The Wisdom Wall

“There's, there's been this correlation between the number of folks at a company and the output from the company Uh, for, you know, decades and decades. I think that basically broke the first week of December, and what we were seeing is that one or two engineers, or a designer and an engineer who was on the tools, quote…”

“If any company in the world has a marketing team and the head of marketing does not have a 100,000, at least a 100,000 followers. Like, you, you need to replace them. Like, like the game has changed.”

“there is a case for a company that raises a hundred million dollars, uses it productively and in a focused way, and is able to deliver a different value proposition than the very same team would be able to do with 20.”

“the intelligence now doesn't need like the graph and the multi-agent lattice work and harness because it will constrain this level of intelligence.”

“I like to go to the line. We should be like a few inches off the line. That's a good thing to do.”

“Being canceled is a choice. You can choose to get canceled and retreat from the public eye, but I think if you're still out there, at some point, honestly, the haters kind of give up.”

“If your group's entire purpose for existing is to fight an enemy, then you need to make sure that that enemy exists. To do that, of course, of course, of course, you would fund them.”

“What I don't like about drawdown funds, the traditional private equity fund is, you know, you commit to them, they charge your management fees for a while, They find a deal, they draw it down, so your money's not been in the ground for a few years. And then a few years later, if it's a successful deal, they sell it for…”

“my theory is the, the ability to directly participate in online virtual combat, uh, is shunting away a lot of the energy that in the past would have, would have translated to street violence.”

“AI doesn't take your job, AI makes you the CEO.”

“investing in technology, uh, you really have to be in deep in the tech to be good at it. And I think that it's very possible as, you know, people get older, you know, they get less into it sometimes. And so, you know, at that point we got to make a change.”

“You know, they, they call it privacy, but that's just a cover for, I don't trust the police to do their job effectively.”

“You can't, you're not going to solve the prison problem by not enforcing the law. Like, that's just, you just make the, you punish the victims on behalf of the criminals at that point, which is, I think that's a very dangerous thing for a society to do.”

“At the end of the day, though, it's just like, does not, there seem to be enough solutions, especially if you are as willing to pay as you, as people are in AI, that I don't really expect it to be a significant bottleneck.”

“As you go from what we have today, which is mostly a very specific intelligence, not a general intelligence, but as labs succeed at their goal of building this general intelligence, We really need to stop using the steering control paradigm.”

“It is definitionally impossible, given what we just discussed, to succeed in AI without gross margin pressure, and I do not know why they have concerns, because we have an existence proof that a software company can deal well with declining margins in Microsoft and Adobe to the whole AI thing came along.”

“if you read the history of, of those people at that time, specifically people like Oppenheimer, you realize, yeah, the people who invent the technology have no special moral claim. Uh, they have no special predictive power for the consequences of the technology, and they have no, no basis for a superior moral claim on…”

“So I think if you're going to reinvent an industry, you don't want experts.”

“All the knowledge in a company is with the individual contributors are doing the work and the customers that like, there's no There's no knowledge with the people who are talking to you as CEO who are on your staff.”

“Don't try and fix them. They can't be fixed. It's not going to happen.”

“a low number of really exquisite weapons is not gonna win a conflict, and I think Ukraine kind of definitively showed that to the world, that that is not a winning strategy.”

“the incentive systems around a lot of these things are, like, if you go to the doctor, parents, one, want these ADHD, you know, Want, want, want these diagnoses. And the reason the parents want them is because it gives you extra time, it gives you extra resources, you can then opt your kid out of certain things in…”

“my underlying belief is early markets are actually pretty darn efficient, a lot more efficient than, than people realize.”

“I think in these times where you don't know the TAM and things are moving quickly, you definitely want to pick the best team. I don't think you should overthink the space, but like asking questions about like TAM or valuation or value makes a lot less sense because that's actually what's uncertain.”

“anybody listening who, who is a VC, you can take as many board as you want. The actual board work itself is not. The question is, can you still be available to founders and add value whether you're on the board or not?”

“there is no, as far as I can tell, as far as we can tell, there is no inherent endemic mode In the technology stack to AI, other than just overcoming the bootstrap problem.”

“Until we've solved cancer, every month that we're not rushing to the frontier of accelerating biological discovery or scientific progress is a month that millions of people are suffering from disease that we could be solving with the eye.”

“The actual regulator of taxis is Uber or Lyft because they have real-time tracking. They have star ratings on both sides. Or your, um, you know, some, you know, the FCC or something like who issued licenses. Well, the actual regulator of speech or what have you said, YouTube or Facebook.”

“the actual way of regulating big companies is with a thousand startup piranhas, not by this regulation.”

“this is actually what led to wokeness because it was, you know, you've heard the saying go woke, go broke, right? Yeah. But in their case, it was actually go broke, go woke. Okay. Not my original coinage, but applied to this graph, it's relatively original because wokeness was, what happened was they just fell off a…”

“Blue America was disrupted by the internet, and so they began wokeness to take a piece of Red America's pie, and the tech clash to try and take back control from the internet. Red America was disrupted by China, so the trade war was against China, and Trump was against Blue America.”

“Build your own distribution to avoid distortion, ok? That is to say, any content you have, it should be posted on your feeds. Why would you go and feed it to some journo? You've got some scoop. You don't need them for distribution anymore.”

“And those are honestly amazing businesses because the demand is so absurdly high that no matter what, when you get more, uh, supply of therapists, even if it means that you're just paying therapists less and you're getting more therapists and you're getting less qualified people, like you can build a business.”

“there's a crypto angle to at least three of them, because obviously there's a crypto angle to money, there's a crypto angle to AI in terms of constraints. This crypto angle to the drones, because you're going to want the control plane for the drones to be on chain, since that's the part that can't get hacked, whereas…”

“I feel like you don't, like if your distribution is strong enough, you don't need to like talk to your users. You just need to look at their data.”

“professional CEOs Are often very good at running a status quo business, but when the, when, when things change, they have a very hard time reacting.”

“what's happening is what Ted Kaczynski Famous philosopher, Takasinski, what's called over socialization, right? People are too worried about what everybody else thinks, right? Uh, and so people, people get too wrapped up in the sort of mob scapegoating cycle of, you know, being able to, and, and, and sort of this, this…”

“new people entering the workforce had a much worse experience, right? And so the, the COVID was sort of a, there was this sort of dichotomous experience where if you were like a 20 or 30 year veteran of a company and like a high ranking official, it was Absolutely fantastic to work remotely because you got to go to…”

“Well, if the AI has reason, then it won't do what the human can do, because we're not reasonable. The, the all breakthrough come from what's, what's not reasonable, or what's not supposed to work. It's figuring out the thing that can't be done, and allowing it to be done. The AI can't invent flight. For the Wright…”

“I believe that the audience comes last, and the artists should be true to themselves, and that ultimately is in service to the audience. The audience is best served when they get the real version of you. If you start watering down the real version of you to do what you think they want. It's a recipe for disaster.”

“maybe the single best, uh, thing that you can do to improve the world is to build a company that, you know, Deliver some product or something that, that improves the world. Like that is actually, that's the thing. It's actually better than, um, has a better impact than any kind of activism or, uh, political activity or…”

“So every piece of network infrastructure you need to make AI work, Is, is in this new technology that's invented”

“oftentimes, you know, your choice if you want to take out a target is dropping a 500 or 2000 pound bomb, which is going to cause widespread destruction, uh, and a lot of collateral damage. Uh, and so, an AI system that might be, uh, using autonomy to, like, a pretty intense degree to figure out, like, where something…”

“probably the, the one thing we could do to lose to China is try and use policy to beat China.”

“the way around that for America is highly automated factories, which are almost impossible to build if you have union labor.”

“You can take, you know, an end-to-end model, go from sensor to, you know, trajectories or actuation. Uh, you know, typically you don't build them in one stage, you build them in stages, but, you know, you can do, like, backprop through the whole thing, so, you know, the concept is very, very valid. Uh, you can, you…”

“And then basically within like five years, it was like Facebook is like the living demon incarnate on earth and destroying everything. Right? It went from like useless to evil in one step. Right? You know, the headline never gets never just like, oh, this is kind of cool. Right? Like that's never the story. Like this…”

“maybe it's actually easier to automate the manager job than it is the worker job.”

“with, with sort of sufficiently advanced, ubiquitous video conferencing, Um, all of a sudden immigration law becomes less relevant, right? Because like if, you know, it's perfectly legal to stream bits across the border, um, even if it's not possible for people to move across the border.”

“we want to do well and do good is probably a pretty dangerous phrase in retrospect in that, um, You know, you kind of want those things aligned. So whatever the company produces is going to make the world a better place like that. And you kind of want to stop it there. If it's like, okay, we're going to go get the…”

“I think there are like a few kind of important issues around that. One is, you know, really arrogance in that, um, I think, you know, we in tech and a lot of the people who went in are like, oh, we're the good guys. We're for the good and everybody will love us when we get there and we can just push our agenda on, you…”

“why I won't destroy all the jobs and, and the sort of short version of it is because it's illegal, uh, to do that because so many jobs in the modern economy require licensing and are regulated.”

“a lot of the areas that, um, You know, are going to succeed are often the ones that are less popular. It's areas working on weapons. Like, this is not an area that a lot of companies are interested in doing, but I think to succeed in the defense space, you have to be willing to actually solve the core problems of this,…”

“AI is what communism actually is a massive concentration of power and blockchain is what communism is advertised to be, which is power to the people.”

“the only correlation between a drop in crime is an increase in clearance rate.”

“If you take out every possible restriction on what people can do and you let them fully express themselves, then what you're left with is pure choice. Right. Right. Right. Right. And so, and so at that point, the, the differences in the outcomes that are based on pure choice maximize. They don't minimize. They…”

“So a cynic, a cynic would say, a cynic would say, That the A number one reason why a big company would be lobbying aggressively to try to get basically government regulatory protection, um, is if internally they are convinced that it's gonna be a race to the bottom.”

“a society built in the idea that love scales becomes an Incredibly dark, dystopian, hostile, you know, and ultimately murderous, murderous place because, because love doesn't scale.”

“from a societal standpoint, the last thing on earth you would have wanted to do is to slow down the progression of the introduction of the car to protect the blacksmith jobs. Like that would have made everybody worse off, right? To protect a very small, what ultimately turned out to be a small number of blacksmith…”

“If this is like, you have to fill out a thousand pages of paperwork and get 15 different licenses from different bodies to, you know, to make an AI system, that's, that's never gonna work, that's gonna slow things down, you know, other adversaries You know, authoritarian countries will get, will get ahead of us.”

“Saying you love lidars and hate cameras or vice versa is saying you love one wavelength versus the other wavelength, right? It's not a fundamental thing.”

“an AI that's smart enough to, like, turn every atom on the planet into paperclips is not going to turn every atom on the planet into paperclips. Like, it's going to be smart enough to be able to say, why am I doing this?”

“The fundamental observation is that full-time employment is a bug. It's not a feature of the market. The reason why full-time employment exists is because the transaction cost of doing something is really high.”

“if you're an adventure guy or woman, and you're trying to get a 10 X, Get the hell out of media. Go to, go to healthcare, or go to energy, or technology places that you're much more likely to get that kind of return.”

“We absolutely do not need AR and VR, uh, in order to build the metaverse.”

“a lot of the, the controversy that has, uh, that we've been seeing the last, you know, year or so on social media, you know, arises from kind of this friction between the, the sort of Almost ideological commitment to crypto and interoperability that you see on the part of crypto bros clashing with, with designers who,…”

“America actually is not a single nation. At a minimum, it's binational.”

“The FDA free zone alone. I mean, there's so many other reasons I think that we need innovation on sovereignty, but that alone is how we get to life extension. That alone is how we get to transhumanism.”

“You want self-driving? You're going to need sovereignty. Do you want drone delivery? You're going to need some form of sovereignty. Do you want like, you know, to be able to have nuclear power? You're probably going to need some form of sovereignty. Like Wyoming is pushing some of this. There's, there's good things…”

“I really no longer believe in the concept of institutional reform. Like, I think fundamentally it doesn't happen, um, or it's so rare as to, like, be, be basically something that you can't ever count on. I, I think basically progress happens by starting new things, right?”

“Nothing kills a great idea faster and has been more of a disaster to innovation, certainly in my industry in drug discovery, than the idea of market projections.”

“And the key thing to make this strategy work is you actually want to over allocate on process and you want to weight outcome to almost zero because you're buying the outcomes that were in fact zero.”

“if you have a fork that's like 10% of the size of a project, and this fork could potentially impact the price of the larger project, it's absolutely in the interest of that larger project to launch attacks on that base layer proof of work vote and do things like censorship, use some small percentage of their hash power…”

“they were basically like, they were rounding errors on everything. Like, there weren't 10, there weren't 20, there weren't a hundred, there were three or four. And there were two preconditions for them. One is, they all were offshoots of monopolies... And then the other precondition was they were all pre 1975, 1980.…”

“look, I believe in artificial deadlines. I believe in playing one against the other. And I believe that you have to do anything and everything possible short of Immoral or illegal to get the damn deal done.”

“well, if you go and do well in your Aristotle class, your foreign language class, this, though useless in terms of what you will do, it is impressive. It is convincing. It's a way of selling people on your value as a worker.”

“the signaling function of education is on an emotional level, at least very deeply grasped, and so there's almost no pressure from anyone in the system to actually improve genuine learning.”

“because the skills that you develop as an operating executive have nothing to do with identifying which company might succeed.”

“there's a really weird thing about information security in that it's an industry that, for the most part, shouldn't exist. If you bought a car and your car dealer made you pay an extra 200 bucks to not have your car go up in flames, you'd be able to sue them. And the cybersecurity industry To some extent is filling…”

“My experience with engineering has always been that if you freeze engineering hiring for a year, engineering gets way better.”

“I actually have reached the point where I think that all of the ideas in tech and all the ideas in Silicon Valley are good ideas. I don't think there actually are bad ideas. I mean, maybe we could, we could pick a couple, but they actually all seem to be good ideas. The reason I say that is they all do seem to actually…”

“only the paranoid survive. That was Andy Grove's quote. And the problem with that is it over, it encodes the truth in a way, but it oversimplifies, because the paranoid are delusional. They see crazy things all the time, ok? And business isn't that easy. You don't get to just imagine everything is going to crush you,…”

“Now we are, we are entering a post-humanist era. Authority is shifting away from humans. If in the last 300 years we saw authority descending from the clouds, To earth, to humans, now authority is shifting back to the clouds, but not to God, but to the Google cloud, to the Microsoft cloud.”

“religion is a virtual reality game that provides people with meaning by imposing imaginary rules on an objective reality.”

“But if you don't need them as soldiers or workers, then the incentive to build hospitals and schools and so forth diminishes.”

“climate change cannot be solved by any means except technology, because a political solution is clearly not going to happen.”

“if you look at these examples of quote unquote successful open source companies, in almost every case, they were going after and cannibalizing an existing market, which is way more difficult than actually doing new innovation, right?”

“It does get down to, look, there are times when you need so much precision that, like, if you're gonna lose somebody in the whatever war for talent because they got their feelings hurt, well, that's just gonna be how it's gotta be, because you need to Teach that Sun Tzu object lesson, like, somebody's getting their…”

“And a lot of people come out of Stanford or, you know, uh, Berkeley Computer Center don't understand what that means, um, and don't understand what it means to do something bigger than yourself. But if somebody's been in the Marines, um, or, you know, and decided to do that, like, you really know what that means, and…”

“most people who are non-violent offenders, if you sold drugs, you had to be violent at some point to protect your interests. You just didn't get caught for a violent crime.”

“So, if you come from a services background, you treat services as a revenue stream. You start building products that require lots of services, and that's a bad product. Like, it turns out that that's almost the definition of a bad product, is it's really heavy to lift it in, and it takes a ton of care and feeding.”

“The great challenge in nonprofits, and you see this in international aid environments, is that, you know, country governments where the budget is largely composed of aid from richer countries are much more accountable to the donors than they are to their own people, which is why we see these really perverse outcomes…”

“Arguably it's better for one benevolent, um, person to end up with all this money instead of just distributing it evenly among all the employees. Right. Um, so that they can all buy, you know, slightly nicer cars and other, you know, materialistic things.”

“the quality of your idea doesn't matter, um, just looking at the, the content, uh, aspects we could predict.”

“for Google, Android is a tactic, and cloud is a strategy, whereas for Apple, the device is a strategy, and the cloud is a tactic. The cloud is a feature of the core strategy, whereas for Google, the device is a feature of the core strategy.”

“If you're an entrepreneur or founder, you always want to build a monopoly.”

“there can be perverse incentives because now there's so much money and they're not actually able to ship a product.”

“I very much believe this, and I think we all do as a firm, a blockchain without a cryptocurrency literally makes no sense, because there's no incentive for their participants to host the blockchain.”

“the battle between every startup and incumbent really comes down to this one thing, which is whether the incumbent gets innovation before the startup gets distribution.”

“There is no substitute. The big takeaway, and this applies to everything we fund, and it's what we work on all the time, there is no substitute to the first five years of people in the office, 18 hours a day, six days a week, grinding away. Um, and it's grinding away on making the product work for customers, and then…”

“But right now, especially with the labs for the first time, you know, in my career, you can take capital and throw it at a company and it compounds their advantage.”

“venture-like returns are possible in late stage, but your best company should be five, 10% plus of your fund. That way you can actually return the fund on a single company. Fund returning math in late stage didn't exist before. It now does.”

“Coding hit, but that's kind of a head fake, right? Like coding, coding is perfectly documented, right? So it has like perfect data. Uh, it's verifiable and it's simulatable, right? And so, um, like most tasks in business do not share those three attributes. And so, you know, maybe the diffusion into other knowledge…”

“You can't just throw an operating partner at the company and say, let's put AI on it. It just doesn't work. You need to completely, and if you have, by the way, if you do have a founder mentality at the management team, like, it is possible, but the board has to be aligned, all the investors have to be aligned, and you…”

“the U.S. doesn't have a problem with energy generation. It has a problem with speed to power. That's permissioning transmission. That's regulatory.”

“if you look at auditing as an industry, you end up with issues like Enron, where, um, if you have the same group who's responsible for doing the audit, as well as also consulting and supporting the company, you have a mixed incentive structure, and, uh, and then it just becomes pay to pass the audit, or, or in this…”

“reconstruction is just like generation with a really long context and you put a lot of stuff in it. Yeah. Right. Like that's the way to actually build this continuum where you kind of bridge between those two things.”

“I mean, Casper got to scale, but lots of CAC, lots of CAC, and that's why a lot of these business models, they tend to degrade over time, because all the economic value goes to Google and Facebook.”

“you're just simply not going to be the low cost provider. If you're not vertically integrated, if you don't own your own compute over the very long. Long-term.”

“what we have seen over the last three years is the steady increase of the capabilities of the models. Where the model is no longer the bottleneck. And in fact, using AI, these models are getting better faster and faster and faster. Now the bottleneck is all what I call south of the model.”

“Any problem that you have can be solved with enough infrastructure and money. And so until we run out of problems, we're not going to run out of demand.”

“the one thing we all knew in startup world is that if I have a two-year lead on you, and you try and cast me by hiring a thousand engineers, you're gonna wreck your company. Like, that never works. It's a mythical man month. Nine women can't have a baby in a month. Like, that's it. Like, that never works. Ok, now that…”

“GrokBot got really right is, no, how about if it's actually an employee? So now you have this thing that's an entity, and it doesn't have, like, special access to your keys or whatever. It has its own computer, and it has its own browser, and because these are the smartest models in the world, it can do whatever an…”

“today to build a frontier model costs, let's say, three to five billion dollars, right? So, and let's, you know, and, and that's to train it. And so the inference has to pay back at least that. Of course, right? You know, in order for any of this stuff to be viable, let's say two times that. So let's say that now…”

“it turns out that you don't speak to these models in code. You speak to these as humans in natural languages. Therefore, the, the model has to, like, understand, like, the, the breadth of human experience and language, right? And so, you know, that means in order to build a coding model, you're actually building, you…”

“Like if you really believe in the product, you would never do a plugin because then you're part of somebody else's product. If you really believed in the product, you wouldn't do enterprise early, you do it later, because the product itself is the kind of go-to-market motion”

“We really think that the kind of rational architecture and the one that is emerging is that for jobs that, um, have unlimited upside, like sales or product, you always want to use frontier tokens. And the reason for that is you just don't know what the value of the new product feature or closing a customer account is.…”

“if you actually have a problem that you can specialize the model around with your reasoning traces, You can start to create this compounding advantage in your domain for your customer base, where you're able to kind of shape the intelligence to be better than any general intelligence for your problem.”

“This is a place where the application layer really shines because labs, of course, are both incentivized and structurally only able to provide their own in-house models. You as an application sort of aggregator can provide the best of breed.”

“There's no app store for AI. So this actual product cycle for consumer looks more like Web two point O, where you have to kind of build the channel alongside the product and less like mobile, where you actually have the central point of distribution for the entire ecosystem.”

“I think if we lived in a world of twenty-twenty-three when it was one model to rule them all, it wouldn't even matter if you had permission, because the labs would just take a hundred percent of your gross margin over time. But now, because you've got many options at all points in the Pareto frontier, you know, the…”

“if I'm writing a program, I'll like, whatever, I'll use a cloud database, I'll use storage, I'll use networking, you know, whatever it is, but like correctness and logic for the program is under the programmer's control. Maybe I'll use a third party library. But again, like, I'm choosing the library. I know the inputs.…”

“And right now, if I give. 20 people a billion dollars, they can actually use it usefully. It's very, so it's like, it's like we've kind of moved the industry from like this engineering bound problem to a capital problem.”

“this is not a magic to discover drugs, because the hard part of drugs has always been candidates. Not candidates. It's always been efficacy and safety. The candidates have, since the eighties, have been able to develop more than we could test. It's human patients, and it's very, very, very hard.”

“No, you're building the machinery. I'm saying in this case, if you're like, I want to exhaustively explore every protein combination. Right. We can just turn that into a money problem. Yeah, yeah. It's kind of very strange.”

“So I think microtransactions will just be necessary for that economy to exist, the sort of agentic economy. And then on the flip side, they're now eminently possible. Because of stable coins.”

“How do you lead 20,000 people and get them oriented towards a beautiful mission? And those 20,000 people are great minds, optimistic, good people. How do you do that? You can't do that by being a bad person. But you have to be hardcore.”

“there's a moment right now to go sell You know, big software again. Right. And, and, and to go sell platforms and, and it's because of this moment. Um, and so I think people need to be studying kind of the 15 years ago models of like how people build this and the, and the ecosystems around it, um, to go and have…”

“from Palantir, I definitely learned that everything that's awesome in my life is kind of a cult. That starts with some sort of truth or belief that flies in the face of an orthodoxy, which is very punk, actually.”

“a markdown file is an employee. And it's an employee that, uh, will do the job perfectly every single time. And it'll do it like as many times as you want.”

“Like you could release something in a day or two that would have taken like six months. And if you could like multiply that out across, like the thing is like probably an org like Microsoft can't. But, like, a startup can. And every startup must.”

“The first, and this is where I think many companies are stuck right now, is the first instinct is, okay, let's adopt AI. And you, you basically leave your structure as it is and just give ChatGPT or Claude to, to your team. And then there's no efficiencies. Your customers have the same problems and nothing happens,…”

“most of these agentic systems in production right now, like large scale agentic systems, usually if an agent hits a wall or can't perform anymore, it'll like send the, this case or this customer to a tier two support and forget about it. That doesn't really work because you don't close the loops. So you don't generate…”

“The first is it has to be top down because of this. Like, if you just get adoption, it won't go anywhere because it's hard to generate this taste or strategy for people to, to bottom up, decide what to build and whatnot, and come up with something that works for the company. So the transformation has to be top down and…”

“Uh, but I think the point is that, um, at a certain point, it's strictly better to use open source models because when your use case is solidified and you're in production at scale and you're pretty sure this is the, the sort of shape of the agent, then there's no reason not to use open source because you, you get…”

“If we ran our entire business exclusively on frontier models, I would care a lot about costs, and I would think a lot about that, but once you're already, once you've already made the jump to saying, okay, now, now we know how to think about open source models, we know how to decompose upon, we know how to, uh, build,…”

“I'm sort of teaching the AI my own procedures. And again, that doesn't happen through fine tuning. That, that happens like in context, because if you were to fine tune on that, you would have to reverse it every single time. You know, you change your procedures, which doesn't make sense.”

“forward deployed engineers are necessary or newly necessary for early stage AI companies because the workflows are new. Right? If you're building a SaaS company five years ago, most SaaS products are pretty well explored, right? Like you roughly know what the user's trying to do, and your job is maybe come with a…”

“Everybody has access to these tools, and so if our competitors are going to use them and build more things, we need to build more things, right? If somebody else said, oh, here's our roadmap, and now we can get through it in, you know, a third of the time, and then they just stop hiring, we would just take that to…”

“that's where the market just expands massively once you throw in labor, because it's like you have this tiny, tiny market for software, which by the way is not that tiny. It's like a trillion dollars. Um, concentrically around that, you have like, you know, big on financial transactions. That's even bigger. That's why…”

“So I kind of think the same rules apply, you know, the battle between every startup and incumbent comes down to whether the startup gets the distribution before the incumbent gets the innovation. But with many changes, one change is the incumbent can get the innovation much more quickly.”

“Cruise was going after the robotaxi concept, but GM makes its profits from personal car ownership.”

“and I, I think if, if you build a company that's successful right out of the gate commercially, like you might not need to do all of that, but if you're, Trying to meme a dream into reality. I think you have to play that game.”

“this middleware layer, it's always, always very unstable. It looks great in a network hierarchy diagram of the OSI levels of networking, but it's just never that stable.”

“the universal truth for enterprise software is the, the most difficult thing to do that happens to be the dumbest is to attempt to just compete head on with, with an existing category. And, and by head on, I mean, not just the same category, but doing it the same way. The biggest opportunity right now is always, always…”

“When AI becomes the interface for everything, every tool we have, every technology we use is going to be built on these models, and the models are not objective. They have opinions. They have strong opinions on things like history, on culture, on values, and those are very different depending where you are.”

“ deterrence is no longer just the size of a military. I think as we see in the Strait of Hormuz, or we saw in the Red Sea, it's the pace of innovation in reacting to cheap software-built tech that adversaries may be using, whether they're a country, criminal, or terrorist group.”

“I personally believe, I'm focused on cybersecurity for a moment, that they help far more in defense, because defense is far harder. You have to be, you know, defending a broad expanse, an attacker has to find one way in.”

“if you kind of lock that up, lock that capability to understand vulnerabilities, then the only people who will be able to understand vulnerabilities are the bad actors who jailbreak the system.”

“most people are just totally irrelevant. Like you need to do something a little bit risky to even get to a place to be cancelable. Most people can't even be canceled because you didn't exist. So the, the general advice, like, yo, just live more and do more shit in general, because You're nothing, and you maybe can get…”

“you really want people to hate you and you want people to love you, but you don't want to be neutral. You don't want to be lukewarm. Um, because then you're uninteresting.”

“chip companies didn't capture the value. Um, ISPs didn't capture the value. Mobile network operators didn't capture the value. Windows and iOS did, but they were doing something else.”

“the models are kind of diff commodities, and the chatbot isn't the right UI or the right product, and the companies aren't going to be able to build all of that stuff themselves, so therefore they're low-level infrastructure.”

“these things become competitive necessities, and everybody has to buy it and use it. Um, but the cost saving or the productivity gain that you get from it just kind of gets competed away, so you don't get to charge more for it.”

“We are not limited ultimately by capital. We are limited by our capacity to create.”

“You can't create people with 15 years of experience overnight. So it goes back to the first principles approach design of the ship. Design the ship so you actually Don't need 15 years of welding experience to build it in the first place.”

“If somebody can't come out of somebody who's building a car for Ford or General Motors or an airplane at Boeing or a rocket ship for SpaceX, And we can't get them highly effective building ships quickly. That's our fault. That means our design isn't simple enough.”

“But, you know, we have a lot of risk takers. And I would say that the biggest challenge for management is the risk management side, which is really getting people to, um, refrain from risk. Which is about a third of the time when you're in that business, and by the way, not the most important part, but probably the…”

“You're not gonna get people to be stupider than they are or unlearn things they've already learned. You can wish that atomic, that the atom had never been split, because maybe the adverse consequences of atomic bombs are worse than the benefits of nuclear power. But guess what? You're not gonna unlearn it, so don't…”

“if you increase marginal productivity of the worker, you don't have a diminishment of human work, you have an expansion of human work.”

“NASA can do the, uh, What probably others wouldn't want to take on the liability of, of launching a nuclear reactor with power and propulsion so we can get to Mars someday and actually bring our astronauts back home.”

“This is not a business, Tony. It's an oil well. It's depleting to zero. And I said, well, What about the Wall Street Journal? Well, that's not a newspaper. That's a trade journal.”

“putting more controls in doesn't necessarily, doesn't necessarily protect you. That a lot of it, if you have good people, and you trust them, and you hold them to very high ethical standards, and that becomes the behavioral norm, That's much better than having lots of watchers and watchers of watchers trying to check…”

“if you force companies to disclose aspects of the AI that let foreign adversaries, for example, reverse engineer the model weights, you're really compromising the American and the Western competitive edge.”

“And this, the thing that's not different about AI, and that agents don't fix, that nothing fix, is that any enterprise of a thousand people or more, or that's older than 10 years, is just a mass of stuff that's sitting there waiting to be integrated. And, and you can't just say it's gonna integrate. AI actually doesn't…”

“I sit in these boards too. So the board goes to the CEO. What does the board say? We need more AI. And what does the CEO said? Oh, okay. I'll get like a consultant to do more AI. And then they have some centralized project that nobody knows how it works. They haven't aligned their operations and those things will fail.”

“What we're seeing instead is instead of viewing AI as software, like just view it as a user. And so instead, like take your product Make it a CLI tool and then have the AI be an agent that actually uses this. You're not fusing the two. You're just making it more useful for AI. This is a very, very significant…”

“these models don't integrate well with software. Actually, I think it turns out and what we're learning as an industry is if you view them more like humans and you draft on the, um, mechanisms we put in place for humans, they are much easier to integrate.”

“the idea that there's like whatever rhetorical combat or like that there's like this virtual environment, um, in which there's lots of points of view and there's lots of, you know, again, both lots of truths, lots of lies. Like this is not new. In fact, this, this, this is the natural order of things. We just live…”

“If you have enough money, uh, and some good data, you can buy enough GPUs and solve basically anything in software.”

“So it's very easy to replicate the code. It's very easy to move the data, and then it's not even going to be a human talking to your software. It's going to be an AI, and, you know, AIs are really flexible on how they use user interfaces, so That mode is gone.”

“their fear is, like, unleashing, not just the agents themselves, but humans to do integration. Because you put, People creating new integrations. And you just say, please break my system of record.”

“You know, we've actually built in a lot of these permission systems. You have to treat it like a human. As a separate human, and then instead of like building another auth layer”

“So you can't fully treat them like humans because here's the thing. And with regular humans, you don't get to look at the Slack channel of the person that, that is working with you or working for you. You don't get to log in as them. You don't get to oversee them. You are, they are accountable for their own set of…”

“The risk is like a thousand times greater. Like these people, like they will just leak your information whenever they want. Like they will happily just go and send some email to somebody because they got prompt ejected.”

“And so then, so then thus, if anything can ever enter that context window, because they have access to a resource, then in theory you should assume it can be, you know, prompt ejected out of the, of the context window, and I don't know that we know of a way to solve that at the moment.”

“And at the end of the day, like, it's the semantics that end up mattering a lot more, right? And so, like, the agents, in my recollection, or in my experience, are very, very good at picking the right Back end for whatever they're doing. So they don't, they're not like, oh, like the interface for this is very good. The…”

“Like, for me right now, I'm like, yeah, we should probably waste a lot of tokens, because that means that we're, like, trying new things.”

“Now with AI, many companies can do something like that. Like a non-Chinese tech company can be like a Chinese tech company where it can have a lot more, let's call it digital autarky.”

“verification is actually harder in the digital world than is in the physical world, which means reinforcement learning and training is much easier, in my view, in the physical world with robots and self-driving cars, drones, and so on and so forth.”

“for an AI to actually reproduce itself, it would need to have physical robots going and mining Or, and constructing data centers, and making chips, and handling that full supply chain, and, uh, and then The AI brain, like the queen of an ant colony, would have to give instructions to all those robots to do things.”

“AI within the trusted tribe increased productivity. Between trusted tribes decreased productivity. So you make more money perhaps within the tribe, but then you have to spend it on verifying stuff between tribes. So crypto is for between tribes and AI is within tribes.”

“Another one is AI doesn't take your job, AI takes the job of the previous AI. Claude took ChatGPT's job, right? Um, just like Midjourney, you know, took, uh, took Dali's job, took Stable Diffusion's job.”

“AIs are really good at programming humans. Much better than humans are at programming AIs.”

“Um, it's like the classic Jevons paradox thing where I, I think that there's probably now just a super set of things that, that can be built. Um, so I don't know, you know, a given tech company might be, might be way smaller, but there might be 50 or a hundred more tech companies, or you're gonna start getting this…”

“I don't think that if we ask customers to, to like, prompt these tools themselves, they're gonna necessarily know the right prompts and come up with the right answers, so we've invested massively on the proactive intelligence side, where what we've found, especially as it relates to money, is like, we need to be…”

“Uh, I like to say like any of us can, can create a peer to peer app in probably a week. Uh, no one's going to vibe code, you know, 50 or sixty million monthly actives who are actually using that.”

“Um, and what we've landed on is that, uh, you have to go full bore kind of like leveraging the advances in autonomy, um, in automotive and humanoid robots to go and apply that to refineries and to, to mining operations.”

“you should not be vertically integrating just because you can save five percent or 10% or 20, even 50%, um, on a subcomponent that does not kind of check the box of like, does the company exist or not?”

“The competition has always been the services competing. It's been competition inside of government that drove innovation, not competition from industry.”

“the Department of War is a monopsony. Um, the nature of the monopsony is it forgets It starts imposing all sorts of constraints on its suppliers and how they behave and what they need to look like. And that led to, it's like putting, we put all of these companies on the Galapagos Islands. They're not on the mainland…”

“Only a founder can do that. Because if you're a professional CEO who's employed, that's, you can't, you can't survive that.”

“the whole motion of the old world Will kill you in the new world. And, and vice versa.”

“And so what he says is if you, if you can be sustainably faster at running your decision loop, You can get inside the other guy's loop in a, in a systematic and perpetual way. And basically the result of that is psychological breakdown. Uh, basically you, you destroy the psychology of the other side because they just…”

“in the old world of software, the one thing you knew is you couldn't throw money at a problem and now you can throw money at a problem.”

“All of deep learning is, ah, doing correlations. It's not doing causation. Causal models are the ones that are able to do simulations and intervention.”

“I think deep learning is still in the Shannon entropy world. It has not crossed over to the Kolmogorov complexity and the causal world.”

“to get to what is called AGI, I think there are two things that need to happen. One is this plasticity. Which has to be implemented through container learning. Secondly, we have to move from correlation to causation.”

“And that's how this country, by the way, if we are going to outperform the rest of the world, uh, our single advantage is to augment neurodivergent It highly individual people to be their absolutely unique best and protect their first, second, fourth, and fifth amendment rights so that they don't get screwed.”

“I've generally been A little more cautious about the very, very horizontal consumer AI apps, just because it's probably both in scope for the, for the bigger companies, and they have the advantage of already having IT approval and enterprise contracts and all of that.”

“to me, businesses are a set of processes. They're not a system of record. Like, these are all process-based systems, right? Everything Alex has just said is totally true, but there are processes like reference checking or other things, and your ability to coordinate a set of processes to happen As cheaply and…”

“But then I have kind of output-constrained work, if I think about anything creative, marketing, I would argue software development, technology, where I can theoretically do an unlimited amount of tasks, right? I'm constrained by my creativity, if you like, and how many things I can think of to do, how much value I can…”

“That actually makes Workday stickier in the enterprise and more valuable because you can build all these applications on top, which is the power of AI and vibe coding and creativity to make it more tailored for what I need.”

“The problem with say customer service where I've saved you, you know, you used to spend 20 dollars on customer service with our tool, you'll only spend 10. That's a great sales pitch in year one. In year two, the customer goes, but I only spend 10. Now I want to spend five. Otherwise you didn't deliver any value.”

“These are all, I would argue, a fundamental foundational design and experience problem. They're not a technology problem, right?”

“because if you're building a business from the ground up in twenty-twenty-five, you're starting a brand new company, you're not going to do things the old way. And, and you think about with the tools that are available now, like that means just by default, you can say instead of us having a, you know, a hundred person…”

“so in an odd way, if we see an AI pitch and the gross margins are super high, we're a little bit skeptical because that may mean that the AI features are not actually what is being bought, uh, or used by the customers.”

“I would say my observation from our companies, even the AI native ones is they run leaner. Partially because they've just grown so quickly and the demand is so strong. I wouldn't say yet we're at the point where companies have fully reimagined the way they run themselves.”

“I think that that is, like, an important idea that people don't think about enough is just how do you shape your environment so that it's naturally health promoting, as opposed to, uh, you know, having to, like, Discipline yourself and doing all the stuff that most people, frankly, fail at, myself included.”

“And it turns out that the best time to raise money is when nobody has money.”

“With AI, if you have proprietary data and you have enough GPUs, you can solve, like, almost any problem.”

“with AI, if you have data, you know, particularly proprietary data, and you have enough GPUs, you can solve, like, almost any problem. It is magic. But it means that you can throw money at the problem. And we've never had that in tech.”

“I think for startups to win, I think you need all three, um, and that's just in the head-to-head stuff.”

“Category two is arguably the biggest, which is basically, it's not competing with the software market at all. Um, software is starting to eat labor. You're basically selling software that does the job of what people would do before. This is arguably a much, much bigger market.”

“the best companies have hostages, not customers.”

“if you're a corporate attorney, you know, and your, you know, uh, you know, junior, uh, attorney is 50 times more productive, you've just eroded some of the revenue that you can actually charge to your end client. Again, in this case, if you can make your attorneys, you know, five X more productive, You can potentially…”

“What makes, it actually increases the peril for anybody who's built a software product that has an enormous margin pool. You know, your margin is my opportunity. Well, I can vibe code, I can vibe code against your opportunity. It has to be very, very sticky. It has to have some unique competitive advantage, and data is…”

“my view on, on everything that's happening in AI right now is it's one of these weird situations where it's very different than cloud, where Most on-prem software providers were like, cloud is stupid. Most potential customers were like, cloud is stupid. It's not safe. I don't trust it. I want to host things. Like you'd…”

“There is a strategy that we think is very interesting, which is instead of having a sales team, you buy one.”

“in many categories, being an aggregator of models is actually preferable to consuming just a single model. And the metaphor that we're all familiar with here, of course, is airlines. It's much more useful to search for a flight from SF to New York on kayak, because I can look across the inventory of every airline.…”

“the biggest mistake we make is we get too wrapped around the axle about some weakness that a company has, as opposed to focusing on what they're great at, um, and how great they are.”

“an investing team, like, shouldn't be too much bigger than a basketball team. You know, a basketball team is like five people who start, uh, And the reason for that is the conversation around the investments really needs to be a conversation.”

“investing is hard enough without like introducing other criteria other than is this thing gonna be a giant company and make a lot of money.”

“right now, currently, I would say that, uh, the complexity of the application itself is very high and is not subsumed in the foundation model. And so I think these things are not as, uh, straightforward as they appear and the benchmarks can be misleading.”

“These are trillion dollar questions, not answers, but once somebody proves that it's capable, it seems to not be that hard for other people to be able to catch up, even people with far less resources.”

“If you run a survey or a poll of what, for example, American voters think about AI, it's just like they're all in a total panic. It's like, oh my god, this is terrible, this is awful, it's gonna kill all the jobs, it's gonna ruin everything. If you watch the revealed preferences, they're all using AI.”

“high prices are really underappreciated. High prices are often a favor to the customer. It's actually really funny. A lot of like, The naive view on pricing is the lower the price, the better is for the customer. The more sophisticated way of looking at it is higher prices are often good for the customer because a…”

“the best of, the best of the AI application companies are actually, they are actually full-fledged deep technology companies actually building their own AI.”

“the more out there we are, and the more outspoken we are, and the more controversial we are, the better for the business in the sense of the entrepreneurs love it. Uh, the, the, the, the founders want to work with, this is very clear at this point, the founders want to work with, uh, uh, people who basically are brave…”

“AI brings a net new user interaction that may be difficult to replicate and may not natively live in the platforms of the product. And insofar as there are net new user interaction models, insofar as there is net new creative outlets and atomic units that look different from what's available in current platforms, My…”

“And so again, while AI is helping automate a lot of the drafting and reasoning work that they do, ultimately it's, it's really about enabling them to take on more clients and make more money. So it doesn't erode, you know, the billable hour. It really reinforces their business model.”

“I would almost argue the application layer companies, like the, the Kriyas, the Hedras, the Higgsfields of the world, popularized that template format. And did it first and did it better.”

“for labs, um, certainly for big tech, and perhaps increasingly for labs, the priorities get set in their promo committee, always. And if you're a PM, and it's always the sort of mid-career PMs, and I've been one of these, and you, like, the incentives are always to get promoted, and the way to get promoted is to build…”

“This is the first time that we've seen a genuine threat to that. And that's because the distance between intent and execution is collapsing. And that's creating not a 20 to 50% better experience for the user, but how you get to that magical TEDx.”

“Why not create a Teach for America for law enforcement? Where you say, look, if you've got student debt, and you go serve in your community for two years, four years as a patrol officer, crime analysts, like, there's a... Yeah. Ton of roles you can have in a police department. Great. We'll retire student debt. So…”

“you want to ultimately be transparent, but if I just walked into a new class and say, hey, the business might be dead, uh, it's not gonna work, and so you have to kind of get a handle of the situation, uh, you know, the people who need to know it, know it, and then once there's a plan and a solution in place, then the…”

“I think the biggest companies aren't incubated because the biggest companies have a founder who is monomaniacal from the beginning and compounds it for 50 years.”

“A neural network is actually a stochastic machine, and so why are we using the substrate that is highly precise and deterministic? For something that's actually stochastic and distributed in nature.”

“Um, but the source, the AI-ness of that capability, in my opinion, is not a source of defensibility. It, it's largely differentiation. The defensibility of a software product resides, in my opinion, you know, from owning the end-to-end workflow, you know, from the context in which that it's, it's applied.”

“nobody can because it's so easy to actually produce software, and that's kind of the, that's the double-edged sword of AI, is that it's very, very easy to produce software. Um, everybody can go do something that is a very obvious idea, because it's obvious everybody's gonna go build it, But can you get to the type of…”

“And there are some companies that operate in this Goldilocks zone of irrelevance, like these janitorial services, where even if you have nine million competitors, Like, they're just not gonna go anywhere, which is why, like, a lot of the strategy that we talk about internally is Greenfield, right? It's like, those…”

“In lots of areas of legal, um, if you make your employee 50 times more efficient, you're eroding your billable hour. In their business, they operate on a contingency basis. Meaning, You know, they only get paid if they make, if they win. So there's no sort of limit to the amount of AI that they want to adopt.”

“momentum has the highest chance of getting you to gravitational scale where you do have a moat. And if you don't do that, by contrast, you're just gonna get eaten alive.”

“the model company is the most cutthroat, because, like, unless you're, if you're state of the art minus, minus, minus, And you're trying to earn a living. It's just like that. That's just not going to work. So that game is super cutthroat.”

“the worst way to measure productivity is I'm going to send a survey to my employees and say, do you feel more productive today from using ChatGPT? First of all, there's a definition issue. Second of all, people are going to answer the way you hope they'll answer. But third, you have no idea if they're actually using…”

“To the extent that AI is going to drive up your margin, that will be all of your competitors opportunity to be less profitable and compete with you.”

“On, uh, outcome-based pricing, um, it sounds very appealing, like, if it can work, it can work, but one thing that, uh, we've started realizing is, um, it actually ends up correlating quite a bit with usage-based pricing, especially with test time compute. Like, if the thing is just, like, thinking quite a bit, like,…”

“a great Investing strategy is to, if you see a company with a five-year plan, short it, um, and it works every time, and yet they still do it, and I, it just comes, it really has no basis in anything.”

“You didn't give the AI a goal. You gave the AI a description of a goal. A description of a thing and a thing are not the same.”

“I think what's happening is that there's something deeper than a goal. And deeper than a value, which is care. We give a shit. We care about things and care is not conceptual. Care is nonverbal. It doesn't indicate what to do. It doesn't indicate how to do it.”

“care is basically like, uh, reward. Like, like how much does this state correlate with survival? How much does this, this, this state correlate with your inclusive, your full inclusive, uh, reproductive fitness? For a, for a somewhat thing that learns evolutionarily, or for a reinforcement learning agent like a LLM,…”

“one of the things I would want to go look for, which you could totally do, is I want to go look in the manifold of its, the belief manifold, and I want to go, See if that belief manifold encodes a sub-manifold that is self-referential, and a sub-sub-manifold that is the dynamics of the self-referential manifold, which…”

“There's a, there's a power of tool that, that just should not be built. Generally, um, because we, it's, it is more power than any human's individual wisdom is available to harness.”

“You have to get it to, it has to be trained on the full manifold of every possible game theoretic situation, every possible team situation, every possible making teams, breaking teams, changing the rules, not changing the rules, all of that stuff. And then, and then it has a really, it has a strong model of, of theory…”

“And as a result, a lot of the techniques we use are like, basically we're just deeply under regularized. Like the models are super overfit. The, the clever trick is they're overfit on the domain of all of human knowledge, which turns out to be a pretty awesome way to get something. That's like pretty good at…”

“so I think many new modalities will be bottlenecked by the Beatles, better than the Beatles problem.”

“the payer that pays the most for, um, diseases of aged population is Medicare, um, which kicks in after 65. Before then, uh, we have a multi-payer system where, um, patients Tend to rotate their insurance every few years. And so there's really not much incentive for someone to cover preventative medicine early in life.”

“And I think biology is like one of the areas where you really don't want to be contrarian. You don't want to be the only company pursuing some obscure mechanism. Uh, usually you want to have some literature validation, but in both of these cases, it was either a big modality differentiator in case of Humira or big, uh,…”

“there's the blockchain stuff that doesn't really make sense to me. Um, I'm a big, big fan of crypto, but it's like, crypto exists, like, if you're trading things that are purely in, uh, that are purely digital, works great. If you're, if you're saying, I'm going to represent something that's physical, where I need to…”

“I think today there's almost a separate trap, which is, uh, being an AI native startup, you can lean a lot on just the AI models and assume they're going to get better and better, which they will, but then that almost becomes a crutch because you're trying to anticipate, well, I don't want to sort of, you know, Create…”

“You think of these public clouds as they have it all together, uh, but really it's a very small set of customers for the highest, highest level of scale, and they're figuring it out on the fly.”

“one, one definition I, I, I kind of like is, um, if you say that you have a remote worker, a human, any, any job that can be done by someone whose job can be done remotely, um, that, that's AGI.”

“In the, in the true pre-training scaling era, you know, GPT-II, three, 3.5, maybe up to four, um, it, it felt like you, you can just, uh, put more internet data in there and just, it just got better. Uh, whereas now it feels like there's a lot of labeling work happening, there's a lot of contracting work happening, a…”

“just to me, this is the biggest risk of AI actually is, it's not It was not described by James Cameron. It was described by George Orwell. You know, it's, in my view, it's not the Terminator. It's 1984 that, um, you know, that as AI eats the Internet and becomes the main way that we interact and get our information…”

“ideas that we thought would for sure get subsumed by the big models are becoming amazingly differentiated businesses, um, just because the fat tail of the universe is very fat and you need really kind of specific understanding of certain scenarios to build an effective model.”

“And so, so therefore, it's just like this assumption that you kind of hear in some of the AI circles, uh, which is like inevitably the smart, you know, kind of thing is going to govern the dumb thing. Like, I, I just think that's like very easily, uh, it's just sort of very easily and obviously falsified. Like,…”

“a bubble is a psychological phenomenon, um, as much as anything, and in order to get to a bubble, everybody has to believe it's not a bubble.”

“we always look at the gross margins, and it's become like a badge of honor for them to actually have low gross margins, because we're like, oh my god, people are actually using your AI stuff. But if you show up, and you're like, I'm an AI company, and it's like, I got 82% gross margins, you're like, I don't think…”

“Because people overvalue systematically their ability once they've acquired a customer through Google to keep it as an organic customer. So they systematically overpay. And they continue doing that. That's why Google never went to outcomes or marketplace, because advertising leads to the advertisers systematically…”

“I'd say the, the, the big shift and what I would urge startups to do is don't build thin wrappers around models that are other people's models. I think the, the, the combination of a model working very closely with the product and the model getting better as there's feedback in the product is going to be super…”

“I think there's a very direct analogy of, let's call it, 1964 to 1972 or something, um, with the reelection of Richard Nixon. Um, like, that cultural revolution that happened during the sixties, kind of the hippie, Vietnam, you know, kind of revolution. Um, I, I think what we, what we in America have been through the…”

“syntax is just an unnatural thing for people. So ultimately English is the programming language.”

“Which is why coding is moving faster than any other domain, is because we can, We, we, we can generate these problems and verify them on the fly.”

“transfer learning is the ability of the machine to, right, to be an expert in one domain and then, and then generalize that into another domain. My answer to that is, like, have you met people? Um, and how many people do you know are able to do transfer learning?”

“we've had 40 years of technology that extended our intellect and our minds. Right, but most of the human experience is actually emotional and it's subjective. Now we have a technology that extends our emotions, our subjective experience, and it can address that.”

“So culture is such an important part of music that you don't just get a model that knows all the music we're ever going to make. Like our lived experiences are what informs the next genre of music. And that's why I'm so bullish on it.”

“simply doing prediction and getting that prediction right, and by the way, it doesn't have to be right a hundred percent of time, it has to be right, like, one percent of the time, because you can validate the other 99% were it right, and then finding that one thing.”

“growing GDP sustainably, durably, is the way to solve America's problems. Austerity is like a temporary hack, and you're kind of holding everything together and making life painful.”

“Hey, you can do whatever you want. You can sell. You're going to make a lot of money and you'll be super successful in life. But you know, if you're like me, you're going to look back the rest of your life thinking, you know, I missed that one shot. That was the one thing I should have taken it all the way. And now…”

“But the problem was that, uh, you know, as it is often with open sources that you just, uh, um, everyone's just downloading the open source version. Actually your biggest enemy Uh, is your open source project.”

“they'll do well in their own archetype. When they have to step outside of their own archetype, they make a mistake. They hire people that are like their own archetype in other roles where that could be lethal.”

“generally, like you, if you're a CEO, and you don't fly low and fast, It's going to be a mess because you don't, you never get the truth because the truth never makes it to you, like through your people.”

“Look, you got to get them to put enough, they're such a big company that they're going to lose interest many times. So if you don't have them write you such a big check that somebody in there is going to get fired if it doesn't go well, you're, it doesn't matter if you get the deal, you're going to lose the deal.”

“Usually founders, usually only one of the co-founders contributes long-term.”

“So, uh, you know, uh, another phrase that, uh, uh, we have been using, uh, recently is, you know, the output of the LLM is the inductive closure of what it has been trained on.”

“there has to be an architectural lead that is able to create these manifolds and just throwing new data will not do it. It'll just smoothen out the already existing manifolds.”

“And, and I, I think that, so research benefits aside, I'm a big believer that society and technology have to co-evolve. It's, you can't just drop the thing at the end. It doesn't work that way. It is, it is a sort of ongoing back and forth.”

“A really good research culture looks much more like running a really good seed stage investing firm and betting on founders and sort of that kind of, than it does like running a product company.”

“there's some quantum, and I would call it 10%, where if you get 10% of all of the supply, You get 100% of the demand.”

“You only buy homes that are mispriced, but turns out not that much in the world is mispriced for a very long time. And that just becomes a, I mean, it may become a good business, just won't become a big business.”

“So I think like there's been three like structural flaws that have basically killed every attempt to solve this. The first one is like, let me solve a part of it. Let me solve the tiniest, like most profitable part of it first, which I objectively has failed like every single time to try it. Like, like when Amazon…”

“So, you know, socialism with Chinese characteristics also means to me, um, investors and, uh, companies kind of lose. They're pretty miserable. Consumers win, and the national government wins.”

“the way that I would characterize, um, China's foreign policy is that it's very engineer-driven. Um, they, the way that they conduct a lot of diplomacy in Africa and Southeast Asia, as well as Latin America, is just to go up to these countries. Build some roads. Build some light rail. Build some ports.”

“nothing has actually gotten that much more efficient, because the filing cabinets were read by humans, the digital records are read by humans, right?”

“And the reason why this is so important to understand is because the whole business model of software has to change. It has to change.”

“we're seeing a lot of companies doing this, they peruse Craigslist, they look for a job listing, and they're like, hey, Plazalane Optometry, I want to apply for the job. And, you know, the optometry's like, Okay, that sounds great. Tell me about your qualifications. Where did you work before? It's like, I know this…”

“we've kind of just embedded in the culture, this idea that risk is like this pass fail thing. Like you either have risk or you don't have risk. And if you have risk, God forbid, like, you should ever go there, basically, and so that to me was like the most eye-opening thing, and I guess I wasn't, I shouldn't have been…”

“there's no concept I would say of a portfolio. Everything is a one-off project. And when a one-off project fails, then whichever party didn't sponsor it says, those guys are idiots.”

“the other problem we have with getting, you know, engineers here is many jobs in, in the engineering organizations are managing contractors. Right. And so if you're some like, you know, super smart person, you know, whatever coming out of, you know, MIT computer science or something, like you actually want to do…”

“However, as we've seen with things like ChatGPT and other things, when you pre-train on the data, when you actually encode the knowledge into the weights, it's not just a retrieval system, you have a richer, deeper understanding of the material.”

“industrial production has a deterrence factor all itself, and that it is critical for winning a conflict.”

“Um, and the only way to replace that skill in my mind is basically high levels of automation to kind of make those jobs simpler so that a new workforce can ramp up in a month.”

“I just think people forget manufacturing is this like real brute force equation where I think you just have to pick like Four to five companies and let them have at it at a scale that's never been seen before and let them figure it out.”

“the whole point of crypto, of crypto is actually to protect consumers from software developers.”

“In some sense, like fertility is declining because the opportunity cost of this suffering is massive, right? It's a massive cost. Now it's totally worth it. I know the hackneyed thing you hear from parents, like all the time, like it's horrible. Also, it's great. Do it. And you're like, that's like, that's like the…”

“So like, what do people, in my experience, the more senior folks actually use AI for code? It's like documentation, writing, testing. I mean, it's a lot of the other stuff that may not actually like increase like the shipping schedule, but like you get a lot more robust code, a lot more maintainable code, a much better…”

“This, this is this really, um, counterintuitive thing where, and you've talked about specialization on the last one is like it, the, the, the, uh, biggest gains of AI go to people who have some degree of expertise in an area to know what is actually true. What is not going to work? What should I, What should I…”

“And to your point, it's actually very rare whenever we say these companies disrupted themselves, it's almost never the case that they disrupted themselves. It's the case that they, they went after a market that they had no, No actual, you know, take the, you know, market share in. And it just worked.”

“If you wanted to be a new systems integrator, and your whole point is, is that we are a systems integrator, but we use cloud code, or we use cognition, or we use, you know, cursor to, to get the output, you will have such an advantage over any incumbent because the incumbent is not going to be able to rebuild that.”

“Also, to build on your point earlier, like, about distribution, and the reason that distribution is not the advantage that it used to be is because It's already exists on seven billion phones.”

“Honestly, last quick point on this is, like, for the first time in a very long time, we're seeing brand effects with an early technology, and what I mean by that is, if you look at, like, whatever, the major model providers, you know, how much better are they from each other? Like, you know, maybe, you know, a little…”

“There's no return for it. Like, if I, if I spend my career checking other people's work, I'm not going to get a professorship at a fancy university.”

“the fact that it's published in a journal doesn't mean it's right. It doesn't mean it's true.”

“How you buy GPUs is like buying cocaine. You call up a couple people, you text a couple people, you ask, yo, how much you got? What's the price?”

“Do you invest in data centers and energy? Yeah. Uh, do you invest in, because that's the bottleneck for your growth really, um, is, is A, well, how much people want to spend and can spend, and B, the ability to actually put them in data centers.”

“And I would say similarly with a virtual cell model, it may not tell me literally why, just like a alpha fold doesn't tell me literally why did the protein fold this way and how, but it just told me the end state and it was reasonably accurate. I think that would already be very important.”

“I've always suspected in tech, we, in Silicon Valley, kind of, we underestimate the power of just kind of brands and consumer inertia, and I think you're sort of seeing that today with ChatGPT of just like such a Household name, like overnight almost, um, that even though it doesn't have in this sort of technical…”

“maybe a lot of the, the network effect has been externalized to the internet, right? Um, and so the idea being, you know, your cursor, and then suddenly all of the, you know, it becomes popular, um, or, you know, mid journey, let's say, right? And then you get all of these, You know, mid journey influencers, YouTube…”

“the capital effects in AI, right? You, you do well, you raise the most money. Um, I, you know, I assume the people raising a billion have already, you know, proven a bunch of things, and at some point the capital becomes a moat, right?”

“That, if you have your one generational player, you pay them whatever it takes. Because, because of the, the max contracts, they by definition are undervalued.”

“Regulate use. Do not regulate development.”

“Eventually, you have to get to consensus. If you're dependent on capital markets, it's very hard to keep the company alive if nobody wants to fund it.”

“you shouldn't be looking for good deals with respect to other investors. You should be looking for good companies and price shouldn't sway you from that.”

“most companies fail from indigestion, not starvation, which is they just raise too much money too easily. They don't listen to the actual market, which is, you know, the customer base. And as a result, they just have a bunch of bad practices and end up running out of money.”

“I really believe the best companies themselves are non-consensus to customers. I just think that the investing market is, is, is, is different than that. Like they kind of understand that and therefore a comment on investors being consensus is very different than a product or being consensus.”

“venture capital uses like the same. Model that you'd use for like a dentist office or something, like a partnership where everybody's equal, right? Like that makes sense for a small service organization, but you can never scale that.”

“So if the use case is the model is creating content, And that content is whatever. It could be language. It could be image that clearly works, right? So, okay, if the model is automating something a human being would do, and we conflate these two things all the time, that's totally different, right? So if I'm like…”

“The only sin is picking the wrong company in a certain space because because that conflict thing, because you're conflicted out of the winner, like investing in a space that doesn't work is fine. I mean, like, there's no way you can actually predict whether a space is gonna work or not.”

“And I think there are definitely like the, the thing that would, would break Substack is if we just looked at the same way that the legacy social media things built advertising and said, oh, we're just going to copy that. Like that's going to work because And if we did that, the thing we would be doing is importing the…”

“What does the end-to-end solution look like? Which only 30% of it might involve AI, and the rest is that schlep that you just have to do to have an impact.”

“I think that's a huge mistake that a lot of the early stage companies can make is you go out there and again, like, you know, you're like, I don't want to charge for implementation. Like these big companies, they won't like that. Like, you know, they just want to pay like one flat fee, right? When in reality, like if…”

“I think that the, I would say my hottest take is, I think the right way to think about foundation labs is that they're like movie studios. And what I mean by that is they invest a ton of money in blockbusters that have a relatively short time spend to pay them back. So you think about, you know, Marvel, you know,…”

“for the first time I can recall, programs are abdicating logic to a third party. Like, we've always abdicated resources.”

“if you just said, here's my entire code base, you know, go run wild, you know, to one agent, it's, it will just, you know, produce worse and worse code over time because it's going to have context rot.”

“Like, I, I don't know how anybody would set up a company to be, to be able to beat 50 startups across 50 different domains.”

“there's this kind of thing in venture capital that a lot of venture capital will say, well, spend all your time with your winners. Um, like we don't believe in that at all. Now, like if you look at a spreadsheet, that's the exact right thing, right? Like because the, the, whatever, three winners are going to produce…”

“we're in this phase where it's such a profound change that anything you do like will work at least for a while. And so it's kind of hard to pass on any deal in, in that way.”

“It's pretty impossible to do drug to fund drug development if there's no M&A market.”

“Every SaaS company under the sun has launched an AI product. They're not just sitting on their hands, and you'd think that they'd have a huge advantage given distribution, but we're just seeing classic innovators dilemma.”

“GPT wrapper was this, like, derogatory term. Like, we've kind of come to the conclusion, like, that's not even a thing. Like, when someone writes software on, like, whatever the cloud, you don't call it a cloud wrapper.”

“Conflicts really matter in this space, and so if you're too aggressive early and you don't really think through things, it can really keep you from investing in the one that's winning.”

“we've kind of come to the opinion that there is no AI. There's like a bunch of subspaces that are totally different that all require their own strategy.”

“like, like anybody that likes decried, oh, defensibility isn't going to work has been wrong. Um, anybody that's decried, like, it's all going to aggregate has been wrong. So zero sum thinking has been wrong.”

“You have complex workflows and a ton of customer data where deep integrations actually are necessary to get that last, uh, last mile, uh, value for the customer. This is where the specialized AI apps are sort of crushing, crushing any, uh, either foundation model layer or otherwise, um, company in the market.”

“AI actually solves that problem. It just solves the bootstrap problem. It's like, these models are so magical. So, like, You know, you wrap one of these models, uh, you know, you make it available, and people think it's amazing they show up. But what's also clear is, like, that doesn't solve your retention problem if…”

“what Noah had just said about some of these other factors that, that I think are really important as well, because, uh, you know, I think both in terms of what it takes on tariff policy, if, if you really want these tariffs to work, stability and certainty is, is, is incredibly important. And, and this is a criticism…”

“Because the supply chain stuff means that five X actually turns into a two and a half X, and then NVIDIA can compress their margin a little bit if you're actually competitive. And then that two and a half X becomes like a 50% better. And then, yeah, so it's like it ends up being way too difficult to, um, and the…”

“No, actually it's not that hard to distill on the outputs of our labs.”

“the status quo is you do not hold scientists liable for downstream uses of their technology. That's absurd. That's, that's a great way to shut down the entire innovation ecosystem and start throwing literally like researchers in jail.”

“open weights is not the ability to produce the weights. But open software is the ability to produce the software. Like if, if you give me open software, I can compile it, I can modify it, whatever, but giving open weights, you don't have it. You don't have the data pipeline, you know, when you're talking about open…”

“closed source and open source are not just differences in technology, but completely different markets altogether. They serve different types of customers. And so, and I, and I think if you believe AI is this sort of explosive new platform shift, then there'll be winners in both.”

“And I think that necessitates vertical integration, not just in the core model layer, but also in like how you interact with the models.”

“Oh, a hundred percent. But I think that the bigger limitation is not the video model right now. I think the bigger limitation is actually that I think large language models still have a lot of work in terms of making people feel very authentic.”

“in general, a company usually needs, in my view, and you may disagree, it needs to be about a hundred exercise of the small company in order to acquire them.”

“startup, startup deals never work in general, because neither of them have money. Maybe one of them can shut down and join the other one. That sometimes works once in a while, but in general, they don't work. Um, that's why M doesn't work, but A works. Like merger usually doesn't work, but acquisition works.”

“About two thirds of the time, the big company says, wow, the leader is really expensive, but we have our magic distribution beans. So we're gonna, we're gonna pick the number two or number three. That's like way, way cheaper and get it in a fire sale. And of course that never, ever, ever works.”

“We need to have a non-key man provision. Which is, this is the designated executive who, uh, is in the event of an aquifer-like thing, and we can decide how to describe it. It's not an acquisition because there was an acquisition, then you have all the FTC, blah, blah, blah, blah, blah stuff, right? But in the event of…”

“tooling itself is never a really huge business because you have to have it. And so it's sort of this, well, if you have to have it, then there's going to be, there are going to be many alternatives and there are going to be some low priced ones, some high priced ones, but the people that want to have the predominant…”

“What China is trying to do is commoditizing our strength. And so the, the, the release of a bunch of open, pure open source, open weight models coming from China is specifically designed to go after a rigid or complacent American view of AI, which is, you know, cloud hosted by a few big players, you know, closed source”

“I feel like we've almost saturated a lot of these evals, and the real, like, metric of, like, how good our models are getting is, I think, gonna be, like, usage, right?”

“tasks matter more at this point, given the fact that we have such a strong algorithm, um, so I think the data, creating data and figuring out, like, the best tasks to train on is, like, the, One of the big questions we have.”

“There are some classes of innovation where you can get by by what I would say is close adjacencies, where the If you're building a product that's closely adjacent to your original product. You can take it from zero to one through the existing Salesforce. Yeah. Right? Because it's going exactly at the same user. But if…”

“no one wins by playing catch up. Right? And so the asymmetry of coming at it from a different angle and being at least an order of magnitude better, uh, is something that you have to continue to keep pushing at and be the biggest skeptic. Because most people will tell you that they're tenx better when they're actually…”

“So if you, if you have an incumbent that's already in the market that, that's entrenched pretty well, uh, a large company will typically try to say, okay, I'm going to go out and build something for the greenfield with an entire platform. But the reality is the market's not greenfield, it's brownfield. So you have to…”

“founders will get much more time. The market gives them more time. The board gives them more time. The employees give them more time. Non-founders for various reasons, and I don't know if you agree, you don't get the same amount of time.”

“And the one lesson I've learned in that is when someone owns more than 20% share in the market and you don't integrate with them, you're just excluding yourself from the market. You're actually not doing anything to displace that vendor.”

“if, if you don't like populism, the way that you get populism is slow growth, right? Because you, because slow growth puts everybody in the mind of, of, of zero sum, right? Everything becomes a zero sum. You know, kind of fight for resources, um, because people don't think that there's a lot of opportunity in the…”

“I think the, the problem has been that people underestimate how many things are truly needed to automate human labor, and so they think, like, we've got reasoning, and now that we've got reasoning, we've, like, this is what it takes to take over a job, and I think, in fact, there's much more to a job than is assumed.”

“One is even if your labor is not worth that much, um, the property you own is potentially worth a lot, right? Like if you own an S&P 500, And there's been explosive growth. You're like a multi multi-millionaire or the land you have is like worth a lot.”

“comparative advantage is totally consistent with like, uh, with human wages being below subsistence.”

“I think it's less like the nuclear bomb where there's a self-contained technology that is so obviously, um, relevant to specifically this like offensive capability. And you can say, well, like there's nuclear power as well, but like neither of those three, like nuclear power is just like this very self-contained thing.…”

“If you have higher inference capacity, not only can you deploy AIs faster and you have more economic value that's generated, but, um, you ha you can have a single copy, sorry, a single model, uh, learn from the experience of all of its copies. And you can have this basically broadly deployed intelligence explosion. So,…”

“SpaceX's network, NASA's state. Uber is the network. Taxi medallions are the state. Uh, Bitcoin is the network. The Fed is the state and so on and so forth, right?”

“the state is someone should pass a law and the network is someone should write some code, right? The state is everybody who is directly or indirectly paid by essentially either the US government or a government more generally. And the network is all those people who are directly or indirectly monetize and make their…”

“And the founding creator is as important as the founding engineer, because the founding engineer is implementation, but the founding creator is the distribution, right? The, the, the founding engineer is the how, but the founding creator is the why.”

“So that's why a small amount of capital, when 20 people compete for it, strengthens the whole ecosystem. But a small amount of aid, when 20 people compete for it, weakens the entire ecosystem.”

“And we've got, actually, we're flanking mainstream media with tweets and podcasts, right? Ultra short form and ultra long form content, where they don't have as much establishment oomph, right? We figured out the formula that works, which is individual over institutional.”

“they will actually, the, the blue journalists will actually deny that sub stack is journalism, right? That Ben Shapiro is journalism, because even if Ben Shapiro has like millions more followers than they do in a much larger audience and so on and so forth, even if he's smarter than they are in many ways, and you know,…”

“Where you have a formal state license to be a blue journo is to have an informal state license, right? Why is it informal? Because if they were formally state license, they could say that it's a state controlled press. So instead, what they get is a White House press pass, and, you know, it's a press control state.”

“People don't like that. That's not what's helpful. That's not what builds a therapeutic alliance. What builds an alliance is being challenged. It's, uh, finding a new way to look at your problem.”

“This is the center of society and they've got their deterministic law with cryptocurrency and their probabilistic guidance with AI and their social network that binds the whole thing together. So those three technologies are like social technologies that are almost like the reactor core of the, of the network state of…”

“It turns out that model was trained on everything humans have ever gathered. So is, like, the incremental experiment gonna update that? Maybe. Information theoretically says probably not.”

“AI makes everything fake and crypto makes it real again. Because AI is a probabilistic technology and crypto is a deterministic technology.”

“But as soon as you're doing RL, where you're training it in a specific domain with a specific verifier, you're likely losing, you know, other areas. So you make it really, really good at playing chess. It's going to be less good at, you know, something else like that, you know, like, you know, writing general scores or…”

“at least right now, I think, I'm not sure if you agree with this, AI doesn't really take your job, it allows you to do any job. Because you can get to like an okay level as like a user interface designer or sound effects or something like that, but you need a specialist for polish”

“So the internet actually introduced the notion of asymmetry, which is the more that you rely on it, the more vulnerable you are. So to wit, the United States is more vulnerable than, like, you know, random, you know, you know, random federal country.”

“And so I think that the real check on something like that is going to be going to have to be cryptography, exit, you know, and so on and so forth, which is ultimately like get out of the jurisdiction, you know, have, have property that they cannot actually seize.”

“The, the social media is many things, but one of the things it is, is it's an x-ray machine.”

“under peak centralization, you're gonna have maximum trust because you're gonna have basically the most information control, right? You just, you're not gonna have the alternative point of view, and so you're just, everybody's, you know, the theory goes, at least you're gonna have a much higher level of unanimity,…”

“the rise of the new technology equals the rise of a new audience equals the rise of an, of a new belief system that inherently is, is, is not trusting.”

“The cliche of our time is that attention spans are collapsing, and this is the rise of, of social media and TikTok, um, and short form video. And so the, the cliche is, you know, and you, you hear this constantly as, you know, kids only want to watch two minute videos and the whole thing. Well, so it turns out, I think…”

“if I was running the democratic party, I would say, okay, look, we either have to, we gotta go full rebel and like, then we can't be like, we, we actually have to be against CNN as well as Fox news and all this shit. And the New York Times and the Wall Street Journal and every single one of them. And we got to be like…”

“I would never, like, I think there's no reason to like tell your primary story through the press. I think that's very dangerous.”

“politics, Um, would be the number one thing I would say don't try at home, uh, in that, uh, it's so, it's almost always, you know, and I'm saying this as a kind of firm that's gotten very involved in politics, um, but it was, unless it's necessary, uh, It's, it's very tough.”

“But you'll end up doing a lot of pilots, um, and because they don't build enough plants, um, kind of sequentially, like, they'll build one big mine every five years, um, if, if that. And the, like, there aren't just, there just aren't a lot of opportunities to get into a commercial scale application, and if you don't…”

“we actually do believe you have to control every single piece of the entire journey, the entire life cycle of, you know, an, an atom of metal end to end to actually be able to build a tech company here. This is not about, you know, a point solution for one particular part of the process in order to actually capture the…”

“Software was always the disruptor. One of the most exciting thing about the AI wave is, like, software's being disrupted.”

“I don't remember ever in, like, the history of computer science where we've, like, from an application standpoint, we've abdicated logic. Like, actual, like, like, like, like, application. Like, in the past, we've abdicated resources. Like, you were like, give me compute. Give me swords. Like, you know, these…”

“Infra creates TAM.”

“AI is changing the value prop of software. You know, it's kind of doing two things at once. One, it's like, it's like the software is not doing work on your behalf. And so the core value of the software is not who in my org has access to the data. It's like, what can this software do for me? Like, you know, can it…”