we're doing here is we're wrapping GPT-4.0 with Glue features that we've implemented to get the most out of the conversation. There are things we have to do to scope the prompt, and then we're using a retrieval-augmented generation service called RAG, which does RAG as a service. That basically slurps in our transcripts and makes them accessible to the AI. So, that's basically the stack that we're using. But as the models get better and better, Glue just gets better and better. Again, can I Chamath Palihapitiya (00:35:49): Can I just make a comment on this? It's just so clean. JCal was the key for me in abandoning Slack. He told me two or three years ago, he called me and he said, you can tell me the exact channels. I eliminated some channels that were random. There were, like, two or three channels David Friedberg (00:36:09): that you had. Jason Calacanis (00:36:09): The random channels. Chamath Palihapitiya (00:36:10): Yeah. Slack instance wasn't allowed to have. And I was like, this is genius. And I went in, and I was like, all of our companies should just eliminate these channels. And we could only get, like, 20% or 30% compliance, but it really started to turn me off Slack because I would get caught in these threads that were just so totally useless. And I thought, why aren't people working? And this is really great because you cannot blather on about nonsense in Glue, which I find really useful. Well, this is what happens with Slack. We used it at 80 90, just so you know. So we were the when we got into the early Jason Calacanis (00:36:42): Get into Slack too much. People start to think Slack is the job. And replying to Slacks and having conversations is the job when there's actually a job to be done. Chamath Palihapitiya (00:36:50): There's a job to be done. Jason Calacanis (00:36:51): Yeah. And so it's important. And what I liked about this implementation was it's like the ability to make a feed or a data source inside of your communication platform. So the fact that you imported all of the episodes and their transcripts is great. But what I want is like our HubSpot or our sales CRM. I want our Zendesk. I want our LinkedIn jobs and our LinkedIn job applications. I want our Notion. I want our code to each have the ability. And when I was using it last night, what you do is you use the at symbol to evoke and to summon in a way. It's like summoning Beetlejuice. So you summon your AI but then you tell it what data set you wanna go after. So you say, you know, at AI, let's talk about, I don't know, how do you manage your deal flow at Craft? Do you use software like CRM software to manage deals? Brian David Sacks (00:37:45): We just do it all. But we do it all in Glue, so it's ready right there. But you're right. So the first thing that Glue AI has access to is all of your chat history, which is amazing, because you get, like, you know, that we can look at all your attachments. And we've got, I think, six integrations at launch, and there'll be more. So, yeah, like, all of your enterprise data will be there. In the short term, you're right. You have to summon the repository by at mentioning because the AI needs a little bit of help of where to look, but in the future, it's gonna figure it out on its own. It's gonna become more and more seamless, but Jason Calacanis (00:38:15): It'll insert itself. So we have a discussion about sales, and then you might have a sales bot that says, hey. By the way, nobody's called this client in three months. David Sacks (00:38:23): Well, that's where I wanna go with it is I call that promptless, which is I want the AI just to chime in when it determines that it has relevant information and can help the team. Amazing. Even if it hasn't been summoned yet. But we need some model improvement for that frankly. I mean, we'll be able to get there by GPT-5, but that's totally where this is headed. I'll show you just one more fun example, if I could. Let me just show you this. So I asked it to write a letter to Lina Khan, to be a guest at the All-In Summit. And I told it, mention positive things we've said about Lina Khan in episodes of the All-In Pod. And so it wrote this letter, dear, Chair Khan. We hope that this message finds you well on behalf of the host, y'all in PAW, your site sent an invitation for you to speak. At the upcoming all in summit. And then it says in our conversations, we have frequently highlighted your impressive credentials and the impactful work you've undertaken. For example, in episode 36. We acknowledge your trailblazing role. And so the letter was able to quote episodes of the All-In Pod just without anyone having to go do that research and figure out, like, what would be the best? Cause I told it, only say positive things. Don't say anything negative. And then it said warm regards and it said who the four besties were again. We never told it who the besties are. We just said, write us a letter. So it's pretty incredible. Now this is just an example of the All-In Pod or think about any work context where the AI has access to your previous work documents. It's pretty amazing what it can do. Jason Calacanis (00:39:53): Well, I mean, it is kind of in the name. Like, this is Glue. Put you together, and Slack is where you slack up. Makes total sense. The brands give you a little bit of a tip. We should have seen it coming with Slack. David Sacks (00:40:04): Totally. Jason Calacanis (00:40:07): We have a breaking news story. It's a breaking news story to roll in exclusive today on the program. I got breaking news coming in. Your life's work. Sacks did his product review. Now it's your turn, Friedberg. We got breaking news coming in. David Friedberg (00:40:23): I did promise you that when Ohalo decides to come out of stealth and explains what we've done and what we're doing, I would do it here on the All-In pod first. Before the All Jason Calacanis (00:40:35): In exclusive. David Friedberg (00:40:36): exclusive. So, basically, by the time this pod airs, we're gonna be announcing what Ohalo's been developing for the past five years and has had an incredible breakthrough in, which is basically a new technology in agriculture, and we call it boosted breeding. I'm gonna take a couple minutes just to talk through what we discovered or invented at Ohalo and why it's important and the kind of significant implications for it. But basically, five years ago, we had this theory that we could change how plants reproduce. And in doing so, we would be able to allow plants to pass 100 percent of their genes to their offspring rather than just half their genes to their offspring. And if we could do that, then all the genes from the mother and all the genes from the father would combine in the offspring rather than just half the genes from the mother and half the genes from the father. And this would radically transform crop yield and improve the health and the size of the plants, which could have a huge impact on agriculture. Because yield, the size of the plants ultimately drives productivity per acre, revenue for farmers, cost of food, calorie production, sustainability, etc. So this image just shows generally how reproduction works. You've got two parents. You get a random selection of half of the DNA from the mother, and a random selection of half the DNA from the father. So you never know which half you're gonna get from the mother or which half you're gonna get from the father. That's why when people have kids, every kid looks different. And then those two halves come together and they form the offspring. So every time a new child is born, every time a plant has offspring, you end up with different genetics. And this is the problem with plant breeding. Let's say that you have a bunch of genes in one plant that are disease resistant, a bunch of genes in the other plant that are drought resistant, and you wanna try and get them together. Today, the way we do that in agriculture is we spend decades trying to do plant breeding where we try and grind all these different crosses, find the ones that have the good genes, find the other ones that have the good genes and try and keep combining them. And it can take forever, and it may never happen that you can get all the good genes together in one plant to make it both disease resistant and drought resistant. David Friedberg (00:42:52): So what we did is we came up with this theory that we could actually change the genetics of the parent plants. We would apply some proteins to the plants, and those proteins would switch off the reproductive circuits that caused the plants to split its genes. And as a result, the parent plants give 100 percent of their DNA to their offspring. So the offspring have double the DNA of either parent. You get all the genes from the mother, all the genes from the father. And finally, after years of toiling away and trying to get this thing to work and all these experiments and all these approaches, we finally got it to work. And we started collecting data on it. And the data is ridiculous. Like, the yield on some of these plants goes up by 50 to 100% or more. Just to give you a sense, like in the corn seed industry, breeders that are breeding corn are spending $3 billion a year on breeding and they're getting maybe 1.5% yield gain per year. With our system, we are seeing 50 to 100% jump in the size of these plants. Pretty incredible. Here's an example. This is a little weed that you do experiments with in agriculture called Arabidopsis, so it's really easy to work with. And you can see that what we have on top are those two parents A and B, and then we applied our boosted technology to them and combined them, and we ended up with what we call boosted ABs. You can see that the plant on the right is much bigger. It's got bigger leaves. It's healthier looking, etc. David Sacks (00:44:15): Friedberg, can I ask you a question? Does that mean that the boosted one has twice the more chromosomes as A and B? David Friedberg (00:44:21): Yeah. Exactly right. David Sacks (00:44:22): So is that like a new species then? Yeah. So, how does it survive with twice as many chromosomes? David Friedberg (00:44:28): Yeah. It's called polyploidy. So we actually see this happen from time to time in nature. For example, humans have two sets of chromosomes, right? So does corn, so do many other species. Somewhere along the evolutionary history, wheat doubled and then doubled again, and you end up actually in wheat having six sets of chromosomes. Wheat is what's called a hexaploid. Potatoes are a tetraploid. They have four sets of chromosomes, and strawberries are an octaploid. They have eight. And some plants have as many as 24 sets of chromosomes. So certain plant species have this really weird thing that might happen from time to time in evolution where they double their DNA naturally. And so what we've effectively done is just kind of apply the protein to make it happen and bring the correct two plants together when we make it happen. And so, so David Sacks (00:45:16): This could only happen for a plant, right? This could never happen with an animal? David Friedberg (00:45:20): It wouldn't work in animals. It works in plants. David Sacks (00:45:22): Okay. David Friedberg (00:45:23): And one way you can think about plant genetics is all the genes are sort of like tools in a toolbox. The more tools you give the plant, the more it has available to it to survive in any given second. To deal with drought or hot weather or cold weather, etc. And so every given second, the more tools or the more genes the plant has that are beneficial, the more likely it is to keep growing and keep growing. And that plays out over the lifetime of the plant with bigger, bigger leaves and bigger, you know, grows taller. But more importantly, if you look at the bottom, the seeds get bigger. And in most crops, what we're harvesting is the seed. That's true in, you know, corn and many other crops. And so seeing over a 40 percent increase in seed in this little weed was a really big deal. But then we did it in potato. And potato is a crazy result. Potato is the third largest source of calories on Earth. And so we took two potatoes that you see here in the middle, A, B, and C, D. We applied our boosted technology to it, to each of them. And put them together, and you end up with this potato A, B, C, D. That's the boosted potato. And as you can see, these were all planted on the same date, and the boosted potato is much bigger than all the other potatoes here, including a market variety that we show on the far right. That's what's typically grown in the field. Now here's what's most important. When you look under the ground and you harvest the potatoes, you can see that that A, B, potato only had 33 grams. D had 9 grams. So each parent had 33 and 9 grams of potato. But the boosted offspring had 682 grams of potato. The yield gain was insane. And so you could see this being obviously hugely beneficial for humanity. You know, potatoes being the third largest source of calories, Indian potato farmers are growing 1 acre of potato. In India, they eat potato two meals a day. In Africa, potato is a food staple around the world, we've had a really tough time breeding potatoes and improving the yield. With our system, we've seen incredible yield gains in potato almost overnight. And David Sacks (00:47:16): the other key David Friedberg (00:47:18): Those are normal-sized potatoes that you see there. Those are, like, you know, table potatoes. Basically, that looks like a russet potato right there. That's like a David Sacks (00:47:27): It started like a little creamer potato, basically, and you blew it up into a russet potato. Is that Jason Calacanis (00:47:32): right? David Friedberg (00:47:32): So the genetics on A B you can see they're like little purple, tiny, little purple potatoes, the genetics on CD are like these little white, you know, tiny, little ball potatoes. But when you put those two together with boosted and you combine all the DNA from AB and all the DNA from CD, you get this crazy high-yielding potato, A B C D, which by the way is higher yielding than the market variety that's usually grown in the field on the far right. David Sacks (00:47:55): So why not just grow russet potatoes then? David Friedberg (00:47:58): We are. And so we're working on doing this with russet. We're working on doing this with every major potato line. Sorry. The the the improvement you'll see is actually yield. So it's not the size of the potato. It's the number of potatoes that are being made. And so you'll Jason Calacanis (00:48:10): see acre or something like that, like the, you know, projects in the 1960s to 1970s. You know, how you can tell Friedberg's onto something here? He got David Sacks to pay attention during it. Chamath Palihapitiya (00:48:21): You got yeah. Jason Calacanis (00:48:22): This is a it's gonna be a decker court and Sacks is awake. Sacks is like, how do I wet my beacon? Sacks is Chamath Palihapitiya (00:48:27): interrogating the potato lines. I've never heard Jason Calacanis (00:48:29): of what's going on. I David Sacks (00:48:30): think Janice is interesting. But so have you tried these potatoes? Do they taste different? David Friedberg (00:48:34): Oh, no. They're awesome. Yeah. They're they're potatoes. And we do a lot of analysis. David Sacks (00:48:39): Any horns yet or anything like that? David Friedberg (00:48:40): No. I mean, again, one of the other advantages of the system that we've developed let me go back here. And I just wanna take two seconds on this. One of the other things this unlocks is creating actual seed that you can put in the ground in crops that you can't do that in today. So potatoes, the third largest source of calories. The way we grow potatoes, you guys remember the movie in The Martian, you chop up potatoes and you put them back in the ground. Because the seed that comes out of a potato, which grows on the top and the flower, every one of those seeds is genetically different because of what I just showed on this chart, right, you get half the DNA from the mother, half the DNA from the other. So every seed has different genetics. So there's no potato seed industry today. And a potato is like a $100 billion market. With our system, not only can we make potatoes higher yielding and make them disease-resistant, what we also make is perfect seed. So farmers can now plant seed in the ground, which saves them about 20 percent of revenue, takes out all the disease risk, and makes things much more affordable and easier to manage for farmers. So it creates entirely new seed industries. We're gonna be applying this boosted technology that we've discovered across nearly every major crop worldwide. It'll both increase yield, but it will also have a massive impact on the ability to actually deliver seed and help farmers and make food prices lower. It's actually cheaper. So higher costs. Do you need more water? Less water, less land, less energy. Chamath Palihapitiya (00:50:09): Do you need more fertilizer? David Friedberg (00:50:12): Fertilizer usually scales with biomass, but these sorts of systems should be more efficient, so fertilizer use per pound produced should go down significantly as we get to commercial trials with all this stuff. And we're doing this across many crops. So there's a lot of work to do in terms of how do you scale Chamath Palihapitiya (00:50:28): and tell us production in the field? Tell us about the patents and how important patents play a role in this because isn't it like one of Monsanto's big things? Like, they just go and sue everybody into the ground or whatever. Like, I'm David Friedberg (00:50:42): gonna answer you one second. I'm just gonna switch my headset just wow. Jason Calacanis (00:50:45): We went from Sacks' bots to Friedberg's crops. I'm glad we're doing David Sacks (00:50:50): I'm glad we're doing him second because all of a sudden, like, group chat doesn't seem very important. Jason Calacanis (00:50:55): Yeah. Wow. He just he just solved the whole Ukraine crisis here. I wouldn't be able to grow wheat in the desert. And then the rainforest. David Sacks (00:51:03): Solve the world food problem. Jason Calacanis (00:51:05): Yeah, Sacks. What if you what did you do for the last six months? Yeah. David Sacks (00:51:07): We made an enterprise chat a little better, but Jason Calacanis (00:51:10): We added AI to enterprise gen. We cleaned up your Slack. David Friedberg (00:51:13): So, yeah, when you invest, we've invested a ton of money. This was stealth for five years. We put a ton of money into this business. When you invest, like, I mean, north of fifty. North north of fifty. Yeah. Jason Calacanis (00:51:26): million dollars, five years, and you don't have a product in market yet. Wow. That's some Do David Friedberg (00:51:29): We actually have some product? Yeah. So I haven't talked about the way we've been making money in some of the business. We've been doing David Sacks (00:51:34): me just make sure I this is, like, clear. So that last photo you showed with the different types of potatoes, you had created the super huge ones, but you're saying that the yield benefit here is just you create a much bigger, hardier plant that's capable of producing Jason Calacanis (00:51:51): many potatoes. But David Sacks (00:51:52): the size of the potatoes doesn't change. David Friedberg (00:51:54): You can control for that when you breed. So the selection of what plants you put together in the boosted system allows you to decide you want small, medium, large. That's all part of the design of which plants do you want to combine. David Sacks (00:52:04): Okay. Cause your goal is not to turn, like, a Russet potato in, like, a watermelon or something like that. David Friedberg (00:52:09): No. No. The goal is to make more russet potatoes per acre so that we use less water. We use less land. Farmers can make more money. People pay less for food. That's the goal. And so it's all about yield. It's not about changing the characteristics. There are some crops where you want to change the characteristics, like you might want to make bigger corn kernels and bigger cobs on the corn, which is another thing that we've done. And that's actually been published in our patent. And the reason, by the way, I'm talking about all this is some of our patents started to get published last week. And so when that came out, the word started to get out, and that's why we decided to go public with what we've done. Because it's now coming out in the open. Jason Calacanis (00:52:43): You mentioned something briefly there about where different crops can be planted. You know, we had these big talks about wheat and corn. They're only available in very specific parts, you know, north of the equator. The champion juncoes can't be in obviously polar or desert extremes. So if you're successful, what would this do for on a global basis where these crops are made? Because remember this whole discussion about Ukraine, like, the wheat belly of Europe, the cradle of wheat. Yeah. David Friedberg (00:53:16): It's a great question. I'm so glad you asked it because that's one of the key drivers for the business is that we can now make crops adapted to all sorts of new environments where you otherwise can't grow food. Today, there's close to somewhere between 800 million and a billion people that are malnourished. That means they are living on less than 1,200 calories a day. For more than a year. But on average, we're producing 3,510 calories per person worldwide in our ag systems. The problem is we just can't grow crops where we need them. And so by being able to do this sort of system where we can take crops that are very drought resistant, or can grow in sandy soil or very hot weather and adapt cooler climate crops to those regions, but through the system, we can actually move significantly where things are grown and improve food access in regions of need. Chamath Palihapitiya (00:54:03): How frequently when you look at a potato, how do you figure out what part of their DNA is the drought-resistant part? David Friedberg (00:54:10): Yep. And then how do Chamath Palihapitiya (00:54:11): you make sure that that's turned on? So even if you inherit that chromosome, is there some potential interaction with the David Friedberg (00:54:17): Generally, if we can, so these are what are called markers, genetic markers. And so there are known markers associated with known phenotypes. A phenotype is a physical trait of a plant. And so we know lots of markers for every crop that we grow. Markers for disease resistance, drought resistance, markers for big plants, short plants, etc. And so what we do is we look at the genetics of different plants that we might want to combine into the boosted system. And we say, these ones have these markers, these ones have these markers. Let's put them together. And then that'll drive the results. One of the other interesting things we're seeing, which I didn't get too much into in the slides, it's not just about combining traits. But it turns out when you add more genes together, biology figures out a way to create gene networks. These are all these genes that interact with each other in ways that are not super well understood, but it makes the organism healthier and bigger and live longer. This is like when you bring, like, why mutts are healthier and live longer than purebred dogs because they have more genetic diversity. So there's a lot of work now in what's called quantitative genomics where you actually look at the statistics across all the genes, you use a model, and the model predicts which two crosses you want to make out of 100, or 1,000, or 1,000,000 of potential crosses that the AI predicts. Here's the two best ones to cross because you'll get this growth or this healthiness. So how do you wanna Chamath Palihapitiya (00:55:42): How do you wanna make money, Friedberg? Are you gonna sell the seeds? Are you gonna become the direct farmer? Are you gonna become food as a service? Like, how do you make the most money for this? David Friedberg (00:55:52): We're not gonna farm. Farmers are our customers, and so there are different ways to partner with people in the industry who already have seed businesses or already have genetics and help them improve the quality of their business. And then there's other industries like in potato where we're building our own business of making potato seed, for example. So every crop and every region's actually quite different. So it becomes a pretty complicated business to scale. We're in the earlier days. We're already revenue generating. I would like a sweeter blueberry. No comment. No comment. Yeah. Chamath Palihapitiya (00:56:25): I get tilted by the quality of the Driscoll's blueberries. David Friedberg (00:56:28): Let me tell you something about the Driscoll's blueberries. Chamath Palihapitiya (00:56:30): Also, the Driscoll I've had only one batch of a Driscoll strawberry that was just off the charts, and every 19,847 other batches I bought have been total. Yeah. Jason Calacanis (00:56:41): Now you want the European small ones or the Japanese ones from Hokkaido? David Friedberg (00:56:46): Cause they're Jason Calacanis (00:56:46): rich and sweet, and they're not these like monstrosity of giant flavorless strawberries. Can you do a seedless? Chamath Palihapitiya (00:56:53): Could you do a seedless mango? Yes. Jason Calacanis (00:56:56): No. Cut it. Chamath Palihapitiya (00:56:57): She can just cut it. Jason Calacanis (00:56:58): Oh Chamath Palihapitiya (00:56:58): my God. How great would that be? Jason Calacanis (00:57:00): Smooth out work per height on a mango. It's like the worst ratio. Chamath Palihapitiya (00:57:04): Yeah. Well Jason Calacanis (00:57:06): Somehow we made it about us. David Friedberg (00:57:07): Yeah. No. No. Look. I think that's it is all about you guys. No. Chamath Palihapitiya (00:57:10): Tell us about the blueberries. Sorry. David Friedberg (00:57:11): Well, now, every year, Driscoll's puts out a special labeled package called Sweetest Batch, and they just had the Sweetest Batch of strawberries and blueberries. I don't know if they're still in the stores, but they only last for like a week or two. And that's the best genetics. Only grown on a small number of acres. Chamath Palihapitiya (00:57:28): Really incredible bonus as soon as this is done. David Friedberg (00:57:31): See if they have it. So I got it a few weeks ago. It was quite delicious. Anyway, we know, let's just say we know the berry market very well. My co-founder, CTO, Judd Ward, whose brilliant idea of boosted breeding was many years ago, who I met because they had a New Yorker article on Judd. I cold-called him and said, hey, will you come in and give us a tech talk? We started talking, and Judd came up with this idea for boosted breeding. And so we started the business with Judd, and Judd ran molecular breeding at Driscoll's. So a lot of Driscoll's people that work in Ohalo. We know the market really well. Chamath Palihapitiya (00:58:01): Can you go back to the patent stuff? Like, are you Oh, yeah. Sorry. No question. So we spend David Friedberg (00:58:05): we spend 50 million bucks on, you know, plus on this business to date. So we have filed for IP protections that people can't just rip us off. But I would say I think that the real advantage for the business arises from what we call trade secrets, which is not just about taking patents and suing people. That's not a great business. The business is, how do you build a moat? And then how do you extend that moat? The great thing about plant breeding and genetics is that once you make an amazing variety, the next year, the variety gets better, and the next year, the variety gets better. And so it's hard for anyone to catch up. That's why seed companies generally get monopolies in the markets because farmers will keep buying that seed every year provided it delivers the best genetics. And so our business model is really predicated on how do we build advantages and moats and then keep extending them rather than try to leverage IP. So I'm a big fan of, like, building business model advantages. Jason Calacanis (00:58:58): It's gonna be incredible, Sacks. If you think about, you know, geopolitically, what's going on in Somalia, Sudan, Afghanistan, those places have tens of millions of people, I think hundreds of millions collectively who are at risk for starvation. If you could actually make crops that could be farmed there, Friedberg, you would change humanity. And then all these people buying farmland in America, that could devalue that farmland if that wasn't as limited of a resource. You have Friedberg? David Friedberg (00:59:25): Like no. Jason Calacanis (00:59:26): I think, David Friedberg (00:59:26): So first of all, like farmland in America is mostly family-owned, that's 60% rented, actually. So a lot of families own it, and then they rent it out because they stop farming it. But the great thing that we've seen in agriculture historically is that the more calories we produce, the more food we produce, the more there seems to be a market. It's like any other economic system. Wheat and rice. Yeah. So those are calorie sources one and two, and there's certainly opportunity for us to apply our boosted systems there. The big breakthrough with potato is we can make potato seed using our boosted system in addition to making better potatoes. Chamath Palihapitiya (01:00:03): McDonald's is the largest buyer of potatoes. Yeah. David Friedberg (01:00:05): So in the U.S., 60 percent of the potatoes go to French fries and potato chips. McDonald’s buys most of the fries, PepsiCo under Frito-Lay buys most of the potato chip potatoes, 40 percent for table potatoes. In India, 95 percent of the potatoes are table potatoes. They're eaten at home. And the Indian potato market's three to four times as big as the U.S. potato market. In Brazil, it's 90 percent table potato. So all around the world, potatoes are different. The U.S. is, you know, unusually large consumers of French fries and potato sauce. Chamath Palihapitiya (01:00:36): I speak on behalf of Jay Cal, and I said, we will gladly invest $1,000,000 at a 10 cap in both of your businesses. Jason Calacanis (01:00:44): Absolutely. Yes. We will work our way into it. Chamath Palihapitiya (01:00:47): Jay Cal. Jay Cal and I will do the deal. We'll pay the money. Wire the money, a little million to each of you guys at a 10 cap. Thank you. Absolutely. David Sacks (01:00:54): You're in. It may not be a 10 cap, though, but yes. Jason Calacanis (01:00:58): And Jay Cal have secured the bag. So breaking news, Chamath and Jay Cal have secured the bag from the besties actually doing work. Yeah. David Friedberg (01:01:06): Well, I appreciate you guys letting me talk about that. Jason Calacanis (01:01:09): I I'm excited to share. David Friedberg (01:01:10): both of you. I love it. It's been, yeah, building stuff is hard. There's always risk. It's a lot of work and a lot of setbacks, but, man, when you get stuff working, it's great. Speaker 6 (01:01:21): We're gonna be doing David Sacks (01:01:22): the things we do best. Friedberg is solving the world's hunger problem and I'm cleaning up your Slack. Jason Calacanis (01:01:29): Right? We can't, you know, make David Sacks (01:01:31): your enterprise channel a little better. Jason Calacanis (01:01:33): All progress counts. Alright. Stanley Druckenmiller has got a new boyfriend, Druckenmiller's got a boyfriend, and his name is Javier. And they've eloped to Argentina. Druckenmiller professed his love with Tom Cruise on Oprah's couch. In a CNBC interview this week, the only free market quote, leader in the world right now, bizarrely, is in Argentina of all places. He cut Social Security 35% if he came to office. They've gone from a primary deficit of, like, 4 or 5% to a 3% surplus. They've taken a massive hit in GDP, basically a depression fourth quarter, and his approval rating has not gone down. Druckenmiller has explained how he invested in Argentina after seeing Milei's speech at Davos, which we cover. Here's a 30-second clip, play the clip, Nick. Speaker 10 (01:02:23): way, do you want to hear how I am thus in Argentina? It's a funny story. I wasn't at Davos, but I saw the speech in Davos and it was about 1 o'clock in the afternoon in my office. I dialed up Perplexity and I said give me the five most liquid ADRs in Argentina. It gave me enough of a description that I followed the old Soros rule, invest and then investigate. I bought all of them. We did some work on them. I increased my positions. So far, it's been great, but we'll see. Jason Calacanis (01:02:53): Yeah. That's quite interesting. Quick note, you hear Druckenmiller mention ADRs. For those of you who don't know, and I was one of them, they stand for American Depository Receipts. Basically a global stock offered on a U.S. exchange to simplify things for investors. Yeah. I mean, he didn't sign a prenup here. He just went all in, and he bought the stock, Chamath, and then he's gonna figure it out later. Tell us your thoughts Chamath Palihapitiya (01:03:19): on this love affair. This bromance. There's a great clip of Milei. He goes on this talk show in Argentina, and the talk show host, she's just so excited. And greeting, and then they start making out. Have you guys seen the spots? Jason Calacanis (01:03:34): They're just, like, full on. Chamath Palihapitiya (01:03:36): Full-on French kissing each other. It's hilarious. Yeah. I mean, like, Soros has been very famous for this invest and investigate thing. It's a smart strategy for very, very liquid public market investors that have the curiosity that he does. I mean, I don't have much of a reaction to that. I think that the thing with Argentina that's worth taking away is when you've spent decades casting about and misallocating capital and running your economy into the ground, the formula for fixing it is exactly the same. You cut entitlements and you reinvigorate the economy. And so the thing we need to take away is if we don't get our act together, that's probably what we're gonna do. Jason Calacanis (01:04:22): Sacks, the influence of Milei on American politics. Will there be any? It seems like he has paralleled what Elon did at Twitter, and Zuck did at Facebook. Do you think that this, you know, experiment he's doing down there of just cutting staff, cutting departments will ever make its way into American politics? David Sacks (01:04:47): Probably not. I mean, not until we're forced to. But what Milei did, he comes in and they've got a huge budget deficit and they've got runaway inflation and they're debasing their currency and this practically overnight, he just slashes government spending to the point where he has a government surplus. And then as soon as he gets credibility with the markets, that allows him to reduce interest rates, inflation goes away, and people start investing in the country. Jason Calacanis (01:05:10): Magic. It's magic. David Sacks (01:05:11): So there Jason Calacanis (01:05:12): is a path. David Sacks (01:05:12): It's obvious. Listen. I mean, you can't run deficits forever. You can't accumulate debt forever. It's just like a household. If your spending exceeds your income, eventually, you gotta pay it back or you go broke. And the only reason we haven't gone broke or experienced hyperinflation is because we're the world's reserve currency. So there's just a lot of room for debasement, and there's not a ready alternative yet. I mean, everyone's trying to figure out what the alternative base is. So we've been able to accumulate more and more debt, but it's it's reaching a point where it's unsustainable. And what we've already seen is that the Fed has had to jack up interest rates from very low, practically nothing to 5.5 percent. And that has a real cost on people's well-being because now your cost of getting a mortgage goes way up. I mean, mortgage rates are over, what, 7.5 percent now? Jason Calacanis (01:06:03): Yeah. 6, 7 percent depending on how much net worth and your credit rating. David Sacks (01:06:08): Right. And so it's much harder to get a mortgage now. It's harder to make a car payment if you need to borrow to buy a car. And if you have personal debt, the interest rates can be higher. The inflation rate actually doesn't take into account any of those things. Remember, Larry Summers did that study where he said the real inflation rate would be 18 percent or would have peaked at 18 percent if you include the cost of borrowing. That's why people don't feel as well off as the unemployment rate would normally suggest. So people are hit really hard when interest rates go up in terms of big purchases they need to make with debt. And then, of course, it's really bad for the investment environment because when interest rates are really high, that creates a higher hurdle rate and people don't want to invest in risk assets. And so eventually the pace of innovation will go down. Druckenmiller made this point in his next set of comments. He said that Treasury is still acting like we're in a depression. It's interesting because I've studied the Depression. You had a private sector crippled with debt, basically, with no new ideas. So intervention policies were called for and were effective. He said the private sector could not be more different today than it was in the Great Depression. The balance sheets are fine. They're healthy. And have you ever seen more innovative ideas that the private sector could take advantage of, like blockchain, like AI? He says all the government needs to do is get out of the way and let them innovate. He said they've spent and spent and spent. And my new fear now is that spending and the resulting interest rates on the debt that's been created are gonna crowd out some of the innovation that otherwise would have taken place. I completely endorse Druckenmiller's view of Bidenomics. And, actually, I mean, this is what I said way back in 2021. Jason Calacanis (01:07:50): Victory lap. Here we go. Well, David Sacks victory lap. We need a little graphic for that. David Sacks (01:07:55): Druckenmiller used the word Bidenomics. And said I give these guys an F because they're still printing money and spending money like we're in a depression, even though we're in a rip-roaring economy. And when they started doing this back in 2021, you know, I tweeted Bidenomics equals pumping trillions of dollars of stimulus into a rip-roaring economy. I'm not gonna pretend like I know what's gonna happen next, but we've never tried this before. What happened next was a lot of inflation, and that backed up interest rates. According to even Keynesian economics, the reason why you have deficit spending is because you're in a recession or depression. And so you use the government to stimulate and balance things out. You still don't Jason Calacanis (01:08:28): do that. Yeah. David Sacks (01:08:29): You don't do deficit spending when the economy's already doing well. So this spending, there's no reason for it. Jason Calacanis (01:08:34): Yeah. It's like showing up to, like, a party that's going crazy and being like, put gasoline on the fire. Yeah. David Friedberg (01:08:39): I mean, more importantly, it should limit the approval or action of certain programs that you might otherwise wanna do in a normal environment, but in an inflationary environment, you don't have the flexibility to do them. Student loan forgiveness is a really good example. Is now the time? Jason Calacanis (01:08:59): Of course not. David Friedberg (01:09:00): To do student loan forgiveness, or do we wait for inflation to temper a bit? Is now the time, you know, so, so there's just a lot of these examples that it actually, the opposite should be true. Jason Calacanis (01:09:09): Yeah. But all of those things get you votes. David Sacks (01:09:12): Before we move on from this, look, what we have coming out of Washington here is a contradictory and therefore self-defeating policy. You've got the Fed jacking up rates to control inflation. You move across town, and you've got Capitol Hill and the White House spending like there's no tomorrow, which is inflationary. David Friedberg (01:09:27): Right. David Sacks (01:09:27): Why would you do both those things? Choose what your policy is gonna be. Right? Jason Calacanis (01:09:30): It's like driving with your foot on the brake and the gas at the same time. It's not a great idea for the car. David Friedberg (01:09:35): Let me just make one comment, J Cal, before we move on about the Druckenmiller investment statement. Jason Calacanis (01:09:39): Of course. Speaker 5 (01:09:39): And I David Friedberg (01:09:40): I just wanted to say, like, I think what it highlights about Druckenmiller and call it a rift in investing philosophy or skill is the difference between precision and accuracy. What I mean by that is precision really references that you do a lot of detailed analysis to try and make sure you understand every specific thing that is going right or could go wrong. But the problem and so that means you, for example, might do a ton of diligence on a company and make sure you understand every dollar, every point of margin, all the specifics of the maturation of that business and where they are in their cycle. But you could be very precise, but be very inaccurate, for example, if you miss an entire trend. Someone could invest in Macy's, back when Amazon was taking off and have done a lot of precise analysis on Macy's margin structure and performance and said, this is a great business. But they missed the bigger trend, which is that e-commerce was gonna sweep away Macy's. And consumers were simply that's not possible in the analysis, but they were doing. Jason Calacanis (01:10:40): Let's be honest, Friedberg. Nobody can make that stupid of a trade to say Macy's versus Amazon over the next 10 years. David Friedberg (01:10:46): Well, yeah. And so, like, and then, and Jake helped. Jason Calacanis (01:10:49): Do you David Sacks (01:10:49): wanna show that? Jason Calacanis (01:10:50): Who No. No. No. No. Do not poke the tiger. Let's not get into it with other podcasters. The bar spread trade in history. Yeah. David Friedberg (01:10:57): Let me just finish the statement. Yeah. But the other one is being accurate and accurate means you get the Jason Calacanis (01:11:02): Yeah. David Friedberg (01:11:02): The right bet, the right sentence. The right friend. The problem with being accurate, you could have said in the year 2000, hey, the internet's gonna take off. And you could have put a bunch of money in, but the problem was you were right. You just had to have the necessary patience. And so accuracy generally yields better returns, but it requires more patience because you can't necessarily time how long it will take for you to be right. So a guy like Druckenmiller is making an accurate bet. He bets correctly on the trend on where things are headed. He doesn't necessarily need to be precise, but he has the capital and his capital structure that allows him to be patient to make sure that he eventually gets the return. Jason Calacanis (01:11:44): And to build on your thoughts, having watched this movie a couple of times, and, you know, I overthought the Twitter investment as about one example. I had the opportunity to invest in Twitter when it was like a single-digit millions company. I just thought, you know what? This thing is only like the headline, and I told that, like, it's the headline. It's not like the entire blog post. It may be like a lot of idiots, this thing is gonna be chaos. And I was right, but I was wrong. Right? Great bet, but my wrong analysis. Right. And so you can add precision to other aspects, like when you sell your shares or when you double down, but you have to get the trend right, which is Evan Williams, great entrepreneur, Jack, great entrepreneur, Twitter taking off like a weed, just make the bet. Right? No. David Sacks (01:12:24): The problem is you knew too much about journalism. You knew too much about the space they were trying to disrupt, and that could be a mistake. Correct. Yeah. PayPal. No one knew anything about payments. That was one of the reasons we were successful. All the payments experts told us it couldn't be done. Jason Calacanis (01:12:38): Right. Absolutely. David Sacks (01:12:39): So that happens a lot. Chamath Palihapitiya (01:12:41): I had never even, no, I didn't even know what a Facebook was when Jason Calacanis (01:12:44): I joined Chamath Palihapitiya (01:12:44): Facebook. It's an American college phenomenon. No, seriously. You don't have that in Canada. Jason Calacanis (01:12:48): But you knew Zuck and you saw some growth charts and you saw some precision in his ability to build product. And that's the way to go. David Friedberg (01:12:55): Which is a great thing. The great thing about network effect businesses is there's a trend line that sustains because it builds if it's an appropriate network effect. So you can be accurate about buying into the right network effect business. You don't need to use all of this diligence to be perfectly sound around the maturation of the revenue and the margin structure and all that stuff as long as the trend line is right, and you're willing to be patient to hold your investment. I think Druckenmiller's point is incredible. He took a look. He very quickly made a macro assessment. From a macro perspective, what Milei is doing is significantly different than what we're seeing in any other emerging market, let alone mature market with respect to fiscal austerity and appropriateness in this sort of global inflationary environment. Jason Calacanis (01:13:38): And he David Friedberg (01:13:38): said, you know what? I don't see any other leader doing this. This is a no-brainer bet. Let me make the bet. And as long as he's willing to hold this thing for long enough, eventually the markets will get there and call it a spread trade against anything, he'll be proven right. Chamath Palihapitiya (01:13:52): Well, and so not to but speaking of bets, JCal, you told me this week that you just made your largest investment ever. Tell us about that. Jason Calacanis (01:14:00): Yeah. So I've gotten very lucky now because a lot of my founders from the first couple of cohorts of investing I did when I was a Sequoia scout, have come back and created second and third companies. And so, you know, that happened with TK, Uber, and CloudKitchens, that happened with Rahul from Reportive then Superhuman. And then it happened recently just in the past year. My friend Jonathan, who's the cofounder of Thumbtack, asked me to come to dinner and said, hey. You know, you were the first investor in Thumbtack. Will you be the first investor in our next company, Athena? And I said, what do you do? And he explained it to me. We put a 7-figure bet in, which is rare for us as a seed fund. Right? Normally, our bet sizes are $100K, $250K, you know, it's a $50 million fund. Chamath Palihapitiya (01:14:41): Why did you do it? Jason Calacanis (01:14:42): Yes. It's very simple. It's the fastest growing company I've ever seen, and I'm including Uber in that. It has been growing at, you know, a rate that I'll just say is faster than Uber and Robinhood when we were investing in the tens of millions of dollars. It's a very simple concept. When Thumbtack was building their marketplace, they used researchers in places like Manila, etc., in the Philippines, knowledge workers. And what they realized was the top 0.1% of those knowledge workers were as good or better than, say, Americans at doing certain jobs. And so they've created this virtual EA service. You can go see it at athenawow.com. And we now have two of them inside of our company. It turns out Americans don't want to do the operations roles. So it's kind of like AWS. You just give them $36,000 a year. They give you essentially an operations or an EA. And they have ones that are kind of chief of staff-ish, and this company is growing like a weed. So I am working with them on the product design as well. So imagine having, you know, two or three of these incredibly hardworking people who are trained with MBA-class level curriculum. They spend months training these people up. They pay them two or three times what they would make at any other company, and then they pair them with executives here. And it's kind of been an underground secret in Silicon Valley because it's only by invitation right now because they can only train so many people. But if you've tried to hire an executive assistant, I don't know if anybody's tried to do that recently. Chamath Palihapitiya (01:16:19): You hook me up so I will be guinea-pigging the service. Yes. Jason Calacanis (01:16:22): Soon. And I have two of them. And so it is just the greatest that you can have an operations Chamath Palihapitiya (01:16:28): person on board by AI tools as well. Jason Calacanis (01:16:31): Yeah. So that's the kind of secret sauce here is they're training them and they watch you work and then they will learn how you do your job and then how quickly you can delegate and get stuff off your plate is the name of the game. So we have an investment team with researchers and analysts in it. We have a due diligence team. And then you have, like, executive functions in our fund. They have now started shadowing, you know, highly paid Americans in an investment firm, ours, and then train them up to now our due diligence, our first level screening, you know, and our tracking of companies is being done by these assistants for what I'll say is a third to a fourth of the price I was paying previously. So what that does in an organization is we're just delegating away and then moving our investment team to doing in-person meetings and doing higher-level stuff. And so Chamath Palihapitiya (01:17:21): You're you're 80, 90. So at 80, 90, we have this funny thing where we've made a deverb. Whenever you see somebody doing high-quality work at a quarter to a tenth of the cost, we say, Oh, you just 80, 90 did. Jason Calacanis (01:17:33): Correct. Chamath Palihapitiya (01:17:33): So you're you're $89,000,000 in the investment team. Jason Calacanis (01:17:36): I'm $89,000,000 in the investment team. And you know what? It was scary as hell for them because they're like, am I gonna lose my job? Chamath Palihapitiya (01:17:40): I was like, no. Jason Calacanis (01:17:41): You now get to instead of doing a check and call once a month, you can do a check and call every other week or every week. Or instead of doing 15 first round interviews a week, you can do 25 because you have this assistant with you doing all the repetitive work. Chamath Palihapitiya (01:17:55): The way that companies will work in five to 10 years, I don't think, guys, any of us are gonna recognize what it's gonna Jason Calacanis (01:18:01): look like. David Friedberg (01:18:01): Yeah. So anyway, I mean, like, watching Sacks's demo earlier, how much progress and how seamless that product works with the features it has enabled by the underlying models, you just get to thinking how all of these vertical software applications become completely personalized and quickly rebuilt around AI, you know, it's so obvious. David Sacks (01:18:26): Like how long it would have taken John to write a letter to Lina Khan, but, like, if we said, John, invite Lina Khan, but be sure to reference all the nice things we said about her on episodes of the pod. Jason Calacanis (01:18:36): Ten hours of work. Gotta go find out the final that said, yeah, David Sacks (01:18:39): listen to him to figure out what the best quotes are. David Friedberg (01:18:41): You've got the and you got it done in five seconds. It's incredible. Totally. And imagine building that same sort of capability into a very specific vertical application that's specific to some business function, and you can probably spend a couple minutes or an hour building that function, and then it saves you hours a day in perpetuity. Jason Calacanis (01:18:59): Yeah. So David Friedberg (01:19:00): and I think I think that's why these tools companies or the tools products that Google, Microsoft, Amazon, and a few others are building are actually incredible businesses because so many enterprises and so many vertical application builders are going to be able to leverage them to rewrite their entire business functions. Chamath Palihapitiya (01:19:19): I got myself and my co-founders at 8090. We get the stream of emails of companies that are like or people that are like, we have this product idea or we have this small product. One of the emails I got, this is crazy, was from a guy that's like, oh, we've 80, 90 Photoshop. So, like, we have like a much, much cheaper version of Photoshop, and the guy was doing like a few million bucks of ARR and growing really nicely. But then it turned out that somebody saw that and then 80, 90 did. So there's a short version of that. Think of and so to your point, Friedberg, none of these big companies stand a chance. David Friedberg (01:19:55): Yeah. Yeah. Jason Calacanis (01:19:56): It's every meeting is not because Chamath Palihapitiya (01:19:59): they're not because the products aren't good, but, like, Jason's gonna go off and experiment with this. Sacks is gonna go off and build a product. You know, every time that you're at a boundary condition, we're all gonna explore, well, maybe we could do this with AI. Maybe we shouldn't hire a person. Not because we're trying to be mean about it, but it's because it's the normal natural thing to do. And the OpEx of companies is just gonna go down, which means the size of companies are gonna shrink, which means the amount of money you need is gonna go down. And that's just going to create the ability for these companies to sell those products cheaper. So it's a massive deflationary tale. Jason Calacanis (01:20:35): We had the same thing happen with compute, and now it's happening inside of organizations. I wrote a blog post about this on my called ADD. This is the framework I came up with. I told my entire team, look at what you got done every week, and I want you to ask three questions. How can I automate this? How can I deprecate this? How can I delegate it? And, you know, the automate part is AI and what you're doing, David. The delegate part is AthenaWow.com. And then the deprecate is, hey, just be thoughtful. What are you doing that you don't need to do? And that's 80, 90 something. Like, there are things inside these products that you don't actually need. What's the core functionality of the product? You know, make it as affordable as possible. And then what's gonna happen for people who think this is bad for society? You've got it completely wrong. We're gonna have more people be able to create more products and solve more problems. The unemployment rate's gonna stay very low. We're just gonna have more companies. So the idea, like, there was somebody who was working on very small, like, software. I get pitched on very niche ideas. I wanna create something where people can find people to play pickleball with, right, like a pickleball marketplace. Now that wouldn't typically work because you would need $5 million a year to build that product. If you can build it for $500,000 a year, well, now you've only gotta clear that number to be profitable. So a lot more smaller businesses, a lot more independence. All these little niche ideas will be able to be built. And a VC who says, I'm not giving you $5 million to build that app. We'll be like, but I will give you $500,000. And that's what I'm seeing on the ground in startups. The same startups that had a request of $3 million in funding five years ago are now requesting $500,000. Jason Calacanis (01:22:12): To $1 million. It's deflationary all the way down. Chamath Palihapitiya (01:22:16): Did you guys see David Friedberg (01:22:16): the incredible? Chamath Palihapitiya (01:22:17): Did you see the Google thing? Did you guys see the Google AI Gemini stuff? Jason Calacanis (01:22:21): ChatGPT-4 launched at the same time or perhaps strategically right before Google dropped its latest AI announcements at I/O. The biggest announcement is that they are going to change search. This is the piece of a puzzle on the kingdom that they have been very concerned with and they're going for it. The new product and they have, like, 20 different products. You can see them at labs.google when they put all their different products, but this is the most important one. They call it AI overviews. Basically, it's Perplexity for most users by the end of the year. They're gonna have this. Here's how it works. And you can see it on your screen. If you're watching us, go to YouTube. Here, they gave an example. How do you clean a fabric sofa? This normally would have given you 10 blue links. Here, it gives you a step-by-step guide with citations and links. So they're preempting, you know, the issue of people getting upset. And as I predicted, they're gonna have targeted ads. Here's the things you need in order to clean your couch. You can only use this if you're using your Gmail account. If you use, like, a domain name, on Google Docs, it won't work there. So go to labs.google. But they're doing citations. And I think that we're gonna see a major lawsuit here. Those people who are in those boxes are gonna look at the answer here and realize maybe they don't need to click through and that this answer was built on that. And now we're gonna have to have a new framework. There's gonna need to be, Sacks, a new company that clears this content so that Google can do answers like this. Chamath Palihapitiya (01:23:48): The workflow stuff in Gmail also kicked ass. The demo that they showed was you get a bunch of receipts and the person giving the demo, she said something to the effect of, "Well, wouldn't it be great if, like, you know, the AI assistant were able to find all the receipts and then aggregated them and it puts them in a folder and then also actually generated an expense report or like a spreadsheet on the fly?" It's crazy. Yeah. I gotta say, I think that it's free to change your mind. And so it's good to do that. Jason Calacanis (01:24:19): Oh, and I think that Chamath, in a rare moment of reflection, might do a, "Are we gonna have a re-underwriting? Is this a Chamath Palihapitiya (01:24:27): I change my mind all the time. I just, I mean, you know, because I'm... I was... Jason Calacanis (01:24:31): Gentlemen, break. Use Chamath for after writing his Google training. Sorry. I know, but to blow your ears out. Chamath Palihapitiya (01:24:39): I think the Google thing is pretty special. Between last week's announcement of Isomorphic Labs, which, let's be honest, that's just a multi-$100 billion company. Jason Calacanis (01:24:50): So you're saying there might be many. Think about it this way. Right? David Friedberg (01:24:53): Multi-billion-dollar Jason Calacanis (01:24:54): opportunity sitting there dormant inside of Chamath Palihapitiya (01:24:56): Google that AI unlocks. Look at a company like Royalty Pharma. So Royalty Pharma, with a pretty... it's a phenomenal business run by a phenomenal entrepreneur. But what is that business? That's buying 2 and 3 percent royalties of drugs that work. And you can see how much value that those guys have created. Which is essentially 90% EBITDA margin business. It's outrageous because they're in the business of analyzing and then buying small slivers. I think something like Isomorphic ends up being of that magnitude of margin scale, but at an order of magnitude or 2 orders of magnitude higher revenue. So if you fold that back into a Google, if you think about what they're doing now on the search side, these guys may be really kicking some ass. Jason Calacanis (01:25:40): Oh, yes. Chamath Palihapitiya (01:25:41): I think that the reports of their death were premature and exaggerated. Absolutely. Jason Calacanis (01:25:48): of their death, Friedberg, were based upon people not needing to click on the ads. But as I said on this very podcast, my belief is that this is going to result in more searches and more knowledge engagement because once you get how to cook your steak and get the right temperature for medium rare, it's going to anticipate your next three questions better. So now it's gonna say, hey, what wine pairing would you want with that steak? Hey, do you need steak knives? And it's gonna read your mind that you need steak knives, and Chamath likes to buy steak knives, but maybe you like to buy mock meats, whatever it is. It's gonna drive more research and more clicks. So while the monetization per search may go down, we might see many, many more searches. What do you think, Friedberg? You work there? And, when we look at the future of the company and the stock price, Nick, we'll pull it up. Man, if you had held your stock, mhmm, yeah. I don't know. Did you hold? I think David Friedberg (01:26:41): I bought some, Jason Calacanis (01:26:43): No. Your original stock. During the weekend. David Friedberg (01:26:46): Oh, no. I sold all my stock back when I started Climate because I was a startup entrepreneur and needed to live. So which, you know, I I I recently did. I did the math on it. It was pretty it'd be worth It'd be worth a lot. Jason Calacanis (01:27:02): It would be worth a billion or 10 billion? David Friedberg (01:27:04): No. No. Would it would have been Jason Calacanis (01:27:05): a billion or no? Okay. David Friedberg (01:27:07): No. No. David Sacks (01:27:07): Okay. David Friedberg (01:27:08): No, I was not like a sup I was not a senior exec or anything. I I think what you said is is probably true. So that's accretive. I think the other thing that's probably true is a big measure at Google on the search page in terms of search engine performance was the bounce back rate, meaning someone does a search. They go off to another site and then they come back because they didn't get the answer they wanted. And then the one box launched, which shows a short answer on the top, which basically keeps people from having a bad search experience because they get the result right away. So a key metric is they're gonna start to discover which vertical searches, meaning like, hey, cooking recipes, that kind of stuff, like the referencing travel. There are lots and lots of these different types of searches that will trigger a snippet or a one box that's powered by Gemini. That will provide the user a better experience than them jumping off to a third-party page to get that same content. David Sacks (01:28:03): And David Friedberg (01:28:03): then they'll be able to monetize that content that they otherwise were not participating in the monetization of. So I think the real victim in all this is that long tail of content on the internet that probably gets cannibalized by the snippet one box experience within the search function. And then I do think that the revenue per search query in some of those categories actually has the potential to go up, not down. Jason Calacanis (01:28:26): Explain. Explain. Give me an example. David Friedberg (01:28:27): You keep people on the page so you get more search volume there. You get more searches because of the examples you gave. And then when people do stay, you now have the ability to better monetize that particular search query because you otherwise would have lost it to the third-party content page. So for example, selling the steak knives is another, you know, it's kind of a good example or booking the travel directly and so on. So by keeping more of the experience integrated, they can monetize the search per query higher, and they're gonna have more queries and then they're gonna have the quality of the queries go up. So I think it's all in. There's a case to be made. And I haven't done a spreadsheet analysis on this, but I guarantee you going back to our earlier point about precision versus accuracy, my guess is there's a lot of hedge fund type folks doing a lot of this precision type analysis, trying to break apart search queries by vertical and try to figure out what the net effect will be of having better AI-driven one box and snippets. And my guess is that's why there's a lot of buying activity happening in the stock right now. And I think they're probably all missing Chamath's point, a lot of these call options, like Isomorphic Labs. I can tell you Meta and Amazon. What Meta and Amazon do not have is an Isomorphic Lab and Waymo sitting inside their business. That suddenly pops to a couple hundred billion of market cap. And Google does have a few of those. Jason Calacanis (01:29:46): So other bets could actually pay off. These are more. David Friedberg (01:29:50): There may be, look. I mean, there's Calico. No one talks; I don't know what's going on Jason Calacanis (01:29:53): out there. Yeah. Let me get Sacks involved in this discussion. Sacks, when we show that example, it's obvious Google is telling you where they got these citations from and how they built their how to clean your couch, how to make your steak. They were in a very delicate balance with content creators over the past two decades, which is, hey. We're gonna use a little bit of your content, but we're gonna send you traffic. This is gonna take away the need to send traffic to these places. They're gonna benefit from it. To me, this is the mother of all class action lawsuits because they're putting it right up there. Hey. We're using your content to make this answer. Here's the citations. We didn't get your permission to do this, but we're doing it anyway. What do you think is the resolution here? Does all this content go away? Because there's no model. Does Google try to make peace with the content creators and cut them in or license their data? What's gonna happen to content creation when somebody like Google is just gonna take Wirecutter or these other sources that are not behind a paywall and just give you the goddamn answer. David Sacks (01:30:53): Well, look, this is the same conversation we've had two or three times where we're gonna need the courts to figure out what fair use is. And depending on what they come up with, it may be the case that Google has to cut them in by doing licensing deals. We don't know the answer to that yet. By the way, I do know a founder who is already skating to where the puck is going and creating a rights marketplace so that content owners can license their AI rights to whoever wants to use them. I think that could David Friedberg (01:31:19): be very interesting. Jason Calacanis (01:31:20): Yeah. I had a call with him yesterday, and you and I will be on that cap table together once again. Yeah. David Sacks (01:31:26): I don't wanna say who it is because it's gonna let him announce his own round, but we're only participating in the seed round. Look, stepping back here, it's interesting. If you go back to the very beginning of Google, the OG Google search bar had two buttons on it. Right? Search and I feel lucky. I feel lucky was just tell me the answer. Just take me to the best result. And no one ever did that because it kinda sucked. Then they started inching towards it with one box, but you didn't get the one box very often. It's very clear now that Gemini-powered one box is the future of Google search. People just want the answer. I think that this feature is gonna eat the rest of Google search. Now it's a little bit unclear what the financial impact of that will be. I think like you guys are saying, there'll be probably more searches because search gets more useful. There's fewer blue links to click on, but maybe they'll get, you know, compensated through those, like, relevant ads. Hard to say. You're probably right that Google ultimately benefits here. But let's not pretend this was a deliberate strategy on their part. They got dragged kicking and screaming into this by innovation of Perplexity and other companies. Yep. They had no idea. They got caught completely flat-footed, and they've now, I guess, caught up by copying Perplexity. And sucks for Perplexity. I think they're kinda screwed now unless they get Yeah. Jason Calacanis (01:32:43): It's over. David Sacks (01:32:43): An acquisition deal. Jason Calacanis (01:32:45): Exactly. David Sacks (01:32:45): But Perplexity came up with the idea of having citations in Jason Calacanis (01:32:51): your comprehensive search results. Yeah. Which was something David Sacks (01:32:55): Search result with citations and related questions. And Right. They did it extremely well. And quite frankly, all Google had to do was copy them. Now they've done that. And I think it does look like a killer product. Jason Calacanis (01:33:04): And by the way, this was all something that I saw 15 years ago when I did Mahalo, which was my human-powered search engine, and which I had copied or been inspired by Naver and Daum in Korea. They were the first ones to do this, you know, it jumped up because there were only three other markets where Google couldn't displace the number one. Korea, Russia, Japan. Russia had, what was the Russian search engine? God. I can't remember it now. Japan had Yahoo Japan. Which Masayoshi Son had carved out. He wasn't a part of it, and they were loyal to that and very nationalistic Koreans and very innovative folks at Daum and Naver just made search that was so amazing. You do a search and it'd be like, here's music, here's images, here's answers, here's Q&A. It was awesome. But, you know, it just shows you, like, you need to have a lot of wherewithal and timing as everything as an entrepreneur. My timing was 10 years too early and the wrong technology. I used humans, not AI, because AI didn't work 15 years ago. David Sacks (01:34:01): One thing I would say about big companies like Google or Microsoft is that the power of your monopoly determines how many mistakes you get to make. So think about Microsoft completely missed iPhone. Remember? And they, like, they screwed up the whole smartphone mobile phone era? Chamath Palihapitiya (01:34:18): And it didn't matter. David Sacks (01:34:19): Didn't matter. Satya comes in blows this thing up to a $3 trillion public company. Same thing here with Google. They completely screwed up AI. They invented the transformer completely missed LLMs. Then they had that fiasco where, you know, they had Chamath Palihapitiya (01:34:33): George Washington. David Sacks (01:34:34): But George, Washington doesn't matter. They can make 10 mistakes, but their monopoly is so strong that they can finally get it right by copying the innovator. And they're probably gonna become a $5,000,000,000 company now. Sorry. $5,000,000,000,000 company. Jason Calacanis (01:34:46): Reminds me, you know, the greatest product creation company in history. I think we all know who that was and take a look down memory lane. Here are the 20 biggest failed Apple products of all time. The Apple Lisa, Macintosh Portable, we all remember the Newton, which was their PDA, the 20th Anniversary Macintosh, super sexy. People don't remember. They had their own video game. David Friedberg (01:35:10): I was at a conference a couple years ago that Jeff Bezos spoke at, I think he's given this talk in a couple other places. You could probably find it on the internet. Yeah. But he talks about Amazon's legacy of failure and how they had the Fire Phone and the Fire this and the Fire that. And he's like, our job is to fail. Jason Calacanis (01:35:29): Big swings. David Friedberg (01:35:30): We have to make these blunders, but what makes us successful is that we learn from the failures and, you know, we make the right next decision. David Sacks (01:35:38): Yeah. Yeah. But I say you're a startup and you make big failures, you usually just go out of business. David Friedberg (01:35:42): One and done. Yeah. But this is how Chamath Palihapitiya (01:35:43): you say this David Friedberg (01:35:44): is how you say this is how you stay competitive if you're a big founder-led tech company. The only way you're gonna have a shot at staying relevant is to take big shots that you're gonna fail at. I just don't know. You you can't just you have to do things that you're gonna Jason Calacanis (01:36:00): fail at. Right. Remember this boom bust? This is David Friedberg (01:36:01): one of David Sacks (01:36:01): The huge difference between startups and big companies is that big companies can afford to have a portfolio of products. They have a portfolio of bets. Some of them will work, and that keeps the company going. A startup really has to go all in on their best idea. I always tell founders just go all in on your best idea. They're always asking me for permission to pivot. And I always tell them, go for the best idea. Don't hedge. Don't try to do five things at once. Just go all in on your best idea. Chamath Palihapitiya (01:36:26): Yep. Yeah. And if it doesn't work out, Jason Calacanis (01:36:28): you reboot and start with a new table. You're gonna go all in. So to speak, another amazing episode is in the can. The boys are in a good mood. You got your great episode. No guest this week. Just all bestie all the time. And very important, the march to a million continues halfway there. You got us there, fans. We hit 500,000 subbies on YouTube, which means y'all earned a live Q&A with your besties coming at you. In the next couple of weeks, we're gonna do it live on YouTube. So if you're not one of the first 500,000, get in there now so you get the alert. We're gonna take your questions live. It's gonna be dangerous. Any questions? No questions. Who knows what could happen on a live show? And by the way, I just wanna let you know that Phil Hellmuth breaking is Phil Hellmuth and Draymond Green resigned from OpenAI. We didn't get into that, but the OpenAI resignations continue. Phil Hellmuth has tweeted he's no longer with OpenAI. Chamath Palihapitiya (01:37:27): You guys like my baby cashmere pink sweater? Jason Calacanis (01:37:29): That's pretty great. We're gonna get summer Chamath soon, and now the button's coming down. Are you gonna go ahead and linen? When does linen Chamath show up? Chamath Palihapitiya (01:37:36): The unbuttoning is about to happen in the next two to three weeks. Jason Calacanis (01:37:39): The great unbuttoning. The great is how you know, it's kinda like Groundhog Day. You know that summer's here when you choose to lock buttons almost. David Sacks (01:37:46): Memorial Day. When after Memorial Day, the button can come down. David Friedberg (01:37:51): Yeah. We're gonna go three buttons down. Jason Calacanis (01:37:53): I'll see you wearing my black tee. Sacks will still be blue blazer, blue shirt, red tie, and Friedberg in fields of gold. Look at Friedberg in fields of gold, taking us out, staying fields of gold coming at you, two for Tuesday. See you all in the next All-In pod for the Sultan of Science, the Rainman, David Sacks, and Chairman Dictator. I am in the Z100 Morning Zoo DJ. We'll see you next time. Chamath Palihapitiya (01:38:18): Love you, boys. Speaker 9 (01:38:19): Bye. Chamath Palihapitiya (01:38:20): Bye. Bye. Bye. Speaker 5 (01:38:23): Bye. The legend Jason Calacanis (01:38:26): Rainman, David Sacks. Speaker 5 (01:38:27): I'm going home. And it said we open source it to fans, and they've just gone crazy. David Sacks (01:38:39): Like, what Speaker 5 (01:38:47): I'm missing your driveway. Chamath Palihapitiya (01:38:55): Thank you, J Cal, because they're all just useless. Speaker 9 (01:38:57): It's like this, like, sexual tension, Chamath Palihapitiya (01:38:59): and they just need to release their mouths. David Friedberg (01:39:03): Flirt. You're happy. Jason Calacanis (01:39:19): And now the plugs: The All-In Summit is taking place in Los Angeles on September 8th through 10th. You can apply for a ticket at summit.allinpodcast.co. Scholarships will be coming soon. You can actually see the video of this podcast on YouTube, youtube.com/@allin. Just search All-In Podcast, and hit the alert bell, and you'll get updates when we post, and we're gonna do a party in Vegas. My understanding when we hit 1 million subscribers. So look for that as well. You can follow us on x.com/theallinpod. TikTok is all_in_talk, Instagram, theallinpod, and on LinkedIn, just search for the All-In Podcast. You can follow Chamath at x.com/chamath, and you can sign up for a Substack at chamath.substack.com. I do. Free bird can be followed at dotcom/freebergandohaloishiring. Click on the careers page atohalogenetics.com, and you can follow sacks@x.com/David sacks Sachs recently spoke at the American moment conference and people are going crazy for it. It's pinned to his tweet on his ex profile. I'm Jason Calicanis. x.com/jason. And if you wanna see pictures of my bulldogs and the food I'm eating, go to instagram.com/jason. In the first name club, you can listen to my other podcast, This Week in Startups. Just search for it on YouTube or your favorite podcast player. We are hiring a researcher. Apply to be a researcher doing primary research and working with me and producer Nick, working in data and science, and being able to do great research, finance, etcetera, all in podcast.co/ research. It's a full time job working with us, the besties, and really excited about my investment in Athena. Go to Athena Wow. See you in a wow.com and get yourself a bit of a discount from your boy, Jakeowel.com. We'll see you all next time in the all in podcast.