WEBVTT 00:00:02.880 --> 00:00:06.637 Tell me the history of Muse. The first thing to start with is probably like 00:00:06.720 --> 00:00:08.957 what our vision was when we started Meta 00:00:09.040 --> 00:00:13.917 Super Intelligence Labs. And when we started it, Mark published this memo 00:00:14.000 --> 00:00:16.797 that myself and Natt Friedman and a bunch of others like worked together 00:00:16.880 --> 00:00:18.557 with him closely on called Personal 00:00:18.640 --> 00:00:20.317 Super Intelligence. And the whole idea 00:00:20.400 --> 00:00:24.237 was like how do we make 00:00:24.320 --> 00:00:29.277 AI how do we build powerful AI that actually makes all of our lives better? 00:00:29.360 --> 00:00:32.957 like how do we build something that actually like lifts the human experience 00:00:33.040 --> 00:00:36.637 broadly speaking. You know, in that memo, we we spoke to a bunch of the 00:00:36.720 --> 00:00:39.117 ideas and we basically alluded to I 00:00:39.200 --> 00:00:41.597 think a lot of what muses ultimately 00:00:41.680 --> 00:00:44.157 become, but we were gesturing at it like 00:00:44.240 --> 00:00:46.957 very very early like this was in you 00:00:47.040 --> 00:00:50.797 know June 2025 um before really agents 00:00:50.880 --> 00:00:55.277 had happened in any meaningful way you know in the summer of 2025 like AI was 00:00:55.360 --> 00:00:58.717 just you know in a totally different spot. But that's where what we wanted to 00:00:58.800 --> 00:01:02.797 build towards. That was the mission of MSL and that was really the where we 00:01:02.880 --> 00:01:07.037 wanted to take everything. Then we kind of just got to work building models 00:01:07.120 --> 00:01:11.037 because you know the endto-end process of developing any of these frontier 00:01:11.120 --> 00:01:15.597 models is many many many months. Um like you have to go through you have to build 00:01:15.680 --> 00:01:19.517 your entire stack you have to produce the data sets you have to you know 00:01:19.600 --> 00:01:22.077 pre-train the model then you have to do pro training the model you have to build 00:01:22.160 --> 00:01:26.077 a whole stack for that. like is an it's a very lengthy end to- end process. Um 00:01:26.160 --> 00:01:28.157 and kind of this whole um production 00:01:28.240 --> 00:01:31.997 line at the start of this year um you 00:01:32.080 --> 00:01:35.437 know Opus 4.5 obviously like really 00:01:35.520 --> 00:01:39.197 caused many people including including us to sort of like see what the future 00:01:39.280 --> 00:01:41.917 of agents could be or was and really 00:01:42.000 --> 00:01:44.637 brought the sort of like potential of 00:01:44.720 --> 00:01:48.157 agents to the foreground and openclaw 00:01:48.240 --> 00:01:50.157 was happening at the start of the year. 00:01:50.240 --> 00:01:52.237 Natt Freriedman, who you know I work 00:01:52.320 --> 00:01:54.957 closely with, was I think the first 00:01:55.040 --> 00:01:59.037 person within MSL to uh work with 00:01:59.120 --> 00:02:02.397 OpenClaw and try OpenClaw. He had an 00:02:02.480 --> 00:02:04.957 experience I would I would describe as 00:02:05.040 --> 00:02:07.437 somewhere between terrifying and 00:02:07.520 --> 00:02:11.517 euphoric. Like I think he he really like 00:02:11.600 --> 00:02:15.597 I think went all in. He trusted entirely. you know, he's told some of 00:02:15.680 --> 00:02:19.917 these stories that uh he had like a a Stripe Sessions interview where he 00:02:20.000 --> 00:02:23.677 talked about some of this where, you know, he told his open claw that he 00:02:23.760 --> 00:02:28.557 wanted to drink more water and then, you know, his openclaw would like watch him 00:02:28.640 --> 00:02:31.677 in security footage to make sure that he was drinking water and tell him good job 00:02:31.760 --> 00:02:35.997 and he had like many many crazy stories like this and it also makes you realize 00:02:36.080 --> 00:02:40.237 like, oh, there's something quite quite magical about what's happening here. And 00:02:40.320 --> 00:02:43.437 I remember one of the things that that Nat told me and this is one of the 00:02:43.520 --> 00:02:47.917 things I remember the most is he said I think there really is something here and 00:02:48.000 --> 00:02:52.797 I don't know like early adopter Nat Freedman is like a pretty good indicator 00:02:52.880 --> 00:02:56.077 cuz he was one of the he was one of the first customers of Stripe back in the 00:02:56.160 --> 00:02:59.997 day and then was really ear early on a lot of open source developer stuff and 00:03:00.080 --> 00:03:04.397 he also you know helped create GitHub copilot and many other things. So that 00:03:04.480 --> 00:03:07.437 was like a clear signal. It's like, okay, there's clearly something pretty 00:03:07.520 --> 00:03:12.077 magical about personal agents generally speaking. But you know, when you 00:03:12.160 --> 00:03:15.997 actually kind of trust the personal agent and and really bring it into your 00:03:16.080 --> 00:03:20.397 life, it is the ability to do so many things. And after hearing all of these 00:03:20.480 --> 00:03:24.477 crazy stories, I start using it very deeply. You know, I think for many weeks 00:03:24.560 --> 00:03:28.157 I think both of us were sort of like every spare moment we would just spend 00:03:28.240 --> 00:03:32.317 talking to our open claws and like you know, you know, really immersing 00:03:32.400 --> 00:03:34.717 ourselves in like what personal agents could be. 00:03:34.800 --> 00:03:38.237 >> What was your initial reaction? Euphoria or terror? 00:03:38.320 --> 00:03:41.197 >> Nat at the time he had this um he had 00:03:41.280 --> 00:03:43.997 built this prompt to do this like 00:03:44.080 --> 00:03:46.717 basically like full psychoanalysis of 00:03:46.800 --> 00:03:51.677 yourself. There were like multiple waves of like deep psychological probing. He 00:03:51.760 --> 00:03:55.117 basically wrote this prompt for you to to use an open claw. And I think that 00:03:55.200 --> 00:03:58.317 was like one of the very first things I did with with my own openclaw. It was 00:03:58.400 --> 00:04:02.877 maybe the equivalent of like 3 years of therapy all, you know, compressed into 00:04:02.960 --> 00:04:06.317 this one moment. >> Wait, I'm curious why when you get your 00:04:06.400 --> 00:04:09.677 first shot at open call, this is one of the first things you do. 00:04:09.760 --> 00:04:13.197 >> You know, I downloaded it. I installed it. And then Nat was like, "Oh, you 00:04:13.280 --> 00:04:15.917 should do this." I was like I was like next to him. He was like, "Okay, I'll 00:04:16.000 --> 00:04:19.597 just do the thing you're telling me to do." And then >> okay um 00:04:19.680 --> 00:04:21.677 >> and then you know this crazy experience. 00:04:21.760 --> 00:04:23.837 >> Wait say more about the crazy experience though. 00:04:23.920 --> 00:04:28.077 >> Oh well it was just like very intense. I mean it was like again it was like 3 00:04:28.160 --> 00:04:32.557 years of therapy because it was just like successive rounds of like deep uh 00:04:32.640 --> 00:04:36.397 psychological probing cuz it had other information about me. It had you know 00:04:36.480 --> 00:04:39.917 access to my email and access to a bunch of my photos and like it could like go 00:04:40.000 --> 00:04:42.637 and look up other information and sort of pull it all together. I was very 00:04:42.720 --> 00:04:44.717 vulnerable through that process. This 00:04:44.800 --> 00:04:47.917 was in February. uh both of us I think 00:04:48.000 --> 00:04:49.917 had come to this like joint conclusion 00:04:50.000 --> 00:04:54.077 that this is this is clearly like you 00:04:54.160 --> 00:04:56.077 know there's clearly something really 00:04:56.160 --> 00:04:58.877 special happening here and we both wrote 00:04:58.960 --> 00:05:03.197 memos at that time I wrote a memo about and both of these we you know kind of 00:05:03.280 --> 00:05:06.717 like sent up through the meta board um cuz you know for whatever reason it's 00:05:06.800 --> 00:05:09.917 board season I wrote a memo about how I 00:05:10.000 --> 00:05:12.957 felt that in many ways this was like the 00:05:13.040 --> 00:05:15.757 final consumer product and that the 00:05:15.840 --> 00:05:17.837 super agent was something that was like 00:05:17.920 --> 00:05:20.877 as far as consumer products went like a 00:05:20.960 --> 00:05:25.277 clear end point in many ways. And then he wrote one called he wrote actually 00:05:25.360 --> 00:05:27.277 two uh which had very memorable blames 00:05:27.360 --> 00:05:29.757 the clause the law and trust is a must 00:05:29.840 --> 00:05:34.477 which ultimately defined um a lot of the a lot of the core of how we built these 00:05:34.560 --> 00:05:36.477 products. So we came to this conclusion. 00:05:36.560 --> 00:05:39.997 This was in February. There was something very important here. Um and 00:05:40.080 --> 00:05:42.397 then we built you know our teams we 00:05:42.480 --> 00:05:45.597 built a prototype within 00:05:45.680 --> 00:05:49.117 like a a week maybe two weeks. Um and 00:05:49.200 --> 00:05:53.277 demoed it to the board like it was this like very fast cycle and many of the 00:05:53.360 --> 00:05:55.037 decisions that made its way into the 00:05:55.120 --> 00:05:57.437 final Muse product were present in the 00:05:57.520 --> 00:06:02.157 first prototype. The Jollybot was there know the early version of the Muse Charm 00:06:02.240 --> 00:06:06.397 was in that first that very first meeting. Not all of the ideas and not 00:06:06.480 --> 00:06:10.637 all the tabs were there but a lot of them were present. Then from there, you 00:06:10.720 --> 00:06:12.877 know, obviously that was February, we 00:06:12.960 --> 00:06:14.957 launched the product 7 months later was 00:06:15.040 --> 00:06:18.237 like the really really hard process of 00:06:18.320 --> 00:06:21.917 taking this, you know, prototype where 00:06:22.000 --> 00:06:26.237 it it like sometimes worked and was magical but most of the time didn't work 00:06:26.320 --> 00:06:29.517 and would constantly break and like going from there to something that would 00:06:29.600 --> 00:06:33.117 be highly reliable like you know kind of 00:06:33.200 --> 00:06:35.357 like a beautiful crafted consumer 00:06:35.440 --> 00:06:37.997 product that we felt confident as a 00:06:38.080 --> 00:06:42.877 contribution to the consumer world. And that that was like this very lengthy and 00:06:42.960 --> 00:06:45.597 quite painful process. And >> before we get there, I want to go back 00:06:45.680 --> 00:06:47.517 to the idea that you wrote the prototype. You guys created the 00:06:47.600 --> 00:06:49.997 prototype in like a one to two weeks. 00:06:50.080 --> 00:06:51.837 You show to the board. What's Mark's 00:06:51.920 --> 00:06:53.437 role in all this? 00:06:53.520 --> 00:06:57.037 >> He was like experiencing the sort of 00:06:57.120 --> 00:07:01.597 like euphoria/ terror of the product at the same time as all of us. I think he's 00:07:01.680 --> 00:07:05.037 talked about this in various interviews, but like I think he he used it for a 00:07:05.120 --> 00:07:07.037 bunch of things around the home and he 00:07:07.120 --> 00:07:08.877 was using it with his kids and he used 00:07:08.960 --> 00:07:12.397 it in his um uh in his MMA training. 00:07:12.480 --> 00:07:15.437 Like he had it watch videos of his of 00:07:15.520 --> 00:07:19.517 himself doing MMA training and like it would give him tips on like how to 00:07:19.600 --> 00:07:23.517 improve. there were like some staff meetings and by the way like you know 00:07:23.600 --> 00:07:25.357 this kind of became a thing for for much 00:07:25.440 --> 00:07:28.797 of the the meta team um was like having 00:07:28.880 --> 00:07:30.877 these having a staff meeting and saying 00:07:30.960 --> 00:07:35.837 like what were the what were the crazy experiences that you had with um with 00:07:35.920 --> 00:07:39.357 with your agent again this was back in February. It's very interesting to kind 00:07:39.440 --> 00:07:42.477 of um think back to this whole cycle 00:07:42.560 --> 00:07:46.237 because obviously the openclaw phenomena 00:07:46.320 --> 00:07:50.957 it peaked and then disappeared. Yeah, there were, you know, it didn't it 00:07:51.040 --> 00:07:54.077 didn't persist and and I think a lot of people kind of had this experience which 00:07:54.160 --> 00:07:58.077 was it was just like you'd have these magic moments but for the most part 00:07:58.160 --> 00:08:00.637 didn't like work that well and there were all these problems and then you 00:08:00.720 --> 00:08:02.717 know it just kind of like died down and 00:08:02.800 --> 00:08:08.317 so part of I think the building the product a lot of the focus of the 00:08:08.400 --> 00:08:09.597 product was just like how do you make 00:08:09.680 --> 00:08:12.477 this into an actual reliable consumer 00:08:12.560 --> 00:08:15.437 experience that you feel comfortable 00:08:15.520 --> 00:08:19.837 putting in front of over time billions of people and you can feel confident 00:08:19.920 --> 00:08:21.597 that like most of those people have an amazing experience. 00:08:21.680 --> 00:08:24.957 >> Why were you able to do that and the developers open call were not? 00:08:25.040 --> 00:08:29.517 >> Well, first of all, I think we respect Peter Seinberger and everyone who's 00:08:29.600 --> 00:08:32.877 contributed to OpenClaw immensely. Like I think Peter is definitely a visionary. 00:08:32.960 --> 00:08:37.997 I think whatever happened in his beautiful brain that created OpenClaw is 00:08:38.080 --> 00:08:39.677 magical. And I think maybe the 00:08:39.760 --> 00:08:42.797 unglamorous part of what made made Muse 00:08:42.880 --> 00:08:48.237 so magical was just really grinding out 00:08:48.320 --> 00:08:50.637 and sanding out all of the details to 00:08:50.720 --> 00:08:53.997 make the combined model plus product 00:08:54.080 --> 00:08:58.157 experience really amazing and we did 00:08:58.240 --> 00:09:00.477 many many iterations on the model 00:09:00.560 --> 00:09:03.757 itself. We aimed the overall model 00:09:03.840 --> 00:09:07.677 roadmap towards building towards personal super intelligence broadly 00:09:07.760 --> 00:09:11.437 speaking but certainly personal agents as like >> say more about that and how that 00:09:11.520 --> 00:09:13.997 differentiates between your other competitors. 00:09:14.080 --> 00:09:16.557 >> Yeah. Well, this is this is kind of >> that's actually really interesting. 00:09:16.640 --> 00:09:18.557 >> Yeah. It's actually useful to go back to 00:09:18.640 --> 00:09:22.077 the moment because if you go back to the 00:09:22.160 --> 00:09:26.877 start of the year, February, Opus and Claude were way ahead in coding, way 00:09:26.960 --> 00:09:30.717 ahead of anyone else and they were exploding. It was the beginning of the 00:09:30.800 --> 00:09:34.797 sort of like you know enthropic takeover in the whole industry and it was a very 00:09:34.880 --> 00:09:37.917 scary moment I think generally speaking which is like oh wow this is you know 00:09:38.000 --> 00:09:41.597 they're so ahead on coding and and then obviously they had mythos and you know 00:09:41.680 --> 00:09:43.677 they were they had like this clear lead. 00:09:43.760 --> 00:09:47.277 I think the the conventional wisdom in 00:09:47.360 --> 00:09:51.917 the industry was that coding agents were the only form factor that mattered or 00:09:52.000 --> 00:09:54.477 maybe the only technology that mattered or only the only capability that 00:09:54.560 --> 00:09:58.717 mattered for advanced AI. you know there's this incredible force in the 00:09:58.800 --> 00:10:01.277 industry to just like go and crowd that 00:10:01.360 --> 00:10:03.357 use case and compete with enthropic on 00:10:03.440 --> 00:10:06.877 coding agents because of our experiences 00:10:06.960 --> 00:10:10.637 and our conviction that I think formed around this idea of personal agents and 00:10:10.720 --> 00:10:15.757 the sort of taking a step back I think the reason why we did personal super 00:10:15.840 --> 00:10:20.077 intelligence in the first place was that it was also what we believed meta was 00:10:20.160 --> 00:10:23.597 uniquely positioned to do and we can talk more about that but like I think 00:10:23.680 --> 00:10:25.357 because of that >> well let's talk about that now 00:10:25.440 --> 00:10:27.037 >> okay we'll just do do the weave. I think 00:10:27.120 --> 00:10:29.597 one of the first things I told Mark even 00:10:29.680 --> 00:10:32.557 before the whole scale and meta deal 00:10:32.640 --> 00:10:34.477 happened and and everything, one of the 00:10:34.560 --> 00:10:37.277 first things I told him was I think Meta 00:10:37.360 --> 00:10:39.757 is actually a really special company if 00:10:39.840 --> 00:10:42.797 we have AGI or ASI or whatever you want 00:10:42.880 --> 00:10:45.677 to call it. Because part of the promise 00:10:45.760 --> 00:10:49.997 of AI is that for all of us, we have a mix of things that we want to do and 00:10:50.080 --> 00:10:52.397 things that we have to do. And the 00:10:52.480 --> 00:10:55.677 promise of AGI is that we're going to 00:10:55.760 --> 00:10:59.437 spend almost none of the time doing things that we have to do because the 00:10:59.520 --> 00:11:03.197 AIS will just start doing a lot of that for us and we can spend all of our time 00:11:03.280 --> 00:11:05.197 on things that we want to do. And for 00:11:05.280 --> 00:11:09.117 me, Meta's products represent the things 00:11:09.200 --> 00:11:12.397 that we all want to do. Like my use of 00:11:12.480 --> 00:11:16.637 Instagram or WhatsApp or Facebook like these are products that represent what 00:11:16.720 --> 00:11:20.637 I, you know, I follow my interests. I follow my hobbies. I follow all people 00:11:20.720 --> 00:11:22.877 that I like my friends and the people I 00:11:22.960 --> 00:11:26.477 care about like the product span of meta 00:11:26.560 --> 00:11:28.957 encapsulates the sort of like what we 00:11:29.040 --> 00:11:30.877 want to spend our times doing sort of 00:11:30.960 --> 00:11:34.397 realm. I think on a long arc meta is an 00:11:34.480 --> 00:11:36.077 incredible winner through all this. 00:11:36.160 --> 00:11:39.197 That's like one of the first things that 00:11:39.280 --> 00:11:43.837 I believed about meta and super intelligence. And then ultimately a lot 00:11:43.920 --> 00:11:47.837 of those ideas made its way into this concept of personal super intelligence. 00:11:47.920 --> 00:11:51.597 You know, you can kind of look at it a few ways. One is what's the heritage of 00:11:51.680 --> 00:11:54.157 the company? The heritage of meta is 00:11:54.240 --> 00:11:58.717 connecting people with their friends and family. Um, helping them discover their 00:11:58.800 --> 00:12:02.317 interests. Like these are things that are very human. And then there's also 00:12:02.400 --> 00:12:06.557 the sort of like you can take the business lens which is Meta has 00:12:06.640 --> 00:12:08.237 distribution to three and a half billion 00:12:08.320 --> 00:12:11.917 people using the apps every day. And the 00:12:12.000 --> 00:12:16.797 natural area where Meadow was going to be 00:12:16.880 --> 00:12:19.997 successful is consumer AI. >> It's not work. It's not coding. It's 00:12:20.080 --> 00:12:20.637 personal. 00:12:20.720 --> 00:12:24.717 >> Yeah. Exactly. So personal agents and 00:12:24.800 --> 00:12:28.077 personal AI and and consumer AI, broadly 00:12:28.160 --> 00:12:32.237 speaking, were always the zones where we always felt that Meta was going to do 00:12:32.320 --> 00:12:33.517 some be able to do something very 00:12:33.600 --> 00:12:37.117 special. So that brings us to uh okay, 00:12:37.200 --> 00:12:40.157 early 2026. Now you have the prototype, 00:12:40.240 --> 00:12:44.157 you have the buy in, everybody knows Mark's on board and now you're doing the 00:12:44.240 --> 00:12:47.997 seven months of the difficult work of refining this product. 00:12:48.080 --> 00:12:50.237 >> Yes. There was kind of the spectre which 00:12:50.320 --> 00:12:52.317 was like oh personal agents like they 00:12:52.400 --> 00:12:57.277 came and went and we always felt that it was because you know the the experience 00:12:57.360 --> 00:12:59.997 wasn't perfect. It wasn't polished yet you know all that kind of stuff. But 00:13:00.080 --> 00:13:03.837 like we knew that that was something that we had to like think about as you 00:13:03.920 --> 00:13:05.677 know as we were developing this product. 00:13:05.760 --> 00:13:08.477 We focused our model development road 00:13:08.560 --> 00:13:11.997 map. The top focus uh or one of the top 00:13:12.080 --> 00:13:14.477 focuses was towards building on personal 00:13:14.560 --> 00:13:16.797 agents and we had like these long 00:13:16.880 --> 00:13:19.757 spreadsheets of hundreds or even 00:13:19.840 --> 00:13:22.397 thousands of different not thousands but 00:13:22.480 --> 00:13:27.197 but more than a hundred like specific behaviors and capabilities that we knew 00:13:27.280 --> 00:13:31.517 the model needed to have to be able to build the ideal personal agent. Like it 00:13:31.600 --> 00:13:35.997 was very very meticulous. And then over time we built evals for all of those. We 00:13:36.080 --> 00:13:38.877 figured out how to train the model to be really really good at every single one 00:13:38.960 --> 00:13:41.357 of those things. And we have these reviews where we like look at that 00:13:41.440 --> 00:13:44.077 spreadsheet and we figure out the things that we're bad at and make sure that 00:13:44.160 --> 00:13:48.237 we're improving on them. Like we did this like very meticulous process to 00:13:48.320 --> 00:13:52.637 slowly just keep improving and sanding out all the edges of the model so that 00:13:52.720 --> 00:13:56.077 it could power the product. >> Is that really the culture of meta too? 00:13:56.160 --> 00:13:59.277 This like constant like sanding of the products? 00:13:59.360 --> 00:14:02.237 >> Yeah. A lot of the culture of meta is like once you know what you're 00:14:02.320 --> 00:14:06.637 measuring, there's very talented people and you can optimize for those things, 00:14:06.720 --> 00:14:09.517 you can send out all the details, you can make those things amazing. I think 00:14:09.600 --> 00:14:14.317 in the sort of like preAI consumer um product era, like a lot of this was 00:14:14.400 --> 00:14:18.477 around how do you optimize your conversion? How do you optimize the 00:14:18.560 --> 00:14:21.917 referral rates? How do you optimize like every part of the onboarding flow? Like 00:14:22.000 --> 00:14:26.077 how do you make all that stuff just perfect? And then I think the AI era I 00:14:26.160 --> 00:14:28.637 think uh a lot of that especially from 00:14:28.720 --> 00:14:33.117 Muse has been in the form of sanding out every detail of the model to be perfect 00:14:33.200 --> 00:14:35.837 for the product. >> I want to tell you about the presenting 00:14:35.920 --> 00:14:40.077 sponsor of this podcast ramp. I have been reading a lot about SpaceX lately. 00:14:40.160 --> 00:14:43.037 SpaceX is one of the most valuable businesses in the world and one of the 00:14:43.120 --> 00:14:47.197 main themes in the history of SpaceX is constantly attacking and questioning 00:14:47.280 --> 00:14:51.037 your cost. Ramp helps many of the most innovative businesses in the world do 00:14:51.120 --> 00:14:53.037 exactly that. The median company running 00:14:53.120 --> 00:14:58.397 on ramp cuts their expenses by 5%. And one thing SpaceX has demonstrated is 00:14:58.480 --> 00:15:01.917 that a religious dedication to controlling costs can help actually 00:15:02.000 --> 00:15:06.237 increase revenue because you can pursue opportunities you couldn't otherwise. 00:15:06.320 --> 00:15:08.797 And we see that in the RAMP data, too. 00:15:08.880 --> 00:15:14.077 The median company running on ramp also grows their revenue by 16%. So, when 00:15:14.160 --> 00:15:17.437 you're running your business on ramp and your competitors are not, you have a 00:15:17.520 --> 00:15:21.677 massive competitive advantage that compounds over time. Ramp is the only 00:15:21.760 --> 00:15:25.837 platform designed to make your finance team faster and happier. Many of the top 00:15:25.920 --> 00:15:29.357 founders and CEOs I know run their business on RAMP. I run my business on 00:15:29.440 --> 00:15:32.077 RAMP and you should too. Go to ramp.com 00:15:32.160 --> 00:15:36.477 to learn how it can help your business save time, save money, and grow revenue. 00:15:36.560 --> 00:15:40.717 That is ramp.com. Deal is how the best founders turn the 00:15:40.800 --> 00:15:44.157 world into their talent pool. I've been studying how history's greatest founders 00:15:44.240 --> 00:15:47.437 operate for a decade. And one thing they all have in common is they understand 00:15:47.520 --> 00:15:51.837 that recruiting and hiring the very best talent is your most important priority. 00:15:51.920 --> 00:15:55.677 A players recognize other A players, which is why top companies like Ramp, 00:15:55.760 --> 00:15:58.317 Shopify, 11 Labs, Uber, and Door Dash 00:15:58.400 --> 00:16:02.797 all use deal. Many of the top founders I know have personally invested in Deal 00:16:02.880 --> 00:16:05.917 after using their product. And what they discovered is that Deal is the best 00:16:06.000 --> 00:16:08.717 company in the world at building infrastructure for global hiring. Deal 00:16:08.800 --> 00:16:12.957 will help your business hire, pay, and manage any worker anywhere in the world. 00:16:13.040 --> 00:16:16.637 So you can retain the best talent anywhere and spend the rest of your time 00:16:16.720 --> 00:16:20.717 focusing on what you do best, delivering value to your customers. The founder of 00:16:20.800 --> 00:16:24.717 11 Labs has a great description of the value Deal can give your company. He 00:16:24.800 --> 00:16:28.877 said, "We built 11 Labs to break down language and communication barriers. 00:16:28.960 --> 00:16:33.437 With deal enabling us to hire and support exceptional talent anywhere, we 00:16:33.520 --> 00:16:35.277 can accelerate our innovation and bring 00:16:35.360 --> 00:16:38.077 more voices, stories, and ideas to every 00:16:38.160 --> 00:16:43.277 corner of the world. Deal is trusted by over 40,000 companies and growing fast. 00:16:43.360 --> 00:16:44.957 Learn how they can help your business by 00:16:45.040 --> 00:16:51.240 going to deal.comra. That is deal.com/enra. 00:16:51.520 --> 00:16:56.237 >> Did you see this profile that's written on Zuck in Colossus written by Jeremy 00:16:56.320 --> 00:17:00.797 Stern? >> I saw it. If I'm being honest, it was like too long. Didn't read. But 00:17:00.880 --> 00:17:04.877 >> it's 15,000 words. I just had Jeremy on the podcast. He's he's my favorite. 00:17:04.960 --> 00:17:08.557 >> I saw I saw I saw that. He's he's just I think he's writing the best profiles on 00:17:08.640 --> 00:17:11.437 on entrepreneurs and investors in the world right now. I think he's literally 00:17:11.520 --> 00:17:12.557 the best in the world at what he does. 00:17:12.640 --> 00:17:16.637 But in that Zach says he was talking about the advantage that he has relative 00:17:16.720 --> 00:17:20.317 to I think Elon and like Sam Alman and he's just like listen my skill set is 00:17:20.400 --> 00:17:23.517 very unsexy but I'm really good at building teams and then just improving a 00:17:23.600 --> 00:17:28.397 product slowly over a long period of time and relative to Elon and and Sam 00:17:28.480 --> 00:17:31.357 Alman and others. like I'm not going to lose control of my company and I don't 00:17:31.440 --> 00:17:34.397 need to raise any more money and so I can just do this for an excessively long 00:17:34.480 --> 00:17:36.877 period of time. That section of the 00:17:36.960 --> 00:17:41.517 profile is popping in my mind when you're talking about this 7-month uh 00:17:41.600 --> 00:17:44.557 stretch of trying to refine and build muse. 00:17:44.640 --> 00:17:49.437 >> I think that's exactly right. Like I think that it took immense patience. I 00:17:49.520 --> 00:17:53.037 mean 7 months like you know maybe it doesn't sound like that long but AI is 00:17:53.120 --> 00:17:55.277 like fierce. seven months in AI time. 00:17:55.360 --> 00:17:58.877 >> Seven months in AI time, which maybe like a decade and like nonAI time. And I 00:17:58.960 --> 00:18:02.717 think it it for all of us. Um but but I really give Mark a lot of credit for 00:18:02.800 --> 00:18:05.837 this. Like it took a lot of patience and 00:18:05.920 --> 00:18:08.637 restraint and a lot of trust in the 00:18:08.720 --> 00:18:10.477 process. And like you'll you can go back 00:18:10.560 --> 00:18:13.437 and look at the you know what Mark said 00:18:13.520 --> 00:18:16.157 in earnings calls for this entire period 00:18:16.240 --> 00:18:18.557 from February through through um 00:18:18.640 --> 00:18:23.997 September and he will allude to the fact that we were building personal agents 00:18:24.080 --> 00:18:27.437 and we weren't really trying to hide it like we were we were building personal 00:18:27.520 --> 00:18:31.277 agents and we were talking about it but I think you know there's a lot of doubt 00:18:31.360 --> 00:18:35.837 in the company for that entire period like people obviously the one of the 00:18:35.920 --> 00:18:39.837 dominant Wall Street narratives is like oh my god Gosh, Meta is just burning all 00:18:39.920 --> 00:18:43.037 this money. Like, are they even going to pull anything off in AI? 00:18:43.120 --> 00:18:47.197 >> It's more than that. It's like Zuck cannot win AI. It's said over and over 00:18:47.280 --> 00:18:50.157 again for a long period of time. And again, going back to Jeremy Stern's 00:18:50.240 --> 00:18:53.357 profile on him. He then off the record 00:18:53.440 --> 00:18:57.437 starts talking to his competitors, including one former researcher that is 00:18:57.520 --> 00:19:01.837 now at a competitor that used to work for with Zuck and is now at another lab. 00:19:01.920 --> 00:19:04.477 And he goes, "I don't want to have to compete with them cuz Mark's like the 00:19:04.560 --> 00:19:08.077 [__] I think the line is Mark is the [__] Terminator." And it was like he 00:19:08.160 --> 00:19:13.117 just does will not stop. And he's like uh then he talks about um one of his 00:19:13.200 --> 00:19:15.917 superpowers that he's never happy with the state of the product. So he always 00:19:16.000 --> 00:19:19.757 wants to constant prove but he's not an [__] Like he's not you know dict 00:19:19.840 --> 00:19:23.757 he's he is definitely a dictator but like he's not you know rude and like 00:19:23.840 --> 00:19:26.717 kind of will destroy the chemistry of a team. He actually talks about in the in 00:19:26.800 --> 00:19:30.477 the piece that team cohesion is excessively important to him. 00:19:30.560 --> 00:19:35.037 >> I think it's all right. I think one of the magical things about Muse was that 00:19:35.120 --> 00:19:38.877 we really like took quite a bit of time to 00:19:38.960 --> 00:19:43.357 sweat all the details before ultimately coming out with a product because I 00:19:43.440 --> 00:19:45.357 think that that personal agents was one 00:19:45.440 --> 00:19:50.557 of these product areas that kind of rested on a knife's edge, which is if it 00:19:50.640 --> 00:19:53.837 can reliably deliver these magical experiences, it has the potential to be 00:19:53.920 --> 00:19:56.957 one of the greatest consumer products ever. But you're dealing with models, 00:19:57.040 --> 00:19:58.557 you're dealing with something nondeterministic, you're dealing with 00:19:58.640 --> 00:20:03.037 agents, they can break, they can be uh unreliable. Like if people have these 00:20:03.120 --> 00:20:06.557 unreliable experiences, they think it's trash. It was one of these things where 00:20:06.640 --> 00:20:08.957 like getting to the product to a point 00:20:09.040 --> 00:20:11.917 of like quality and value such that it 00:20:12.000 --> 00:20:16.157 resonated with so many of the people who downloaded and used it. I want to talk 00:20:16.240 --> 00:20:19.917 more about this process. I'm curious how much resources, how big the team, how 00:20:20.000 --> 00:20:23.357 many people working on it. Do you guys have any data on like okay I'm using 00:20:23.440 --> 00:20:26.717 Muse nine great experiences two like 00:20:26.800 --> 00:20:30.477 inconsistent or you know shitty experiences and then I turn out like do 00:20:30.560 --> 00:20:33.517 you have any information on this? You can see this from like a lot of the 00:20:33.600 --> 00:20:37.517 other agent products at the start of the year like almost everyone turned out of 00:20:37.600 --> 00:20:38.637 those and it's because they were 00:20:38.720 --> 00:20:41.997 unreliable. We did a lot of user testing 00:20:42.080 --> 00:20:46.317 of Muse. Like we did many many rounds of 00:20:46.400 --> 00:20:49.277 new people trying to like download it 00:20:49.360 --> 00:20:51.757 and and use it from scratch and just 00:20:51.840 --> 00:20:55.037 seeing what that experience was like. A 00:20:55.120 --> 00:20:57.597 huge part of the premise of Muse and I 00:20:57.680 --> 00:21:00.717 think what's really landed is that it is 00:21:00.800 --> 00:21:03.837 like a different thing from what most 00:21:03.920 --> 00:21:05.677 people have experienced with consumer 00:21:05.760 --> 00:21:08.477 AI. Like it is not a chatbot. It is an 00:21:08.560 --> 00:21:10.957 agent and it can do agent things for 00:21:11.040 --> 00:21:15.757 you. And like I think to a developer this is kind of like old news because 00:21:15.840 --> 00:21:18.397 they've had to like you know download agents and have been using them for a 00:21:18.480 --> 00:21:20.557 while. But for most people like just 00:21:20.640 --> 00:21:22.717 even that graduation process like you 00:21:22.800 --> 00:21:25.517 know it needs to be explained and needs 00:21:25.600 --> 00:21:28.077 to be onboarded on on a very thoughtful 00:21:28.160 --> 00:21:32.877 way. And then for sure we definitely have seen like when because we've done 00:21:32.960 --> 00:21:36.717 tests with like different versions of the model and various model versions and 00:21:36.800 --> 00:21:40.477 continuously do those AB tests. You know the product experience is very very 00:21:40.560 --> 00:21:42.717 sensitive to model quality. It's very 00:21:42.800 --> 00:21:46.317 sensitive to how reliably the models can 00:21:46.400 --> 00:21:49.757 undergo these tasks, how reliably they communicate with the user about them, 00:21:49.840 --> 00:21:52.877 how reliably they like, you know, check back with the user if there's things 00:21:52.960 --> 00:21:54.637 that they require guidance on. 00:21:54.720 --> 00:21:56.557 >> But how'd you know it was ready to be 00:21:56.640 --> 00:22:00.797 released? Part of it was was pretty 00:22:00.880 --> 00:22:05.437 numerical which is like we had set you know you go back to that spreadsheet of 00:22:05.520 --> 00:22:09.437 100 plus things. >> We had set for each of those rows what 00:22:09.520 --> 00:22:13.277 threshold is launch blocking and above what threshold is good. We had many 00:22:13.360 --> 00:22:17.357 checkpoints that like were like green on 80 of the rows but red on 20 of the 00:22:17.440 --> 00:22:18.877 rows. We knew that wasn't good enough. 00:22:18.960 --> 00:22:23.357 And we had our first model checkpoint which was green across the board which 00:22:23.440 --> 00:22:25.517 was a specially trained version of 00:22:25.600 --> 00:22:28.317 Musepark 1.3. So there was like the very 00:22:28.400 --> 00:22:33.517 numerical part of it and then there was just the experiential part which is we 00:22:33.600 --> 00:22:36.797 tried the model. We could tell that it was a lot better than than anything we 00:22:36.880 --> 00:22:39.677 had iterated with before. So going back to that seven-month period you were 00:22:39.760 --> 00:22:41.517 talking about, how many people were working on the product 00:22:41.600 --> 00:22:44.237 >> between the model work, model work and 00:22:44.320 --> 00:22:49.677 the product work? Like under 200 people total, which I guess sounds like a lot, 00:22:49.760 --> 00:22:52.637 but in big company land is >> it's actually a lot smaller. 00:22:52.720 --> 00:22:55.917 >> It's not very many. That was kind of cool looking back on this and thinking 00:22:56.000 --> 00:23:00.717 about it is like we knew this was going to be the most important product that 00:23:00.800 --> 00:23:02.717 the whole executive team, everyone knew 00:23:02.800 --> 00:23:06.157 that this was probably going to be one 00:23:06.240 --> 00:23:08.077 of the most important products that Meta 00:23:08.160 --> 00:23:11.757 would ship this year, but we had the 00:23:11.840 --> 00:23:15.437 restraint organizationally I think to make sure that like it was a really 00:23:15.520 --> 00:23:17.597 small focused team working on it and 00:23:17.680 --> 00:23:22.477 that it didn't balloon to this like oh you have like you know huge swasts of 00:23:22.560 --> 00:23:25.437 the company working on because we knew it was a single work of art. Like it 00:23:25.520 --> 00:23:29.037 required strong point of view and strong 00:23:29.120 --> 00:23:33.277 focus to like pull that through into one beautiful experience. Say more about 00:23:33.360 --> 00:23:36.477 this. Why did you just call it like a single piece of art and you need a 00:23:36.560 --> 00:23:40.637 single perspective or point of view? I think one 00:23:40.720 --> 00:23:42.957 um 00:23:43.040 --> 00:23:45.757 criticism of products that come from 00:23:45.840 --> 00:23:47.917 larger organizations and I think this 00:23:48.000 --> 00:23:52.237 like is even now starting to plague 00:23:52.320 --> 00:23:54.317 a lot of the AI labs is that they become 00:23:54.400 --> 00:23:57.037 kind of this Frankenstein 00:23:57.120 --> 00:24:00.397 almost like Croninberg amalgamation of a 00:24:00.480 --> 00:24:02.157 bunch of different people's points of 00:24:02.240 --> 00:24:05.997 view and visions and and beliefs and 00:24:06.080 --> 00:24:08.557 it's sort of like it's sort of PM hell 00:24:08.640 --> 00:24:13.037 in some sense where like every PM has like a thing that they're trying to get 00:24:13.120 --> 00:24:15.757 into the product and jam into the product and then all that kind of like 00:24:15.840 --> 00:24:18.077 blends together into this like like 00:24:18.160 --> 00:24:21.917 smoothie almost. People feel it as users 00:24:22.000 --> 00:24:25.757 and consumers like you feel you can feel when a product is just like it feels 00:24:25.840 --> 00:24:28.877 like lots of different people worked on it and there were like clearly different 00:24:28.960 --> 00:24:32.157 people who were gold differently so they just kind of like jammed it all 00:24:32.240 --> 00:24:34.397 together. Muse had to be the exact 00:24:34.480 --> 00:24:39.677 opposite, which is like there's one clear point of view. There is one 00:24:39.760 --> 00:24:43.917 cohesive experience that we're trying to deliver um to the world. I love that you 00:24:44.000 --> 00:24:46.797 said that and it just happens to be pure coincidence. Like we're I don't know 00:24:46.880 --> 00:24:48.877 when this is going to come out, but we're recording this on the 15th 00:24:48.960 --> 00:24:52.317 anniversary of Steve Jobs death. And when you think of somebody like a pro 00:24:52.400 --> 00:24:56.077 consumer products that had a single point of view, you obviously think of 00:24:56.160 --> 00:24:59.517 Steve first. I've read every single book on Steve and the history of Apple and 00:24:59.600 --> 00:25:03.117 I've done like I don't know 15 or 20 episodes on them uh for my other podcast 00:25:03.200 --> 00:25:07.437 founders, but one of my favorite lines to describe Steve's impact was it says 00:25:07.520 --> 00:25:09.517 that um Apple is just Steve Jobs with 00:25:09.600 --> 00:25:14.477 10,000 lives. The products are to his taste, his perspective, his point of 00:25:14.560 --> 00:25:18.477 view. What he wanted to see in the world is exactly what the the the end user 00:25:18.560 --> 00:25:21.117 got. In Muse's case, like Nat Friedman 00:25:21.200 --> 00:25:25.517 deserves a lot of credit. like his taste and sensibilities ultimately dictated 00:25:25.600 --> 00:25:27.437 and shaped what that product became. 00:25:27.520 --> 00:25:30.077 >> Do you talk about how many people are using Muse? Are you guys talking about 00:25:30.160 --> 00:25:31.357 this publicly or no? 00:25:31.440 --> 00:25:36.237 >> We have said millions and then I have 00:25:36.320 --> 00:25:40.637 retweeted other people who have shown 00:25:40.720 --> 00:25:43.357 graphs where we would imply that Muse is 00:25:43.440 --> 00:25:48.077 the fastest growing consumer app of all time. Consumer AI app of all time. 00:25:48.160 --> 00:25:49.437 >> So you won't say a number though. 00:25:49.520 --> 00:25:53.357 >> We've said millions. Yeah. Well, millions could be hundreds of millions. 00:25:53.440 --> 00:25:56.877 That's also millions. So, you're just not adding the the the beginning to 00:25:56.960 --> 00:26:00.157 that. I want to like go into like you've been kind of going like uh and I mean 00:26:00.240 --> 00:26:05.677 this in a loving way like unhinged on X. So, you talk about like your is 00:26:05.760 --> 00:26:08.317 this like marketing strategy? Like what is going on? Like I feel like you're 00:26:08.400 --> 00:26:13.357 there's a distinct difference in your public tweeting since Muse launch. What 00:26:13.440 --> 00:26:15.677 are you doing? >> There's a few pieces of what's happening 00:26:15.760 --> 00:26:18.877 here. This has got to be more fun than scale AI when you were building scale. 00:26:18.960 --> 00:26:22.237 It has to be. We'll talk about that later, but it has to be the past few 00:26:22.320 --> 00:26:27.357 weeks have been some of my most fun uh building like yeah certainly my career 00:26:27.440 --> 00:26:31.037 and I think there's a few things happening. So, one is it felt important 00:26:31.120 --> 00:26:33.917 that we figured out a way to make Muse 00:26:34.000 --> 00:26:38.477 break through. It's a pretty like torrential going back to this like this 00:26:38.560 --> 00:26:40.397 torrential stream of updates. Like for 00:26:40.480 --> 00:26:43.117 most people, there's just so they're 00:26:43.200 --> 00:26:45.997 just like constantly being like like 00:26:46.080 --> 00:26:50.397 kind of bombarded with new AI products and startups and releases and things and 00:26:50.480 --> 00:26:53.357 models and here's another model and here's another product and here's 00:26:53.440 --> 00:26:57.597 another feature and it's just like this torrential stream. We had actually 00:26:57.680 --> 00:27:00.557 meeting large meetings where we talked about this as a team. It was like we 00:27:00.640 --> 00:27:04.637 need to break through and obviously like meta we have lots of distribution like 00:27:04.720 --> 00:27:07.837 we can make sure everyone knows about it but that's not enough to like make it 00:27:07.920 --> 00:27:11.357 break through in an interesting cultural 00:27:11.440 --> 00:27:14.077 or fascinating way for the world to kind 00:27:14.160 --> 00:27:17.437 of like you know pay attention to. I 00:27:17.520 --> 00:27:21.117 separately in my life have happened to 00:27:21.200 --> 00:27:25.277 have met and um become friends with like 00:27:25.360 --> 00:27:29.917 memers and and uh I have these uh I 00:27:30.000 --> 00:27:33.437 affectionately call them my skitso friends but but people who like 00:27:33.520 --> 00:27:37.197 literally they spend all their day like thinking about memes in the internet and 00:27:37.280 --> 00:27:41.037 and understanding that I have people that are obsessed with Muse and they're 00:27:41.120 --> 00:27:44.637 also obsessed with your X feed and they're convinced that you have like a 00:27:44.720 --> 00:27:49.277 hundred muses coming up with memes. to to be used. 00:27:49.360 --> 00:27:52.637 Um, turns out you just have a collection of schizophrenic friends that are 00:27:52.720 --> 00:27:55.597 terminally online. >> The X feed, just for what it's worth, is 00:27:55.680 --> 00:28:00.157 like literally every one of those tweets I write. There were like a few a quite a 00:28:00.240 --> 00:28:03.517 period where I was like, it was like really hard for me to pay attention in 00:28:03.600 --> 00:28:06.317 meetings because I would just like think of a meme and I'd be like, "Oh god, I 00:28:06.400 --> 00:28:10.237 have to get this out." And I would like it was like moments of inspiration. 00:28:10.320 --> 00:28:13.277 There was like um I was uh I was with 00:28:13.360 --> 00:28:17.517 one of my like I was trying to have like dinner with one of my co-workers and 00:28:17.600 --> 00:28:20.877 then during the dinner I was like wait I figured it out and then I like posted a 00:28:20.960 --> 00:28:24.237 different me. It was like it was almost I almost felt like I was like an 00:28:24.320 --> 00:28:27.997 engineer solving bugs again cuz it was like you know you could tell when 00:28:28.080 --> 00:28:30.957 something was going to click. But yeah so one things we wanted to break 00:28:31.040 --> 00:28:32.957 through. Another thing that like I just 00:28:33.040 --> 00:28:34.877 thought was important is like how do we 00:28:34.960 --> 00:28:38.477 make meta cool again? cuz like for 00:28:38.560 --> 00:28:41.197 whatever reason despite having like 00:28:41.280 --> 00:28:42.877 these incredible pieces of cultural 00:28:42.960 --> 00:28:45.997 software um like Instagram and and uh 00:28:46.080 --> 00:28:49.117 Facebook historically meta was not cool 00:28:49.200 --> 00:28:53.037 for a while and so wanted to make meta 00:28:53.120 --> 00:28:54.877 cool and interesting and kind of like 00:28:54.960 --> 00:28:57.757 funny and then the unhingedness was 00:28:57.840 --> 00:29:00.637 something that happened kind of like 00:29:00.720 --> 00:29:04.957 naturally over time when Muse first came out I was just like mostly focused on 00:29:05.040 --> 00:29:08.797 promoting all the interesting use cases that I saw I was like retweeting like 00:29:08.880 --> 00:29:11.837 tens of interesting use cases that I saw 00:29:11.920 --> 00:29:16.317 of people using Muse. And then at some point it was like because I was just 00:29:16.400 --> 00:29:20.077 like tweeting so much. There were like a few tweets where I was just like didn't 00:29:20.160 --> 00:29:23.197 think that much about it and just said [__] it post. And then some of those 00:29:23.280 --> 00:29:26.877 tweets like did really well. Have you seen this like time spent thinking about 00:29:26.960 --> 00:29:29.197 it and then banger you seen that? Great. 00:29:29.280 --> 00:29:30.637 >> I haven't se but I agree with that. 00:29:30.720 --> 00:29:34.477 That's like I just some of these I didn't spend that much time on. I like 00:29:34.560 --> 00:29:36.797 tweeted them. I didn't even check my phone for a few hours and I was like, 00:29:36.880 --> 00:29:38.397 "Oh, wow. That one went really well." 00:29:38.480 --> 00:29:41.517 And then I kind of like remembered something about the internet which I it 00:29:41.600 --> 00:29:45.037 wasn't super intentional, but I remembered like the internet rewards 00:29:45.120 --> 00:29:48.397 risk and it rewards like surprise and 00:29:48.480 --> 00:29:52.077 like things that like people don't expect. And so then I just kind of like 00:29:52.160 --> 00:29:54.557 went off. A lot of my co-workers uh I 00:29:54.640 --> 00:29:59.757 think were somewhat surprised at uh something that um Nat said for for the 00:29:59.840 --> 00:30:02.797 whole week when I was posting all the memes. He was like, "You know, I thought 00:30:02.880 --> 00:30:06.317 we would learn." He said this like in a bunch of meetings. He was like, "I 00:30:06.400 --> 00:30:10.557 thought we were going to learn a lot about um you know, Muse users and like 00:30:10.640 --> 00:30:14.077 how people were using Muse, but actually like we're just learning a lot about 00:30:14.160 --> 00:30:17.037 Alex's mind." 00:30:17.120 --> 00:30:19.037 It was a lot of fun. I felt like I was 00:30:19.120 --> 00:30:21.277 drawing on all of my internet knowledge 00:30:21.360 --> 00:30:26.397 for years and years and years to like pull into those moments. And I think it 00:30:26.480 --> 00:30:28.557 actually, you know, I don't know, we don't have the tracking and it's like 00:30:28.640 --> 00:30:32.317 hard to tell, but I actually think it actually genuinely made an impact on 00:30:32.400 --> 00:30:34.157 grow growth, which is kind of hilarious. 00:30:34.240 --> 00:30:38.557 >> I definitely think it did. I I can't stop hearing about it. So, and all the 00:30:38.640 --> 00:30:41.757 people I'm hearing about it from also downloaded the app and are now using it. 00:30:41.840 --> 00:30:42.957 Go back to what you were saying though. 00:30:43.040 --> 00:30:45.517 It's like, hey, we have distribution, but that doesn't make mean that we can 00:30:45.600 --> 00:30:48.237 make something that people are going to come back to and actually sticky. So, 00:30:48.320 --> 00:30:49.837 like, how did you do this with Muse? 00:30:49.920 --> 00:30:53.197 >> I think it was like, how do you make this into how do we how do we make this 00:30:53.280 --> 00:30:55.597 breakthrough as a product? And I think there were like a few parts of that, but 00:30:55.680 --> 00:30:57.837 I think one is how do you get people to 00:30:57.920 --> 00:30:59.597 a match the experience? Like how do you 00:30:59.680 --> 00:31:03.517 get someone to to a wow moment with the 00:31:03.600 --> 00:31:07.357 product as early as possible? And this product I think naturally lends itself 00:31:07.440 --> 00:31:11.437 to that because I think you know there's it's like it's AI. It can do a lot. It 00:31:11.520 --> 00:31:16.717 can there's like if if someone has never used an AI agent, there's a lot that it 00:31:16.800 --> 00:31:20.477 can do that will surprise you. So that I think was a big part of it. I think 00:31:20.560 --> 00:31:23.357 another part of it was like how do you 00:31:23.440 --> 00:31:26.077 harness all of those stories of people 00:31:26.160 --> 00:31:28.717 having these wow moments and like use 00:31:28.800 --> 00:31:33.117 those to help make the product sing and make the product fly. 00:31:33.200 --> 00:31:34.477 >> How are you using the stories though? 00:31:34.560 --> 00:31:36.557 Cuz you you guys aren't running ads for me, are you? 00:31:36.640 --> 00:31:39.037 >> We ran a few ads. I mean honestly like I 00:31:39.120 --> 00:31:43.677 think at first it was just the fact that I would retweet so many of them. Like I 00:31:43.760 --> 00:31:45.677 was just >> anytime I saw someone do something 00:31:45.760 --> 00:31:49.037 interesting about news I was like quote tweet. I couldn't understand why both 00:31:49.120 --> 00:31:52.637 Enthropic and OpenAI's um ads are so bad 00:31:52.720 --> 00:31:56.397 because remember Anthropic was doing this huge like outdoor campaign and it's 00:31:56.480 --> 00:32:00.797 all about them and same thing with um Chet. It's like they'd have like the the 00:32:00.880 --> 00:32:03.757 icon which I don't even think people recognize that as a logo and it just say 00:32:03.840 --> 00:32:06.637 CEX and just be two people sitting at a desk. I'm like what the [__] does this 00:32:06.720 --> 00:32:10.237 mean? Where it's like you have all these use cases. All you have to do your ad is 00:32:10.320 --> 00:32:12.477 to show the benefit of your product. 00:32:12.560 --> 00:32:15.277 That's it. >> You have millions of them. Run those 00:32:15.360 --> 00:32:18.957 ads. It's it's so like elementary that it's interesting that people make this 00:32:19.040 --> 00:32:21.357 mistake over and over again. >> Yeah. I think one thing that we came 00:32:21.440 --> 00:32:24.637 back to a lot in in both building news 00:32:24.720 --> 00:32:29.037 as well as just like building stuff in general at me was just like why would 00:32:29.120 --> 00:32:31.997 someone care? They don't care about you or your company. They care about 00:32:32.080 --> 00:32:36.077 everybody's self-interested. This is like I'm there's this very old book. 00:32:36.160 --> 00:32:39.677 It's called Scientific Advertising. It was written in the early 1900s. is now 00:32:39.760 --> 00:32:42.877 it was so valuable that the guy that made the most money as an advertising 00:32:42.960 --> 00:32:47.277 agency founder in the like 1920s and 30s he it was written by this guy named 00:32:47.360 --> 00:32:50.397 Hopkins who was this best copyriter he's like these are all the ideas how I'm 00:32:50.480 --> 00:32:53.917 making this money I'm putting this manuscript in a safe for 20 years right 00:32:54.000 --> 00:32:56.477 but essentially it's not about advertising it's about human nature 00:32:56.560 --> 00:32:59.757 which doesn't change and his whole point was just like your ads are about you 00:32:59.840 --> 00:33:03.117 your ads about your product your ads are like hey I have this thing buy people 00:33:03.200 --> 00:33:05.437 don't care about that they're self-interested and if you just all you 00:33:05.520 --> 00:33:10.157 have to do is just do the work of what this does if for you and you just lead 00:33:10.240 --> 00:33:13.997 with that and there's people that took that ideas those ideas were written in 00:33:14.080 --> 00:33:17.517 the early 1900s and went on to build massive companies advertising agency 00:33:17.600 --> 00:33:20.317 foundations in the ' 50s ' 60s7s and even to this day 00:33:20.400 --> 00:33:23.357 >> yeah 100%. I think there are a few 00:33:23.440 --> 00:33:25.757 insights. Let's say first is most people 00:33:25.840 --> 00:33:31.277 in the world, they don't give a [__] about like OpenAI or the research or you 00:33:31.360 --> 00:33:34.317 know these other companies or the research or even us or our research or 00:33:34.400 --> 00:33:36.077 whatever. They care about like what it 00:33:36.160 --> 00:33:38.877 can do for them. We had this gift of the 00:33:38.960 --> 00:33:42.317 mascot which was the little jolly guy. I 00:33:42.400 --> 00:33:46.637 remember the tweet where I where I realized I was like, "Oh, wow. This is 00:33:46.720 --> 00:33:48.397 mileage." because I was basically I was 00:33:48.480 --> 00:33:52.317 um I was driving into work uh and while 00:33:52.400 --> 00:33:56.157 I was on the way I like generated a a 00:33:56.240 --> 00:33:58.237 version of the guy holding a briefcase 00:33:58.320 --> 00:34:03.357 and I tweeted MFW I go to work on Monday. I remember thinking like this is 00:34:03.440 --> 00:34:06.717 like one of my best tweets in a while and then I kind of realized like you can 00:34:06.800 --> 00:34:11.277 put him into a lot of situations. One thing I like kind of realized as we were 00:34:11.360 --> 00:34:16.157 going through that process was like how do you create surprising moments that 00:34:16.240 --> 00:34:19.517 are very different from any other marketing that you've seen of like a 00:34:19.600 --> 00:34:21.677 product. I don't like really remember 00:34:21.760 --> 00:34:23.677 the last time someone like you know 00:34:23.760 --> 00:34:28.797 leaned into racy memes to like market a product. Um I don't actually know if 00:34:28.880 --> 00:34:31.997 it's been done before but I kind of like realize like oh wow this is like a thing 00:34:32.080 --> 00:34:34.797 that we can do with this little guy. And that's something that like is 00:34:34.880 --> 00:34:38.237 interesting. That's something that like drives intrigue to like the average 00:34:38.320 --> 00:34:41.917 person. >> Okay. But that might get them raise awareness, might even get them to 00:34:42.000 --> 00:34:45.917 download the app. But then how do you get them to stay engaged once the app is 00:34:46.000 --> 00:34:48.157 they the app is on their phone? There's 00:34:48.240 --> 00:34:52.237 a cycle that people who who really love 00:34:52.320 --> 00:34:55.037 the app and and we even like we back in 00:34:55.120 --> 00:34:58.397 February had which is you try to give it 00:34:58.480 --> 00:35:02.637 a little thing and it does it. You're like, "Oh, wow. It did that." And then 00:35:02.720 --> 00:35:05.837 you try to give it a little bit something like a little bit more and 00:35:05.920 --> 00:35:09.197 then it does it and you're like, "Oh, wow. It can do that. It could do that 00:35:09.280 --> 00:35:11.917 first thing. It could do the second thing." And then you give it like a 00:35:12.000 --> 00:35:14.877 little bit bigger of a of a problem. And 00:35:14.960 --> 00:35:17.197 it's kind of this like trust fall almost 00:35:17.280 --> 00:35:20.237 that you have with an AI agent, which is 00:35:20.320 --> 00:35:24.237 at first you like don't really trust it, but you're like kind of intrigued. So 00:35:24.320 --> 00:35:26.957 you're going to give it a little bit nibble and then you give it like 00:35:27.040 --> 00:35:29.117 something a little bigger and something a little bigger and something a little 00:35:29.200 --> 00:35:32.637 bigger. Certainly like when I talk to like college students or whatever like 00:35:32.720 --> 00:35:36.397 some of the ways that people use AI it's like a full trust fault. They just like 00:35:36.480 --> 00:35:39.997 are recording voice notes and brain dumping to the AI and they send it and 00:35:40.080 --> 00:35:43.677 then they just trust the AI to like organize all the thoughts and give them 00:35:43.760 --> 00:35:48.077 like clear plans and think clear things to do etc. I think for the successful 00:35:48.160 --> 00:35:50.157 users of Muse it really is like getting 00:35:50.240 --> 00:35:53.197 them onto this trust fall process of 00:35:53.280 --> 00:35:57.197 using it for slightly bigger and bigger and bigger things and then having it 00:35:57.280 --> 00:35:59.677 work every one of those times. 00:35:59.760 --> 00:36:02.637 I found one of my all-time favorite quotes when I was reading the book 0ero 00:36:02.720 --> 00:36:06.557 to1. The quote says, "The single most powerful pattern I have noticed is that 00:36:06.640 --> 00:36:11.277 successful people find value in unexpected places." And they do this by 00:36:11.360 --> 00:36:15.117 thinking about business from first principles instead of formulas. That is 00:36:15.200 --> 00:36:19.117 exactly what Apploven has done with their advertising platform. Apploven 00:36:19.200 --> 00:36:22.637 connects you with over a billion potential new customers in mobile games. 00:36:22.720 --> 00:36:27.357 Apploven allows you to capture undivided attention. Apploving ads are full screen 00:36:27.440 --> 00:36:31.757 videos that are watched for an average of 35 seconds. That is retention that 00:36:31.840 --> 00:36:36.237 blows other ad platforms out of the water. And you can launch on Apploven in 00:36:36.320 --> 00:36:40.557 minutes. You set the goal and Apploven achieves it. No complex setup, no 00:36:40.640 --> 00:36:44.957 expertise needed. And Apploven scales quickly. They can put your ads in front 00:36:45.040 --> 00:36:47.277 of over a billion potential customers. 00:36:47.360 --> 00:36:51.197 Other businesses have seen immediate results scale to hundreds of thousands 00:36:51.280 --> 00:36:55.437 of dollars of spend per day and increase their revenue by millions. So, you want 00:36:55.520 --> 00:36:59.277 to get started quickly before all of your competitors are on AppLoven. And 00:36:59.360 --> 00:37:01.997 you can do that by going to apploven.com. 00:37:02.080 --> 00:37:04.877 That's apploven.com. 00:37:04.960 --> 00:37:09.837 I wonder how much of this is actually driven by word of mouth. So, obviously 00:37:09.920 --> 00:37:12.477 talking to you, I was going to have to download the app no matter what. But 00:37:12.560 --> 00:37:15.597 what really piqued my interest is a friend of mine was sending me 00:37:15.680 --> 00:37:18.717 screenshots of how he was using Muse. 00:37:18.800 --> 00:37:21.517 So, like when you download the app, it's it doesn't really tell you like too much 00:37:21.600 --> 00:37:24.557 of like what you can do. It was kind of like open-ended like a you know like a 00:37:24.640 --> 00:37:27.437 open text box but a friend of mine who's 00:37:27.520 --> 00:37:32.237 running like a hundred billion dollar company he left his ID on a flexjet and 00:37:32.320 --> 00:37:35.517 then he went through and he's just like well I want to he tell me I don't want 00:37:35.600 --> 00:37:39.037 to be involved in this process at all you have to figure out how to get the ID 00:37:39.120 --> 00:37:41.677 that like go pick up the ID this is where it's at and then this is where I'm 00:37:41.760 --> 00:37:44.477 at and then figure out all the steps along the way and then like arranged the 00:37:44.560 --> 00:37:48.637 courier went there got to his security in his office somehow and got upstairs 00:37:48.720 --> 00:37:52.797 to his office like and it was like holy [__] this is the best marketing. It's 00:37:52.880 --> 00:37:56.397 just three or four screenshots of Muse going back and forth and him saying, "I 00:37:56.480 --> 00:37:59.117 don't want to be involved. You have to do everything." And it figuring out on 00:37:59.200 --> 00:38:04.317 the way. Yeah. Know, this is actually one of the things about um uh I don't 00:38:04.400 --> 00:38:08.877 know how it was like somewhat intentional, but definitely like modern 00:38:08.960 --> 00:38:12.477 product marketing is like very screenshot driven. Like I think to have 00:38:12.560 --> 00:38:14.877 a modern consumer product like it has to 00:38:14.960 --> 00:38:16.317 work on the screenshot. like the 00:38:16.400 --> 00:38:20.397 screenshot has to has to fly so to speak 00:38:20.480 --> 00:38:24.397 either through like group chats or word of mouth or like online. And I think 00:38:24.480 --> 00:38:27.597 that's been a big part of it. This wasn't intentional or engineered, but I 00:38:27.680 --> 00:38:29.837 think the fact that we had the little 00:38:29.920 --> 00:38:33.037 guy in the screenshot is like was a big 00:38:33.120 --> 00:38:36.397 deal because I think having Jolly or 00:38:36.480 --> 00:38:40.317 whoever your muse is like in the screenshot immediately communicates this 00:38:40.400 --> 00:38:43.597 is a Muse screenshot. It's just like the best marketing you could possibly have. 00:38:43.680 --> 00:38:46.637 So, are you on your Muse like all day long? like what is your own personal 00:38:46.720 --> 00:38:47.597 usage? 00:38:47.680 --> 00:38:49.837 >> I mostly like set up a lot of workflows 00:38:49.920 --> 00:38:54.237 using it and then I um I guess my 00:38:54.320 --> 00:38:59.357 meetings a lot of the time but um so in practice like I think in my job 00:38:59.440 --> 00:39:02.717 quoteunquote I'm supposed to like pay attention and and be pretty present and 00:39:02.800 --> 00:39:06.477 interact with people. I set up a bunch of workflows and like kind of treated as 00:39:06.560 --> 00:39:10.237 like a second brain. >> Yeah. You said that in the post that you 00:39:10.320 --> 00:39:13.117 wrote and it was like really short and I think I saw last time it was like 5 and 00:39:13.200 --> 00:39:16.637 a half million views so far but you said uh I love how you describe Muse like 00:39:16.720 --> 00:39:18.157 well if everyone had a second mind 00:39:18.240 --> 00:39:21.597 beyond their own. I think this is 00:39:21.680 --> 00:39:26.397 philosophically speaking one of the most interesting questions about like what 00:39:26.480 --> 00:39:28.157 does it mean to be human when you have 00:39:28.240 --> 00:39:31.357 like powerful AI and what is that 00:39:31.440 --> 00:39:35.677 relationship like and and how exactly does that look and different people have 00:39:35.760 --> 00:39:38.877 different takes on what exactly that looks like. I think the sort of like 00:39:38.960 --> 00:39:42.877 fear cases obviously that you know AI is above us and that we're just sort of 00:39:42.960 --> 00:39:46.797 like doing what they do and there's like other worlds where that relationship is 00:39:46.880 --> 00:39:49.437 different. And before you go on, what is your own personal view though? 00:39:49.520 --> 00:39:53.117 >> I really believe in muse. I believe in the personal agent as like a long-term 00:39:53.200 --> 00:39:55.597 form factor of just it is human nature 00:39:55.680 --> 00:39:58.157 to have wants and desires and dreams. 00:39:58.240 --> 00:40:01.997 And I think like AI and personal agents 00:40:02.080 --> 00:40:07.037 like muse will be the mechanism and the bridge that enables us to accomplish 00:40:07.120 --> 00:40:10.557 those things and continue accomplishing them. And then like humans will dream 00:40:10.640 --> 00:40:11.917 bigger and bigger and bigger and bigger. 00:40:12.000 --> 00:40:15.437 Kids will dream of like having their own star systems or whatever. And like Muse 00:40:15.520 --> 00:40:16.717 will help them do that. Like there's 00:40:16.800 --> 00:40:21.037 this thing that like um Bezos uh wrote 00:40:21.120 --> 00:40:25.197 in like well I think it was one of his shareholder letters which is like the 00:40:25.280 --> 00:40:29.597 beautiful thing about uh consumers is that they're like always beautifully 00:40:29.680 --> 00:40:33.037 unhappy or something like that or like beautifully unsatisfied with the options 00:40:33.120 --> 00:40:37.357 they have. That is like one modality of our relationship with AI. I think that 00:40:37.440 --> 00:40:41.757 there's other modalities like I think that AI will start doing more and more 00:40:41.840 --> 00:40:46.077 of the scientific discovery process. I think AIs will start inserting more and 00:40:46.160 --> 00:40:48.477 more into various parts of the economy, 00:40:48.560 --> 00:40:51.677 but I do believe in this mod in this 00:40:51.760 --> 00:40:54.477 like form factor long term of everyone 00:40:54.560 --> 00:40:58.877 having an AI that supports them and what they want. So you see AI as kind of like 00:40:58.960 --> 00:41:02.477 removing all like the cludge of life, right? The stuff that we don't want to 00:41:02.560 --> 00:41:05.357 do. I think in the the piece you said something like the world is full of like 00:41:05.440 --> 00:41:09.437 gatekeepers and obstacles and like Muse can get around this for you without 00:41:09.520 --> 00:41:11.597 spending any more of your time so you can actually focus on the stuff you want 00:41:11.680 --> 00:41:15.437 to do. I think Muse can 00:41:15.520 --> 00:41:19.517 give everyone the adulthood they dreamed 00:41:19.600 --> 00:41:21.597 of in childhood. Everyone when they're 00:41:21.680 --> 00:41:24.237 kids, they have like big ideas and big 00:41:24.320 --> 00:41:29.437 dreams and when they like imagine their life and they sort of like, you know, 00:41:29.520 --> 00:41:33.037 play it out. They they imagine this life where they can like be an astronaut or 00:41:33.120 --> 00:41:36.077 they can save the planet or they can change the world or, you know, whatever 00:41:36.160 --> 00:41:38.157 it might be. It's like a big they they 00:41:38.240 --> 00:41:41.597 think big and then for, you know, a 00:41:41.680 --> 00:41:44.477 variety of reasons by the time people 00:41:44.560 --> 00:41:48.877 finish school, they enter the workforce, by the time they get to their, you know, 00:41:48.960 --> 00:41:50.957 have been in a job for a while, it's 00:41:51.040 --> 00:41:56.557 that like all that hope and ambition and and ability to dream has been sucked out 00:41:56.640 --> 00:41:59.117 of them. >> No, it's been beaten out of them. The 00:41:59.200 --> 00:42:03.677 life is beaten beaten out of them. I think as adults, I don't think we like 00:42:03.760 --> 00:42:08.077 just like appreciate the degree to which we're all zombies. We lost our agency 00:42:08.160 --> 00:42:10.637 and we lost our I think ambition is 00:42:10.720 --> 00:42:12.797 really a good word for it. One of the 00:42:12.880 --> 00:42:16.077 promises of AI broadly speaking and I 00:42:16.160 --> 00:42:20.637 think through products like Muse is to um keep it going from when you're a kid. 00:42:20.720 --> 00:42:24.317 Like you're a kid, you have big dreams, use AI, help make those things happen. 00:42:24.400 --> 00:42:27.357 Then like dream bigger, make those things happen, dream bigger, make those 00:42:27.440 --> 00:42:31.277 things happen. and kind of like experience this like escalator of 00:42:31.360 --> 00:42:34.077 agency, an increase in agency throughout 00:42:34.160 --> 00:42:38.637 your life versus a sort of like you have lots of agency and then it gets like 00:42:38.720 --> 00:42:42.077 crushed. You know what I was thinking of when you were speaking just now? You you 00:42:42.160 --> 00:42:45.037 mentioned earlier this this experience that you and Nat Freeman went through 00:42:45.120 --> 00:42:47.517 where you had this like disturbing uh 00:42:47.600 --> 00:42:52.077 psychological audit and like coming in waves from this AI. It's like yeah, what 00:42:52.160 --> 00:42:53.757 if we had that but it's like the opposite of what you're saying. It's 00:42:53.840 --> 00:42:56.717 like actually gave you more self-confidence like gave you 00:42:56.800 --> 00:42:59.197 understanding like the world is malleable and if you push on it hard 00:42:59.280 --> 00:43:01.517 enough you go after it with enough energy and drive you can actually change 00:43:01.600 --> 00:43:04.077 it around you. >> Yeah. I think this is something that 00:43:04.160 --> 00:43:08.397 like certainly the most impressive entrepreneurs exhibit which is like I 00:43:08.480 --> 00:43:12.557 think they like dream big they accomplish that then they dream bigger 00:43:12.640 --> 00:43:16.317 and then maybe they work on that for like a decade but then like you know if 00:43:16.400 --> 00:43:19.277 they accomplish that then they dream bigger. Like Elon is obviously a great 00:43:19.360 --> 00:43:22.797 example of this. Um I think Mark is a great example of this. I think but I 00:43:22.880 --> 00:43:24.557 think a lot of like the iconic 00:43:24.640 --> 00:43:26.797 entrepreneurs this is like like kind of 00:43:26.880 --> 00:43:30.077 like their lived experience and I think 00:43:30.160 --> 00:43:31.837 there's a version of that that should be 00:43:31.920 --> 00:43:34.557 true for every person and I think what's 00:43:34.640 --> 00:43:38.157 kind of like tragic is I think for most 00:43:38.240 --> 00:43:40.637 people you have like bigger dreams big 00:43:40.720 --> 00:43:43.917 ideas you enter the corporate workforce 00:43:44.000 --> 00:43:46.557 you're you kind of become a zombie maybe 00:43:46.640 --> 00:43:50.957 at some point you want to go tackle your dreams but then it's like really really 00:43:51.040 --> 00:43:54.077 hard cuz you all these commitments. You maybe you have a family, maybe you have 00:43:54.160 --> 00:43:56.237 whatever it is. It's just like it's 00:43:56.320 --> 00:44:01.117 almost you become trapped in the sort of like zombiehood. There's one more thing 00:44:01.200 --> 00:44:05.997 on miuse before I want to get to scale AI and like the the partnership that you 00:44:06.080 --> 00:44:10.237 do that you did with Meta and making that decision. But in this essay that 00:44:10.320 --> 00:44:12.637 you wrote or this like short postage you wrote, I love what you said. said, "The 00:44:12.720 --> 00:44:15.597 world until now has been shaped by the small number of fanatics who have 00:44:15.680 --> 00:44:19.437 somehow found a way to make their wants real, but we've never seen humanity with 00:44:19.520 --> 00:44:21.437 every single person's agency fully 00:44:21.520 --> 00:44:23.997 switched on." I think the promise of 00:44:24.080 --> 00:44:27.117 this world that I'm describing where 00:44:27.200 --> 00:44:31.597 every single person has this like agency increase throughout their lifetime and 00:44:31.680 --> 00:44:33.997 they have the ability to accomplish 00:44:34.080 --> 00:44:36.157 their wants and their dreams. I think 00:44:36.240 --> 00:44:39.037 that world looks crazy in a very good 00:44:39.120 --> 00:44:44.317 way. Like I think it will be very interesting. It'll be artistic and cool 00:44:44.400 --> 00:44:47.677 and you'll like go into different pockets of the world and it will be very 00:44:47.760 --> 00:44:52.397 like diverse. It'll be kind of insane to think about what exactly that looks like 00:44:52.480 --> 00:44:56.317 where literally billions of people have, you 00:44:56.400 --> 00:44:59.437 know, all of a sudden because of abundant intelligence like the resources 00:44:59.520 --> 00:45:04.717 to make incredible things happen. I think that's part of the promise. Like I 00:45:04.800 --> 00:45:08.557 think in my head it's almost like the Rick and Morty interdimensional cable 00:45:08.640 --> 00:45:11.197 but but just like somehow manifested 00:45:11.280 --> 00:45:13.677 into like reality for humans. Like it 00:45:13.760 --> 00:45:19.037 could just be really awesome. So I keep hearing uh this line from that show 00:45:19.120 --> 00:45:22.317 where it's like the universe eats smart people. 00:45:22.400 --> 00:45:27.757 >> Um okay so I want to go back. What I'm personally curious uh about is you're 00:45:27.840 --> 00:45:31.357 you founded Scale AI, you're running Scale AI, and then one day Mark 00:45:31.440 --> 00:45:34.557 Zuckerberg reaches out and says, "Hey," I assume he says, "Hey, I want to talk. 00:45:34.640 --> 00:45:37.917 Can you tell me about this?" >> Yeah, I think the exact message was 00:45:38.000 --> 00:45:39.997 something like, "Hey, do you have time for a call?" 00:45:40.080 --> 00:45:42.637 >> Did you have a relationship with Mark previously? Did you spend any time with 00:45:42.720 --> 00:45:47.277 him? Like what was what's the background there? I had a friend who uh this guy 00:45:47.360 --> 00:45:49.357 Alex Schultz uh who I've known actually 00:45:49.440 --> 00:45:52.877 for many many years um from when scale 00:45:52.960 --> 00:45:54.957 was maybe a year old. I met him in San 00:45:55.040 --> 00:45:56.797 Francisco is currently the chief data 00:45:56.880 --> 00:46:00.237 officer at Meta but was longtime Meta 00:46:00.320 --> 00:46:04.477 executive lieutenant and he introduced me to Mark 00:46:04.560 --> 00:46:06.317 I want to say in about like I think it 00:46:06.400 --> 00:46:11.757 was 2021 or 2022 to talk about AI because I think at that time Mark was 00:46:11.840 --> 00:46:14.397 getting a lot more interested in AI scale. We were doing a lot of stuff in 00:46:14.480 --> 00:46:16.477 AI. My memory is that I think it took 00:46:16.560 --> 00:46:22.157 like 1 year to schedule that first meeting. Like from from the intro to the 00:46:22.240 --> 00:46:25.757 meeting being scheduled, I think it was like a full year. Why? Mark obviously 00:46:25.840 --> 00:46:28.797 has like an insane calendar and like has like a bajillion different things to 00:46:28.880 --> 00:46:31.437 deal with. And he's also very good at 00:46:31.520 --> 00:46:34.877 prioritization. Like he will spend a lot 00:46:34.960 --> 00:46:39.197 of time on the things that are really really important in some ways like too 00:46:39.280 --> 00:46:42.237 much time on the things that are like really really important. the only way 00:46:42.320 --> 00:46:45.677 he's able to do that and like be a good dad and be a good partner and all this 00:46:45.760 --> 00:46:50.237 other stuff is to just like ruthlessly prioritize. I had spoken to him a few 00:46:50.320 --> 00:46:54.317 times kind of after that and had gotten his advice as a founder to get advice 00:46:54.400 --> 00:46:58.317 from you know Mark Zuckerberg obviously like a really big deal. Um and I 00:46:58.400 --> 00:46:59.837 actually remember at that time like he 00:46:59.920 --> 00:47:03.197 he wrote these like like I would ask him 00:47:03.280 --> 00:47:07.597 a question on WhatsApp and then he would write these like really long responses 00:47:07.680 --> 00:47:11.277 and I was really confused. I was like, how does he have time to write these 00:47:11.360 --> 00:47:12.877 like really long responses? But now I 00:47:12.960 --> 00:47:15.917 realize he like is just like really fast 00:47:16.000 --> 00:47:20.157 and rigorous at writing to like a lot of WhatsApp messages. He'll write like 00:47:20.240 --> 00:47:23.917 write a very rigorous like long response very quickly. It's like a very 00:47:24.000 --> 00:47:26.797 impressive skill of his. >> Gives you a sneak peek in what's going on his mind. 00:47:26.880 --> 00:47:29.597 >> Yeah. Like yeah, there's racing. Meta 00:47:29.680 --> 00:47:32.317 was a important customer of scale. Um, 00:47:32.400 --> 00:47:37.357 and I would say like prior to this instance, maybe we spoke once every like 00:47:37.440 --> 00:47:41.837 six months or so. Maybe that was like the cadence of our interactions. And 00:47:41.920 --> 00:47:43.997 then yeah, I think he he had called and 00:47:44.080 --> 00:47:46.717 I think it was it was actually like not 00:47:46.800 --> 00:47:50.317 obvious at first what the like end point 00:47:50.400 --> 00:47:54.957 was going to be because the first conversation was just like, you know, 00:47:55.040 --> 00:47:59.437 this was after Llama 4 and it was kind of just like asking what do you think we 00:47:59.520 --> 00:48:02.397 should be doing? Why is that an important point that this is this 00:48:02.480 --> 00:48:04.237 conversation is happening after Llama 4? 00:48:04.320 --> 00:48:09.197 >> Llama 4 was not on the trajectory that Meta wanted as a company. It definitely 00:48:09.280 --> 00:48:12.797 was like I think internally speaking a disappointment and it was also at a time 00:48:12.880 --> 00:48:17.837 where AI was clearly becoming just like increasingly increasingly important. It 00:48:17.920 --> 00:48:20.637 was clear that AI was going to be like really critical to the future of meta. 00:48:20.720 --> 00:48:24.797 So yeah the first few conversations we spoke on the phone and he was sort of as 00:48:24.880 --> 00:48:27.997 saying you know asking for advice on like what do you think we should be 00:48:28.080 --> 00:48:30.397 doing? What do you think we should be focused on? This is what I've heard 00:48:30.480 --> 00:48:34.317 about him privately that he has this insane it's not a board of directors 00:48:34.400 --> 00:48:37.437 like a meta board although there's some people on the board that he also does 00:48:37.520 --> 00:48:41.197 this with but he's got this group of world-class entrepreneurs around him and 00:48:41.280 --> 00:48:43.597 I've spoken to some of them like you don't understand like he asked for 00:48:43.680 --> 00:48:46.877 advice constantly he's like hitting us up he's like here's what's going on in 00:48:46.960 --> 00:48:50.477 my life or the work would you do and he'll do that over and over and over 00:48:50.560 --> 00:48:54.237 again I think this is like quite an impressive trait yes because it takes 00:48:54.320 --> 00:48:58.637 humility obviously like continue doing that even you know after all the 00:48:58.720 --> 00:49:01.597 incredible things that, you know, he's been able to accomplish. I mean, the 00:49:01.680 --> 00:49:04.557 whole, you know, timeline from when he 00:49:04.640 --> 00:49:07.917 first reached out to when we announced 00:49:08.000 --> 00:49:12.237 the deal was like five, six weeks. Like, it was like pretty quick. Another thing 00:49:12.320 --> 00:49:15.277 that world-class entrepreneurs have in common. We had Jonathan Ross, the 00:49:15.360 --> 00:49:19.197 founder of Grock, on this podcast. The first time we ever spoke uh publicly 00:49:19.280 --> 00:49:23.197 about the $20 billion deal he did with Nvidia was on the show. And he's like, 00:49:23.280 --> 00:49:25.437 "From the time first call from Jensen to 00:49:25.520 --> 00:49:27.437 money's in my bank account was 3 weeks." 00:49:27.520 --> 00:49:30.237 >> Yeah. That's amazing. Um, I just had 00:49:30.320 --> 00:49:34.717 lots of ideas on like what I'd be doing if I choose and like how I'd be thinking 00:49:34.800 --> 00:49:37.677 about Meta's AI strategy. One of the first things I said was what I said 00:49:37.760 --> 00:49:43.117 earlier, which is like, hey, I actually think because Meta is like heritage wise 00:49:43.200 --> 00:49:48.397 focus on the things that people want to do is actually one of the most specially 00:49:48.480 --> 00:49:49.997 placed companies with this incredible 00:49:50.080 --> 00:49:53.357 shift AI. I like had lots and lots of 00:49:53.440 --> 00:49:57.917 ideas that I sort of sent his way and then that was kind of this like 00:49:58.000 --> 00:50:02.237 conversation happening in parallel with this like oh yeah maybe we should 00:50:02.320 --> 00:50:04.557 potentially consider if there's a way to work together. 00:50:04.640 --> 00:50:07.997 >> So when he says that what do you think cuz this kid you this was not 00:50:08.080 --> 00:50:10.957 predictable to you. >> Not predictable. No. The whole thing was 00:50:11.040 --> 00:50:15.757 like it was like a really crazy um sequence at the time. I was just like, 00:50:15.840 --> 00:50:18.237 "Oh, that's nice, but like there's no way that 00:50:18.320 --> 00:50:19.437 >> Well, you know what that means when they 00:50:19.520 --> 00:50:21.997 say that, right? They want to buy you. 00:50:22.080 --> 00:50:24.637 It's always the same thing, but they say it's like, "Oh, we got to find a way to 00:50:24.720 --> 00:50:27.597 work together." >> Yeah. I don't think I'd had that that 00:50:27.680 --> 00:50:30.957 kind of coaching at the at the time. One 00:50:31.040 --> 00:50:34.797 thing that I explicitly felt at the time was like, "Yeah, like it probably 00:50:34.880 --> 00:50:38.317 doesn't make that much sense even." Like I think I personally was like, "Yeah, 00:50:38.400 --> 00:50:43.357 like you know, maybe he's like, you know, teasing it or or like throwing it 00:50:43.440 --> 00:50:45.597 out there, but like does he even like make that much sense?" 00:50:45.680 --> 00:50:47.037 >> Why wouldn't you think it makes sense? 00:50:47.120 --> 00:50:50.797 >> I mean, obviously like a deal ended up happening, so you know, what do I know? 00:50:50.880 --> 00:50:52.797 But um >> No, no, but back then, not we know what 00:50:52.880 --> 00:50:55.917 happened, but like I'm very curious your thinking as you're going experiencing 00:50:56.000 --> 00:50:59.117 this like the fact that you said, "Oh, yeah, you said this was like that 00:50:59.200 --> 00:51:01.197 doesn't make sense." Like why wouldn't it make sense? 00:51:01.280 --> 00:51:05.677 >> Taking a step back. Obviously the the like iconic meta acquisitions have been 00:51:05.760 --> 00:51:07.837 Instagram and WhatsApp and these were 00:51:07.920 --> 00:51:11.037 like very 00:51:11.120 --> 00:51:14.877 clear you know product logic driven 00:51:14.960 --> 00:51:19.037 decisions. Scale is like you know a deep 00:51:19.120 --> 00:51:21.917 enterprise government sales like you 00:51:22.000 --> 00:51:23.517 know a entirely different kind of 00:51:23.600 --> 00:51:27.437 business from from Meta. And so the the 00:51:27.520 --> 00:51:29.757 logic that I had thought at the time 00:51:29.840 --> 00:51:35.037 would have made sense is if meta wanted to get into those things that then I 00:51:35.120 --> 00:51:38.237 think you know like you know maybe that's that's the industrial logic that 00:51:38.320 --> 00:51:39.677 would have made sense but this wasn't 00:51:39.760 --> 00:51:42.317 that cuz I think Mark clearly was like 00:51:42.400 --> 00:51:47.357 predominantly interested in getting the llama program and over his overall AI 00:51:47.440 --> 00:51:48.717 program on >> he wanted talent. 00:51:48.800 --> 00:51:52.077 >> Yeah he wanted talent. It was kind of shrouded and unclear the whole time, you 00:51:52.160 --> 00:51:55.597 know, a lot of the way. >> When did it become clear? After the 00:51:55.680 --> 00:51:58.637 money hit the account when, you know, 00:51:58.720 --> 00:52:00.477 when we started talking numbers, I was 00:52:00.560 --> 00:52:06.237 like, "Oh, wow. Okay. All right." Um, that's when things became like more 00:52:06.320 --> 00:52:09.437 clear. >> At first, you're like, "Oh, this this will make sense." Then you guys keep 00:52:09.520 --> 00:52:13.517 talking. And how fast did he convince you like this is a good path for you? 00:52:13.600 --> 00:52:14.877 >> There was a part of it was just like, 00:52:14.960 --> 00:52:18.077 "Wow, this is like a really fascinating 00:52:18.160 --> 00:52:20.637 deal construct." We ultimately landed on 00:52:20.720 --> 00:52:25.037 this deal where you know Meta invested 00:52:25.120 --> 00:52:28.797 uh owns 49% of scale scale continues. 00:52:28.880 --> 00:52:31.437 Myself and a few people join join Meta. 00:52:31.520 --> 00:52:32.957 What I was kind of incredulous about at 00:52:33.040 --> 00:52:35.757 the time was like it's such a weird kind 00:52:35.840 --> 00:52:39.597 of deal but is one that I think 00:52:39.680 --> 00:52:43.437 genuinely kind of was like this very 00:52:43.520 --> 00:52:46.317 interesting winwin win so to speak. 00:52:46.400 --> 00:52:49.997 There's a win for the shareholders of scale because all shareholders of scale 00:52:50.080 --> 00:52:52.877 got a great deal, benefited a lot. Scale 00:52:52.960 --> 00:52:57.117 continues and I truly believe the best days of scale are ahead of it. My 00:52:57.200 --> 00:52:58.557 question to you is like or what I'm 00:52:58.640 --> 00:53:00.477 personally interested in is like okay is 00:53:00.560 --> 00:53:04.317 this he can explain this to you and you 00:53:04.400 --> 00:53:08.157 come around to that perspective in one phone call. Is this just like wait Mark 00:53:08.240 --> 00:53:10.157 I got to think this is so [__] crazy. 00:53:10.240 --> 00:53:13.197 I got to think about this for a few days. Like explain this process as much 00:53:13.280 --> 00:53:16.157 as you can. many weeks of being like, 00:53:16.240 --> 00:53:18.877 "This is insane. Is this even real? Does 00:53:18.960 --> 00:53:21.277 this make any sense? If it does make 00:53:21.360 --> 00:53:24.157 sense, like how do I feel about it?" The 00:53:24.240 --> 00:53:28.877 predominant emotion was more like this is kind of insane. And then as I 00:53:28.960 --> 00:53:31.357 thought more about like, oh, I actually have to make a decision of whether or 00:53:31.440 --> 00:53:33.837 not I do this. That was it in and of 00:53:33.920 --> 00:53:38.797 itself. There were like a lot of like conversations there with the people at 00:53:38.880 --> 00:53:41.677 scale, with our investors. Like there were a lot of interesting conversations 00:53:41.760 --> 00:53:44.557 in in that uh on that side. I mean it's 00:53:44.640 --> 00:53:49.277 obviously like really hard to to like let go of or give up your baby 00:53:49.360 --> 00:53:50.557 >> cuz you worked on scale for how long? 00:53:50.640 --> 00:53:53.677 How many years? >> Yeah. 9 years from founding until the 00:53:53.760 --> 00:53:56.717 >> And you're still young. You're not even 30 yet. So this is like a third of your 00:53:56.800 --> 00:53:58.557 life. >> There was something that Paul Graham 00:53:58.640 --> 00:54:02.237 said for a long time which is like you know if you think of your company as 00:54:02.320 --> 00:54:04.637 your life's work like you will operate 00:54:04.720 --> 00:54:09.677 differently. I did really genuinely think of of scales my life's work for 00:54:09.760 --> 00:54:13.037 the whole time that I was I was working on it. That was a tough emotional 00:54:13.120 --> 00:54:14.957 process. And then I think what 00:54:15.040 --> 00:54:16.797 ultimately got me was like a combination 00:54:16.880 --> 00:54:21.277 of wow this is like a win-win winwin 00:54:21.360 --> 00:54:24.477 like kind of all um it's it's a good 00:54:24.560 --> 00:54:26.797 deal for all parties and also just this 00:54:26.880 --> 00:54:29.197 like I saw the potential of what could 00:54:29.280 --> 00:54:32.237 happen at Meta. And you know, this is a 00:54:32.320 --> 00:54:36.317 point at which like I think meta on AI 00:54:36.400 --> 00:54:40.557 certainly looked like damaged goods in many ways 00:54:40.640 --> 00:54:42.637 and like was maybe not the most 00:54:42.720 --> 00:54:46.397 appealing place to work on AI, but I 00:54:46.480 --> 00:54:51.277 think that kind of like triggered my entrepreneur side where I was like 00:54:51.360 --> 00:54:54.557 >> you can essentially rebuild it. It's like a refounding of the lab. 00:54:54.640 --> 00:54:59.037 >> Yeah. Yeah. Like I think this is this is one of the things that ended up being 00:54:59.120 --> 00:55:03.037 kind of fascinating which is like because 00:55:03.120 --> 00:55:05.037 it was you know it was clear that there 00:55:05.120 --> 00:55:07.357 was so much work to do. It really like 00:55:07.440 --> 00:55:12.477 cleared the way for me to be able to to kind of build a lot of stuff up from 00:55:12.560 --> 00:55:16.077 scratch, set up the right principles, build the right culture. Like it it 00:55:16.160 --> 00:55:19.437 created a the conditions for us to build something amazing. 00:55:19.520 --> 00:55:20.877 >> And is that how you and Mark discuss it? 00:55:20.960 --> 00:55:24.957 It's like hey we have to hit a reset here. Like obviously like we're not if 00:55:25.040 --> 00:55:28.077 we keep on this path like we're essentially going to throw out what we 00:55:28.160 --> 00:55:32.957 have and redo it with a completely different uh like level of talent. Are 00:55:33.040 --> 00:55:36.157 these the conversations that are that occurring between you and him? It didn't 00:55:36.240 --> 00:55:39.357 seem like he wanted like a minor adjustment. You understand what I'm 00:55:39.440 --> 00:55:42.637 saying? He's just like, "Oh, this is not working at all. It's not going to work. 00:55:42.720 --> 00:55:46.557 Then let's, you know, rip it down to the foundation and rebuild." Correct. It was 00:55:46.640 --> 00:55:49.757 an evolving conversation where I think there were like like one of the things 00:55:49.840 --> 00:55:51.917 that I believe really strongly I think 00:55:52.000 --> 00:55:54.397 Elon is like has has demonstrated this 00:55:54.480 --> 00:55:57.437 in in various times in his career where 00:55:57.520 --> 00:56:01.757 you just have like a small incredibly cracked highly technical team that's 00:56:01.840 --> 00:56:05.597 like very flat and like if you do that you can accomplish a lot very very 00:56:05.680 --> 00:56:08.317 quickly. So that was something that I 00:56:08.400 --> 00:56:12.877 had a lot of conviction in and like I believed in AI like that is the right 00:56:12.960 --> 00:56:16.157 approach especially if you had to you know do what we had to do which is like 00:56:16.240 --> 00:56:18.957 move very very quickly and I think early 00:56:19.040 --> 00:56:24.317 on that was one of the things I talked to Mark a lot about and I think he was 00:56:24.400 --> 00:56:25.917 excited about as well that ended up 00:56:26.000 --> 00:56:28.237 becoming one of the tent poles of our 00:56:28.320 --> 00:56:31.437 overall strategies like small very flat 00:56:31.520 --> 00:56:35.517 highly technical team very high talent density how quickly can you move if you 00:56:35.600 --> 00:56:37.997 have those ingredients From an outside perspective, it just 00:56:38.080 --> 00:56:41.917 seems like Mark like empties the clip is the way I think about this, right? Where 00:56:42.000 --> 00:56:44.077 um have you ever read this book called The Mind of Napoleon? 00:56:44.160 --> 00:56:45.677 >> No. >> Okay. So, mind and Napoleon published in 00:56:45.760 --> 00:56:49.757 1957. Um I found it because I saw this interview. I think Tyler Cowan 00:56:49.840 --> 00:56:53.037 interviewed Sam Alman. And uh I think this is 2018 and he asked him like 00:56:53.120 --> 00:56:55.277 what's the most important book that you read this year? And he said the mind of 00:56:55.360 --> 00:56:58.717 Napoleon. So it's very hard to find. I buy the book. It's 300 pages of just 00:56:58.800 --> 00:57:01.997 Napoleon his own words organized by topic. And in that he talks about over 00:57:02.080 --> 00:57:03.757 and over again that hesitation is fatal. 00:57:03.840 --> 00:57:07.357 But he says before you engage in a course of action, there's a lot of 00:57:07.440 --> 00:57:11.037 deliberation. He'll like study from every angle, make sure he's making the 00:57:11.120 --> 00:57:15.437 right decision. I feel there's like a echo with the way Mark operates. Does a 00:57:15.520 --> 00:57:18.397 lot of deliberation, make sure he's on the right path. But once he makes that 00:57:18.480 --> 00:57:21.837 decision, he just goes all in. He won't like he just empties the clip. And 00:57:21.920 --> 00:57:25.757 that's the way I felt about like what was was occurring right at this point 00:57:25.840 --> 00:57:28.237 where he comes and gets you and then you guys start rebuilding this entire 00:57:28.320 --> 00:57:30.877 organization. I think he really has 00:57:30.960 --> 00:57:33.517 internalized that failure doesn't 00:57:33.600 --> 00:57:35.757 matter. It matters like when you win, 00:57:35.840 --> 00:57:40.557 how big you win. And I think this overall concept I think enables him to 00:57:40.640 --> 00:57:44.717 operate I think truly without fear in a lot of circumstances because I think he 00:57:44.800 --> 00:57:46.557 is just very comfortable taking lots of risk. 00:57:46.640 --> 00:57:48.477 >> He's got that great line where it's like 00:57:48.560 --> 00:57:51.597 in a world that's changing all the time, the biggest risk is not taking any. 00:57:51.680 --> 00:57:53.597 >> Yeah, exactly. I think he's internalized 00:57:53.680 --> 00:57:56.797 this quite deeply. Like there's certainly like versions of the world 00:57:56.880 --> 00:58:01.757 where Meta looks a lot more like Google in some sense like it's just like you 00:58:01.840 --> 00:58:06.237 know more slowm moving and more bureaucratic and just like looks very 00:58:06.320 --> 00:58:09.837 different as a company but I think he because he's internalized some of these 00:58:09.920 --> 00:58:11.757 lessons so deeply keeps the thing 00:58:11.840 --> 00:58:14.397 dynamic and moving and live. talk me 00:58:14.480 --> 00:58:18.877 through the difference between how like how you were as like you're you're the 00:58:18.960 --> 00:58:20.797 founder, you're running scale, now 00:58:20.880 --> 00:58:24.157 you're refounding this lab, working with 00:58:24.240 --> 00:58:28.317 Mark, like the different ways that you approach your work in those two uh 00:58:28.400 --> 00:58:33.117 environments at scale. One of the things like I really 00:58:33.200 --> 00:58:36.237 internalized and I don't even remember where this advice came from, but it was 00:58:36.320 --> 00:58:40.157 kind of one of these ideas that like if you want to be a good founder, you have 00:58:40.240 --> 00:58:43.677 to be able to do every job. had a very strong point of view about how like 00:58:43.760 --> 00:58:47.197 every little thing at the company should happen and should work. And that 00:58:47.280 --> 00:58:49.117 resulted in a company that was like very 00:58:49.200 --> 00:58:53.277 much so a product of my effort and like 00:58:53.360 --> 00:58:56.637 my point of view and it like expressed itself in all sorts of ways throughout 00:58:56.720 --> 00:59:01.197 the company but also made me the bottleneck in very big ways. There's 00:59:01.280 --> 00:59:04.957 like pros and cons of that for a company. I think definitely I think 00:59:05.040 --> 00:59:07.837 that's like you know if you were to start a company and it's your first time 00:59:07.920 --> 00:59:09.757 starting a company you should like definitely take that approach because 00:59:09.840 --> 00:59:12.637 it's like much more likely to be successful than if you don't take that. 00:59:12.720 --> 00:59:15.197 >> Scale your first company. >> Scale is my first company. 00:59:15.280 --> 00:59:17.517 >> Yeah. But at Meta like it was a very 00:59:17.600 --> 00:59:19.917 different assignment because we had to 00:59:20.000 --> 00:59:23.357 accomplish so much in so little time. 00:59:23.440 --> 00:59:27.437 Like I didn't have the luxury of being able to quote unquote do every job 00:59:27.520 --> 00:59:30.237 within the lab. I 00:59:30.320 --> 00:59:32.877 uh certainly am not an AI researcher. 00:59:32.960 --> 00:59:35.757 I'm also not a like product designer. 00:59:35.840 --> 00:59:37.197 I'm not you know I'm not all these 00:59:37.280 --> 00:59:40.477 things that are like totally pivotal to 00:59:40.560 --> 00:59:43.037 um the organization being successful. So 00:59:43.120 --> 00:59:48.317 at Meta very much so I adopted this like 00:59:48.400 --> 00:59:52.077 philosophy around just how do we create 00:59:52.160 --> 00:59:54.477 an environment where we're going to hire 00:59:54.560 --> 00:59:57.277 incredible people like it was like job 00:59:57.360 --> 01:00:01.197 number one hire brilliant and incredible 01:00:01.280 --> 01:00:04.637 people like who are at the top of their field across every discipline. 01:00:04.720 --> 01:00:07.197 >> Why do you think you're able to do that at Meta? Is it just because you have 01:00:07.280 --> 01:00:09.037 more resources? Like why can you do 01:00:09.120 --> 01:00:13.597 that? I think one of the things that was 01:00:13.680 --> 01:00:15.757 charis very charismatic and attractive 01:00:15.840 --> 01:00:20.637 about the founding moment of MSL was 01:00:20.720 --> 01:00:23.837 there was an opportunity to build a lab 01:00:23.920 --> 01:00:29.357 from scratch with a really really small team and we had a strong point of view 01:00:29.440 --> 01:00:33.437 of where we're going personal super intelligence that I think resonated with 01:00:33.520 --> 01:00:37.037 people like I think everyone who works on AI wants to help people like wants it 01:00:37.120 --> 01:00:40.397 to ultimately like mean something to their mom or to their grandma or to 01:00:40.480 --> 01:00:44.157 their grandpa wants it to like be something that's like meaningful to 01:00:44.240 --> 01:00:48.237 every single person in the world because of the fact that we're doing this with 01:00:48.320 --> 01:00:51.997 the resources of meta. There was an opportunity to we weren't going to be 01:00:52.080 --> 01:00:54.957 bottlenecked by comput. We weren't going to be bottlenecked by like 01:00:55.040 --> 01:00:57.837 infrastructure. We weren't going to be bottlenecked by by distribution. like we 01:00:57.920 --> 01:00:59.917 weren't going to be bottlenecked by the things that like startups often are 01:01:00.000 --> 01:01:03.677 bottlenecked, but it was an opportunity to like really build something that you 01:01:03.760 --> 01:01:08.477 could like put your stamp on and have like a lot of influence over and really 01:01:08.560 --> 01:01:11.117 shape. I think that was like quite exciting for a lot of people. 01:01:11.200 --> 01:01:14.397 >> And so that's why he chooses to spend an insane amount of money, what looks like 01:01:14.480 --> 01:01:18.717 an insane amount of money, getting a handful of top talent because then he 01:01:18.800 --> 01:01:23.357 knows that that those top talent will recruit in turn other top talent. But 01:01:23.440 --> 01:01:25.277 you need that as a starting point. 01:01:25.360 --> 01:01:27.517 Correct. >> Yeah. Yeah. And I think one thing that 01:01:27.600 --> 01:01:29.117 like is maybe somewhat underappreciated 01:01:29.200 --> 01:01:33.277 about that specific part is that like I 01:01:33.360 --> 01:01:36.397 mean the stock prices of OpenAI and 01:01:36.480 --> 01:01:40.157 Anthropic like ran up a lot. So it just 01:01:40.240 --> 01:01:42.317 happened to be the case that like you 01:01:42.400 --> 01:01:44.557 know if you looked at how much these 01:01:44.640 --> 01:01:47.597 people were making if they stayed at 01:01:47.680 --> 01:01:52.237 Open Eye or Enthropic like it also was like it was a lot of money. So I think 01:01:52.320 --> 01:01:55.437 for a lot of the talent it ended up being like kind of net neutral in many 01:01:55.520 --> 01:01:59.437 ways. um compwise, but obviously I think it's kind of like the same thing like 01:01:59.520 --> 01:02:02.797 oftentimes we don't we don't a lot of us don't think about how much like the 01:02:02.880 --> 01:02:05.437 early employees or the early researchers 01:02:05.520 --> 01:02:07.997 at these like labs that have had immense 01:02:08.080 --> 01:02:10.637 stock uh stock value appreciation you 01:02:10.720 --> 01:02:13.037 know what their comp looks like we 01:02:13.120 --> 01:02:15.117 started this like fully on the belief of 01:02:15.200 --> 01:02:17.917 having and hiring like most brilliant 01:02:18.000 --> 01:02:20.557 people that's one Slar personal super 01:02:20.640 --> 01:02:23.037 intelligence um those two how do we 01:02:23.120 --> 01:02:25.437 create the environment where people do 01:02:25.520 --> 01:02:30.077 their best work at Meta and in MSL. I'm 01:02:30.160 --> 01:02:34.317 doing much more of like quote unquote traditional management, so to speak. Cuz 01:02:34.400 --> 01:02:36.797 these are the things I'm like thinking a lot about is like how do I create the 01:02:36.880 --> 01:02:39.037 environment where all these brilliant 01:02:39.120 --> 01:02:42.237 people who are exceptional and very very 01:02:42.320 --> 01:02:46.077 special can all operate and achieve the 01:02:46.160 --> 01:02:48.157 greatest expression of their talent. I'm 01:02:48.240 --> 01:02:49.837 thinking a lot more like a almost like a 01:02:49.920 --> 01:02:53.437 coach in some in some ways versus at 01:02:53.520 --> 01:02:57.117 scale it was like very different like I was the aur I was like the person whose 01:02:57.200 --> 01:02:59.837 like point of view had to be expressed in everything that we did. 01:02:59.920 --> 01:03:01.277 >> Say more about that you're you feel like 01:03:01.360 --> 01:03:05.757 a coach. I think that like my job is to 01:03:05.840 --> 01:03:09.677 a like help set the like north star and 01:03:09.760 --> 01:03:13.917 the like directionality of where we want to go, which again personal super 01:03:14.000 --> 01:03:16.157 intelligence, winning on consumer AI, 01:03:16.240 --> 01:03:18.957 like building AI that means something to 01:03:19.040 --> 01:03:22.237 every person and then spotting in the 01:03:22.320 --> 01:03:26.317 team special talent and then giving 01:03:26.400 --> 01:03:30.797 people the environment and resources to be able to express that maximally. Like 01:03:30.880 --> 01:03:35.117 I think one thing that big companies often mess up is that there's like 01:03:35.200 --> 01:03:39.117 exceptional people at most big companies but they just like are not empowered to 01:03:39.200 --> 01:03:42.637 do their best work and that's why they often leave to go to smaller companies 01:03:42.720 --> 01:03:44.557 or to startups or whatever it might be. 01:03:44.640 --> 01:03:47.517 You mentioned Jeff Bezos's shareholder letters earlier. That's something he 01:03:47.600 --> 01:03:50.237 would repeat as well and all the books I read about him but also shareholder 01:03:50.320 --> 01:03:52.717 letters like if you have great people that can't build they're going to leave. 01:03:52.800 --> 01:03:55.517 >> Yeah, exactly. And AI is such like a 01:03:55.600 --> 01:04:00.637 it's like such a special thing. Like I think you know this term research now it 01:04:00.720 --> 01:04:02.157 it feels like it almost doesn't mean 01:04:02.240 --> 01:04:05.677 anything but like by definition humanity 01:04:05.760 --> 01:04:09.277 doesn't know the limits of what these 01:04:09.360 --> 01:04:12.797 like models are capable of and what are the things you can do with them. Like 01:04:12.880 --> 01:04:15.597 they really are these like unknown 01:04:15.680 --> 01:04:19.997 artifacts that we don't understand super world. So like it really is science and 01:04:20.080 --> 01:04:24.237 research. Like we are exploring the limits of what you can do with this 01:04:24.320 --> 01:04:25.437 technology and what are the things that 01:04:25.520 --> 01:04:27.997 can be built and it is a process that 01:04:28.080 --> 01:04:33.277 requires lots of trial and error. It requires like brilliant people to have 01:04:33.360 --> 01:04:37.037 incredible insight. Like it's one of these things that you can't think about 01:04:37.120 --> 01:04:39.757 this process like a very opishly intense 01:04:39.840 --> 01:04:42.957 process. Like people need space and time 01:04:43.040 --> 01:04:47.677 and resources to be able to like you know do their best work. And I think 01:04:47.760 --> 01:04:49.837 about in many dimensions, right? Like I 01:04:49.920 --> 01:04:53.197 think Nat Freiedman is absolutely 01:04:53.280 --> 01:04:58.077 brilliant and as I mentioned like so much of Muse if there were a single 01:04:58.160 --> 01:05:02.317 person whose like single point of view came through on Muse it is now it's 01:05:02.400 --> 01:05:05.917 important for him to be in a position where where he can like fully express 01:05:06.000 --> 01:05:10.157 all of that. And then we have some brilliant researchers. They have like 01:05:10.240 --> 01:05:12.877 very very exciting ideas. How do we create an environment where they are 01:05:12.960 --> 01:05:16.477 able to explore those? And like most of them won't work out. Some of them will 01:05:16.560 --> 01:05:18.157 work out incredibly well. The cases 01:05:18.240 --> 01:05:20.637 where that works out justify all the 01:05:20.720 --> 01:05:23.757 investment. So there's two kinds of 01:05:23.840 --> 01:05:27.997 problems so to speak and that analogy is like diamond mining or building 01:05:28.080 --> 01:05:32.557 skyscrapers and it depends on what the distribution of the payoff of the things 01:05:32.640 --> 01:05:37.117 that you're working on are. So there's like some areas where the distribution 01:05:37.200 --> 01:05:40.797 is super power. Like most of the things you do will be kind of useless, but then 01:05:40.880 --> 01:05:44.157 some of the things you do can be like literally a million times more valuable 01:05:44.240 --> 01:05:47.677 than everything else. VC I think is a great like diamond mining kind of 01:05:47.760 --> 01:05:49.837 industry to give a sense. It's just all 01:05:49.920 --> 01:05:54.877 about like being able to identify the special ideas and like see those through 01:05:54.960 --> 01:05:58.717 >> and that's easy to do. Just look at just go to your 19th birthday party. 01:05:58.800 --> 01:06:01.437 >> Yeah. I mean that's like unbelievable. 01:06:01.520 --> 01:06:02.877 Um >> do people listening that don't know what 01:06:02.960 --> 01:06:06.157 the hell I just referenced? Can you talk about real quick what who was at your 01:06:06.240 --> 01:06:08.797 19th birthday party and why that was would have been a good bet if they just 01:06:08.880 --> 01:06:10.957 invested everybody around the table? 01:06:11.040 --> 01:06:14.557 >> This is like a really crazy thing to think about. my 19th birthday party. I 01:06:14.640 --> 01:06:18.637 was a freshman at MIT. It was in January. I was doing a win internship, a 01:06:18.720 --> 01:06:21.517 winter internship, win internship at a 01:06:21.600 --> 01:06:26.477 trading firm called HRT. There were like 10 of us, 12 of us, um, something like 01:06:26.560 --> 01:06:30.877 that. Almost all of whom were students at either Harvard or MIT. Most of us 01:06:30.960 --> 01:06:35.197 already knew each other from various math and science or computer science 01:06:35.280 --> 01:06:36.877 competitions. Some of the other people 01:06:36.960 --> 01:06:40.557 were Jeffrey Yan who went on to start 01:06:40.640 --> 01:06:44.077 Hyperlquid very successful. Um Scott Woo 01:06:44.160 --> 01:06:46.397 went on to start Cognition very 01:06:46.480 --> 01:06:48.957 successful. Another person there was uh 01:06:49.040 --> 01:06:51.837 uh Jesse Zang went to start Decagon also 01:06:51.920 --> 01:06:54.077 quite successful. Um another person was 01:06:54.160 --> 01:06:57.197 Vicky Yei who is I think a brilliant 01:06:57.280 --> 01:06:59.517 researcher works at Anthropic. Everyone 01:06:59.600 --> 01:07:01.357 was like incredibly talented. I actually 01:07:01.440 --> 01:07:04.557 talked to the uh the founder of HRT 01:07:04.640 --> 01:07:09.837 recently and HRT itself by the way now is like crushing it. Unbelievable. 01:07:09.920 --> 01:07:10.957 >> What does HRT do? 01:07:11.040 --> 01:07:13.037 >> HRT it's a proprietary trading firm. So 01:07:13.120 --> 01:07:17.037 they they trade their own money in Q2 of 01:07:17.120 --> 01:07:18.717 2020 01:07:18.800 --> 01:07:23.037 um 6 had 11 billion of revenue and like 01:07:23.120 --> 01:07:25.437 I think 8 billion of profit is the the 01:07:25.520 --> 01:07:30.077 record number. This is a I think it's under a thousand people work at HRT. I 01:07:30.160 --> 01:07:34.157 mean, it's like unbelievable. But I I recently uh met up with the founder of 01:07:34.240 --> 01:07:36.557 HRT and they were talking about like, "Oh yeah, we're starting to do some 01:07:36.640 --> 01:07:38.237 private investments." I was like, "Just 01:07:38.320 --> 01:07:42.077 invest into the interns. All the interns 01:07:42.160 --> 01:07:45.277 get seed checks." 01:07:45.360 --> 01:07:50.557 That's great. >> I grew up doing uh math and uh science 01:07:50.640 --> 01:07:52.237 and computer science competitions and 01:07:52.320 --> 01:07:55.437 Olympiads and got to know like all sorts 01:07:55.520 --> 01:07:59.757 of people from all around America doing this kind of stuff. It's like quite 01:07:59.840 --> 01:08:04.477 surreal to see the people that I did competitions with become so successful. 01:08:04.560 --> 01:08:07.677 >> It's going to be fascinating what all of you do the next few decades. Before I 01:08:07.760 --> 01:08:10.637 interrupted you, did you have more to say about diamond mining and skyscraper 01:08:10.720 --> 01:08:14.637 stuff? Basically, it's like does it have a power law payoff where some things 01:08:14.720 --> 01:08:17.117 just like pay for like literally like a 01:08:17.200 --> 01:08:21.677 million times the investment or is it something more like building skyscrapers 01:08:21.760 --> 01:08:26.717 which I would I would describe data annotation or or or data businesses as 01:08:26.800 --> 01:08:30.157 more like this which is like Amazon Prime deliveries is another example 01:08:30.240 --> 01:08:33.837 where the payoff is pretty linear and you have to just get develop a process 01:08:33.920 --> 01:08:37.037 where you're like really really amazing at doing it. If you're building 01:08:37.120 --> 01:08:40.237 skyscrapers, then it's like a very operational business and it's all about 01:08:40.320 --> 01:08:43.677 like squeezing every ounce of efficiency 01:08:43.760 --> 01:08:47.437 out of that process and you just look at it and you figure out like I can make it 01:08:47.520 --> 01:08:51.517 like 2% more efficient this way and 1% more efficient that way and half a 01:08:51.600 --> 01:08:54.957 percent more efficient that way and you just try to squeeze every ounce of 01:08:55.040 --> 01:08:58.157 optimization out of it. And then there's 01:08:58.240 --> 01:08:59.757 diamond mining where you're kind of 01:08:59.840 --> 01:09:03.517 like, "Hey everyone, just like you know, 01:09:03.600 --> 01:09:08.157 go forth, try all your crazy ideas and like you're just nurturing the ideas 01:09:08.240 --> 01:09:10.637 that seem promising." >> That's what you feel you're doing at Meta right now. 01:09:10.720 --> 01:09:12.157 >> That's what I feel I'm doing at Meta. 01:09:12.240 --> 01:09:15.837 And and and the building server stuff generally speaking is what I felt I was 01:09:15.920 --> 01:09:20.957 doing at at scale. And that's like this big paradigm um paradigm difference. 01:09:21.040 --> 01:09:24.237 Going back to the coaching and the way the difference of how you're essentially 01:09:24.320 --> 01:09:26.797 managing now at meta compared to way scale. Why do you have so many direct 01:09:26.880 --> 01:09:29.437 reports? You have like what 100 direct reports? How many direct reports? 01:09:29.520 --> 01:09:33.357 >> Yeah, I think more than 200. Yeah. H >> how and why? 01:09:33.440 --> 01:09:35.357 >> Going back to it like one of the one of 01:09:35.440 --> 01:09:39.277 the founding thesis of MSL and and and 01:09:39.360 --> 01:09:40.957 what we want to do and like rebuilding 01:09:41.040 --> 01:09:44.397 this this lab was how do you have like 01:09:44.480 --> 01:09:49.597 really really high talent density? How do you make it like the best place for a 01:09:49.680 --> 01:09:53.997 lot of these brilliant people to do their best work and how do you build an 01:09:54.080 --> 01:09:57.997 organization that's like technically focused where the technical people are 01:09:58.080 --> 01:10:01.837 doing their best work, you know, have have the opportunity to really do their 01:10:01.920 --> 01:10:05.357 life's work. And one of the design principles around this was making it 01:10:05.440 --> 01:10:08.957 very flat. you know, effectively all the researchers that we hired into this 01:10:09.040 --> 01:10:13.597 group called TBD, they report directly to me. And it's more of a statement of 01:10:13.680 --> 01:10:17.837 the fact that like this is like anti-bureaucracy. Like TBD is like 01:10:17.920 --> 01:10:20.557 literally anti-bureaucracy. The point is 01:10:20.640 --> 01:10:23.437 we're hiring brilliant people and you 01:10:23.520 --> 01:10:28.237 all have like demonstrated clearly that you can do brilliant, brilliant work. 01:10:28.320 --> 01:10:31.757 You have like done it before in your careers. So I don't need to be 01:10:31.840 --> 01:10:34.237 micromanaging you to do brilliant work. 01:10:34.320 --> 01:10:38.397 you're going to be able to do that on your own. We need technical leadership. 01:10:38.480 --> 01:10:41.277 So, some of you are going to be responsible for like setting clear 01:10:41.360 --> 01:10:44.797 technical directions for your groups and we have this kind of like pod structure 01:10:44.880 --> 01:10:48.797 where we have various pods and there's various technical leads of the pods who 01:10:48.880 --> 01:10:52.397 who tie break on the technical decisions. So, there's like a technical 01:10:52.480 --> 01:10:56.477 leadership structure in place, but from a quote unquote people management 01:10:56.560 --> 01:10:59.757 perspective or from a bureaucracy perspective, we're just 01:10:59.840 --> 01:11:02.397 anti-bureaucracy. And then the other 01:11:02.480 --> 01:11:04.637 part of this is like we do make a lot of 01:11:04.720 --> 01:11:07.037 group decisions. Like we want to debate 01:11:07.120 --> 01:11:11.757 because it is like a very talent dense group of like brilliant people. We want 01:11:11.840 --> 01:11:15.677 to have those conversations where we're all discussing what do we think the 01:11:15.760 --> 01:11:19.117 right path is. Everyone in the lab has done brilliant things. Like we want to 01:11:19.200 --> 01:11:22.317 hear all these opinions. We want to arrive at things that we all believe are 01:11:22.400 --> 01:11:26.317 the best path forward. So it's certainly like an unconventional way to run the 01:11:26.400 --> 01:11:28.317 team. And like I would be lying if I 01:11:28.400 --> 01:11:30.797 said I was like the best manager to more 01:11:30.880 --> 01:11:33.597 than 200 um people. But this part of the 01:11:33.680 --> 01:11:36.317 point is that like they don't need 01:11:36.400 --> 01:11:38.957 managers, they need a great environment. 01:11:39.040 --> 01:11:43.277 They are all brilliant. They're all extremely capable. They just need the 01:11:43.360 --> 01:11:45.597 room to cook. >> Yeah. And so maybe not a coach. You're 01:11:45.680 --> 01:11:48.397 kind of like the steward of a great environment so they can do great work. 01:11:48.480 --> 01:11:51.277 >> Yeah. Exactly. >> Alex, this was awesome, man. Thanks for 01:11:51.360 --> 01:11:53.917 making taking the time. >> Yeah. 01:11:54.000 --> 01:11:56.957 >> I hope you enjoyed this episode. Please remember to subscribe wherever you're 01:11:57.040 --> 01:12:00.157 listening and leave a review. And make sure you listen to my other podcast, 01:12:00.240 --> 01:12:04.557 Founders. For almost a decade, I've obsessively read over 400 biographies of 01:12:04.640 --> 01:12:07.917 history's greatest entrepreneurs, searching for ideas that you can use in 01:12:08.000 --> 01:12:09.917 your work. Most of the guests you hear 01:12:10.000 --> 01:12:15.160 on this show first found me through founders.