Skip to content

Podcast

The Stablecoin Future, Milei's Memecoin, DOGE for the DoD, Grok 3, Why Stripe Stays Private

All-In with Chamath, Jason, Sacks & Friedberg

Source ↗ ← All highlights
  • Missed Opportunity
    • Chamath missed out on investing in Stripe early on.
    • He met the Collison brothers 18 years prior when they were working on their previous startup. Transcript: Patrick Collison The last time we met Chamath was 18 years ago when we were working on our prior startup, Octomatic, with Harj and Kul Tagar. You were how old? They were 17, 18, 19. It was one of these san francisco setups where it was like a two-bedroom apartment there was a few of us living there i think maybe six people working out of there and a normal number exactly Normal numbers to load up a two-bedroom apartment with and then chamath you came and visited this is what’s so brutal about this okay i could have invested a dollar one single dollar Chamath Palihapitiya And I would have made a billion dollars. I remember meeting these guys. And I was with Alan Morgan, who was my boss at the time. I was a junior principal at Mayfield. And I think we tried to, guys, I don’t know if you remember Patrick and John, I think we tried to invest in the business or it didn’t happen, or then you ended up shutting it down. But right away, you spun back up and started Stripe. And I just watched from the sidelines the whole way. It is such a… First of all, it’s an amazing… It’s totally… Well, no, it’s an amazing place for Silicon Valley where you can see these people just keep pushing the boundaries up and up and up, number one. Number two, the thing that is such a learning for me is like, why didn’t I just pick up the phone and call them at any point in the last 17 years? What am I thinking? It’s so brutal. (Time 0:01:10)
  • Stripe’s Target Market
    • Stripe initially targeted startups because they believed payments were broken for them.
    • They later realized the issue affected everyone, expanding their services to large enterprises. Transcript: John Collison Two things. One, first, you probably don’t remember this, but I remember that meeting that we offered you, do you want something to drink? We did not have a broad selection. I think we had water or milk in the fridge and you asked for a glass of water. And so I went over to the sink and I realized that we hadn’t really been on top of the washing up. So I had to sort of gingerly wash a glass for you to get your glass of water, which I can’t remember if you touched it over the course of that meeting. But then secondly, like when we started out with Stripe, like the fintech sector basically didn’t exist. I mean, the word hardly existed and… Exist. Yeah. People just didn’t think that, I mean, you know, teenagers weren’t actually teenagers at the time, but, you know, people in their early twenties, college kids taking on, you know, PayPal or the incumbents or regulated financial services or whatever, you know, people just didn’t think it was a good idea. So I don’t know, you, you, you certainly were like the vast majority of investors we spoke with in the first year or two of Stripe turned us down. So you were not anomalous. (Time 0:02:50)
  • Stablecoin Use Cases
    • Stablecoins are becoming increasingly useful, particularly for cross-border transactions and treasury management.
    • They offer US dollar services globally and address the limitations of Bitcoin as a payment method. Transcript: Jason Calacanis Stablecoins, we’ve got to talk about stablecoins. Stable? Patrick Collison Why’d you do a stable Stablecoins are finally happening and they’re really useful. Like we followed crypto for a long time. The Bitcoin white paper dropped in 2008, the year before we started working on Stripe. And so it’s been funny where Stripe and crypto have grown up together. And, you know, we tried to make Bitcoin happen as a payment method on Stripe. Just wasn’t that good as a payment method. I mean, it’s good as a store of value, as kind of a gold substitute. But transactions are slow. Transactions are expensive. You never know exactly how much you’re going to get because it’s not denominated in dollars. The stablecoins are now really good. If you look at, you know, something on an Ethereum L2 or Solana or something like that. And so we bought a company called Bridge late last year, who is building the stripe of stable coins. And so if you’re, I mean, I think you guys have talked about them a little bit, but, you know, people like SpaceX using them for treasury management, people using them to offer US dollar Services to people all around the world. Just stable coins are, I think, the first really big payments use case. And I think it’s finally coming because the tech is good enough. Chamath Palihapitiya Is there a moment, guys, where is it a regulatory event where you’ll say the Visa MasterCard duopoly can get challenged? Is there a set of boundary conditions that you have written down where when you can check a few of these boxes, you know that it’s time for those companies to get dismantled? Patrick Collison The behavior we’re seeing right now is that stablecoins are most interesting and seeing most adoption where there is some cross-border component. And so you need to manage corporate treasury around the world. You want to send remittances to people in other countries. Often it’s people in other countries want to hold dollar balances or things like that. What we’ve always seen is that, I don’t know, in the US, things work pretty well. In Europe, things work pretty well. And so we even see this pre-crypto where the way people pay for stuff has been radically changing. UPI in India, PIX in Brazil, you have all these designed by central banks, actually really good kind of government-run Venmo solutions. Those have all happened in emerging markets broadly, and not in the US and Europe. We certainly keep our eyes peeled for that changing at some point. But I think right now, I don’t know, Patrick, would you characterize it that way, that a lot of the interesting stuff we see is happening internationally? Yes. John Collison So with respect to Visa and MasterCard, I think an important thing to keep in mind is that most of the interchange fees that are charged to merchants, and you mentioned what we charge The All In podcast, the vast majority of that flows right back to the issuing banks in the form of interchange. And almost all of that flows right back to the consumers in the form of the lending that the cards themselves represent, but then also in card rewards. Card programs are not actually big profit pools for most of the major banks. And so I think any substitute for Visa and MasterCard in that sense, you know, the sort of a question of, well, are the consumer rewards going to go down? Are the consumer protections going to go down? We’ll be extending less consumer credit. And maybe other points in that space are viable, but, you know, it is a set of trade-offs and it’s not as simple as this enormous rent extraction happening. John’s totally right. I think the interesting use of stablecoins is cross-border is outside the US. I mean, the big use case that’s taking off right now is consumers in other countries seeking to hold dollars. You know, we in the US here today, you know, we obviously benefit from being able to do that. The vast majority of people in the world have kind of a, are subject to a worse currency, worse in the sense that it’s less stable, it’s more inflationary, you know, storing savings is Much less favorable. You know, if you look at the Naira, for example, there are a lot of people in Nigeria and the currency there has devalued by a factor of, you know, three or four over the last couple of years. And so that use case of consumers being able to store dollars is really exploding. And we think about this really as kind of an analogy to the euro dollar system, where the euro dollar system in the 70s and 80s, this was a way for companies to store dollars and to have something More stable and reliable and so forth. But it was only, I mean, I think the minimum transaction size was like a million dollars. And so there’s kind of a very high barrier to entry. Whereas with stable coins now, you can be a consumer in Ecuador and you can have like a 10 US dollar balance. And that was just not a product that was accessible to you before. And so I think it’s a really big deal, certainly for people in those countries. And in some sense also for the US, because the dollar status as the world’s reserve currency, I think, is in the process of becoming much more deeply established. (Time 0:06:45)
  • Remote Work Pros and Cons
    • Remote work has benefits like access to a wider talent pool.
    • It can be detrimental to early-career individuals needing mentorship and social interaction. Transcript: Jason Calacanis Right, Freberg, before we get into the dock, you got any question for the boys here? David Friedberg If you were to kind of build the financial system from scratch today, we’ve got SWIFT, we’ve got banks that store assets, we have credit cards, and these credit card networks, then we’ve Got transaction service providers that sit on top of this. What’s the right solution if we were to build a financial system for the world from scratch today? And can you guys see a world where we bridge away from the credit card networks, where we move out of some of these legacy systems? Are they so deeply ingrained and everything that it’s it’s going to continue to be this thing where we’ve got to build these complicated solutions into and around the legacy of financial Patrick Collison Infrastructure i’ll give my view and then i’m curious what um patrick has i would say firstly there is just general tech scalability. The finance industry has its version of the mineshafts, for sure, where everything should be highly scalable in real time. And I think in a way, stablecoins are solving something that you don’t technically need full decentralization to do, but the ability to make real-time payments any hour of the day or Night is a useful property. And again, some private systems have also built that. I think a big one for us is trust. And the fact that the fraud problem hasn’t really been solved in online payments, a big reason people come to Stripe is basically we are a reputation network across the internet economy. And so when someone comes and buys something from a Stripe user, 93% of the time, we have seen that card before. And so the merchant can know something and know that they can trust this end user. And it’s gotten to the stage now where if someone comes along and buys with a credit card, if they’re signing up with an email address or a phone or something that we haven’t seen before, That is just ipso facto suspicious because, you know, they are coming along and maybe trying to, it’s a stolen credit card or something like that. And so a big part of what Stripe ends up doing is acting as a reputation network to keep fraud out of the system that maybe you would have wanted to design in from day one. (Time 0:16:56)
  • Dimon on Bloat
    • Jamie Dimon believes excessive bureaucracy and unnecessary meetings lead to inefficiency.
    • He claims he could reduce staff by 10% and maintain efficiency. Transcript: Jason Calacanis Right, Shamath, I want you to respond specifically to this next clip. Let’s play the second clip about organizational bloat. Every area should be looking to be 10% more efficient. Chamath Palihapitiya If I was ready to depart 100 people, I guarantee you, if I wanted to, I could run it with 90 and be more efficient. I guarantee you, I could do it in my sleep. Jason Calacanis And the notion, these bureaucracies, I need more people, I can’t get it done. No, because you’re filling out requests that don’t need to be done. Your people are going to meetings they don’t need to go to. Someone told me to approve some of his wealth management that they had to go to 14 committees. I am dying to get the name of the 14 committees. And I feel like firing 14 chairmen of committees. I can’t stand it anymore. All right, Shemath. The bloated bureaucracy at big companies, your thoughts? (Time 0:25:19)
  • Software and Bloat
    • Chamath suggests organizational bloat stems from rigid software-defined roles and reliance on off-the-shelf solutions.
    • Custom-built internal tools can increase efficiency. Transcript: Jason Calacanis Well, you know, there’s that adage that says something akin to 50% of advertising is useless. Chamath Palihapitiya We just don’t know which 50%. Yeah. I think it’s probably true for most corporate structures in general, which is that a lot of the organizational bloat has evolved because of the way that people have responded to how You use technology. So meaning, if you look back 50 years ago, if you look at that famous picture of the Microsoft early team, they didn’t rely on software necessarily. There wasn’t Salesforce, there wasn’t Workday, there wasn’t all of this infrastructure. And so instead, they probably organized by what they were good at, and they just tried to do things efficiently. And I suspect that many companies in the absence of technology found a way to just be very efficient. That started to change when you had these rigid demarcations of where one job ended and another job started. And part of why that happened is because you had all this software that went in and convinced people. This will create efficiency, but in return, the chief marketing officer’s job is X, Y, and Z. This is how the roles are defined. This is how people do it. And so I think that the reason why things have become so bureaucratic and bloated is that there is just this propensity to run towards software because you think it’s a solution. At best, it’s a symptomatic aid. It doesn’t address the root cause. And in fact, it promotes bureaucracy and it promotes the bloat that Jamie’s talking about. And if you look at Jamie’s P&L, he spends $16 billion a year on IT. And I suspect that if you streamline that, you’d actually have half as many people because they’d be doing the job in a wholly different way. And by the way, the counterfactual to it is if you look at companies like Facebook, or Google, or Tesla, or SpaceX, who designs and I’m sure Stripe is the same who designs a lot of stuff Internally, that’s custom built for their org. I think the way that you see this in the revenue per employee and a bunch of these other metrics in terms of the efficiency of those companies. So I think what he is talking about is that he is a victim of this push to productivity because he would look like a Luddite if he didn’t adopt technology. But by adopting the off the shelf stuff, he introduces organizational bloat because these things are demarked very, very rigidly. Yeah, you got the marketing team, as you mentioned, using HubSpot, and then you got the sales team using, I don’t know, Salesforce. Tool upon tool upon organizational bloat. The other thing I just want to say on the first topic is I’ve mentioned this before. Other than engineers who are naive but can be extremely productive from day one, there are very few other job types where naivety is an asset. Most people early in their career are in a J-curve where they are negatively contributing. Yes, slowing everybody down. And the whole goal is that you invest in these people so that they come out of the J-curve. There are probably other jobs that are like engineering, but many, many are not. And so I think it’s important to get the kind of mentoring you get by being in an office. And in the absence of that, I think these young people, like Jamie said, are totally lost. (Time 0:26:06)
  • Defense Budget Cuts
    • The Pentagon was ordered to cut its budget by 8% annually over five years.
    • This could lead to $300 billion in savings, raising concerns about audits and technological advancement. Transcript: Jason Calacanis One of them is it’s one-sided. We’re only hearing about, you know, people on the left doing grifts and USAID. The other one is, hey, you’re pointing at little tiny things like USAID. When are you going to get to defense spending and social security? Well, here we are. Washington Post is reporting that in between doing sets of 47 pushups, Defense Secretary Pete Hegseth asked senior leadership at the Pentagon to develop a plan to cut 8% from the defense Budget each of the next five years. That’s compounding 8%. Here’s a chart. We’re talking about close to $300 billion in savings over five years if they hit, which isn’t a crazy target, 8% a year. It’s just crazy in our country where we haven’t even been able to have our Defense Department pass a basic audit if you’ve seen those type of reports. Let’s pause there and just talk about military spending, Chema. Chamath Palihapitiya I think that military spending needs to sit downstream from technology because if it doesn’t, you’re sort of misappropriating the money. And what I mean is that we’re inventing incredible capabilities in AI and autonomy. I think that you need to take those things first and figure out how to projectize them because I think that builds the kind of modern war machine we need. Otherwise, what happens, I tweeted about this, Nick, maybe you can find it, but the CBO red flagged a project where the Navy was about to appropriate $1.2 trillion to build frigates. Now, there’s a body, I think, of military planning that says this is a projection of power and so you need to spend this kind of money because people want to see the big boats and the big Iron in the water. Okay, and maybe there’s something about that. But the reality is you can’t be spending three or $4 billion a boat and taking, you know, eight, nine, 10 years to build these things. It’s so this is not sustainable. And part of why they do that is it’s not coupled to what’s actually happening with respect to innovation, where there are core pockets of companies. Saronic just announced a $600 million raise today. SailDrone announced hundreds of millions of dollars of contracts with the Navy. Andrel is doing that with the Army. So I think that military spending needs to happen downstream from what’s actually happening in technology, broadly speaking, we don’t have that. What you have instead are system integrators with extremely deep connectivity that are able to contract well, not (Time 0:34:26)
  • Military Spending and Tech
    • Chamath argues military spending should follow technological innovation in AI and autonomy.
    • He criticizes expensive, slow-to-build military projects. Transcript: Jason Calacanis Necessarily to invent well. Freiberg, any thoughts there on cutting defense spending? Obviously, we have, to chumat’s point amazing founders like my guy palmer lucky cutting the cost of very important uh he’s not your guy he hates that’s my guy he hates bestie oh no we it’s All a joke everybody calm down you just have a little literally jason no i talked to listen i know all the board members i got a personal relationship with everybody. He’s going to send one of those drones to your ranch. He loves J-Cow. He loves it. Everybody needs a foil. David Friedberg It’s all sorts of problems for the rest of us. When you go out and talk shit about people for no reason, it’s great. What are you talking about? Jason Calacanis I never talk shit about Stripe. Guys, you don’t have any beef with J-Cow. You just call your employees Mids for no reason. David Friedberg For no reason. You’re just like, oh, what about your mids? That was a hypothetical. Okay. Jason Calacanis I did say that you guys pocketed 500 large. Patrick Collison Look, we can come to the summit and Palmer style just like- Oh my God. Yeah. Long list of things I’ve said about you. David Friedberg Please. And another thing. Please. Yes. Absolutely. Jason Calacanis Shout out to my guy, Palmer Lucky. But what do you think, Freeberg? For serious, let’s get back on track here. David Friedberg Okay. So here’s what I think. If you take the fence down to first principles, There was an excellent tweet that we were all texting about yesterday. It made the observation that Trump’s negotiations with Russia and China, where there’s all of this hemming and hawing about those negotiations being complying with the wants and Needs of dictators, may actually be a shift in strategy on the global relationship the United States has. My has with other global powers. In particular, a shift from the objective being about U.S. Primacy and the U.S. Being kind of the sole great power on earth to recognizing that that’s no longer the case and that in a multipolar world, we no longer need to invest in wars, need to invest in conflicts, Need to invest in defense with supposed allies to try and build up our strength across the globe. And I’m not saying that this is necessarily the right strategy, but it was an observation that maybe the strategic imperative is now to have kind of a multipolar stance in the world rather Than a stance of primacy. And in that framing, you then ask the question, okay, make that the case. Now, if we do agree that we are all going to settle into a new world where China, Russia, the United States are not necessarily equal powers, but shared powers across the globe. In that context, do we need to have as much of an investment in global defense? Do we need to continue to pour dollars into building up arsenals and military bases and troops and stations and positions all around the world? Perhaps not. Perhaps the world gets divided peacefully and we open up global trade relationships and everyone benefits economically from the advances in technology and improvements in productivity And the world order is peaceful but multipolar. Maybe that’s the new era that we’re entering. And in that context, you don’t need as much of a defense. And separately, to Chamath’s point, there’s different technology that’s now in play. We’ve seen it in the Ukraine-Russia context that a $10,000 drone can destroy a $10 million piece of equipment. And China now has drone factories that can output millions of drones each month. So if China develops this new type of arsenal with millions of autonomous flying systems that can go and attack troops and attack expensive pieces of equipment, do we really need aircraft Carriers? Do we really need tanks? And I think that’s the whole Hegseth-led, Trump-led conversation that’s underway in defense right now. Number one, multipolar. Number two, therefore, we don’t need as much defense spending. Number three, maybe the defense spending that we do do should account for the new technology in play in the battlefield. And that really changes the character of how the defense department is structured and how funding is structured. So that’s really, I think, the way to look at it versus, hey, let’s just cut defense spending for cutting sake. And that might be what’s going on right now. (Time 0:37:00)
  • Milei’s Meme Coin Disaster
    • Argentinian President Javier Milei promoted a meme coin called Libra, which crashed 95%.
    • He faces lawsuits, impeachment attempts, and an investigation. Transcript: Jason Calacanis Last week, Argentine President Millay, who is a hero to a lot of people on the right or for government efficiency, promoted a coin it was called libra dollar sign libra and he originally Tweeted this private project will be dedicated to encouraging the growth of the argentine economy with a link to libra for his citizens to go buy it and buy it they did but he deleted that Tweet when this whole thing came apart and said i was not aware of the details of the project and after having become aware of it i decided to not continue spreading it the market cap ripped 4 billion it crashed 95 as these meme coins always do 74 000 traders lost almost 300 million 24 wallets had losses over a million and malay has been sued 100 plus times already and this Just happened last week he’s being investigated by his own government now and uh there is an impeachment uh attempt underway by the opposition malay’s team told cnn that his endorsement Of the coin was a mistake. Really? Oh, wow. Going out on a limb there. According to insiders, Malay never actually owned any Libra and was not associated with the coin. I think family members maybe put him up to it. The details of why he promoted it remain a little bit unclear. There’s a lot of speculation. Jamath, your thoughts? Chamath Palihapitiya It’s kind of crazy. I mean, he was on such a positive upswing of momentum. It doesn’t make much sense why he got embroiled in all of this. The problem with this, though, is I think that the cover-up is always worse than the crime itself. So the first message was very Clinton-esque, like I did not have sexual relations with that woman. Was like i did not endorse it i just shared it was his justification for how he how he um could rationalize what he did the kid that’s behind this thing hayden davis i think is his name he Jason Calacanis Was on coffee zilla was an incredible one hour did you see that the coffee zilla well i saw i saw some the clips on X and it was pretty brazen because he essentially said that he had Javier Chamath Palihapitiya Malay in his pocket. And then there were text messages that use some pretty colorful language to basically say the same thing. Then on top of that, there were some text messages that seemed to implicate Malay’s sister as having got some of the money from all of this. I don’t know. The whole thing just makes absolutely no sense. He was doing so much good, and now he’s going to go through this whole cycle of trying to wash his hands of this whole thing. It’s a complete waste of time and effort. I don’t know why he did this. Jason Calacanis And there was another interesting little tidbit, Friedberg, friend of the pod, David Portnoy, supposedly he’s been getting in on this and he’s a gambler and he loves gambling and he Looks at his gambling, obviously. He had put reportedly millions of dollars into this and this guy we’re talking about gave him his money back. This guy also has something like a hundred million dollars sitting in a bank account anywhere. What’s your take on all these meme coins, Freedberg? David Friedberg I don’t like them i don’t think that they’re like good i don’t think they’re productive all right i think that a bunch of people are going to put money in and lose money and a few people are Going to make a lot of money and you know but at the end of the day it’s no different than the people that sell trading cards or the people that create and sell collectibles and get paid for Them and this is just effectively a digital collectibles business. Unfortunately, I think it’s amplified by like a thousand X because collectibles businesses have friction and they’re manual and you got to ship them. And this creates a bit more of a digital frenzy where you see the social feedback loop happen really quickly in real time. And that drives these things to a high value, which means people have the ability to lose a lot more than they otherwise would. But look, I mean, these are not helping the financial system get rebuilt, as we talked about earlier. They’re not creating productive value. They’re entertainment mechanisms, just like any other kind of gambling system might be. And, you know, people can choose to do that if they want. But personally, I’m not into it. I just think it’s stupid. But whatever. (Time 0:43:57)
  • Arc Institute’s Focus
    • The Arc Institute, a non-profit with around 230 people, focuses on curiosity-driven biological research.
    • It aims to address complex diseases with new technologies like single-cell sequencing and AI. Transcript: Jason Calacanis Yes and we’re the co-founders and then there are two scientists and you guys are funders of it or yeah maybe you guys give us the intro a lot of money into this yeah yeah so uh the institute John Collison Is a non-profit it does basic biology research it’s in palo alto next to stanford it’s about 230 people today and uh yeah john and i are among the funders of us but there’s a bunch of other Very generous donors can you explain how this idea of curiosity-driven research that’s on the website? Yeah, there’s kind of two things behind it. So the first is the vast majority of biology scientists today receive NIH grants doing basic research. And the NIH grants are, one, just hard to get and annoying to get. Scientists spend 40% of their time working on grant overhead and so forth. But worse, even more perniciously, the grants are very restrictive in terms of the kind of science they can do. And so we ran a survey of scientists back a couple of years ago of top scientists, and four out of five, like 79% of them told us that if they could just spend money however they wanted, if They weren’t kind of limited by what they’re prescribed by these NIH grants, four out of five told us they would change the research agenda a lot. And so I think the analogy here is imagine if there was only one VC firm and it was run by the government. How would that change? And that VC firm had strong opinions on what kind of companies people should build. Exactly. Literally, the grant panels at the NIH are consensus-based, like explicitly. They’ve consensus-based scoring mechanisms and they penalize you if you’re doing work outside of your field and so forth. So we go to all this work to train these amazing scientists, and then we don’t let them pursue their best ideas. That’s kind of problem one. And the ARC investigators, they’re fun to do whatever they want, curiosity-driven research. And then the second thing behind ARC is this idea that you can kind of divide diseases into three categories. You have diseases, and we, broadly speaking, know how to generate cures for and treatments for infectious diseases. We have monogenic diseases, like one genetic mutation or something, and we don’t know how to cure those in most cases, but we can at least screen for them and so on. And then we have what the biologists call complex diseases, where there’s some kind of gene-environment interaction. That’s most cancers, most autoimmune diseases, most neurodegenerative diseases, and so forth. Alzheimer’s, things like that. Exactly. And we’ve never cured a complex disease. And many of these diseases are very tragic, precisely because not only have we not cured them, we don’t even have treatments, as John says, in the case of Alzheimer’s, for example. And so the question is, can we do something about this? And what would a research agenda and program that can help shine some light on these complex diseases look like? And our hypothesis, we’ll see how much it’s borne out, but our hypothesis is that we’ve gotten a couple of new technologies over the last couple of years, single cell sequencing. We can sequence the DNA or the RNA just like in one cell. We’ve, you know, fancy new functional genomics and CRISPR technologies. So you can make these, you know, fine edits and perturbations, again, even just in a single cell. And then obviously you have transformers and AI and ML and all this stuff. And this is kind of a new read, think, write loop in biology that just didn’t exist a decade ago. And again, the question is, is this powerful enough now to solve some of these previously intractable diseases? And so yesterday, ARC released this new foundation model for biology. It’s the largest biology ML model ever. It’s actually, I think, the largest open source AI model ever. Jason Calacanis This is Evo 2 you’re talking about. Exactly. Evo, the number two. John Collison Evo 2. Yeah. And so it’s not just open weights, like the DeepSeq model or LAMA or something. It’s actually open source since the training code is public. And people can go read the blog post or the paper or whatever. The thing I find amazing at Evo and that just really surprised me is, so it’s trained on 9 trillion base pair gene tokens. So, you know, chat, GBT, LLMs are normally trained on human language. This is a language model, but it’s trained on DNA, the language of life. And there’s only one human genome in the training set. It’s mostly other species. And even though it’s only seen one human genome, it’s state of the art at predicting the pathogenicity of human genome mutations. And so, you know, a famous mutation is the BRCA mutation for breast cancer. Like it’s state of the art at predicting the pathogenicity, the harmfulness of BRCA mutations. Despite never having seen one in humans. It’s only seen one human genome and that human, you know, did not have these pathogenic mutations. And so it’s kind of learning something deep across the tree of life. And I don’t know, I find that pretty cool. David Friedberg And sorry, is there a data set that’s used in training? So I think like, you know, when you’re typically, right. And so when you’re building models in typical, like genotype by phenotype models, you’re trying to look at the phenotype, the physical characteristics of the organism, what can it Do? What does it look like? What are the features? And then you look at the genome and so that tells you hey these are the specific genes or alterations or mutations that drove this particular phenotype is kind of what the model tries To learn over time with the objective being hey can i ask it to define a genotype or a genome based on a phenotype based on a physical set of characteristics i’m looking for vice versa maybe You can just help us understand what is it trained on and how did that kind of prediction in BRCA, how is that possible? Great question. John Collison So it’s totally unsupervised. That is to say, you’re just showing us lots of genomes and any kind of latent structure that it learns is just based on trying to figure out how to kind of organize that knowledge, but we’re Not showing it any labeled data or phenotypic outcome data or what have you. And so then you can give it a genetic sequence and ask, relative to its understanding of the genetic universe, how likely is this particular sequence? And so then you can do things like predict anomalousness or pathogenicity or whatever. You can also then, kind of using the embeddings of the upper layers, we don’t need to get too technical here, but you can train another model on top of the model. And even if you showed maybe only a couple of examples, it learns very quickly, okay, here’s how the weights of Evo2 correspond to this particular task. And those sort of models trained on top turn out to be really accurate. Chamath Palihapitiya You guys open source the base model or you open source the fine tuned or both? John Collison We open source the base model, but there’s no kind of proprietary reason that we didn’t open source the fine tunes. It’s really easy to produce them. And yeah, if anyone wanted one of them, we’d happily share it. Chamath Palihapitiya Where does it stand in the spectrum of different tools that folks would use to solve these life sciences problems? There’s cell models that are being developed by some, then there’s these protein models. Where does this fit? John Collison This kind of landscape (Time 1:00:24)
  • Pre-training Upside
    • Chamath was surprised by the value gains from larger pre-training clusters, challenging earlier assumptions.
    • Elon Musk’s Colossus project demonstrated this, leading to increased confidence in NVIDIA’s potential. Transcript: Chamath Palihapitiya And so I was not convinced where all this CapEx would go in a productive way. Like, why are you buying all these NVIDIA GPUs? And then I think if you looked at Colossus, Elon’s Colossus. Jason Calacanis XAI built the largest data center over 100,000 GPUs going to 200,000 in 122 days. Chamath Palihapitiya I mean, basically what he proved was that there are still valuable gains in pre-training. And so the larger the cluster, the more value that there is. Now, he also benefits, I guess, from the X feed, but that was really interesting. So now I’m like a little bullish on NVIDIA. I’m like, oh my God, if this is true, then all this CapEx may be justified. You could be buying a lot of stuff. Then, look, I actually also, just to maybe riff on this Grok3 thing for one second, I had three takeaways. My first takeaway was I was sneakily surprised on the pre-training upside on having a larger cluster. So I think that that’s very pro-NVIDIA, actually. And it’s actually also just really good in general for foundational model makers. So I think that’s a really positive thing. The second thing is, I don’t know if you watched the live stream, but did you guys hear some of the stuff that these guys had to pull to pull this thing off? One of the most incredible. So the way that Elon narrated it was, we first had a physical problem. So we just had to search all around the country for one single location where we could actually put 100,000 GPUs, and they found it, which was an old Electrolux factory in Memphis. Memphis. That’s kind of interesting. Then power. He only had like 15 megawatts, and he had to get a quarter gigawatt, and so he had to basically buy every useful generator that was available. But then they had to liquid cool it. And so they bought one third of all the portable liquid cooling capacity in America and located it on-prem. But then they figured out that there was a power problem. So then they took all these Tesla power packs and then had to do power smoothing, which had them had to rewrite all of the power pack firmware. In all of this, you know how we talked about DeepSeek being this moment where we had lost sight in America of capital being the source of innovation? He proves actually a more generalized rule that I took away from this, which is you always have to have a constraint. So meaning, let’s say there’s like infinite capital in his case, and infinite talent, because he can basically recruit anybody he wants. What did he do instead? He created this artificial constraint of time. And so he was just able to say, you’re going to get this done in a moment. And Nick, I saw the third graph that the guys at artificial analysis sent to me. I just want to put it up here because it shows you guys. The quality of Grok 3 relative to the amount of time that they’ve spent on this problem is to me what’s staggering. So if you just sort of project the rate of change of this, and this is without judging OpenAI or Anthropic or anything else, those guys have been doing it for years. These guys have been doing it for a year. And they did all of this MacGyver engineering and were able to pull this off. So that’s my second takeaway is that innovation needs a constraint. Sometimes it’s capital, sometimes it’s talent, and sometimes it’s time. And so if you can basically be just completely rigid on one of those dimensions, you can get a great team to create something. So that was an interesting takeaway. And then the third is, I think what this also speaks to is the notion of like a kuretsu, right? Which is like the Japanese word for like companies that work together while still remaining independent. Conglomerates, yeah. It’s more interlinked companies. It’s interlinked. Sure. You know, Koreans have chaebols, right? Samsung. Japanese have kuretsus. But this is the manifestation of an American kuretsu, which is Elon is able to get engineers from Tesla. He’s not just buying the power packs. He had them re-engineer the actual firmware in real time on site. And so there’s this positive ability to just like organize effort and human capital. Like, look, could we all stand up a data center and go and buy $500 million of power packs from Panasonic? Absolutely. It would take a few months. And then when it looked like we need to rewrite the firmware, it would take another 18 months to your point, Jason. So it’s really incredible what these guys are able to do together. Those were right. It was really, really inspiring. Chamath, (Time 1:19:11)
  • Asteroid Impact Probability
    • NASA lowered the probability of an asteroid hitting Earth in 2032 to 1.5%.
    • Even if it hits, the impact might not be catastrophic due to its size and likely ocean landing. Transcript: Jason Calacanis What do we want to do first, boys? Do you want to talk about this asteroid coming? Dave? Is it the end of the world? If it hits us, what’s going on? NASA dropped the probability of it hitting Earth to one and a half percent. David Friedberg So every day when the sky gets dark, they can do a better job seeing this asteroid that everyone’s freaking out about. So we finally got a good night sky two nights ago. The telescopes were able to get a better trajectory reading on it. And that allows the models to make estimates on the probability of this asteroid hitting Earth in 2032 when it’s projected to cross our orbit. And so right now, the probability is estimated at 1.5% that it will hit the Earth. And based on the size of this asteroid, there’s this range. It goes up to 320 feet in diameter, as small as 80 feet in diameter, which actually can have a pretty big effect on how big of an energy release there would be if it actually hit the earth. So even on the high end, if it was, call it 300 feet, it would be the equivalent of, call it a 20 megaton bomb, which is not insignificant. If it were that big, it would hit the earth. If it was smaller than that, it would probably just detonate the air and create a massive shockwave and firestorm. But the region that it would decimate would be limited to probably a couple dozen miles, up to a thousand miles of effect. And if you look at the total surface area of the earth, you know, we’re talking about 10 to 15% of the earth having people that habitate, you know, enough people to habitate. Probably going to land in an ocean, right? I mean, right. Yeah. So it’s one and a half percent chance of hitting the earth and then call it a 15% chance if it hits the earth, they’re causing loss of life. Jason Calacanis 10 basis points. It hits a city. Yeah. And then, right. David Friedberg And then it’s a function of how big it is. If it’s actually as small as 80 feet, then it’s not going to be that significant, even if it does get close to inhabited areas so yeah i’m not losing sleep over over did you come across in Patrick Collison Your research i feel like this is a real boys are monitoring the situation that moment did you come across the tunguska event yeah research on this so that one it’s incredible yeah so David Friedberg Uh i don’t you want to talk about it go ahead yeah just um no one knows this in 1908 an hit the Earth. Patrick Collison It hit a relatively uninhabited part of Russia. First off, the asteroid did not hit the Earth because it got so hot on reentry, there was an airburst. And it was a thousand Hiroshima’s in size. The explosion, yeah, they have here the 60 meter asteroid and they have the megatonage somewhere. Wow. It’s the largest impact event in recorded history. Obviously, there was stuff before recorded history. It flattened 80 million trees. Weirdly, basically no one was killed because it was so uninhabited. But this is quite comparable to the one that NASA is talking about. That’s right. It’s about the same size. Exactly. And I think you can take a little bit of reassurance, maybe, that we have had similar size asteroids hit before. And there is some existence proof that, despite the giant explosion, you know, it doesn’t show up in the climactic data for 1908. David Friedberg Yeah, the Tunguska asteroid was at like 160, 200 feet. So if this asteroid is in that range and it hits the Earth, you have this kind of explosion in the air. If it gets above, I think, 250 roughly is where they think that it doesn’t burn up fully in the air and it actually will strike the earth. But yeah, that’s, um, this is roughly what we saw happen, what we think the size will be if it hits. Jason Calacanis Is there a counter measure? I don’t mean to get all sci-fi here. Great question. Yeah. David Friedberg No, but is there a counter measure possible and like if this thing was coming let’s say in five years and it was twice the size yeah yeah but relative to the earth this is like tens of thousands Of kilometers an hour right it’s it’s a it’s a very fast moving object it’s pretty small right 160 feet so you’ve now got to figure out the exact trajectory get it perfectly right get a Launch off of the Earth and intercept this thing at the exact moment that you need to to push it off course or detonate something nearby it to redirect it. So technically very complicated, very hard to pull off. But this is exactly why we have this planetary defense funding at NASA, which is to track these objects. And this is another example, by the way, where I would say AI can play an important role. And I’d love Patrick and John to opine on this, but I have a thesis that like AI, more than anything, unlocks deeply complicated projects for humans that would otherwise be kind of infeasible In the pre AI era. I think in the post AI era, we’re going to be like, oh, here’s all these projects that we do that are like, oh, we, you know, we, we, on a daily basis, we mine to the center of the earth and we Get cool like rare earth minerals from like 500 miles down. And we go to space and colonize the moon and all these crazy things because AI unlocks these large scale projects that would require millions of people to do things in a coordinated way. And AI can be very smart in this way. But I think AI could play a role also in these planetary defense initiative concepts, J-Cal, in the future, where you can actually build a complete project model in software on how you Would actually address this problem and then, you know, go execute it with automation. But yeah, there’s a planetary defense function at NASA. They track these objects, and they’re funded to do it. So we hope that NASA continues to get funding to do this work. Very important. And guys, it just came through that NASA just dropped the probability of an impact event to about one third of one percent. So it’s gotten even smaller, which is we can all go to sleep comfortably. All right. (Time 1:30:20)
  • Stripe’s Stance on Going Public
    • Stripe prioritizes remaining private as it benefits their current growth and focus on long-term value.
    • They believe public market discipline is not essential for effective company management. Transcript: Jason Calacanis All right. Now, good morning. Everybody’s been waiting for. Patrick, John, you founded the company in 2010. It’s 15 years later. The entire LP industrial complex and venture capitalists everywhere, I’m sure some employees, are wondering, when will Stripe go public and under what circumstances? And what’s the holdup here? Why aren’t you public already? Patrick Collison Yeah, look, I think people sometimes hold us out to be dogmatic or something on this topic, whereas we feel like so many other people out there in the world are dogmatic. We just try to be pragmatic on it. You know, Keith was on the show and he was saying, you know, he believes companies should go public as quickly as possible. I don’t know. Maybe that’s the right thing for some companies. But in at least Stripe’s case, that hasn’t been the case. I also think the environment has changed quite a bit, where it used to be the case that to do any return of capital to shareholders, or if you needed any kind of large sums of money, you needed The public markets. That’s obviously not true today, where the stable private markets exist. But we look and we say, is Stripe better off at the moment as a private or a public company? And up to this point, we have determined private. That could change at some point. But it’s kind of no dogma from our point. The last thing I’ll just say is, I think Keith made the argument, people generally make the argument that it is critical for discipline to be public and public companies run in a more Disciplined fashion. And I think that’s hogwash. Like if you need a 25 year old fidelity analyst asking you to double click on your CapEx, blah, blah, blah, blah, to run the company with discipline, something is horribly wrong at the Company and you need new management. And so that argument has never really resonated with me. Chamath Palihapitiya Basically, what you guys are saying is, for your intellectual perspective, you get a lot more return on the time you spend talking with the private investors you have and your team and Then- And customers. And customers. And it would just be dilutive and you would have your outcomes, quite honestly, if you had to talk to these other folks who are talking to you and 50 other companies don’t really know much Of anything, maybe very surface level, and then may actually distract you and force you to make decisions you don’t want to make. John Collison We’re not even that negative. We’re not even that negative. I was going to say, but there’s no spiritual status associated with being public. Like why be public? It is a cheaper source of deeper and more liquid capital. And so if you want and more liquid capital, then by all means, go with it. But it’s not more moral. And I think, again, it’s just helpful to get away from that kind of framing. I also think it’s noteworthy that if you look at financial services in particular, and we’re kind of a company at the intersection of financial services and technology, being private For a long time is the norm. So Bloomberg is a private company, Fidelity is a private company, Vanguard is a private company, Jane Street is a private company, Goldman Sachs. Citadel Securities. Citadel. Yeah. Goldman waited 130 years to go public. JP Morgan waited 70 years to go public. Visa waited 50 years to go public. And those are all different times in history. So I’m not saying you can draw definitive decisions from them. But I think it’s a thing of financial services where there’s always a tendency, uniquely here, to be kind of pro-cyclical. And I think you need to be kind of particularly careful as a public financial services company to avoid some of those temptations and some of those tendencies. And so, you know, I think that that’s a unique dynamic that applies in our space. (Time 1:35:39)