YouSaid · the spoken record

Garry Tan

lines on the record: 166
first: 2025-04-29
most recent: 2025-04-29
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“That's right. Yeah, and that's sort of the quandary, right? You have to choose. Do you want the outcomes to be equal or do you want a rising tide to raise all boats?”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I think what are the alternatives, right? Like I think of all the people who, you know, they're locked away in companies, they're locked away in academia, or heck like, you know, these days the wild thing about intelligence is intelligence is on tap now, right? All of the impediments to being able to, all of the impediments to fully realizing what you want to do in the world are starting to fall away. There's always going to be something that stands in the way of any given person. And I'm not saying those things are equal, but through technology and through access to technology, those things are coming down. Like if there's the will, if there's the agency, if there's the taste, like... That's what I want for society. I want them to achieve that.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, totally. I mean, return 600 million 650 million dollars on that investment alone. I was sort of right at the pinnacle of my investing as if you're running my own VC firm. And Paul and Jessica came to me and said, Gary, we need you to come back and run YC. And it was really, really hard to walk away from that. Luckily, I had very great partners, Brett Gibson, my partner, my multi-time co-founder went through YC with me. He actually built a bunch of the software with me at YC before we left. He runs it now. They're off to the races and still doing great work. And I sat down with Paul right after we shook hands. And he's like, Gary, do you understand what this means? It means that if we do this right, we kind of like, I think”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I think looking back, I mean, growing up, I always just looked up to the people who made the things that I loved. And Steve Jobs, Bill Gates, like the people who really created something from nothing. And I just think of Steve saying, you know, we want to put a dent in the universe. And ultimately, that's what I want. Like that's success to me is how do we bring forward this is actually when Paul Graham came to recruit me to come back to YC. I had actually left and started my own VC firm, you know, got to $3 billion under management. Yeah, you did get a bit of”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Like just after the first pitch of the first inning of what is about to be like a great, great time for humanity, for technology. I'm with you. I'm so excited.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“All that being said, it's like we're in the middle of the revolution. This is totally, I mean, it's not even the middle. I think it's like”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I couldn't think of another way to start that conference than have Sam Otman and a bunch of his old, we had Bob McGrew there, we had Evan Morakawa, who was the engine manager who released ChatGPT. Bob McGrew actually worked with me at Palantir back in the day, but he's outgoing chief research officer, Jason Kwan was there. He actually worked at YC Legal before leaving to run a lot of things at OpenAI. And so I had them all stand up. And we had a room full of 290 founders, all of whom were working on things that happened essentially because OpenAI existed. And there was like a standing ovation”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“That happened, that was going to happen, right? Like, I think there's a lot of the Sam Altman haters or the open AI haters out there love to point out like, oh, you know what? The transformer was made by all these teams. Someone was like, these teams absolutely did incredible things. Like you can't take away from that, right? The researchers, Demis did incredible things. But at the same time, it's like they believed a thing that nobody else believed and they brought the resources to bear. And so recently, you know, Sam Altman came back to speak at our AI conference this past weekend. And we, you know,”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“No, I agree. If I had to I agree. I mean, especially to have, you know, I pinch myself every day on the regular, like in the morning. I wake up and it's like, oh, this AI thing is happening. And then somehow I'm filling the shoes of the person who, like, I mean, Sam Altman probably brought forward the future by, you know, five years, ten years, at least 10 years. All of the things that, you know, him and Greg Brockman and all the researches he brought on like were working on.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Together, right Yeah, I mean, I feel like we're in the idea fire hose. So we're just like hearing about all kinds of things that are very promising. And then I think the most unusual thing that I'm still getting used to is. I mean, in full transparency, I mean, probably the median YC startup still fails, right? YC might be one of the most successful institutions of its sort that has ever existed, inclusive of venture capital firms on the one hand. On the other hand, the failure rate is absolutely insane. It is still a very small percentage of the teams actually do go on and create these companies worth $50 or $100 billion. But the remarkable thing is not that it's that low. The remarkable thing is that it happens at all. Like, it's just unbelievable that.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“you Is a little bit less what you might expect. It's like, oh, this is someone with a firm handshake who looks like a certain way and bends the will of the people. Like, you might think of an SBF that's like, that's all artifice. Like, think about that guy. Like, that guy was like full of shakes. And the guy was like on meth, right? Like the guy was, you know, everything about it was an affectation. Like he was a caricature of an autist. Like we see very autistic, incredibly smart engineers all the time. But, you know, for him, it was like, that was part of the act. Like I remember he did a YouTube video with Nasdalia.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“And what I did was I did 30 minute, 45 minute office hours with companies in the YC Winter 11 batch sitting back then as an interaction designer. I used Omnigraffel a lot. And so we just sat there and designed their homepage. And it's like, this is what the call to action should say. Here's, you know, put the logo here. Here's the tagline. Here's the, maybe you have a video here or right below you have how it works. And then what's funny about it is like some people would take the designs we did in those 30, 45 minute things. And like, that would be their whole startup. And I sell those companies for hundreds of millions of dollars years later, which is just fascinating to think about. It's like clear communication, great design, you know, creating experiences for other people. All of those are sort of exercising the same skill. And so that's what a founder really is. It's like, you know, a founder to me.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Oh, yes, I get it. Like, I know what it is, and I know why that's important. I know why I should spend more time with you. That's what a great two-sentence pitch is. And knowing what it is is very hard. That's all of Paul Graham's sort of editing down and whittling down in a nutshell. It's like people do really complex things. How do you say what you do in one sentence? That's very hard, actually. And then the second sentence is like, why is it important? Why is it interesting? Why should I, you know, and then that may well change with like the person that you're talking to? So yeah, to the degree that clear communication is clear thinking, one of the things I did when I first joined YC, I had no intention of ever becoming an investor, ever being a partner, let alone running the place. Like I was just a designer in residence.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Like you're helping people with their pitch that sounds so basic. Yeah, I guess that makes sense. That's what an incubator would do. But the reason why it's very important is that it's actually almost like a mantra. It's like a special incantation. Like you believe something that nobody else believes. And you need to be able to breathe that belief into other people. And you need to do it in as few words as possible. So if you, the joke is like, oh, yeah, like what's your elevator pitch? But like you might run into someone who could be your CTO who could introduce you to your lead investor, who could be your very best customer, and you will literally only have that time. You will only have time to get two sentences in. And so, and even then, I mean, I guess it's kind of fractal. Like that's what I love about a really great interview. Someone comes in and I'm like,”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“A very complex concept, and then trying to say it in as few words as possible. And, you know, the amount of time that Paul spends on his essays is fascinating. Sometimes days, like sometimes weeks, like he'll just iterate and iterate and send it out to people for comment. the amount of time he spends whittling down the words and trying to like combine concepts and say the most with the least number of words it would shock you and then also that is actually thinking like writing is thinking like one of the more surprising things that we do a lot of at YC is we help people spend time thinking about their two sentence pitch so you know you would think that that's oh yeah that's like something you know startup”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Whereas, like, actually, I shouldn't be worried about that. What I should be worried about is, and I spend a lot of time with our YouTube team and our media team at YC talking about this. It's like, if we get too focused on just view count, we're liable to just, yeah, like optimize for the wrong audience. If we're not being authentic to ourselves or if we're just trying to like follow trends or do things that get clicks, it's like, that's not helpful to them either. Like then we're just on this treadmill, right? Yeah, basically trying to be very, very high signal to noise ratio, you know, the thing that I probably struggle with most, and I don't know, maybe some of the listeners here might feel this. It's like sometimes I think out loud and then, you know, really, really great ideas are not like thinking out loud. They're actually figuring out.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I think the number one thing that is very hard but is so, I mean, you can see it and read it in his essays. It's to be plain spoken and to sort of Be hyper aware of artifice, of kind of like bullshit, basically. Like, don't let bullshit. I think it creeps in here and there. I'm like, oh, yeah, I I sometimes am in danger of caring too much about the number of followers I have and things like that.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Gary just told you you got to click like subscribe and hit the bell icon because You want knowledge. You want to be smart, and this is the place to get it.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Well, you do if you are regular, and then the other hack is be very shameless about asking for subs. And then the funniest thing is like subs do very little, actually. There's no guarantee that you show up in people's feeds if someone subs. It helps a little bit. Liking helps more. Watch time helps the most. And then the extreme over-the-top hack that probably you should do here is you should ask for the like, subscribe, and hit the bell icon. Because if you hit the bell icon and they have notifications on, that's the only thing that is almost as good as having their email address and emailing them.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, but once I stopped posting so regularly, then it sort of didn't matter as much anymore. But if you're going to post very regularly, that's pretty important, actually. So yeah, unfortunately, it's clickbait. And then there is an interesting interaction. Yes, you can optimize for better thumbnails and better titles for the click-through, but if it has absolutely nothing to do with the actual body, as you mentioned. You will not get watch time.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“And then you want it to be relatively recognizable. Like, you know, you want some sort of style that when someone sees it, I mean, basically what I was doing at the time was just taking whatever frame that was sort of kind of representative and throwing it in there. But when you train someone to look at YouTube back to back to back every time it shows up, like you sort of want to be highly recognizable.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I think it's clickbait. Unfortunately, and this is the thing, when you're trying to make smart content, it's actually kind of tricky because you don't want necessarily more clicks. You want more clicks from people who are smart.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Oh, yeah, he's great. I mean, he was very brusque with me. He said, you know, look, man, your titles suck and your thumbnails are even worse. And I think that he spent so much time trying to understand the YouTube algorithm and what people want that he just loaded it completely into his brain. What makes it good?”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I mean, I think meta is sort of the dark horse because they are amassing talent and then they have crazy distribution. And I think, you know, I just would never count suck out. I think that it's so crazy that it's super smart that he is on that, he's always thinking about what is the next version of computing, like so much so that he'd probably put more money than he should have into AR. And that was maybe premature. He might still end up being right there. But AI for a fraction of what he's put into AR is likely to push forward all of humanity and accelerate technological progress in a really profound way.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I mean, I think Microsoft has just really, really deep access to open AI, and I think OpenAI is probably, you said public companies. Yeah, yeah. So, you know, I think there's a non-zero pretty large percentage of the market cap of Microsoft that I think is pretty predicated on Sam Altman and the team at OpenAI continuing to be successful.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Engineering, like in theory between robots and more and more intelligence, we could sort of figure these things out, but not if we run out of energy.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Like that alone will, I mean, if anything, that's probably the bigger problem right now. We know that the models will continue to get better. We know that the demand for intelligence will be unending. And then even going back to the robotics question, it's like if we end up making universal basic robotics, the limit will still actually be sort of the climate crisis and the available energy available to human beings, right? And maybe solar can do it, but maybe there are lots of other sort of solves. But I think energy and access to energy is sort of the defining question at that point. Like everything else you could solve, like and everything else you could sort of either, you know, if it's in the realm of science and”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Probably. I mean, NVIDIA just has an out and out for now. They're just so far ahead of everyone else. I mean, it can't last forever. But I think that the demand for building the infrastructure for intelligence in society is going to be absolutely massive. And maybe on the order of the Manhattan Project and we just haven't really thought about it enough, right? Like it's entirely conceivable. say like level four innovators turns out to work like It's sort of the meta project because then it's like the Manhattan Project of instantiating more Manhattan projects, actually. You could imagine if we can, if more test time compute or you could do the work of, you know, 10,200 IQ Einstein's working on bringing us basically unlimited clean energy.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Okay, I guess is it cheating to say I put even more money into the YC funds that I already run, but that's a cup out. That's a cup-out. That goes without saying. I think that it's very unusual just because we end up, like this is the commercialization arm of every AI lab is what I realize. But short of that, I mean...”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“And it's the second best because the first best will be a really great UI made by a really good product designer, who's a great engineer, who's a prompt engineer, who actually creates software that doesn't require copy-paste. It's just like link this, link that. Okay, now this thing is now working. And so I think that those are the moats are not different actually at the end of the day. It's still, you still have to build good software or you still have to be able to sell. You have to retain customers. You have to, but you just don't need like a thousand people for it anymore. You might only need six people.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Just still apply. One thing that a lot of our founders who are getting the five to $10 million a year in revenue with five people in a single year are saying is, yes, there's prompting, there's evals, like there's a lot of magic that is sort of mind-blowing. But what doesn't go away is building a good user experience, building something that a human being who does that for a job sees that knows that's for me, understands how to start, knows what to click on, how to. Software in the world for everything is using ChatGPT because you can basically copy and paste almost any workflow or any data and it's like the general purpose thing that you can just drop data into it.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“For free. Yeah, I mean, I think ultimately the model itself is not the moat. I think that the evals themselves are the moat. I don't have the answer yet. Basically, for now, maybe it's a toss-up. If you're a very, very good prompt engineer, you will have far better golden evals and the outcomes will be much better than what 03 or deep-seek R1 can do because it's specific to your data and it's much more in the details. I think that that remains to be seen. Like the classic thing that Sam Altman has told YC companies and told most startups, period, is you should count on the models getting better. If that's true, then that might be a durable mode for this year, but it might not be past, I mean, 03, we haven't even seen yet the results seem like fairly magical. So it's possible that that advantage goes away even as soon as this year. But all the other advantages”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“And then he could actually see, I think he said pages and pages of the model talking to itself. Literally, does this make sense? Like, can I break it down into steps? So what we just described as a totally manual action that a really good prompt engineer CEO like Jake Keller did, and he sold his company Case Text for almost half a billion dollars to Thomson Reuters. That is actually very similar to what the model is capable of doing on its own in a reasoning model and that's what it's doing when it's taking doing like test time compute. It's actually just spending more time thinking before it spits out the final answer.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“With GPT-4 when it first came out and what 01 and 01 Pro maybe is doing and what DeepSeek R1 is doing clearly because it's open source and you can see it is that those steps like breaking it down into steps and the sort of metacognition of whether or not it makes sense at all of those micro steps that's what in theory this reasoning is actually happening that's actually happening in the background for O1 and O3 and if you use ChatGPT you'll see the steps but it's like a summary of it and so it's you know I just only saw it this morning I mean this is such new stuff like I was hoping that someone would do a open source reasoning model just so we could see it and that's what it was I think Simon's blog post this morning showed here's a prompt”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Oh, I love making stories with them. So, you know, what I find is 01 Pro is actually extra good now. So yeah, actually there's like an interesting thing that's happening right now. And I saw it up close and personal this morning looking at some blog posts about DeepSeek R1, which is deep seek reasoning model. I was reading Simon Willison's blog post about he got deep-seek R1 running. It's the first, one of the first open source versions of sort of the reasoning. And so what we just described with how Jake Heller broke it down into chain of thoughts to make case text work, it turns out that that maps to basically how the reasoning stuff works. And so the difference between what Jake”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Basically, do the work of hundreds of lawyers and paralegals, and it would take a day or an afternoon instead of three months of discovery.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Oh, yeah, it's really straightforward. It's just a test case, right? So given this prompt and this data evaluate the prompt to see if, and it usually maps directly to something that is true, false, yes, no, like something that is pretty clear. Let's say there's a deposition and someone makes a certain statement, right? You might have a prompt that is like, you know, is this what this person said in conflict with any of the other witnesses or I don't know? I'm totally making this example of, but like this is the kind of thing that you can do at a very granular level. You might have thousands of these. And then that's how Jake Keller figured out he could create something that would”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Across the whole industry right now. They literally are taking model out what a human would do in knowledge work and then break it down into steps and then have evaluations for each of those prompts. And then as the models get better, because you have what we call the golden evals, basically you just run the golden evals against the newest model. 4.0 comes out, Clot 3.5 comes out, deep seat comes out. You have evalves, which is basically a test set of prompt context window data and output. And you can actually, you know, what's funny is it's even fuzzy that way. Like you can even use LLMs in the evals themselves to score them and figure out, does it make sense?”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“This type of summarization, he would have to go through and read all the materials. And then this is why apparently lawyers have sort of their many, many different colored little flags and highlighters and things like that. They just get very good at doing a read-through paragraph by paragraph, sentence by sentence, and pulling out the things that are relevant and then sort of synthesizing it. And so, you know, early versions of case text and a lot of it today, I think, is still just doing that. It's like, what is a specific thing that a human does, break it down into the very specific steps that a real human would do, and then actually basically if it breaks, you're just asking in that step to do too many things. So like break it down into even smaller steps. And somehow that worked. And like basically, this is the blueprint that I think a lot of YCE companies and AI vertical SaaS startups are doing.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“And what he needed to do, I mean, I sort of equivalent to Taylor Time and Motion Studies in factories. It feels like that's what he did for what a lawyer does. Let's say you have to put together a chronology of what happened in a case and what a real, he's a real life lawyer, which is sort of unusually perfect to figure out this prompting step. He realized that he needed to look”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“That it would actually work. And what he says he figured out was if GPT force started hallucinating for them, they realized that they were doing too much work in one prompt. They needed to take that thing that they asked GPT-4 to do and then break it up into smaller steps. And then they found that they could get deterministic output for GPT-4 like a human if they broke it down into steps.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I mean, I'm mainly a user. I spend a lot of time with people who spend a lot of time in prompts, probably the person I would most point people to is Jake Keller. So he's the founder of Case Text. He was one of the first people to get access to GPT-4. And we think of him at YC as the first man on the moon in that he was the first to successfully commercialize GPT-4 in the legal space. So what he said was that they had access to GPT 3.5 and it basically hallucinated too much to be used for actual legal work. Like lawyers would see one wrong thing and say like, oh, I can't trust this. GPT-4 he found actually with good evals would actually give, they could program the system in a way.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“So, I think in a lot of domains, and this is sort of the year where maybe there's like 5 or 10% of things that it's sort of hitting the Turing test and really satisfying that. But I think maybe this is a year where it goes from like 10 to 30% and the year after that it doubles again. And the next few years are like actually the golden age of building AI.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Like AGI, the thing is in a lot of domains, it feels like it's here Actually, I mean, can it have a conversation with someone and give incredibly good wine pairing recommendations and have a perfectly fine indistinguishable from a real human sort of or even better than human sort of interaction? And also take orders for very expensive wine and have that just work. Yes, like that's happening right now.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I guess the funniest thing is Microsoft, I think, is defining it when it gets its $100 billion back. But I am sort of skeptical of that because, you know, I think Basically, only Elon Musk then would qualify as a human general intelligence, I think.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Yes, I mean, I funded a company in this very current batch that is called Leaping AI. They are working with some of the biggest wine merchants in Germany, which is fascinating. So, I mean, that's another fascinating thing. Like these things speak all human, you know, they certainly speak all the top languages very, very well and are indistinguishable. And I think 80% of the ordering volume for some of their customers is entirely no human in the loop.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Like it's the speech to text and text to speech, those things are indistinguishable from human beings now. And you can train these things. The evals are good. The prompting is good. Going back to what we were saying earlier, like what we're seeing is like it or not is actually replacing labor.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Well, the demand for encouragement. Basically, I think you can live in two different worlds. One world says, like, SaaS apps like Salesforce per seat per year, that kind of thing. Or we're selling classically Palantir was selling million dollar or $10 million ACV very specific vertical apps. And so all of those things are selling software or hardware. And that's like selling technology. And so increasingly what we're starting to see is like especially the bleeding edge is probably customer support and all of the things that you would use for a call center. Like those are sort of the things that are already so well defined and specific. And there's a whole training process for people in usually overseas to do these jobs. And AI now is just coming. In”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“Well, we just don't have enough GPUs, really. I think what's funny is I'm still pretty bull on NVIDIA and that they more or less have the monopoly on the best price performance.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source
“I mean, the hard part is I spend most of my time not with those people. I spend most of my time with people who are commercializing it. So, the very, very smartest people are clearly the people who are in the AI labs actually actively sort of creating these models. But the people who I know who are in those rooms, I mean, sounds like test time compute is really it. That's the reasoning models are sort of the thing that will really come to bear this year. Like we're sort of understanding that right now. For now, it sounds like pre-training might have hit some sort of scaling limit of which I don't understand yet. There's a lot of debate about it. Will there be new 4O style models that have more data or more compute? And seemingly, there's just rumors of training runs gone awry that basically the scaling laws may have petered out, but I don't.”
2025-04-29 · The Knowledge Project with Shane Parrish · Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders · IDENTIFIED FROM THE TRANSCRIPT · source