The Wisdom Wall

“it turns out that, like, the thing that was hampering the world was being worried about security and privacy and all the things that could go wrong, and when you, like, worry a bit less, you're like, oh my god, these things are unbelievably powerful.”

“you could not really get Gradient descent to encode, uh, uh, sort of like reasoning style algorithms. It was not because the models could not represent these algorithms. It was because gradient descent could not find them, right? So the problem was that it wasn't about deep learning not being trained complete or…”

“It's much easier In my mind, I think, to be a large company executive, because you have all these resources, you have all this leverage, you actually work less hard in your product market fit”

“the place where decentralization is really, really important is at the very bottom of the stack, right? You need decentralization at the Ethereum layer, at base, because what that enables is it enables this global platform that everyone can come and participate on. Now on top of that platform, the thing that we're…”

“obviously domain expertise is still important, But it's not as important as it used to be. It's exponentially less important.”

“my view for what it's worth is I think people should do what, what it takes to, to move to San Francisco. If that's not possible, that's not where your customers are based. At the very least, as you said, move to a tier one city and try and be as close to, uh, you know, folks that are also building stuff.”

“Something that I think is, uh, extremely important in building AI is, is a very rigorous adherence to truth, even if that truth is politically incorrect. Um, I, my intuition for what could make AI very dangerous is if, um, if you force AI to believe things that are not true.”

“people should not spend any time trying to optimize for like a fundraising outcome. And like, you should just basically wait until You're getting investors that are emailing you term sheets that are signed with like a blank for the valuation. And they say, just fill in the valuation and send it back to us. And like,…”

“You know, there's a lot of advice out there to kind of stick with it, Keep going. Don't give up. And I think it's terrible advice. I think if it's not, if it's not working, like give up immediately.”

“what usually happens is they either are, either they joined your startup because they're, they weren't doing well. I'm trying to think of a nice way to say it. Either there's a reason they joined your startup and left Google, or they had expectations that you were farther along or more successful than you actually are.…”

“And in the years since we have created these legal tools that allow firms not just to dominate, but to maintain their dominance, not by making better products, but by establishing a kill zone around their products where it becomes so legally fraught to make compatible products that nobody even tries.”

“If we fix the tech companies, if we say in order to operate a social network, you have to make sure That you are policing hate speech, copyright infringement, terrorism, and every other kind of bad act, then what you're really saying is that you can only have a social network if you're so big that you can afford the…”

“And, and so, if you have great managers, good process people, they will work against allowing the company to become great. They'll take, take a great idea and turn it into a good one, And execute a decent idea at lower risk that's more reasonable, more sensible.”

“So, you know, I think, um, I would probably wipe out most of the structural features. Cause when you think about, you know, the main things are you have a subject, you have a signature, you literally carbon copy and blind carbon copy. And so I think I would remove all of that. So I don't think you need a subject. I…”

“I think if you have worked for a large company for 20 years, you might not be a founder, unless you were forced to for visa reasons, because if you were the kind of person that would, that would make a good founder, you wouldn't be able to stand working for a large company for 20 years, right?”

“Every single problem that companies face, every single problem that companies face is a leadership problem. And if you can come up with an example, if someone wants to text me or hit me up on Twitter with an example that's not a leadership problem, like, oh, we're, we have a problem on our manufacturing line. Guess…”

“There really literally is no such thing as value investing in startups.”

“I think value investing is not a winning strategy when it comes to being an angel investor most of the time.”

“It's probably a bad bet to start a VC-funded company for the vast majority of people. You should not go that route.”

“Another thing that I think is worth hiring earlier, um, that almost no one does, is a full-time fundraiser. Let's say you hire someone like really, really great, and their full-time job is to raise money for the company. Um, you hire them after your B round, and you say, you know what, by the time we raise our C round,…”

“I think the biggest PR hack you can do is to not hire a PR firm. Just pick three or four journalists that you develop really close relationships with, that like you, that understand you, that you get. Um, and then you contact them yourself. They will cover every story you ever give them, and they'll actually pay…”

“When should the founders think about hiring a professional CEO? Never. Um, you, if you look at the most successful companies in tech, um, they are run by their founders for a very long time, um, sometimes forever.”

“Probably one thing is people don't quite realize how possible it is to have, you know, agent based systems that can run not just for an hour or two hours on a problem you care about, but for some problem domains and with highly capable models underlying them, you can get them to run for days or weeks and do really,…”

“If they're completely failing, that's probably a good sign. If they're kind of able to do some of it, but not very well, That's maybe not a great sign because that's probably a sign that the capability is starting to be present in those models and with more training data or larger scale models or, or whatever, it's…”

“I mean, I really think, I think we're at this, like, amazing moment in the world where the bottleneck is not the progress of the AI models. The bottleneck is diffusing that through the rest of the world and, and helping the world adapt to this amazing technology that already exists. Like, I think if the models didn't…”

“And now I actually think with the power of agents, um, and AI broadly speaking, it's much closer to Goliath versus Goliath. Like, I think, but maybe the startup is like a mecha Goliath that is like vastly enhanced by the power of agents and AI, and you know, the, the large companies are the sort of like more…”

“Like all of a sudden the scarce resource isn't gonna be Intelligence or agency. I really think it's going to be vision and ambition.”

“it seems to me like the demand for sufficiently high quality intelligence at a sufficiently low price is effectively uncapped.”

“And so long as you're able to confront the reality, so long as you are able to learn the technology itself actually doesn't matter.”

“whatever it takes to fit the car to you, whatever it takes to fit the organization to you, that's what you ought to do. And the next CEO, whatever the personality is, they can figure it out.”

“Founder mode could scale for 34 years. From zero to five trillion.”

“The narrative about AI destroying jobs is exactly backwards. AI eliminate tasks. AI automates tasks away. But it doesn't necessarily, doesn't necessarily eliminate jobs. And the reason for that is because the, the job of a person has a purpose, and that purpose has many tasks. Some of those tasks could be automated…”

“what we said instead is culture flows from the top. Uh, so we exemplify the culture, we hire, we promote, we hire based on the culture, and today, that still works.”

“Well, what I do is I, I just pick up those random emails, um, or, you know, Slack or whatever, and I just, I just reply with a question, you know. I was like, I don't understand what's going on. And then, you know, and then, you know, the, what happens is that, there's two things that happen. The first one is the…”

“Your job is not to understand what I want, As a boss, your job is to understand what's happening within your span of control, and understand it very, very precisely. What's, what's working and not working?”

“On the other hand, I think it's absolutely critical Uh, to go after the US market as soon as you can. So as soon as you have a product with some form of market fit, uh, go after the US market.”

“The, the one thing, the one lesson I keep learning all the time, so, is that you should always move faster, especially when it comes to hiring and firing.”

“And a new way to do it is actually have the LLMs just write the code you need right now. And then you write 10 or 20 times less code, but it's way more, uh, customizable. It does, you know, more and is more awesome. Between token maxing and even like, you know, don't write code that calls LLMs, write markdown that…”

“And then now it's like, actually what would take a week or a month or a year, like you literally do in, you know, minutes, hours, or, you know, at the worst, like you could build a whole cathedral of like 20 to 40,000 lines of code. Like what would take a typical engineer an entire year? You do it in a few days, right?…”

“if I were to start a company today, I would say, okay, the premise is why can't it be just me? Like, and then you start from there and your token consumption is probably going to be a lot higher than if you said, well, I'm going to have like three people or five people or seven people.”

“The execution is out, right? The execution is gone and the model is going to do that better. The wisdom to choose is still, I think the. The missing bottleneck. And to me, that all comes from which signals are not in the models.”

“And I think that the biggest discontinuities in a positive way that we've had were when we said, hey, let's keep this old way here, keep putting it in a corner. And like, how would we design it if we started the company today from scratch? And then just doing that. It takes a little bit of founder energy to do that,…”

“I think the CEO needs to be the chief AI officer. Like it's not a engineering team thing. It's not like a product team thing. It's like, you have to understand the bounds of the technology better than anyone.”

“how do you build super intelligence inside a company? You do that on everything you do. And it's not more complicated than that. Like, you literally just compose everything that you do, and any given thing that any given person can do, you combine that in aggregate, and in this particular process, and, like, you have a…”

“I think the conclusion that this essay points to is that as we get better at building AI native software. It's going to look a lot more like the agent wrapping software deterministic tools rather than deterministic software wrapping an AI.”

“What I really think is that we shouldn't be just building Investor controlled companies or founder controlled companies. There's a third way. Interesting. We can be building mission controlled companies where the mission itself has sovereignty and these companies can then last a lot longer.”

“you forget that you take a venture investment from a venture firm. You're not taking investment from that individual person. You're taking investment from a company. Then that person leaves. And now you're stuck with some new person and you've given all these veto rights, all these control rights to someone. You don't…”

“Founder control is not that great either, because I know a lot of people who are like basically trapped They can never quit their company because they're like the one, they're like literally the human shield blocking every, that's too much. And of course people die. Then what? So if you want to build a truly long-term…”

“All of the difficulty in energetic engineering today is when people try to do things that should be in markdown in code and it fails because code is brittle.”

“It's very significant for people who are technical because it actually raises the bar on like what you are capable of doing. Like all the people who are attacking me about lines of code, they particularly are, The people who are most likely to get wings if you, like, let it rip and token max. This is sort of like the…”

“One of the key maxims for YC is You know, how do you find good startup ideas, live in the future and build what's missing? Right. And so this is a profound version of that where all you have to do is commit your brain to look at, you know, spending 500 dollars in a single day on tokens and say, actually, like, you…”

“That's the most important part, is that if we had Sudoku, and we know how to solve Sudoku, because like we were just, you know, dumb homo sapiens that didn't know how to solve Sudoku, like it would just have solved it. And that's why it's cool, because it actually is able to discover things without being teacher forced…”

“You have to have an active system, uh, that can actively solve problems for you to get to AGI. That was always clear to us. So agents are that path.”

“The bottleneck here is not Not the model's intelligence. As long as you set the models up right, they are already smart enough to do extraordinary work on your code base.”

“we think that it's going to be more like a peeling an audience analogy, where you start from a really strong base model that have all sorts of common sense knowledge and already works to some extent on your robot, and you have then a Mixed autonomy system. Uh, very similar, for example, to a autonomous driving car…”

“You don't need a incredibly expensive robot that is capable of very precise motion today to be able to do this task. And the reason why is this model really reactive? And so they can compensate for some of the inaccuracy in the actual robot movement”

“this only works for simple cases today, and the reason why that's the case is because I think it's pretty, um, fundamental limitation of the model that we have today, which is that they are not at the core model that take action in the world and see the consequences of its own action, especially action that changes the…”

“if you can detect a microscopic level of DNA and be able to say that that is actually cancer, that is the same really technical problem as being able to detect those in healthy patients or, you know, general population.”

“You don't need a classifier anymore because AI agents are able to read a set of standard operating procedure and reason over an image or reason over sort of unstructured data and know that this is possibly chargeback fraud. You don't need a specialized classifier for it. Um, and you don't need human reasoning anymore.”

“And I think right now we're in this situation where any problem where the solutions you've proposed can be formally verified, and you can actually trust the reward signal. It's not just some guess made by a model. Any domain like this can be fully automated with current technology, with, with the LM-based stack.”

“general intelligence is human level. Skill acquisition efficiency on the, on the same scope of tasks that humans could potentially learn to do.”

“I mean, to me, the fact that you need humans to engineer these harnesses is also a sign that we're, we're short of AGI today, because if we had AGI, you know, AGI would just make its own harness. It would not need to be told how to solve a problem.”

“if you look at, um, Atari games, for instance, or even Dota, you're training, uh, on, on the same environment as what you use for testing. So effectively, you're just trying to memorize the best strategies. You're trying to Uh, at, at training time, explore the full, uh, space of possible game states, and…”

“I do believe the human mind does at the highest level, uh, something that looks a lot like programs in this, like we're currently building Causal models of our surroundings. Like we are, we are describing our surroundings in our mind as, you know, a set of objects and agents and relations between objects, uh, that are…”

“You have to break the combinatorial wall, and the way to do it is to add deep learning guidance. It's actually very similar to the principles that analyze something like AlphaGo or AlphaZero.”

“If you try to scale up genetic algorithms, I mean, I'm sure you can do incredible things with that. You could, in fact, probably do new science. Because that's based on search, and search is the best fit for automating the scientific method.”

“If you're working on something, but the only way to increase the capabilities of the system Is to have human engineers and researchers spend time on it. It will not work because even if the idea is very clever and very elegant and works really well, capabilities are going to be bounded. They're going to be bounded by…”

“if you can solve for verification, which is essentially, you know, you can solve the, the testing part, uh, you can actually automate all the software engineering. That was sort of our key insight that like, you know, verification is the loop, which sort of keeps agent running for a longer, longer period of time.”

“And I think fundamentally it's like, unless you start from, you know, a starting point, which, which, uh, sort of solves all of these problems along the line, the whole software development life cycle, it's actually really hard to come from the other side and solve these problems because you'll make some architectural…”

“brain-computer interface research and development is limited by your ability to record and stimulate these signals. The neuroscience, comparatively, is actually pretty simple. As soon as you can record the signals, we've very quickly figured out what, we talk about neural representations, what they are.”

“especially as we're reviewing YC applications, if you look at a demo and that demo, uh, does not have a base level of quality design, it seems like the person just didn't even try because it's so easy to do it now.”

“historically in machine learning, you always, you know, it's like the, the rule was you have to know your data set really well. Um, but now we're kind of outsourcing that to the AI itself, where the AI is the, it's the AI's job to understand the dataset and figure out where are the failure modes, um, and where are the…”

“There's this other general principle that I think is maybe interesting where you can build for the model and then you can build scaffolding around the model in order to improve performance a little bit. And depending on the domain, you can improve performance, maybe 10, 20%, something like that. And then essentially…”

“If, if you hit this, my recommendation would be delete your QuadMD and just start fresh.”

“I think my big difference is that it actually runs on your computer. Like everything I saw so far runs in the cloud. It's like, it can do a few things. If you run it on your computer, it can do every effing thing, right? So that's really more powerful.”

“Coding is really like creative problem solving that maps very well back into the real world. I think, I think there's a, there's a huge correlation. They need to be really good at creative problem solving, and that's a skill. That's an abstract skill. You can apply to code, but like to any real-world task.”

“If you need to, you can use MCPs on the fly. You don't have to restart, unlike, unlike Codex or Cloud Code, where you actually have to restart the whole thing. I think it's way more elegant, and also scales way better.”

“Segment is a funny business in that, uh, where we started was building these integrations, right? Um, and so it's like, oh, you need to wire up, like, the same data going to, like, Mixpanel and Kissmetrics and Google Analytics, et cetera. And I think just writing that code now, like, that used to be maybe a more…”

“do not give it access to emails would be my number one piece of advice or probably anything. Um, cause it's not clear how safe it is. And it's probably almost certainly gonna probably a lot of people being prompted injected by it right now”

“the models themselves have incrementally improved this year, but there haven't been like major steps forward that have shaken everything up, which is, has a knock on effect. Many episodes ago, we talked about how it was It felt easier than ever to pivot and find a startup idea, because if you could just survive, if you…”

“AI is going to reduce the cost of, like, producing the time to produce things, and so then the expectations from your users and customers will just go up, and you'll need to keep hiring more people to satisfy, like, the growing expectations. I feel like this year has been more in that second camp, and I think that is…”

“Like, it needs to be so weird that it polarizes. Every day we'll get a whole bunch of people talking about that website, and we monitor all our brand mentions online, and it's going to be like either this is literally the best thing I have ever seen, and we also get This website is atrocious. Um, but we feel like we're…”

“with the capabilities of AI, because they're so jagged, you, it's a technology first understanding of what is possible. And so if you go to your customers and tell them and ask them what they want, like, they're not even going to be able to describe what's possible.”

“one of the really clear takeaways from any of these journeys is there is a point that you get to a year, maybe two years in, where From a rational standpoint, you probably should quit, um, but for whatever reason, um, those successful ones don't, and so that is the number one filtering criteria.”

“Even if you look at the successful founder CEOs, after about a decade, most of them leave. And in my mind, it's actually for this very reason, because being a large company executive is different.”

“When you're paying someone else, like the reason why that's interesting for consumer products is, uh, it's a mispriced asset generally still. I mean, Mr. Beast is not a mispriced asset. He's like getting his value from it. Like he's, you know, but It does seem like the mispriced assets are sort of the creators with…”

“today, the actual amount of money that we already know people and companies are willing to spend is the combined salaries of all the people they're currently paying to do the job. And that number is like a thousand X bigger. You pay 20 dollars a month to solve a problem, for example, you know, pay a typical SaaS kind…”

“in a semi-automated world, that's no longer true. And it could be the fact that a single person could make a decision that changes the entire impact of a product. And there's no single person that might be aware of that except themselves. It gets extremely easy for bad actors to do bad things.”

“But I think if you take a step back, and you say, hey, hold on a second, here's two principles, and you can choose the AI that follows this principle or this principle, The first principle is that we're never gonna blow smoke up your ass. Uh, we're only gonna tell you if we like something, or if an idea's good, if it…”

“I, I think that, like, the one sort of general intuition I have is, like, compute is the thing that matters. So, like, I think if you throw enough compute at any of these objectives, you're gonna get something that's probably pretty good, and can kind of be fine tuned to other things, and it's, it's surprising how…”

“and to do a project like training a large language model requires a lot of people to collaborate on like a really complicated piece of infrastructure that isn't going to be a paper, right? Like you're not going to publish like, oh, I got a slightly, I got five percent more efficiency than the next one, um, and it's not…”

“And it might be that like, that data ends up more valuable because you, everything that's linked to a lot, you've already got. Like at some point, you're maybe like going for the tails, or you're going for the stuff that, uh, no one's ever, like, you know, it's only been linked in one place, but it's this like useful…”

“Theoretically, I shouldn't be able to train a better model than that. Like, I'm just going to get the same thing out. So I think that's-”

“Like, it is the case that, like, the labs right now are really driven by getting good eval scores. And it's hard to make them, and anyone can do it. There's no comparative advantage to having the model to making an eval. So I do think it's, it's actually, like, an interesting way to, like, influence the behavior of the…”

“the thing we, like, most need is engineers. Almost always, like, throughout, like, the entire history of this field. It's, like, the case that you throw more compute, the thing kind of works. Uh, the challenge is, like, actually doing that.”

“It's just that scale is easier, and it's more reliable, and I think you, we're still seeing really big gains to that.”

“I keep telling my team, we need to be okay with building crappy products today, ah, because two months down the line, the models will get better, and your business, your product will suddenly become viable.”

“agents that can write code, ah, is actually the easy part. The hard part Is the infrastructure around it. Sometimes I call it the habitat for which the agent lives in.”

“And, and for Replit, and I'm sure a lot of other businesses to survive, at some point Replit needs to stop being focused on making applications and start being focused on solving problems with software.”

“I think there's a limit on how much the training can increase reliability, but I think the environment feedback and the ability to try things really fast is the way to get to the upper echelon of reliability.”

“sales-led product discovery, you're talking to people from the outside. And again, this is important very early on, but it's not as effective as the FDE-led product discovery, where you're solving these problems from the inside.”

“I think one of the other failures, by the way, that's even prior to that, and more, the, the easier failure to become a consulting firm, it's where you build the product in the field that the customers are asking for, rather than the one that's actually valuable to them.”

“my first, second, and third pieces of advice to people who are thinking about trying an FDU strategy is like, don't, Don't do this at home. If you can avoid it, like it's, it's probably bad for you. Probably you're going to end up doing services. And then only if you really try hard not to do it and fail, then, well…”

“In the product market fit strategy, you want to be doing less work for every customer, you want to be driving down costs, you want to keep the contract size the same. In the FTE strategy, you want to drive the contract size up. So you're doing more and more valuable work for this customer and also for future customers.…”

“And the big innovation there does not really come from, you know, how do you prompt and work with the model, but it's how do you make this run at scale? You know, how do you run a 100,000 queries in parallel at the same time and make sure nothing breaks.”

“founders are often very afraid to have a willingness to pay conversation. They feel like if they bring a dollar amount up, it might scare the customer off. But honestly, that's almost always not true.”

“So I'd be pretty flexible on signing, frankly, whatever your early customers want, as long as it's not going to expose you to like unlimited liability or container clause that's going to transfer all the IP in your product to your customer or something company ending like that.”

“if you really believe that development gets easier and it's more simple to create software, it's faster to create software, then like, what is your differentiator? It's design. It's craft. It's attention to detail. It's point of view.”

“as you're, you know, doing, uh, developing a model or you're developing research ideas, You have to have good evals, and usually the researchers are the ones building those, and I think that's kind of just the wrong model. For us, at least, designers, my point of view is that they should be contributing to evals.…”

“We found that there's actually something very, very, very precise and surprising underlying AI training. This really blew us away that there are these nice trends that are as precise as anything that you see in physics or, or astronomy.”

“basically you can Scale up the compute in both pre-training and RL and get better and better performance. And I think that's sort of the fundamental thing that is driving AI progress. It's not that AI researchers are really smart or they suddenly got smart. It's that we found a very, very simple way of making AI better…”

“One is I think it's really a good idea to build things that don't quite work yet. This is probably always a good idea. We always want to have ambition, but I think specifically AI models right now are getting better very, very quickly. And I think that's going to continue. That means that if you build a product that…”

“if scaling laws are failing, it's because we've screwed up AI training in some way. Maybe we got, ah, we got the architecture of the neural network wrong, or there's some bottleneck in training that we don't see, or there's some problem with Precision and the algorithms that we're using. So I think it would take a lot…”

“And so I think the lesson here is that scale is necessary for developing these models that can generalize in open world conditions, but they're subordinate to actually solving the problem. So you need scale, but it's not sufficient for the entire problem.”

“If you get direct answers to these questions with booking links right there, how are you going to mint money from booking and Expedia and Kayak and like, like, you know, all like same, same, same thing for shopping. How are you going to take money from, um, Amazon and like Walmart for the same ad where they're all…”

“no AI product has within app network effect. Like, like it's not like WhatsApp where if you build a WhatsApp rival, Meta has a definitely like a questionable brand, right? Like people don't necessarily trust Meta's products. They think like these are ad products. Despite that, nobody's able to switch up WhatsApp that…”

“that's the key advantage of the browser that you do not have if you commit entirely to just the MCP vision. If you commit entirely to the MCP vision, you require these third-party MCP servers to work reliably. Ah, the data that they send you, ah, on, with the MCP protocol, it has to be perfect. Your chatbot has to…”

“You should assume that if you have a big hit, if your company is something that can make revenue on the scale of hundreds of millions of dollars, or potentially billions of dollars, you should always assume that a model company will copy it. Mainly because they are really looking for revenue. They raise, like, tens of…”

“And even though a lot of the PR excitement and hype has been on these, uh, technology layers, it turns out that almost by definition, the biggest opportunities have to be at the application layer. Because we actually need the applications to generate even more revenue so that they can afford to pay the foundation,…”

“getting data for a lot of startups is a slow mechanism for making decisions. And a subject matter expert with a good gut is often a much better mechanism for making a speedy decision.”

“One surprising thing is, we're used to thinking of code as this really valuable artifact because it's so hard to create, but because the cost of software engineering is going down, code is much less of a valuable artifact as it used to.”

“So choosing the software architecture of your tech stack used to be a one-way door. Once you're built on top of a certain tech stack, you know, set a database schema, really hard to change it. So that used to be a one-way door. I don't want to say it's totally a two-way door, but I find that, um, my team will more…”

“With software engineering becoming much faster, the other interesting dynamic I'm seeing is that the product management work, getting user feedback, deciding what features to build, that is increasingly the bottleneck.”

“safety is not a function of technology. It's a function of how we apply it. So like electric motor, you know, You can't, the maker of electric motor can't guarantee that no one will ever use it for an unsafe downstream task. Like using electric, electric motor can be used to build a dialysis machine, electric vehicle…”

“The technical expertise is now actually really like the missing piece for a lot of these things. Um, and we consistently see, at least in YC, that college students are actually at the forefront of this stuff, like actually understanding how to use the models and how to squeeze the performance consistently out of the…”

“Intelligence is a process, and skill is the output of that process. The skill itself is not intelligence, and displaying skill at any number of tasks does not show intelligence.”

“Constructing world model might be a little more nuanced. The world is more structured. There might be signals that we need to use to guide it. You can call it in the shape of prior. You can call it supervision in your data, whatever it is.”

“The founding fathers of AI who came together in 1956 in Dartmouth, you know, the John McCarthy and Marvin Minsky of them, they wanted to solve the problem of machines that can think, and that's a problem that Turing, Alan Turing, also put forward a few years earlier, 10 years or whatever, earlier than them, and And…”

“The way I kind of like simply think about it is, is that for the last two cycles, let's call it the internet and mobile, we've moved from outcomes and attention into an area of relationships and affection. And I think that's much more dynamic and it's much more in the ordinary course of people's lives.”

“if there's one lesson history has taught us is that if you're going to use energy, you better have social permission to use energy. Uh, so that means you've got to make sure that the output of this AI is socially useful. Uh, in other words, if we really are not creating social surplus, economic surplus, as measured by…”

“So that change management is the real rate remitter, right? Because you're now taking the means of production in an insurance company, in a financial services company, in a healthcare company, in a software company. And saying, we are going to change everything, uh, in the way we work. In fact, we're going to change…”

“I think the key of her is not the little piece of hardware. It's that this thing got to a point where it could run in the background and feel like a sort of AI companion.”

“when the clock cycle of the industry changes this much, startups almost always win, and we've probably never seen it change this much.”

“I, I cannot recall ever thinking until like after starting OpenAI that they were going to be so obviously related. That, you know, the, that energy would be the eventually the fundamental limiter on how much intelligence we could have.”

“LLMs, like, flip, they flip the direction of technology diffusion, ah, that is usually, ah, present in technology. So, for example, with electricity, cryptography, computing, flight, internet, GPS, lots of new transformative technologies that have not been around, Typically, it is the government and corporations that…”

“working with valuable LLMs and so on, I would say, you know, it's less Iron Man robots and more Iron Man suits that you want to build. It's less like building flashy demos of autonomous agents and more building partial autonomy products”

“And then, you know, one thing I will note kind of harkening back to a last response is that even if you had something you could talk to that was human level at coding, Or faster and better, better than a human at coding, you know, sort of the skill of an entire engineering department. I think that the UI of just having…”

“One of the things that characterize search, and I think also characterize our market, is distribution is really helpful for making the product better. And so if you have lots of people using your thing, you have an at scale business, you get a sense of where the product's falling over and where it's doing well.”

“developer tools are power tools, right? Developer tools by definition allow you to, to get to the bare metal, right? To get under the hood of what it, whatever it is you're working with. And so the teams that built these, these tools allowed me Full access to have this agent do whatever I wanted it to do. They didn't…”

“The insight was, yes, let's do 10 minute delivery, but also let's build out the full stack of commerce where we're able to run the logistics, we're able to Have a much better sense and selection on quality control and sort of that audacious thinking Put us in a position where not only will we be able to build a better…”

“Now, obviously at scale, the, the insight is that it's not just important because people are used to it, but it's just, it's a necessity because you've got a lot smaller household sizes in India. You've got a low penetration of four wheelers. You've got much more perishables that people buy. People have got lower…”

“the medicine of the 19 sixties and 19 seventies has become the disease of the 20 twenties. The very same policies that we used to stop the despoliation of the world 60 years ago now stop us from building dense housing, which is good for the environment, solar and wind power, which is good for the environment, nuclear…”

“Institutional renewal is the charge of every new generation. And when we fail to update on the systems that we built 60 years ago, it's not their fault. It's ours. We're the ones deciding to live inside a building. That was built half a century ago that we refused to refurbish, right? That's what relying on 19…”

“where in the, um, places outside of city centers, the suburbs, you tended to have, um, a greater percentage of families, which have more mouths to feed. As a result, you tended to have higher baskets. Higher baskets meant larger revenues. You also tended to have easier ways to find parking. You tended to have a greater…”

“Like it's, it's wild how your coding style changes when actually writing the code becomes a thousand X cheaper. Like as a human, you would never just like blow away something that you'd worked on for a long time and rewrite it from scratch because you had a bug. You'd, you like, you'd always fix the bug, but like for…”

“I think there's gonna be a generation of software engineers that are, like, good enough because it's so easy to retool there with, with all these cogent tools. Like, the barrier is so low. You're gonna be good enough engineers. There's gonna be tons of those. But to be exceptional, like the top one percent, I think…”

“The AI agents, incidentally, will do exactly the same thing. They will absolutely, like, The AI agents will bullshit you just like a human employee will if you don't, like, if you're not technical enough to, like, call them out on their bullshit and be like, like, no, like, you didn't make the change that I had.”

“There are probably only two problems where, uh, you can work on AI and also build product at the same time. One is like search, and the other is self-driving car.”

“there's one way of doing these things where we go to each of these domains and like try to build an index of it and put it into specific formats like tables and then have the LLM, like, like read that. Uh, in a structured language of SQL, or you could do the other way where you just keep it unstructured and expect the…”

“When startups get funded, every single dollar, every single energy effort should go towards trying to find product market fit, which usually is The two founders or the three founders, their computers, their apartment, and maybe one more engineer who's also probably an engineer that you hire. Like that's basically it.…”

“Yeah, I would say if you're, if you're a founder, the, the right approach is to start with the very best model you can, because, you know, your, your startup is only going to be successful if it exploits some, something about AI that realistically is going to be on, uh, you know, the frontier. So start with the very…”

“a lot of the sort of deeper problems within organizations can't really be solved by very narrow point solution software products. And that if you can build A whole suite of really seamlessly interoperable applications. You can build much better products for businesses.”

“And so I think that the fact that these systems can read and their context windows are so large is actually Much more powerful for B to B software than the fact that they can write. Like the generative AI is actually really a misnomer. It's actually the, it's the sort of the ability to ingest all this information.”

“One, I think you need really good executives and you don't want to do the founder mode thing unless something's broken. You know, like you kind of want, cause you can't do it everywhere. Like you need people to help you run the company and like, and so you want to do it when something is broken and when it matters.”

“And finally, some open source companies charge for support and services, but I would actually discourage you to follow that monetization path because your incentive becomes to build the most complex product possible so you can charge for support and services. Which is a bad incentive. And if your product doesn't need…”

“As a rule of thumb, I would wait to get to about one million AR before hiring your first salesperson.”

“Um, you, you can almost Argue that anything that is consumer and publicly available on the internet, that's going to be in the base model. So then your moat ultimately is for all of the other things that are not already online, whether it's, you know, for case techs being a lawyer, or maybe over here on science, or,…”

“Most of the world still does not understand the value of, like, a fairly extreme level of conviction on one bet.”

“I, I had been telling people for a while, I thought that The level two to level three jump was going to happen, but then the level three to level four jump was, level two to level three was going to happen quickly, and then the level three to level four jump was somehow going to be much harder and require some…”

“One thing that we know from running YC for a long time is that whatever the companies in the batch use is a very good predictor of what Like the best companies in the world are using, and therefore what products will be most successful. A lot of YC's most successful companies, you could have basically predicted which…”

“If you treat OpenAI as the Google of the next 20 years, you want to invest in Google and all the things that Google enabled, like Airbnb. Google could do Airbnb, it probably won't. Just from, like, I don't know, Coase's theorem of the firm, probably. It's just, like, too inefficient and too difficult, requires too…”

“One of the most surprising things I learned was that when a customer really wants your product, it's hard to scare them away by quoting a price that's too high.”

“And like, if you play the big company game, they will always beat you. You always have to be thinking about, what would you have gotten fired for at the big company? That should be your playbook at the startup.”

“the skill is often getting the thing taking off to take your money versus someone else's money, not picking amongst things that haven't launched yet and having feces about why one's going to do better than the other.”

“there's so much interest from potential customers to like want a co-pilot that it's actually quite easy to start getting like inbound leads if you pitch this. And if it's even easy to get people to pay you money upfront, but what's really hard is to get them to actually like use the co-pilot because they don't actually…”

“while in the next five or 10 years I think we will get far more used to using it that way, um, I think the, the low-hanging fruit right now is just using the large language model To actually do the sort of knowledge work that a human being could do, and then package it into the UI that, you know, whether it's a mobile…”

“a bunch of us are funding, uh, dev tools companies that sell to AI companies and they're selling tooling, but then they might, you know, they might sell, uh, enterprise contract to someone who also upstream has a fortune 100 that said that they'd pay a 100,000 dollars a year for that contract. And then six to nine…”

“Anything you do with those prompts, you can get your own model to do with a little bit more training.”

“I mean, this is why I think the chat interface is wrong. Like I actually think there is value accrued to really great UX, like good copy, good, um, you know, interaction design, information hierarchy, uh, you know, being able to approach a product and say like, this is the job to be done. And for users to come in and…”

“If your net promoter score, your MPS, is not extremely high as a new consumer company, you're toast. You pretty much have to re-engineer the product to make it something that people love. The reason is it correlates extremely well with word of mouth referral. So I would argue Positive 50 is pretty much a minimum…”

“it's really easy to say, well, you know, I can be in the game in Austin. I can be in the game in Seattle. There are investors there. There are people there. I can be in the game in New York. And I wish that we were more honest that like, yeah, you can play basketball for your high school team, but that's not, that's…”

“they are able to do this task extremely well. And the only way to do that is to have gotten better at under, you know, some form of reasoning and some form of knowledge.”

“One thing that is surprising to me is that the people who say to me they use it the most are some of the people I consider to be better or more senior developers. You might've thought this tool would help juniors more, but I think people who are more accustomed to editing and reading code actually benefit more from…”

“You know, one of the things that we say sometimes at YC is that don't make the investor your customer. Um, but a lot of these conformists who are getting into startups, that's what they do. Like they want an idea an investor is going to like, they want to act the way an investor is going to like, they want to have…”

“you should not use ads as your primary business model unless you expect to be a top 10 site on the internet. Otherwise, it's too hard to monetize and build a huge scale to become a massive company.”

“A lot of the folks who are most approachable when you are a first timer, or when you're just moving to the area, or you're trying to break into the industry, the friendly people that want to talk to you are often The most exploitative, and they're the people whose business model is just, just like, Squeeze you, and…”

“And what they didn't tell that founder is that every subsequent investor would say to themselves, huh, this investor put in a relatively small amount of capital, bought so much of the company that the founders are going to get so diluted that they're not going to end up owning anything. And it's going to shut off later…”

“It shouldn't cost 50,000 dollars to a price round. Like you're just ripping off founders, pure and simple ripping off founders.”

“if your idea for how you start a startup is to first create a pitch deck and go pitch VCs, you're setting yourself up to get screwed. You're setting yourself up to get taken advantage of by people who are gonna have 20 different ways to rip you off, versus if you do something cool, you're gonna have a lot more…”

“What I've also seen for second time founders is that oftentimes for an investor, it's more of a catch. It like helps you build your reputation in your firm, or it helps you get the deal approved. And like, so oftentimes there's like internal political reasons why you might be getting funded that have like nothing to do…”

“We see this a lot with struggling startups. All sorts of people will come out of the woodwork to sell you services or promise they'll connect you with the investors or they'll help you fundraise or you name it. And it's like, it's kind of like moving to Hollywood to become an actor and it's not going well. All these…”

“spending more money allows product market fit to happen faster.”

“No one gets acquired effectively. Like there's an asterisk next to no one. But the, the, the companies, of the companies that attempt, when they're lower on runway, they're, they're, they flew the plane into the side of the, of the mountain. There's no acquisition coming, man.”

“This is what's so weird about this business is like, who has to live with the rest of their lives that that was their startup. Or that they could have done something different. They weren't able to versus the investors like, yeah, whatever. Cool. And they go, you know, they don't think about it at all, ever again.”

“So if a founder actually, like, actions on your feedback in sort of like a material way, like, it makes you less than like, if you're a good investor, I think it makes you less than likely to want to invest. You don't want to invest in someone who is relying on you for product insights.”

“If they can keep you spending your entire salary, and if they can keep the money that you haven't vested yet larger than the amount of money that you have in your savings by like two to three X, you never leave.”

“In many ways, this is a defensive tactic to warehouse talent. Out of the fear that maybe someone else can use you in a way that can hurt the company.”

“literally in any YC batch, I could take every single company and like pick out some like fatal flaw. Um, and like, you know that some of them are going to turn out to be worth like billions of dollars, but you can still do it.”

“A structural advantage with a product that's good enough can win. And man, it really helps when some startups specs the whole product for you. Right? Like Microsoft's move is to not innovate the good enough product. It's to copy it. Maybe five to 10 years later.”

“And if he was an overthinker, he would do, he would rule all the stuff out. All the stuff is like, obviously a bad idea. And anyone in the Midwit territory would rule out all these ideas.”

“I think sometimes counterintuitively investors who know slightly less about your space can be a little bit more grasses greener, like, optimistic than folks who know more about your space.”

“Too often, genius is mistaken for discipline. It's just like, you know, a lot of people think, oh, this person out executed all their peers because they were smarter, and it's like, that assumes the peers were actually working on the thing they said they were working on, as opposed to, like, hiring and going to…”

“if you have a business that is doing something that is legally fraught, but that business is doing something that people love, That you are creating a constituency for a normative shift in how we think about this stuff, and that you are also creating surplus capital that can be used to fund legal adventures to make…”

“I think we could combine the European fair dealing approach where they have enumerated exceptions with the American fair use approach where they have principles that are applied to get a kind of best of both worlds where we might say, okay, here are a bunch of things that are always fair use, right? These, these, these…”

“those conspiracies are necessarily a function of market concentration, because the collective action problem becomes insoluble when you have too many people who have to be inside the conspiracy.”

“Every industry has an incumbent disrupted by a new entrant, not by making a thing that is totally disjoint from the new entrant, not by asking people to leap the chasm from one to the other, to let go of the vine they're holding on before they've got a secure handle on the next one. Everyone figures out how to allow…”

“while launching the product is easy, turning the product into a really big company is actually really hard. Whereas with Boom, it's exactly the opposite. Building a supersonic jet is, like, incredibly hard, but everything else around it is really easy.”

“Most really big companies didn't use all of those channels. They use one or two channels.”

“The idea that it's high, but you don't have any dignity, or, you know, we give you a check to replace lost income because you've been displaced, say, by technology, that that's just as good is probably not true. Probably not just as good, even though, I mean, if you ask somebody, ah, ok, you're making 45,000 dollars a…”

“his job and the way he sees our work at YC is that it's not to figure out what's wrong with the company, but to figure out how it could possibly when. Cause our bets, the ones that when, are the ones that are non-obvious.”

“And so we have to, like, work on our imagination, we have to work on our optimism to figure out, Oh, what is the way that whatever story that they're telling me could become a billion dollar company? And then a great investor pitches that back to the founder.”

“if you have to ask the question, you definitely don't have it.”

“I actually think like the legacy of that growth team, and then like Twitter and LinkedIn and the others that followed is actually maybe problematic when we think about some of the applications of growth later on, because they were applied in this case of like extreme product market fit. And the things you do in that…”

“I also think, like, one of the mistakes that companies make a lot is this, this idea that speed of testing or velocity of testing is the most important thing. Um, it is important to move fast because it is, it is quite hard to have good hypotheses about what's going to work. Like, I am constantly wrong about trying to…”

“I'm now certain that like the majority of people who could be a great software engineer at a really competitive and selective company, Um, the, the majority of those people are not by default autodidacts who could just go to a MOOC and get all the information they need.”

“To say that you can succeed without a degree is on the face of it true, but it is most true if you come from privilege.”

“We think there's basically a market failure where all of society would be richer if there was more efficient mechanisms to like allocate more resources to creative people that are making kind of culture writ large”

“And the advice we give people is basically take your best, most accessible, most focused stuff and make that free. Right? Like make your sort of like your very thoughtful essay on a specific topic free, because that's how people are going to share it. That's how people are going to find out about it and learn that they…”

“Today, the way you break out, it's hard to do that on the product side. You get in front of your competition a little bit, wow, they catch up quickly. The way you get ahead of the competition is create a go-to-market experience that's 10 times better than the competition. That's the way I see companies winning today.”

“I actually think a lot of companies or organizations, the group over optimize for serendipity. Like they're like, what about that chance moment? But to me, like serendipity, like we were like some random thing happened, all of a sudden we have a great idea. Like that's like hitting the lotto. And to me, optimizing for…”

“Ultimately, the challenge is always, especially with consumer kind of focused startup is as soon as you start making money, then VCs will start judging you based on that revenue today. And so strategically, there's some challenges there, which is an unfortunate byproduct of I think how a lot of investors think.”

“because it's so hard to create fake accounts on Facebook, the accounts that are fake might actually be a little bit more powerful because people don't expect there to be as many fake people on Facebook as they do Twitter, and so they might actually have more of an impact in the circles and communities that they've…”

“experience doesn't matter. The rate of learning matters. First principles, thinking matters. Pick for the best athlete, not the person who's the most established wide receiver who knows how to run one pattern and one pattern only.”

“Ah, when you hire a VP of marketing, and I've said this to you before, one question, the functional question is, can you do marketing? But that's not the most important question. If he's one of the top five people in the company, The most important question is, what are the questions he will ask? How will he make the…”

“So I often tell entrepreneurs a business plan is completely irrelevant other than to judge how somebody's thought about a problem, not what they're going to do.”

“I think an investor is an employee who you can't fire. And that's how you should think about it. Otherwise, all the same principles apply.”

“with a sufficient kind of technology of quantum computation and so forth, and if we knew the precise rules, uh, by which black holes evolve, we could reconstruct from the quantum state of the horizon, we could reconstruct what fell in, what's inside, and, uh, and so forth. We could reconstruct that world that fell into…”

“It's, it's not, I don't believe it's the physicist's job to decide what, um, what should and shouldn't be discovered. From a physicist's point of view, everything should be discovered if possible. It is the job of politicians and, uh, other people of that ilk to, uh, to make sure that things are not misused.”

“in very recent years, we found out that entanglement and Einstein-Rosen bridges are the same thing. That, in particular, an example would be if you have two black holes. Black holes have all kinds of internal structure to them. They're quantum mechanical objects. If the two black holes are entangled, they will have an…”

“At Amazon, moving fast and innovating fast, and having a very long pipeline of experiments, that's the more and most important thing, so you're willing to allow, uh, duplication to happen, you're allowed to create technical debt, as long as you know that you have to pay it off.”

“There's a very unique culture at Amazon, I think, that works really well, um, for fast-moving companies, and that is to have, to build your teams as independent as possible, to remove as much hierarchy and structure from your organization as that is possible. Hierarchy, in my eyes, is totally unnatural.”

“if you're a heavy technology company, lots of engineering, there's a, there's a risk that the engineers get in charge. Yeah? If they get in charge, you do not necessarily build products, you build technology.”

“Jeff often makes the distinction between, um, mercenaries and, uh, missionaries. Yeah? Mercenaries are the startup founders that are in it for the money. Yeah? And missionaries are the ones that are in it for the love of the product. Yeah? I want to build this product, or I want to build this business. And you support…”

“The truth is, if with all the knowledge I have about the company, I can't figure out something, I think it's very, uh, rare that someone would come and figure it out.”

“I think that the input is far more important at this stage than the output, because the truth is, If they don't hit their target, first, your target could be wrong, and how could you know what your target should be when you are so early? Um, and two, maybe your product is an issue, or maybe whatever, your pricing is an…”

“It's almost like the, it's not the end user who you have to please, it's the network. Because they understood that the most important feature of a social network is that your friends are there. And so sometimes there's actually a tension between like what's good for the individual and what's good for the network. And…”

“if they get a huge round of funding before they have product market fit, it kills the company. Like it because, because it's just too hard for the founders to resist That, because actually going out and dealing with the fact that you don't have product market fit is really painful.”

“It, I, I don't think it's that you're smarter necessarily. It helps to be smart. It isn't that you're harder working. Like there's a limit to how hard you can work. Um, what I think necessarily to get to that point, you're sitting on top of a big exponential change in reality or in, in, in the world.”

“Yeah, the most common miracle that's quoted today is a data moat. People say, well, we'll generate tons of data, and then we'll be differentiated, and that, I think in genomics something like that could work, but outside of that, I actually think I've never seen a company pull that off in terms of a broader, you know,…”

“I think that, ah, the most successful companies in 95% of the cases are successful despite their investors. I shouldn't say despite. I should say irregardless.”

“Uh, I think the best way is to get an introduction through a founder, uh, that I already know, or somebody else that I already know, so that it's in network. I think if you can't find a way to do that, then you're probably not going to find a way to get customers, you're not going to find a way to get hires, you're not…”

“Magic Leap raised lots of money and spent just 11 years working on a product, versus Oculus, which raised two million dollars on Kickstarter, shipped a, you know, an early development kit that was not amazing, but they iterated and they got through, and now the Oculus, um, The new one, that's like, 400 bucks, doesn't…”

“if you're in a situation where you're building something that's, like, mind-blowingly cool and revolutionary, but you're finding that customers don't want to buy it, pivot your company so you're actually doing what your customers would have done, and just use your pioneering technology to do what they do faster,…”

“If you run into an investor who is asking for five-year projections at seed, you've run into what we colloquially call a noob. They don't know what they're doing. Who, who has five year predictions at this point?”

“generally speaking, I think that sort of pre product market fit metrics are actually relatively unhelpful. Um, uh, I think you sort of, you really want to bias very strongly towards kind of as much, um, sort of inspection and kind of, uh, high throughput qualitative feedback as possible.”

“in, in startup time we really extol and sort of uphold the virtues of, sort of, you know, determination at all costs, you know, never quit and so on. And, you know, that's clearly not the right answer, uh, in that, like, sometimes you should quit.”

“And the, kind of, the growth curve or adoption curve is just, kind of, a function of, of that go to market apparatus. But basically, it's, it's, it's not some, kind of, cosmic trajectory. It's something, kind of, very much of your creation and under your control.”

“as an empirical matter, it seems that somewhere between kind of, you know, two and 10 depending on exactly what you're building is kind of the, um, optimally responsive size.”

“when you hit a certain size, again, maybe, I'm just gonna say 10 people for the sake of simplicity, Maybe it's a bit before, maybe it's a bit after. I think you need to kind of adjust more, more deliberately to this kind of explicit communication model of sort of, um, uh, of, of being sort of quite, um, quite firmly…”

“Frankly, I think the really big product market fit, uh, moments for every company are pretty unmistakable. Like, the Dropbox founders have called it stepping on a landmine. Uh, I just, I, I really don't think you can mistake it. It, it really happens in a way that you, you lose control. It's very obvious. Every metric…”

“My feeling is that that's extremely rare and that, again, this is a really dangerous place to be because you can stay in this iterative mode for years. And it is unlikely that the iterations are going to get you to a good place. So I remember very clearly early on being really inspired by the Airbnb story and it being…”

“The, the problem with the first product And product market fit is that you can move the product and you can move the market because it's fit between these two things. And so what's unclear and they move in some crazy multidimensional space. And so what's the, the issue is that to get them to both match up, um, You can…”

“So we went from literally zero management experience, aside from what had been picked up along the way, growing from zero to 50 people, to having someone who really knew it, or two people who really knew what they were doing. Um, that was hugely impactful, and we should have figured out a way to do that earlier. 50…”

“So strangely, starting with a higher price or a price is almost always better than starting free. Um, almost always better.”

“the kind of community of startup people and or investors usually don't have the problem that you're solving. So if you're using investors as a, as a trigger for, am I solving the right problem? Or like, Do they find this useful? It's almost never the case. I'm almost never the user of a product that comes into YC. And…”

“The outcomes of startups are hard to predict. I think to some degree they're actually indeterminate. There's a huge amount of luck involved, you know, um, and so I think even if, even if Y Combinator were as good at picking startups as you could, as you could possibly get.”

“intelligent people can judge other intelligent people, But trustworthy people cannot judge other trustworthy people. In fact, trustworthy people are often fooled by untrustworthy people.”

“And what do things that don't scale means is do those things early on anyway, because, you know, if you don't do them, you'll never be big. And so you got nothing to lose.”

“and actually it turns out it's not that important to be smart. Um, it's much, much more important to be determined.”

“Ok, so the way you deal with different commitment levels is you ask yourself, um, would I rather have 30% of this person or a hundred percent of some other person? And in my case, I would rather have, like, 10% of Robert Morris's brain than a hundred percent of almost anybody else's.”

“launch as soon as you have a quantum of utility, which means as soon as you're, there's one person in the world who is glad that you launched, because now they can do something that they couldn't do.”

“About 20%. Um, we try to, we try to hire one out of every five or six people, uh, From the industry. We think we need that expertise. My, my principle is five or six people. Cause like most teams should have five or six people on it. And I want one person in the room has done it before, but not two. Cause then they'll…”

“what entrepreneurial sales is very different than regular sales, regular sales. You want to get to know as fast as you can. So you stop wasting your time and get to the next Yes. You want to get to know within five minutes ideally. Yeah. That's great if someone hangs up on you. Cool. I can go to the next one. Totally.…”

“In fact, I think in 2018, at least in Silicon Valley, it's easier to start a hard startup than it is to start an easy startup.”

“all of the really great businesses I know, um, have an answer to this question. And in fact, the better they are, the more they pretend not to.”

“So two people eating the same loaf of, uh, white bread, one would really spike their glucose and one would really stay flat. Um, and that's, uh, that's true even if you normalize the responses to the responses to glucose. So you have, um, just, just to see if, so, so, so even foods are not categorically good or bad. It…”

“fecal transplants have been used, they, they, their claim to fame is, uh, by treating Clostridium difficile infections... you indeed, um, transplant stool into these people, and that transplant works mainly because their microbiome is so depleted and it's like, you know, cultivating, uh, uh, an ecosystem in a place…”

“if you can get away with charging more, then it means that you have a product that people truly want, which means that you can grow faster in part because people truly want your product, and in part because the capital leverage that Mark talks about is actually crucial. In other words, and he makes this point, I think,…”

“If you had two good leaders in a SEAL platoon, the SEAL platoon was going to do awesome. And it didn't matter where they were. If they happen to be the platoon commander, great. If it happened to be the platoon chief, great. If it happened to be the leading petty officer, great.”

“People ask me, what do you do when you have a weak leader? I'm like, I'm happy. If my leader's weak, I'm stoked. I'm going, hey boss, you know, I got this. Hey, thanks for your support. We're going to keep over. We're going to be over here doing what we do. I'll let you know if we need anything.”

“So it's really the entanglement which, um, holds space together, which keeps it from falling apart into little pieces.”

“In some sense, um, ICOs are kind of the gaming use case, right? It's kind of like a gambling, it's a gambling use case, right? And like, that's actually kind of what's happening is people are just gambling as real money.”

“So I essentially think that it needs to be mandatory privacy if you're going to actually have a privacy coin. Um, if, if it's ever, um, if it's ever optional, you get looked at if you're starting to use a privacy feature, like why are you using this privacy feature? Do you have something to hide? So first of all, it…”

“if you're gonna spend money on Customer acquisition reserve at least five million dollars. That was like a year or two ago, I know.”

“reducing the probability of existential risk is super important, even if AI is, uh, you know, decades or centuries away, and even if we can only, you know, decrease the probability, you know, of, of that happening by, like, .1% or whatever in expectation, that's, like, a huge amount of value that you're, uh, uh, that…”

“corrigibility, what he calls corrigibility, and what others have called corrigibility, might actually be, like, a stable basin of attraction, in the sense that if a system, you know, uh, is, uh, designed in such a way that it's able to, like, take critical feedback, and it's able to say, okay, yeah, what I was doing…”

“We, we've noticed that it is almost impossible to, you can make a lot of mistakes as a company, but if you screw up the first five, seven hires, it's almost impossible to get through that.”

“Once you send your kid off to school full-time, parents remain involved in some respects. I think reading is one of the main ones. They still think, feel like it's their job to make sure they're reading regularly to their child and, and helping to reinforce, you know, their ability to read at home. And then it sort of…”

“with psilocybin it's completely different because the expert is themselves. They are the expert because what's happening, the, the, the healing power of this is something in the unconscious mind, their own unconscious mind is writing itself. There's a homeostatic principle at play here, I'm sure of it, so that whatever…”

“If the network's built well, we would argue that no value actually accrues to the company itself. The purpose of the token is to remove the centralized party. And so the, the company itself, the, the LLC or the corporation shouldn't get transaction fees. They shouldn't get subscription fees. They shouldn't be rent…”

“oftentimes the bad ideas, I mean, the good ideas look bad. Uh, and the reason for this is that if the idea were obviously good, like, the opportunity wouldn't exist. Big companies are not that dumb. They're capable of executing. They have lots of resources. They have market power. Like, if they thought there was this,…”

“Any time I've been, like, excited about something, but then it's like, ah, the price is so high, eh, you know, I'll invest in something else. I've always regretted it. Like, so now, actually, in my head, I kind of turn around the rule. I'm like, if I'm dissuaded by the price, I should probably invest more. I just need…”

“Implausibly ambitious or frivolous ideas, and the reason that we're kind of at these two extremes is that if the idea is in the middle, as I mentioned at the start, probably there's like a big company somewhere working on it, or a hundred other startups, and so you want to see something that's either sounds impossible,…”

“I'm very nervous about people who've always gotten perfect grades. They got into all the right things. They did all the right clubs. The people who have the perfect resume. The people who didn't fuck up. Um, because these are people who've never failed. And startup founder Being a startup founder, you're basically…”

“You can almost define, predict the, the, the, I think success of a startup just based on how quickly they're able to iterate. And the really great companies just like iterate really quickly. The ones who, you know, do a release every two years fail. Do zero.”

“So one thing that we hear a lot of, especially for folks that haven't done much of this, is they say, okay, I'm in, once you go find the rest of the round. And this is the cheapest of cheap shots that you can possibly do, um, because what you're actually saying is, I would like an option that's guaranteed in your…”

“an A investor, grade A investor, Um, gives you money quickly, signs the paperwork quickly, shuts the hell up. That's an A investor. There are very, very few A plus investors in the world. There are very, very many less than A investors.”

“the most hyped investment on Demo Day is often not the best one. Very, very often. Um, I think what's interesting is that the crowd on Demo Day is just not as smart as it appears”

“Um, so I would say that it's extremely rare that an investor can have truly big impact.”

“I think investors believe that having the skill to network to an investor is an important characteristic of a good founder, and I think, ah, kind of YC's core position from the beginning is that it's not.”

“The, the cap tables that we see, and we've now seen thousands of cap tables, it's surprising how many times the lawyers actually get the conversions wrong.”

“most of the time you lose one extra money, but occasionally you do get to make a hundred or a thousand X, and that, uh, for the same reason slot machines are so satisfying, is incredibly addictive.”

“I would say 80% of investors outsource 80% of their decision making to what other people think about an investment opportunity, but the problem is everyone does that, and so there's this weird schooling effect where a company gets hot for no discernible reason, or it fits a trend or whatever, and then everybody wants…”

“this idea that we're available and we're open, um, I think this was in the two or three most important secrets of YC. This was something that we did that was really different. Other people hadn't done it before.”

“Speaking of the really big companies, I think, I don't know if this was always true, I suspect it may have been, but I think today, it is easier to start a hard company than an easy company.”

“Yeah, I think, ah, you know, the most interesting companies start with a TAM of nearly zero, and it's like, ah, very bad investors are the ones that are super focused on the TAM of today. Very good investors are focused on the TAM of 10 years from now.”

“the number one most important insight about how to blitz scale is that the good kind of blitz scaling is when you are not having to generate demand as you go, but that you first got the product right.”

“I, you know, you can, people talk about working smarter all they want. There's nothing that makes up for, like, working smarter and a lot of hours.”

“The next piece I would say to people is like, you have to find, you know, be really wary of solving a small, rare problem. You can solve a big problem. You can solve a rare problem. Sorry, you can solve a small problem, fine. A rare problem, fine. But small and rare is just It's, it's never works. I've never, and these…”

“you actually try to make that core be as small as possible because machine learning is really hard. It's a lot of compute. It's really hard to tell what's going on there. And so you want it to be as simple as possible, but then you surround it by as much engineering as you possibly can.”

“The thing about League of Legends is it doesn't run on Linux and, uh, uh, it doesn't have a game API and little things like that actually are the biggest barrier to making AI progress in a lot of ways.”

“getting good basics in linear algebra, in linear algebra and in basic statistics, that's, especially when Doing experiments, it's easy to make like elementary statistics mistakes, uh, and, and, and linear algebra is just kind of most of what you need to know to like basic optimization as well, uh, to, to, to, to follow…”

“And when you meet with people who you think might be interested in buying your company, you tell them you're fundraising. You are never in the process of selling your company until it's been sold.”

“My, Feeling about all this, and just this is a general thing, being a manager, executive, founder, is you should give people as much information as possible. You really should try and be transparent. If that information is actionable, if it can actually cause them to Do something. Create a new venture. Build a new…”

“if you just fund like the same people over and over again, Then yes, they'll be like really efficient. They'll understand what it takes to do this, but you'll also miss out on all of the other innovative, different thinkers who are out there who actually made the discoveries, like the PhD students and postdocs are the…”

“And so they end up having to rely on proxies for that. And a good proxy often is, you know, where's the venture funded companies happening and venture funded companies have tend to happen closest to where VCs happen to live because they don't like to travel that much, um, to go fund companies. And so that tends to be…”

“she doesn't care that she's getting it like directly from us. That's not like the value proposition. Like maybe she even wants to just go get it right, you know, at Sephora like the next day. But what is, you know, really interesting about how we're reinventing the beauty experience is that relationship she feels with…”

“VC is not built for long-term investment in things that are Extremely high risk and building some deep foundational technology. VC is tuned for 10 year returns, which means that in two, three years you have to, like, show very strong, um, you know, sign of, like, massive adoption and a really strong business, which…”

“my, my claim is you couldn't build something as free and open, and that works as well as the internet today, um, because no group would fund it. Um, and, and what you would end up with is a, a massively stunted version of something that is highly centralized and controlled by a couple, by a couple of groups”

“the company is fading away, or not fading away, but rather a new thing has come in, which is the network or a market that, that is not a company, but it functions kind of like a company. And so you can think of Ethereum not as a company, but rather as a, as this network that, that has some shared asset that is, that is…”

“You can't, you can't measure knowledge work in hours. Somebody's investment over a decade can put them into the right perspective to make the right contribution at the right time, that creates something like Ethereum or Bitcoin or whatever. And, you know, how do you correctly reward that? And I think, like, something…”

“I could say that if you're not going to a top five school, You are not technical and you want to get into tech startups. It makes no sense to me why you wouldn't try to get a job in an early stage company in a startup city.”

“people do not see the NBA as a startup. Um, uh, what is that called? It's not seen as like a positive resume item.”

“people that run these perfectly non-chiotic organizations somehow never build great companies. So, so the trade-off is we're going to accept a little bit of chaos, um, in exchange for a shot at one of these massive great companies, and founders just have to sell that to their teams.”

“my, my, my version of this is you want to hire for values first, aptitude second, and skills third. And I think, um, the problem with most executive recruiters is that they reverse that order.”

“The general framework That I think about when building, when adding people to a team, which is I think what defines culture, is what, what is going to be this person's net effect on the output vector of the organization? And there are people who are brilliant and can get a lot of work done themselves and would add to…”

“And, and the, the, when I talk to folks, I'm trying to get them to sharpen their mind, is what, what filter would have A players working at other companies than yours? Which A players do you not want working at your company? And that, and that, that shows you that you begin to have something that's a kind of an edge…”

“Certainly when you get to the stage where people that were used to knowing everything don't know everything, ah, that, I, I think more early employees leave over that than anything else, and, and it's, it's really tough to say what to do about that. Um, it's people that have gone from being sort of absolutely in the…”

“part of what the blitzscaling stuff is, you actually deploy capital in a fast and inefficient way in order to get to scale and or global scale.”

“The one mistake that I think companies really make in this is they decide that they need a lot of people, they can't get them, and so they're going to just, like, double salaries or triple salaries. They're just gonna, like, start way overpaying for people, and that's the one that seems impossible to turn back down in…”

“frequently one of the questions that can be in consumer internet especially, enterprise needs to identify this earlier than consumer, and consumer sometimes will overly obsess with operating margins early, when it's actually in fact get to scale, Then obsessed with operating margins. Because if you're not at scale,…”

“Majority of business problems can be framed as supervised learning, and therefore they can be solved with current techniques, as long as we have sufficient number of input examples and what we want to predict”

“people who are fresh out of university and boot camps actually in many cases, because they've been practicing, are better at the kind of problem that gets asked in an interview than your very senior, you know, eight, 10 years of experience at that large company engineer.”

“if you have five years of experience, it's, it's just flat out easier to pass an interview. You will get a job offer You know, after a worse performance”

“asking pretty easy interview questions is often more generally predictive than asking harder interview questions.”

“If you've worked with someone for a week, you have a far better read of their skill than I think anyone can get during a, during a three to four hour interview. Um, the problem is that there's a pretty strong, um, bias in who's willing to do trial employment. Um, and that's, it's adverse bias. So that, so some of the,…”

“whiteboard coding tends to skew toward the academic. It tends to Give preference to people who are really good at breaking their thoughts down into this sort of structured academic way and writing with a small amount of code. So you often have people who are actually really productive, excellent programmers who look…”

“We tried, we tried scoring engineers, and I had to have them talk about past projects. And scoring them and trying, like, even go, even like, like, like a full hour, like going into depth in the project, talking technical details, scoring it, um, just talking skill. Ability to spin a tail ended up dominating the actual…”

“Founders that were obsessed with their own personal liquidity when the company had nothing turned out to like fail most of the time. And so investors learned that if founders pushed on this in the seed round, It was a very, very bad sign.”

“As a general rule, don't start any acquisition conversation, unless you're willing to sell, ah, for a pretty low number.”

“In fact, a lot of the companies that I've been very involved with that have had a very bad first hire in the first three or so employees never recover from it. It just kills the company.”

“If someone's choosing between like McKinsey and joining your startup, um, very unlikely that that person's going to work out at the startup.”

“Boards add value to business strategy only rarely. Um, but very frequently you can use them as a forcing function to get the company to care about metrics and milestones.”

“I think, you know, not to stress about press too much because press is not a sustainable user acquisition strategy, not even for You know, consumer companies, so it's not something to spend too much time thinking about.”

“And so I would ignore total shipments for VR and AR, and I would wait until it gets to the point where someone builds a device or software that once people get it, they use it a lot of every day.”

“it is harder for existing companies to get good technology people than it is for good technology people to learn about other industries.”

“I do think that working hard early in your career to get the leverage and the compounding effects is underrated and one of the most valuable pieces of advice that I never got.”

“And a framework that I have for when to give up and when to keep working is it should be an internal, not an external decision. Um, if people aren't Using it or if people are saying it's bad, that alone is not a reason to give up. You want to pay some attention to that, they might be right, um, but I think the best…”

“at least in my own experience, what I found is that burnout actually comes from, um, failing and things not working. Momentum is really energizing. The lack of momentum is super draining.”

“I think history belongs to the doers, and I think you should Take a risk. Actually do something. Don't just sit around and talk and organize and bring together groups and go to thought leadership conferences and whatever it is these people spend their time on other than actually building something.”

“I think, um, having a high bandwidth interface to the brain, like, um, we're currently bandwidth limited. We, we, we have a digital tertiary self, uh, in the form of our email capabilities, our computers, phones, applications, uh, we're effectively superhuman, um, but we're extremely bandwidth constrained in that…”

“It feels like technology just automatically gets better every year, but it actually doesn't. It only gets better if smart people work, work like crazy to make it better. That's how any technology actually gets better, and By itself, technology, if, if people don't work on it, it actually will decline.”

“So, so I think if we can effectively, uh, um, merge with, uh, AI by, um, improving that, uh, the, the neural link between your cortex and the, the, your digital extension yourself, which already, like I said, already exists, just has a bandwidth issue. [865] Elon Musk: Um, and then, then effectively, um, [869] Elon…”

“what really matters is the, is the machine that builds the machine, the factory, um, and this, that is at least towards magnitude harder than the vehicle itself.”

“if I had to say the most important traits of the most successful founders, I've already mentioned determination. That is by far the most important quality. More than intelligence. More than previous success, you know, in school.”

“it's such a better idea for founders to just try and if they don't have a co-founder to just do it themselves. It will be very hard, but it is the damage that can be done by tacking on a co-founder that you haven't worked with, that you don't know very well, that you don't think that you are not completely sure shares…”

“If we've, if we were looking at someone, someone's application, and they worked for a long time for a large company, That's, that's actually bad to us, because like, the, the best startup founders probably could not stand that.”

“You know, really big markets are bad for startups, not good.”

“inference time compute to perform search over plausible ways of solving the problem that can get much, much higher performance or much more reliability in, Long-running agent flows.”

“the importance of specifying what you, what it is you want has actually gone up, because before you'd be handing it off to a very intelligent human who maybe has context or can ask you follow-up questions, Um, and agents can sometimes do that, but I, I think clear specifications is, is a really good idea.”

“Um, I mean, I think there's still just, like, astronomical opportunity in, uh, agentic looping, and, and figuring out how you develop systems that enable you to spend, like, 1000 X more or one million X more on tokens to drive an outcome in a, in a continuous feedback loop.”

“every six months, delete your Cloud MD. Delete your skills. Delete your hooks. See what the model does, and it might surprise you. And actually for Opus Five, this is something we really do recommend, is just try deleting all of these things, because the model might really just not need all those instructions that you…”

“The thing that you want to do is you want to run it. And if it's like a custom agentic product that you're building, you want to kind of run the products. Uh, you want to see where it fails with the model. You want to see what it does well. If you're using quad code, you want to see where it does well with your code…”

“I think evals, they outlive the harness a little bit, but not quite that much. Like, an eval might live for maybe one, two, three model generations, but nowadays the, you know, we're on the exponential. The model is improving so quickly, very often we just saturate the eval, and then we have to throw it away, and we…”

“I, I think that nowadays, with modern models, there is so much product overhang that I, I'm not seeing startups capture. And I think there's people thinking about these problems, but there's just a huge amount, amount of opportunity to elicit these behaviors from the model that are just, like, amazing and interesting…”

“I, I think the skill nowadays is less about prompt engineering and more about figuring out how do you give Cloud a hard task that seems a little bit too hard? And then how do you make it possible for Cloud to verify its work along the way? And the verification, I think, is probably the single most important thing that…”

“when I look at engineers that have been, you know, coding for a long, for a long time, you know, like for, for years or for decades, this is a really, really common failure mode is trying to over specify and it's trying to be overly specific and then, you know, get the model to do the, to do the task exactly the way…”

“The big idea of the company that was spot on is that it is possible to augment the CPU to solve problems that otherwise are too difficult to solve. And, and, uh, molecular dynamics is one of them. Image processing is one of them. Inverse physics is another one. And so all kinds of different algorithms, of course, deep…”

“the premise for the company, which was, uh, let's bring Dev and Ops together into one platform, uh, was actually completely central to the explosion of the cloud. Um, and we also completely underestimated the explosion of the cloud with, oh, that's interesting. There's this Amazon thing. Um, but, you know, when you…”

“Another thing I think I, I encourage everyone to do as soon as you start scaling a little bit in your company is to keep in touch with what's actually happening on the ground. And by that I mean the actual support requests, the actual, uh, sales transcripts from the sales conversations, the actual customer feedback,…”

“The thing I got wrong with Tribe was trust. So, so first of all, the thing that I, I, I built a social network before Facebook and I managed to fail, and I just got the trust component wrong, that for people to put themselves on the web, they first had to feel this good container of trust, and that's the thing that…”

“There's a funny thing about, I call it true signal, but, or heat, you know, about the right product, the lightning in the bottle moment. When you have it, you know it. And when you don't, you don't know. What I mean is, you don't know it's not. You know positively when you achieve it. Yeah. Like, from everything. And…”

“great product makers, you know, they're collecting winnings, not making bets. And so, you know, long before you launch that it's a hit or that this, that your users are going to love this. You don't look to see if they like it.”

“you can't create a hundred websites and then expect Google to think that, Hey, you're an authoritative source and just creating content for content's sake is not the goal.”

“every single good AI product you've used is an agent loop with tools. That's it. Like there's no, you try to sort of over-engineer the harness and then do certain things, but at the end of the day, it's skills tools and a model. Like there's not really much else.”

“I always tell people, like, I think if you don't, if you can't minimize your surface area and, and solve the problem with a very clear set of boundaries, you haven't found the right problem to solve.”

“having a sense on ROI on tokens is important, but I think it misses the point that you're standing in the timeline of history, and it's six months after electricity was invented.”

“I think you have to assume that if you're a big, large company that's not AI native, you're doing a turnaround, uh, to some degree.”

“you should separate the agent that, and, and the systems that are actually emitting code from the system that is talking to customers and the system that is reasoning about the conversations with customers and translating into product roadmap. Basically three separate things.”

“when, when all of that context is in one place, with a little bit of additional Uh, information about how the schema is laid out. An agent can go and ask any, or answer arbitrary questions about, about our business.”

“these things are really good at working with MCP and CLI. Like they're a little even better with CLI.”

“that's why you should start a startup, because people are going to be trapped in organizations with people running organizations that are very powerful and have all these resources and all this capital that do not believe what we just said.”

“There are actually choices we can make as founders, and especially choices we make early that can change the trajectory of the company so that, as you say, its longevity could be measured in decades and centuries, not quarters.”

“everyone who's ever studied this and looked at it seriously understands that shareholder value is like the exhaust that comes out of the engine. When you take the exhaust pipe and put it in the intake and make that your explicit goal, now you don't stand for anything anymore. Now, now product quality suffers.”

“Dual class is defeated all the time. Uh, I give a bunch of examples in the book, like, you know, the stock price drops. Everyone panics. Founders forced out anyway, because at the end of the day, like having the votes is not the only thing that matters. There's a bunch of things that matter, including like investors…”

“Should I go there? And the answer is the same now as it always is. Yes, you should. You can go there for a bit and then come back, but you should at least go.”

“Silicon Valley is the place where people go from being nobodies to billionaires faster than anywhere else in the world, and so someone who had a taste for being really nice to nobodies, you know, the kind of person who sees the nobody walking along on the street and just wants to pet it on the head as opposed to, like,…”

“unless you have your own prompts and you can write it for yourself, like you are You know, below the API line for some PM or developer that is not you who like will not understand you will not understand your needs will not understand what you uniquely care about.”

“I think machine learning tends to have a long history of people starting with bio-plausible arguments, and then realizing that there's some variant of them that seems highly bio-implausible that actually works better.”

“If you chain of thought it on all the bubble sort input and output, it will only do bubble sort. In fact, it won't even do bubble That well.”

“continual learning, Long-term reasoning. Uh, some aspects of memory. These are still unsolved. Um, and how to get the systems to be more consistent across the board. Um, I think all of these are going to be required for AGI.”

“really you can think of a lot of The things we're doing today, all the leading models with thinking modes and chain of thought reasoning as aspects of what was sort of pioneered with AlphaGo coming back now.”

“I think that's one of the not having continual learning currently is one of the things holding back agents from doing full, uh, tasks. You know, I think they're really useful for aspects of tasks right now, and you can patch them together and do some really cool things, but they don't adapt well with the context that…”

“the lesson I've learned from all the alpha projects we've done, specifically alpha go and alpha fold is, um, I think the techniques we have and the problems I look like to look for are great in, if this, if the situation can be described as massive combinatorial search space, the more massive, the better in some ways.…”

“the way we would test that is, you know, sometimes call it my Einstein test, which is, you know, can you train a system With the knowledge of cutoff of 19 oh one. And then will it come up with the, you know, what Einstein did in 19 oh five, including special relativity, you know, his Annus Mirabilis. Can it, can it do…”

“The problem with SaaS is you onboard a new SaaS tool that creates a new silo of data that you can't really program. So now they're taking matters in their own hands and building things like, you know, quote configurators and, um, and being able to save the company hundreds of thousands of dollars, if not millions of…”

“Once you solve parallel agents, you've also solved teamwork, because every time someone jumps into the session, you can start an entirely new, you can fork entirely new VM for them, and they can work in parallel to you.”

“The reason Excel sheet agents got better is because coding got better. The reason, like, you know, some commerce agents are starting to work is because they can call APIs and they can, so coding turned out to be a bit of a hack or workaround”

“So you don't need to accept slop and you shouldn't accept slop. You should, you know, hunt for, fight for the right solution.”

“I wouldn't call it an MVP. It's probably like an MVQP, like a minimum viable quality product. Because again, you don't want to lose trust by experimenting with something out in the world.”

“And what this two work really show is that if you start from a vision language model that is really powerful, and you kind of use robotic data to adapt this model to speak robot language, if you will, um, then you see a lot of transfer from the kind of knowledge that exists in the language, in the vision language…”

“And cross embodiment, there is the data collection aspect of, as well, which is to really make sure that your model and your organizations and infrastructure are set up to consume data From many different sources of, uh, of robots, and that actually allows you to scale easier. For example, if I were to contrast our…”

“And the argument is that, you know, single robot is simpler to scale. And actually that's not how it plays out in practice. Like how it plays out in practice is even if you have a single robot that you're optimizing for, over time that Platform is going to drift. You know, maybe you want to make hardware change or you…”

“The interesting thing about the approach is that you're converting it from a very difficult engineering problem into a operation problems of how do I identify the use case, and how do I collect the right data, which is, in some sense, more scalable, because you can build a system that allows you to collect data for…”

“one of the inset that we have here is that you can actually bury the inference time Within the robot control loop because, you know, if I'm a robot, I have enough action for me to execute for the next hundred milliseconds. Like, there's no reason for me to wait until I finish executing that action to ask my model for a…”

“Evaluation is a really hard problem in robotics because it scales super linearly to model capability. Like, let's say you have a model that can perform a two-minute task. Running evaluation for that is very different from running evaluation for a task that's 20 minutes. It's not 10 times harder. It's, it's, it's more…”

“We have found that having that iterative cycle within one scientist actually accelerates the work that they do by an order of magnitude.”

“You know, the, the minimum description length principle that the model of the data that is most likely to generalize Is the shortest. And I think you cannot find a model like this. If you're doing parametric learning, you need to, you need to try symbolic learning.”

“When it comes to, uh, Competency. There's always a trade-off between intelligence and knowledge. If you have more knowledge, if you have better training, you need less intelligence to be competent.”

“Science is not about curve fitting. Science is about finding the equation, finding the most compressive symbolic model of your pile of observation.”

“the more, you know, the more expertise you have, but things like programming, for instance, the better you're able to Use and leverage these tools for your own benefit. And with the right kind of expertise, uh, all this AI progress is actually empowerment. Like, it's something that you can leverage for yourself.”

“our key insight was that to automate all of software engineering, you will have to build a platform that replicates what, what best engineering team do, like code reviews, automated testing, debugging, deployment, security, hosting.”

“our view is that I think, uh, the, the coding aspect is only 20% of the job, right? I think like taking an app to production is like really, really hard.”

“Well, I mean, the brain is a computer, and it's gonna, saying that it's gonna get me yelled at by some corner of, of the field, but I think, like, I think that you can take that, like, almost literally. It's a, it's a very different architecture than, like, a, like a von Neumann architecture electrical computer, but it…”

“But I think one of the interesting things that's come out of progress in artificial intelligence is we're seeing this big unification in neuroscience and, and AI. I think we're actually learning a lot from AI research, more than I think we thought we would learn from AI research. I mean, I can tell you 10 years ago we…”

“Like BCI couldn't have done this on its own, but Apple and Samsung and others have poured epic amounts of money onto making these types of electronics exist in the world so that people like us can use them.”

“Humanity has totally dominated the planet, And we keep our closest living relatives in glass boxes so they don't go extinct. And so there's a real history here of, of greater intelligence being very dangerous.”

“effectively you could say, like, each model or each set of models that we're working with will have their own S-curve. The poetic system, the poetic meta system itself, is also going to have its own S-curve. And so as the poetic meta system gets better, and as the underlying models get better, you, you'll Find that…”

“That will get you some performance improvements, but it's very far from, uh, everything that you can get. Uh, if you actually think about these reasoning strategies that are really going to be written in code rather than in, in just better prompts.”

“At Anthropic, the way that we thought about it is we don't build for the model of today. We build for the model six months from now.”