Scale AI & the Startup Hunger Games with Alexandr Wang

was sort of the original idea for scale and then um and then after we had the idea then i spent like you know a night just looking for a domain um scale api.com was available so uh so bought it um that turned out to be an unusually good decision I think um scale uh has been turned out to be a really good name uh and and then we launched on product ton and it generated like enough traction on product on and then there were probably you know there's probably like after that maybe like four to six months of just sort of like you're kind of in this like wandering mode where you know you have a few customers you're trying to get more customers I remember um I I remember that I would personally talk to anybody who would come into our company's intercom.

So if you were a visitor on scaleAPI.com and you clicked the little chat bubble, you would end up talking to me. And so I remembered there was this period where it was still very unclear if it was gonna work. And then it wasn't until six months later where we started getting, there was one client. you wanted to be really big and then sort of like the idea started to come into its own but um but yeah there's probably you know a year or so of of what are generally called wandering and uh and i think that's on the low end honestly of most startups so so um let's focus a little bit on that angsty part um how did you deal with it did you just because you know it's easy to it's easy to look back on it and to kind of like acknowledge it as like a rite of passage and it's kind of like a learning process but when you're in it it fucking sucks so like did you did you at that point were you able to in some ways like stare the angst in the face or were you just kind of like i am along for the ride and it just up and down every single day every single hour you know one thing i'll say i don't know if it sucked more so than other periods in the future of the company so i think there's like there's plenty of times that suck um when building

a company but i think that the um i think the main thing that sort of got me through all of it was sort of um i think it was i think i think once we arrived on the idea for scale that i did really believe like i get a very high conviction that in the future this thing will exist and it doesn't exist right now so then you have like you have enough of a gradient to go off of at that point and then you can i think have a reasonable confidence in the pathway and like hey at some point i'm going to like figure out how like we go from point a to point b and that's going to be like this long journey um i think that uh i think that so that that i think was like pretty um comforting but i think in the in the in the process itself i think um i remember one of the things that like kept getting in my head was just like the survival odds in yc was like the hunger gains like you know 90 of the companies just die and um but they don't die for years so you're just like you know unlike the hunger games where people die in the first day immediately um for yc if you're gonna die anyway in five days you might as well die now yeah exactly well that was that was kind of the that was the most terrifying thing i think the most terrifying idea was that like

Like, you could already be dead, but you might only find out in three years. Which, but I think that, like, I'm a pretty anxious person. And I think I channeled all of the uncertainty into just, you know, anxiously doing things. And I think if you do that, I think a lot of founders do that. But I think if you do that, then at the very least, like, you feel like you're ruddering and you're making progress in a pretty good direction. did you ever worry i mean coming back to your point about um for any idea that you have you know you can kind of squint and see like 10 people ostensibly doing something similar um and i always say that it's if you tend to get easily psyched out by competition in the early days then you then you're in a hard spot because if you pair that with like extreme paranoia which a lot of founders have that's a deadly combination because you're never actually going to get over it So I guess my question to you here is that Mechanical Turk kind of existed and there were some other, I guess, players.

Did you worry about them or were you like, no, that is just not a good quality product, so I know I can do better? Yeah, there were a bunch of things that existed. And I think the actual answer here is the right attitude is kind of this blustery confidence. And it's actually pretty interesting. i see this uh it's just this conference with uh with this guy from one of the one of the keys that gives a palantir and um he uh he we were both talking to this government customer and he said this right in front of me and he was like you know it's okay that you guys didn't pick palantir but um you know you're gonna come to regret it and you know it would have been nicer if you just pick Palantir from the start and you'd be waiting.

happier and we do a much better job for you than anybody else possibly could but um but it's cool that you guys didn't and uh and you know you'll come around eventually and eventually we'll be friends again but right now you know it was it was it was i'm not even exaggerating it was like a ridiculously aggressive talk track and um and that's actually a part of the success uh recipe for palantir has been this sort of like they this like deep fundamental not particularly grounded i I would say, but this deep fundamental belief that, you know, any software they build is God's greatest software, God's greatest gift to earth, and nobody else can build software as good.

And, you know, there's some of that that is a self-reinforcing loop. And so I mentioned this because obviously, you know, if you're sort of like a rational third-party bystander, you can be like, oh, that seems ridiculous, and that seems kind of, you know, obnoxious. But I think it's actually part of the success story, which is, I think you need some kind of irrational self-belief to believe that you will be able to compete in the marketplace. And over the long run, what ends up happening is you need to believe that you can recruit better than everybody else and you need to believe you'll make better product decisions and you need to believe you'll go the extra mile and maybe other people don't care as much.

And then over the long term, those things will ultimately accumulate to the irrational self-belief that you had. And so one of the things that I believed or that we believed at scale is like, you know, I did a lot of competitive programming competitions and competitive math competitions. And so I felt, you know, if I have to end up competing with people, I'll do just fine. You know, I've done plenty of competitions to date. And so, you know, bring it, we're just, we're just going to build the better thing. And I remember there was like an 18-year-old YC founder I talked to maybe six months ago.

And I asked her the same question. I was like, oh, you know, how do you think about competition? And she was like, I think for competition, just be better. Just do better. Just be better. And I think that is, you know, it's kind of blustery, but I think that really is the answer. You just have to like really believe you're going to be better. um you know that speaking of that palantir uh anecdote they're just really good at they're great yeah this is they're actually very good at making the customers feel as though that without their software their business is actually going to be literally an order of magnitude worse off right um and this is something i struggle with at dropbox because you know at dropbox we were like we would build good software and the software would speak for itself and every time you would launch something somebody else would launch the same launch the same thing two days later and it took us a while to understand it's like oh they weren't actually launching it but they were very good at talking about it so which philosophy do you follow at scale do you like to build the thing or do you like to sell the thing or do you find a happy medium i think you have to do it at both well at least for what we do so at scale we have like small number of

customers um they're very large customers and um and in that world you have to i think it's like one way to think about it is perception is reality a lot of times. And I think probably a lot of the companies that you all have worked with, they're very data-driven companies. It feels like the truth makes its way. The truth, everybody serves a shared sense of reality and whatnot. That's not

Alexandr Wangtrue at most large companies and also not true within the government, unfortunately. But at a lot of large customers, perception is more real than reality. The reality is just so ugly most

of the time, that very rarely do people actually confront reality. And most of the time, they just sort of choose to believe the perceptions that they live in. So what that means is, if you end up doing enterprise sales, or you end up building enterprise products, that you have to, you know, just as much as your job is to improve the reality, it is to shape the perception. And this is, I think, in some ways, Palantir's... one of Palantir's superpowers is that they shape the perception better than most other technology companies do because they think they view themselves, it's like a combination of a sort of acting troupe combined with a software company. And I'm not even saying that, you know, they literally give an acting book to all of the new hires, so they did for a very long time. So I think there's something profound in what you just said, because if If I'm if I'm understanding you correctly, it's that most companies, including the government, they're already living in some perception because the reality is so painful.

So your job is not to actually outsell reality. So outsell shitty perception with an even better one. OK, all right. Sorry, you also should produce value. This is like this is like dark arts on top of if you build a good product. I think that like fundamentally, you also need to build a good product. Maybe switching to you personally a little bit, what would, I mean, you're clearly very competitive, which is great, and you have extreme self-belief, which is amazing, but apart from those, do you have, what else would you self-identify as kind of being your 10x spike? I think for us, the thing, I think my personal one as well as I think it's exhibited in a lot of the key people at scale, it's really sort of creative problem solving, but in the face of kind of like the least sexy problems you can imagine. So, you know, a lot of what we have to do is large-scale operations. We have, you know, hundreds of thousands of contributors all around the world. We have to coordinate this process of them producing high-quality data for models. And it's a very, very unsexy problem, but I think we treat that problem with the same level of respect that we would, you know, let's say a math Olympiad problem,

and we apply the same kinds of techniques and we're like okay like what are the most creative ways that we could solve this problem how could you break it down let's like apply a scientific method towards solving it and we think something a lot of great tech companies do um but we we hold no sort of uh we hold no sort of um elevated sense of purpose of like oh we'll only solve certain problems you know that we think about the impact of the problems and so i think with and then and then we work like hell you know we work very very hard and the combination of these things means that um that's ultimately i think the recipe that allows us to out compete all of our um all of the competition which is that we we're we're smart you know we're good at problem solving and we also will treat the sort of like most lowly problems with the utmost respect um and uh what that means is over time i think we've devised some pretty clever operational solutions pretty clever technology, pretty clever products. And we've all these things have sort of come together in a way that it'd be very hard. You know, I think a lot of times when I look at other companies or other startups, you know, you have one or the other. Either it's an exceptional team, but they

limit themselves or officially towards, you know, esoteric problems that are intellectually interesting. This is a very common pattern. There's a lot of companies where you have a brilliant team but they work on um the most useless problems and then you have and then you have teams which are like scrappy and they'll work on uh they'll work on really um any problem that that is given to them but um but they just don't uh they don't manage to solve the problems in scalable ways or they don't figure out how to like abstract the problems into a way that enables them to solve it effectively so this sort of combination of of you know extremely strong powerful problem solving or extremely good sort of scientific process and technical ability married with sort of you know full humility about the problems to solve how did you I mean that's a that that those unsexy problems and ones that are operationally heavy so there are often this though Ones that are hard to motivate people, particularly in the Bay Area, but I would say largely in Silicon Valley, to get excited about. Have there been cultural things that you have tried to institute at scale, either in hiring or maybe onboarding or the things you talk about? How have you managed

to make that a core part of your DNA? I think the main thing that we encourage people to think about is, you know, uh, we want, we want all of our employees to think about their sort of like their, their net economic impact much more so than their sort of like, you know, the most, the difficulty of their technical problems. And so I think a lot of times like in school or, um, yeah, and particularly in school, you sort of like, you know, you feel like you gain mastery by solving more and more technical problems and that's, that's fine. But I think that, um, uh,

Alexandr WangThere's an inverse correlation between technical difficulty and economic value.

And it turns out like a lot of a lot of very economically valuable problems are not super technically difficult, but they're just like there's just like a lot of hair and you just have to like sift through a lot of mess. And that's I think where most of the good startup opportunities are. And so for us, I think what we really motivated our staff on is like, you know, if you solve this problem effectively, then the net economic impact, if you just think about like net economic impact of problems that you're solving relative to just technical difficulty, that's the ladder you want to be climbing.

And that means that, you know, the unsexy but very valuable problems end up getting solved very effectively. I think Facebook did something pretty similar as far as I can understand. Yeah. I mean, I think that's a good analogy to Facebook. At Facebook, we took great pains. In most companies, like especially on the engineering ladder, you tend to get promoted if you tackle the harder problems, right? Like the harder the problem, you kind of, in some ways, there's a lot of implicit status and respect accorded to it. and at facebook we would often kind of be like no we would actually want to promote the most productive engineer right like simply the person who actually cranked out the most code maybe worked on like the least kind of like sexy problems but maybe like you know so i think it was a very conscious like archetype that we promoted um and a lot of the senior engineers the more traditional senior engineers i think didn't get it they're like why is this person getting promoted they worked on like you know just like kind of you know reducing the number of errors in our logs by like an order of magnitude but actually it's easy i could i mean like yeah but somebody's got to do it you know um yeah that's a that's it's a good lesson from facebook um let's maybe

shifting gears a second let's talk about geopolitics um so what are uh tell us some predictions that you might have for like next year as in how you know obviously the along multiple fronts but certainly on kind of the technology and the hardware and chips kind of like i would say the the cold and the the warm war between kind of like the u.s and china is definitely heating up so how's you know do you have any thoughts on kind of like how that plays out over the next 12 months and how it might affect everyone in silicon valley yeah so i think the first thing to really observe, you know, one of the most meaningful things from a geopolitical standpoint happened a few weeks ago, which is that, you know, OpenAI obviously released L1 some number of months ago, our own preview. And the first replication of the thinking loop, so to speak, or the sort of like test and compute scaling came out of China from DeepSeq as a DeepSeq R1 model. And this is like a very surprising result that the first replication is not from you know Anthropic or Google or any of the American companies it was an open Source model released out of a Chinese lab.

What this tells you is a few things. First, I think there's no gap in research.

Alexandr WangThere's basically no research gap between the leading US labs and the Chinese labs.

So, you know, they're basically caught up in terms of performance. And there's like pretty far-reaching implications for how this plays out going forward. And so that was kind of like a landmark result. I think there's probably a few things to watch. I think the first is which AI technology gets exported more between the US versus China. It will be this race to see which AI stack, so to speak, becomes more globally adopted. So in the first wave of technology, the US came out on top. So Google search is basically globally dominant except for China. social networking, American social networking is basically dominant everywhere except for China.

Then in sort of like the second leg of this race, when it was more of our hardware and telco, China actually came out on top. Huawei technology became pretty widely exported globally. It was sort of packaged in with the Belt and Road initiatives where China sort of pretty quickly became the partner of choice for the majority of countries around the world, actually. And now I think AI is kind of the third phase. So you can see that the U.S. forced the UAE to decide if they want to be on the sort of like China Huawei stack or they want to be on the U.S. stack. The UAE for now is picking the U.S. I think we're going to see a lot more decisions where, you know, countries are going to decide are they on the U.S. AI stack or the the on the Chinese AI stack. I don't think the U.S. even the U.S. from a foreign policy standpoint doesn't even really care about being the AI stack globally. I think we care about being the AI stack to some of our partners, but not all the partners. That'll be one war that happens. And the other one that happens is, the thing to really watch is the export controls. So one of the most important things that the Biden admin did is they launched very restrictive export controls on chips. So there's maybe like

one one hundredth the number of high-end GPUs in China versus the United States because we don't let them buy them. And they also don't have ASML machines and they don't have all the sort of precursors to enable them to actually build the chip industry. With President Trump coming in and with sort of new players in the field, I think one of the questions, I think one of the biggest questions is, are we going to maintain a hard line on the export controls? I think we certainly should. That's one of the biggest things that's enabling us to maintain a strong advantage. And what is that negotiation going to look like? Because there's, the looming threat is by 2027, President Xi and the CCP have said they want to take Taiwan. So, and they've asked the military to prepare um the people of liberation's army to prepare to take taiwan by 2027 so there's this looming date three years from now um uh by which you know china's at least the ccp is at least saying that they'll take um they'll take taiwan and uh as soon as president trump gets into seat and gets inaugurated january 20th that's going to start the clock on the negotiations which is can we prevent China from taking that action or prevent China from blowing out the PSNC

or, you know, prevent the sort of catastrophic scenario and what do we need to give in exchange? And I think that will be, you know, whether it's the headline or it's like a low-grade kind of negotiation over time, that is the big picture thing that we'll watch over the next few years. And ultimately, I think there's like... too much incentive in the world, like World War III will be averted. I don't think there's any interest to get into a hot war, but I think we need to watch like, what are we going to have to give in exchange for them not invading and not taking over the semiconductor industry?

Maybe coming back to your point about the AI stack and what, you know, whether essentially like sovereign states or companies around the world will pick a Western kind of AI stack versus kind of a chinese stack um curious like do you think it's actually possible for us to affect that by anything other than actually innovating in our labs i will admit though to your point it has been a little surprising that after the inference time compute it's kind of been what like eight weeks eight ten weeks that we haven't seen an equivalent release by some of the other major labs which is actually really surprising because my initial take was that this feels somewhat somewhat like easy to replicate it's just like hey just throw a bunch of like more inference time essentially like loop some kind of like um on the various techniques and essentially like you know do some tree climbing and figure out the end result but i haven't seen that be replicated which is odd uh and your deep seek point is a good one so do you think we can we have to just innovate right or do you think we can actually enforce it through policy um well i think the first step is to not squander our open source industry which i think you know at this

point we're through that but definitely there was like there's a lot of conversation at one point and whether or not the U.S. is going to more actively regulate open source models. Now that we're through that, I think that there's, you know, we do have, the United States has interesting levers at our disposal. Most notably, it's the export control. So I think for a lot of countries, we can tell that we can negotiate and say, hey, do you want clusters of NVIDIA GPUs? Well, if you want those, you probably have to, you know, you have to build on top of our stack. And that stack can include the GPUs.

It can include the open source models. include sort of a broad package. I don't know if that's going to be a priority for the new administration. I don't know if this is a priority from a foreign policy standpoint, but I think we have levers to their disposal. Just in the same way that on the flip side, China has their levers. They can offer large infrastructure buildouts. They can offer debt. They can offer a lot of free technology. And that's not stuff that we can match. So it's a, you know, this kind of like give and take of, you know, a few decades ago, China was like, or the U.S. was indisputably the sort of like provider of infrastructure and technology of the world.

Hopefully that continues being the case in the next era. Yeah, I mean, I think to your point about the open source models, I think we're going to look back on the decision by Zuck to actually have like a frontier scale open source model as being pretty critical in that kind of like evolution. Very American decision, for sure. I know, absolutely. Very patriotic, absolutely. Um, coming back to one of your, maybe your other thoughts, I mean, obviously we have seen a lot of, um, kind of like excitement about agentic AI agents kind of means everything and also means nothing right now. Um, what, what do you make of it?

Like, what do you, how do you think in 2025 that kind of plays out both from a consumer perspective, but then maybe also, I think it's a little bit clearer what it means maybe for companies but like for consumers i would still argue that i i mean i haven't used an agent to kind of do anything interesting for me yet uh well i haven't really used one in the company sense either but i'm curious what do you make of that yeah i think there's i think where we are with agents so where we are with models overall is that they're um they're quite good in within one turn you know with one prompt response they perform pretty well and then the performance just goes down a cliff as you increase the number of turns um and this is kind of like i mean nobody will really go out and say it but this is really this is the reality of where we are so where they work really well it's kind of like the Google use case where you ask a query and kind of look at the answer, where they work really poorly is if you actually need to like work with the model and do something more complicated, you know, they just kind of fall off a cliff.

And so I think there's kind of two things. I think the first is the industry is gonna, like the model companies are gonna have to work very hard to improve the reliability with greater numbers of turns and ultimately get these models to have some greater level of internal coherence and and ultimately sort of you know be more like entities right now models don't even know what they don't know and so they kind of and you know in general you're sort of like you know in interacting the model on multiple terms you're sort of like it's like this machine that you know is statistically more likely to be correct than incorrect but it's sort of it's not like a it's not like an entity with any theory of mind that you're interacting with um so i think that'll that'll have to change but i i actually think that like for a lot of for for this like watershed agents moment um the biggest blocker is just um the product design yeah and i think the like you know there is the models are good enough already for there to be some agent product or agent kind of um product experience that actually will will be pretty mind-blowing and pretty great for a lot of people um i think for the people in this room it's probably not obvious because you know maybe

we play with models all the time and so we know that they're actually really good at a lot of things but most people don't even know that the models are that good and so cursor is like a good example of this like i think most engineers didn't actually know that the models were super good and you put into cursor and all of a sudden it's like a part of their workflow it's much easier to use So I think that kind of moment will happen for consumers.

Alexandr WangAnd I think it really is just breaking the models out of the chat paradigm into something that's a little bit more baked into the fundamental workflows.

I think this is one of the big, this is like the biggest startup opportunity in 2025, if I had a guess, is like actually iterating to find the right agent modality. I agree. Great. Maybe let's move on to talking about, obviously, you know, everybody is talking about one of the biggest limitations to building the next kind of generation of frontier models is simply accessibility of like data. How real do you think the data wall is? What's your take on kind of essentially synthetic data for training models? And obviously, I think in this overall conversation scale is a huge kind of like you know uh huge part in the conversation so curious to get your thoughts there yeah i think like i think the the um the frenzy of the last you know 12 to 18 months of conversation was just like who has more compute who has more chips you know the more chips you have you know you're going to win and so it was just this like very un-nuanced conversation around who has the biggest cluster and whoever's the business cluster is going to win and then i think what's like coming to light is, you know, even with a much bigger cluster, you, you were

Alexandr Wanghitting some data limits, like, you know, we've, we're hitting the limits of all publicly available data. And, you know, we need a tandem approach of, you know, production, you know, what are the

specialized data sets in addition to the computational power that's going to yield much greater performance and so i think i mean the this the uh the the data wall and the sort of like the wall on progress that we've hit right now is like certainly real you know i think we've hit some kind of pre-training limits um and a lot of the progress now is coming from post-training that post-training is much more bottlenecked by specialized data sets and and uh and sort of high quality data sets that don't look like the ones that we have on the internet. So data is really critical. I think there was a sort of like, there was a One of the research beliefs was that, oh, the models will just generate large-scale reasoning traces or other kind of data that we'll just pre-train the models on. That hasn't really panned out. I think a lot of the experiments on synthetic data, it just turns out you lose so much of the real richness in the data distribution if you use synthetic data.

So I think the reality of what's going to happen going forward is like, you know, we're going to need to rely on new forms of human generated data to get to where we want to go with the models. But if we scale data, in addition to scaling compute, we can keep making progress. We may not, it may not be necessary to have this sort of like, you know, trillion dollar, five trillion dollar, ten trillion dollar clusters that we were talking about a year ago. But we still probably need the hundred billion dollar clusters. So I think it's kind of, I think there's some kind of normalization to the conversation that's happening.

Yeah, it's kind of like, it was actually all about who had the biggest cluster and then will capitalism work and then we actually have a lot more chips, which is great. But I do think that the new bottleneck is on the data side. So, I mean, I'm curious, how do you think what scale's role in that? Like, obviously you guys play a big role in the post-training part of the world, but do you think that, are there opportunities for scale to even ramp up its kind of its presence on that front? Yeah, I mean, we're having to ramp up data production across the board. So I think that, like, I think that, frankly, what's going to have to happen is, like, if you zoomed all the way out on the, like, path to AGI, so to speak, I think we're going to need, you know, the computers that have to scale exponentially, and then you'll have this, like, smaller curve, but this equivalent curve of, like, data production needing to scale up exponentially.

And so I think we just need to sort of, like, keep scaling up on that curve. and it needs to scale with everything else. So you can't just scale the data ahead of the computer. You can't get, one can't get too far ahead of the other. But yeah, I think our role is really, our key role to play is in producing all the data that actually enables to get to that, get to data scaling in addition to compute scaling. Maybe shifting a little bit to, like a lot of folks out there in the audience are founders of AI companies. What do you think is different about starting a company in the AI space versus all the companies that have been created in the prior generation?

What are some of the notable differences in your mind? So I think there's a few. So one big thing, in any new, technologically new space, there's just a lot of uncertainty. And the flip side of that is a lot of bullshit. so um so ai is no different as an industry um and i think a long time ago probably software as a service was the same where there was you know law of uncertainty a lot of bullshit but but ai is certainly the point right now we're like i don't know 80 to 90 percent of what is out there so to speak is bullshit and uh and a lot of like what people will say or what investors believe or what like other people if you go to a party will tell you like nobody knows what the is actually going on truly truly nobody knows what is actually going on and so um but there's a lot of people who would be confident at wrong and so i think one of the things that's like hard about i'll get i'll tell a funny anecdote um so now we're getting into yeah okay so uh so So when we were raising our Series A, so it was like early 2018, one of our first slides in our deck said something like, data is the lifeblood of AI systems or something like that.

And then first we were pitching Sequoia. This is a whole partner pitch. I get to this slide, and one of the Sequoia partners, I'm not going to say exactly who, very loudly and confidently says, that's not true. Andrew Ring just told me that, you know, we no longer need more data for AI, and so actually, we're going to be fine without more data. This is like two minutes into my presentation. And I'm just like, I don't know what Andrew, I don't know, I was 20 at the time, so I was a little shell-shocked, and I was like, I don't know what Andrew's talking about, I'm sure I could talk to him about it, but you obviously need way more data for AI.

And then Hugh Beast was like, no, I don't think so. And then he didn't pay attention, he was pretty senior, so he didn't pay attention to those presentations, so none of the junior people paid any attention, so I just went through the motions and then you would like you know obviously they said no and that's an example of the kind of thing like there's there's people who are like very very confident about some things which will just prove to be wrong um and that's true in any early industry it's certainly true right now because you know so many people in the tech ecosystem are paid to have thesis to have a thesis or to have theses so it's like if you're like an investor you're paid to have a thesis and a point of view, and you tell your LPs about your thesis.

Or if you're a product leader in a big tech company, you're paid to have a thesis and a point of view. And so, so many people have these beliefs that are not actually truly grounded in that much reality. And so you're just like, I think if you're building in this space,

Alexandr Wangthere's a large premium in you having the ability to figure out what you actually believe and then executing against the flow to accomplish that.

and i think that's probably the that's probably the biggest thing i think that like where you're gonna get fucked is if you follow the trends and like you could you could even raise a lot of money following trends like there's tons of companies if i look at the history of ai there's tons of companies that raised like 50 million dollar series a's or 100 million dollar series a's by like doing something that was really like vogue and in trend but then it turns out that that was just a stupid idea so um i think that's like the that's like the hallmark and ai obviously has gone through many ups and downs and you know um it'll there'll be many more ups and downs in the future but um but the biggest challenge i think for founders is to develop a sense of what you actually believe and then also having a mechanism by which you're constantly learning from the ecosystem versus just digging your heels in a belief and then you get like washed out in the tide um that that'd be my the most important thing yeah you know i'd say that uh so i mean that kind of it helps to be right um but i'd say i mean there's actually a very interesting takeaway particularly for the i think the founders at spc which is that i think that there are very

few categories or verticals that are actually have like any kind of incumbent or even kind of an emerging kind of like winner that's actually winning right now. I think a lot of this shit is still actually up for like grabs because I think that and it's not actually an indicator of like how much money has been raised or even honestly like their usage metrics because everyone's usage metrics are kind of low right now um so i think that you just have to not be psyched out well you also have to be right which is i think a good filter uh but i also just don't think that there's it's i would not get psyched out about essentially like you know the competition at this stage because there's still a ton of opportunity uh and if anything um you know if you have a bunch of people all kind of harping about the same in some ways approach towards a particular idea it's pretty interesting to kind of take a slightly contrarian take on that particular subdomain um yeah yeah i think that i think in general You know, most people in San Francisco don't really believe anything, and they get all their ideas off of Twitter and parties. So I think your mandate is to not be one of those people, and to think independently. And I think that's critical for being able to start a company, because at various points, whatever you build will be popular or unpopular, popular or unpopular.

um and uh and your job is to weather that crazy storm and that's true at the at the at the highest levels like NVIDIA you know has been at various points very unsexy and then all of a sudden very sexy um and they'll go through those waves again and so um that's kind of the uh i think that's the mandate yeah something that you know um something that mark always used to tell us at Facebook is that you're never as as bad as kind of the world tells you but you're never as good as the world tells you right and everybody always takes the you're like oh you know when I'm fucking struggling then like I need to have belief in myself but there's a flip side to it is when the world is telling you that what everything you're doing is right that's also kind of like a sign that like maybe you're not actually thinking critically enough maybe you're not actually being opinionated enough maybe like you're actually converging to like you know that midwit mean, you know? So I do think that there's something to be said about that. It's easier said than done. Like the echo chambers are real. We're all pretty susceptible to it. I'm curious if you agree with this, Alex. I think that one of the mythologies in Silicon Valley has been that

in order to compete in AI, you need a lot of money. I would probably like take the opposite point of view on it. I think there's a ton more opportunity either if you're actually doing on the foundation model space i think the term opportunity in terms of being creative with like algorithmic changes maybe even like in your data sources and certainly like if you're building up and down the ai stack i think you can get a lot further with having like contrarian creative strategies than simply going out and raising boatloads of money um i'm curious is does that resonate so um if your business plan is that you need to raise a lot of money because you need to spend a lot of money that is in business terms a bad business because um because it uh it um yeah obviously you want to build a profitable company you want to build something that makes money so if you have to spend a lot of money you're just deeper in the hole um i think that like i think that oftentimes um silicon valley gets caught up in certain uh sort of uh um races for these grandiose, ambitious objectives.

And so, you know, if you want to be emperor of the world by controlling the foundation model that is God, then I think, yeah, you probably need lots of money. But that's probably not the goal of most people here, or most companies. The goal of most companies is to, you know, build a profitable business that has the ability to keep compounding for a long time period. And so, you know, we watched this pretty closely in the self-driving car debacle, where you know there were um there were hundreds of billions of dollars raised i think close to 100 billion dollars raised to build uh large-scale autonomous vehicles i think like now with the whole cruise thing i think actually like basic like exactly zero of those companies ended up succeeding um so all the money that went into the self-driving cars like you have one winner which is waymo uh and it turns out they have an infinite bank account so they're not really a company you can compete with and i think there's a real analogy to where we are today which is there's like you know if you want to get on the treadmill and compete on trading foundation models you're competing with companies with more money than than you know all 10 countries. It's truly the richest entities that humanity has ever known.

That Microsoft should be in the G8. Exactly. That's not a good strategy. I don't think you should try to compete with an infinite balance sheet. I think that for sure right now, the imperative is what is a creative way? But what is a way in which you can develop something that is genuinely differentiated in the ecosystem and also something that you can keep investing in for a long time horizon?

Alexandr WangLike, I think for at least right now in AI, like, the premium is actually just picking a certain problem to care about for a really long time.

um you know opening eye and anthropic and google these are like very fearsome companies but they also have like literally a bajillion like 10 000 things to care about and worry about and optimize for and um it's very hard for any of them to focus on a particular area perplexity i think is the best example of this like it's still none of the other products are as good as perplexity in just simple search-based LLM output. And so, they haven't spent that much money. They're just way more focused on that problem. So focus always has a big premium. We'll open it up to the audience in a couple of minutes so folks start thinking of your questions.

But before I open it up, do you still write any code? No. No, I used Cursor a little bit the other day just to kind of like see what the hubbub is about um i think it's a great product uh i think it's what i don't know is like it's pretty hard to fully monopolize a workflow yeah um in this way so i think that like if you look at the history of of of comparable businesses like like just owning a workflow is very hard to hold on to over long periods of time because, like, developers, like anybody, you know, they're fickle and they're going to try something new and eventually something new. We'll have, like, one bella worsel that's cooler.

So I think it'll be interesting to see how they choose to, like, sync roots and how they try to lay roots to try to, like, solidify their position. But we built a great thing. Yeah, no, I think developers are super picky and fickle. My take on installing Cursor was also, like, oh, I don't like these three small things and ignore kind of the 10 great things, you know. um who's one person that you think silicon valley should pay more attention to i think in general uh silicon valley is very politically inept so to speak like or like politically i'd say like unsophisticated and so um the uh the broad strokes i mean i think there's this great um email that leaked a long time ago where it was like an email thread is the all this demographic information of the baby boomers and like that's actually the major thread to pay attention to and then i think that in many ways that's like if you get the like broad stroke political threads right then those can be like huge trends for uh for growth and so i'm obviously the major kind of the the major political thread right now is is you know populism and trumpism and all that kind of stuff and um it's one of these things where like i think you know hey i

don't think i tech companies mostly are apolitical and so shouldn't be that shouldn't be that focus on this stuff but um if you were to think like globally what's happening politically uh in terms of greater isolationism greater nationalism greater populism etc etc um there's there's good takeaways for like, what are the, you know, 20 year, um, Tailwinds that might be embedded in all the things that we see. And so I think the sort of like greater political literacy in Silicon Valley is underrated. Ryan, I'll open it up to the audience. I'm Ryan. I just wanted to ask you about your position on synthetic data. So you didn't seem very bullish on synthetic data for post-training. I wanted to hear more about that position. And do you think that is indicative of where we are right now?

and we just need to find the right tricks or is this more of a fundamental thing also understanding that like you said people get paid to have theses you're also paid to have a thesis and this thesis is important for scale oh sorry i think synthetic data works for post-training obviously but it's like i guess like synthetic data it's kind of it's not a philosopher's stone it's not like it's not like this like thing that you just like i'm gonna press the button and keep getting more data out of it it's like you need to come up with all these tricks to get like each trick gets you a little bit more synthetic data and you need to like fundamentally what synthetic data is is like you're leveraging um the structure of the data or you're leveraging all these priors that you can or these sort of like structural you're you're leveraging underlying structures in the data to get you know squeeze out some amount of synthetic data to improve your model like that obviously works as a paradigm but it's not like an infinite pursuit isn't there's not something that like work to infinity and so i guess my overall thought is like it obviously works and it's obviously part of a world-class post-training effort it's just not the like it's not the like

solution to all of our scaling woes hi how are you my name is lorenza and um yeah i would like to um understand you know most entrepreneurs uh at the beginning they're told you know you should target small to medium-sized companies. You shouldn't go for enterprise. You're likely to fail. So what would you say, especially if I understand correctly, from an early stage, you focus on enterprise. Why did you choose enterprise despite the advice that said even from Y Combinator? And yeah, just any advice for entrepreneurs who want to go straight, you know cater to enterprise and not small to immune uh businesses so the issue with most enterprises is that you can't even believe what they tell you because um because there's like then i mean it's just like this it's like this most large enterprises are these cancers and most of the people you'll talk to um have nothing to do with the success of the company but they have some job where they're sort of like, you know, responsible for a thing. And, and so you're like, it's, it's like this whole, uh, rat's nest that you have to learn to navigate and figure out exactly how you leverage it to, to work. And, uh, but if it works, then obviously, you know,

it can be incredibly, it can work incredibly well. All the largest, um, all the largest companies in the world are either enterprise companies or consumer companies. So, um, the sort of like small this is like a really a comment about like very large-scale businesses but small medium businesses you know you tap out at some point now this is not really a statement for startups um so i think i think for like i think for a startup if you want there are some startups that focus on enterprises we were one of them there's there's some others i think have been

Alexandr Wangsuccessful but if you want to do it you have to know that you'll have to spend let's say 30 percent of your time navigating the bureaucracy and bullshit of an organization. Whereas if you

work with small and medium businesses, then they don't even have time to lie to you. So they're just going to, you know, whatever they tell you, you can take at face value, which I think is a great benefit for startups. Yeah. I mean, I think the general advice probably is that like focus on small and medium business. I think if you want to like get liftoff and success as a company, small, medium businesses is definitely the obvious gradient. But I think depending on whatever the ecosystem is like, like if you're in a startup ecosystem where 99% of companies are focused on small to medium businesses, then maybe there's less opportunity there.

And I think this was kind of the, this is the thing that I notice. um if you think if you look at longer themes with startups they're kind of this is what kind of ends up happening which is that um so a bunch of companies start uh half of them are trying to model themselves against the the companies that are successful at that time they all end up being in like perfect competition with one another so none of them get traction then some company those focus on some fringe idea ends up having no competition but that idea ends up working and so then two years later that's the hot company and then all the startups try to copy that you try to model themselves over that company and so on and so forth and so you never want to be like if you look at like all the companies around you you never want to be exactly where all the other companies are focused you want to try doing something that's a little bit fringe because um it's kind of you avoid the the perfect competition so to speak sure hey alex um just a quick question on um your thought on people's opinions and kind of having a specific opinion i think people's opinions are often a combination of their life experiences first principles thinking and your own inputs um now it's difficult to be right

if most of your inputs are bullshit so what are your inputs and how do you filter for your inputs so um what what i choose to like um i try really hard to focus on get it on talking to so the best inputs for me personally are um is like writing on the internet or like talking to specific experts and i try really hard to talk to people who like truly do not care about what other people think right um and you'll you'll find these people because they just really seem to beat to the you know you don't go to the beat of their own drum and you know they like they'll seem very strange and weird but like try to go to try to get the advice of people who like genuinely are uh yeah are independent thinkers and um and also look at the writings of people who are independent thinkers um i think that's that's another thing i think there's a there's a one of the great benefits of i think the modern internet ecosystem is there's there's so much high quality writing on substack that like you can actually learn so much about any individual thing from like experts who post about on substack it's actually like crazy um and so yeah those i think are the the key inputs and then your customers and i think that like um you just have to uh

I adopt a mindset where, like, customers are just always right. And, like, some people were successful in various ways without needing this, but you just have to, like, you have to be super, super subservient to your customers. Cool. I think that's all the time that we have. Thank you so much, Alex, for coming by and sharing your thought today. Yeah, thanks. Thank you. You