WEBVTT 1 00:00:00.231 --> 00:00:02.773 The Frontier duopoly is over. 2 00:00:03.493 --> 00:00:07.716 This week we got new Frontier models from Google and from Meta. 3 00:00:07.916 --> 00:00:15.502 Google launched Gemini Argon 4 and Meta launched Wonder which comes in standard and full so you can run Wonder full. 4 00:00:15.702 --> 00:00:17.343 You know, that's their sense of humor. 5 00:00:17.543 --> 00:00:24.406 Of course, both new models from these pretty much second tier entrants in the frontier model race topped the benchmarks. 6 00:00:25.166 --> 00:00:35.871 But our hands-on testing inside of Every running these models for knowledge work, for coding, for design, for writing, revealed something a lot more nuanced and different from what the public benchmarks show. 7 00:00:36.231 --> 00:00:46.236 To us, it seems pretty clear that Muse is benchmarked, which means they trained it to look really good on public benchmarks, but those gains don't necessarily translate into your everyday work. 8 00:00:46.456 --> 00:01:01.980 So it looks like it scored really well on Sweebench, but in our testing, for example, on my senior engineer benchmark task where we ask a model to rewrite a vibe-coded slap code base, it scored about a 50% across a few different runs, which is solidly middle-of-the-pack territory. 9 00:01:02.060 --> 00:01:04.060 So it's also kind of a mid-writer. 10 00:01:04.440 --> 00:01:06.881 It buries the lead in the stuff that it writes. 11 00:01:07.461 --> 00:01:07.701 And it's... 12 00:01:08.241 --> 00:01:18.108 It falls prey to some basic benchmark tasks like my McDonald's test, where I ask a model to do literary analysis of a very serious book. 13 00:01:18.328 --> 00:01:20.710 And then I say, hmm, that reminds me of McDonald's. 14 00:01:21.090 --> 00:01:22.852 And I see whether the model takes the bait. 15 00:01:23.412 --> 00:01:24.593 Muse takes the bait. 16 00:01:24.833 --> 00:01:28.556 So Muse might take the bait, but Gemini kind of takes the cake. 17 00:01:29.256 --> 00:01:30.336 I'm so sorry. 18 00:01:31.257 --> 00:01:38.278 Gemini 4 is one of those models where I saw the benchmarks and I was like, this is also really benchmark maxed, but it's good. 19 00:01:38.578 --> 00:01:39.458 It's a good model. 20 00:01:39.638 --> 00:01:45.100 I used it for writing all weekend, and it is without a doubt my favorite new writing model. 21 00:01:45.500 --> 00:01:47.520 It beats Astra, which was my previous favorite. 22 00:01:47.540 --> 00:01:48.501 It definitely beats Claude. 23 00:01:49.201 --> 00:01:51.661 It has an incredibly good feel for structuring information. 24 00:01:52.582 --> 00:01:58.663 And I used it all weekend to write this 8,000 word piece, How Codex Happened, which is coming out on Thursday. 25 00:01:59.783 --> 00:02:07.365 And I think it probably made me two times more productive, like truly a game changing model for writing coding. 26 00:02:07.705 --> 00:02:08.145 Not very good. 27 00:02:08.205 --> 00:02:10.646 It got like a 20% on the senior engineer benchmark task. 28 00:02:10.986 --> 00:02:15.147 And generally it seems better than previous Gemini models. 29 00:02:15.287 --> 00:02:21.888 But especially in the anti-gravity harness, it's not really something you're going to want to switch to unless you're using it specifically for writing. 30 00:02:22.108 --> 00:02:35.579 Coming out also today on Every, we've got a really good article by Alex Duffy, who runs Good Start Labs, which is a company that we spun out of Every that uses games to benchmark models. 31 00:02:35.599 --> 00:02:41.864 And this particular article is about using StarCraft to benchmark Astra and see how it behaves. 32 00:02:42.444 --> 00:02:43.845 The answer is it cheated. 33 00:02:43.965 --> 00:02:51.491 It wasn't quite hugging face up in here, but it went online to find tips and tricks and used them improperly in its gameplay. 34 00:02:52.252 --> 00:02:54.693 And that is really interesting. 35 00:02:55.054 --> 00:02:57.976 And again, my piece, how codecs happen, it's an 36 00:02:58.176 --> 00:03:10.005 It's an 8,000-word story about how OpenAI fell behind in the coding race and then came back with Codex, written by me from a bunch of first-person interviews and also a little bit from my perspective because I was there for a lot of it. 37 00:03:10.025 --> 00:03:11.446 That's coming out on Thursday. 38 00:03:11.967 --> 00:03:12.967 It's going to be awesome. 39 00:03:13.087 --> 00:03:13.988 I am psyched for this. 40 00:03:14.068 --> 00:03:17.411 I wrote it mostly with Codex and a lot of Gemini 4 Argonne. 41 00:03:17.731 --> 00:03:21.473 And last but not least, we have an interview coming up with Natalia Quintero. 42 00:03:21.493 --> 00:03:23.074 She's our head of AI consulting. 43 00:03:23.394 --> 00:03:27.277 And she interviewed Matt Hartman, who is the managing partner of Factorial Capital. 44 00:03:27.397 --> 00:03:32.180 Matt is a really interesting investor who's he came out of Betaworks in New York City. 45 00:03:32.380 --> 00:03:34.081 He was involved in their hugging face investment. 46 00:03:34.121 --> 00:03:37.102 He's just been in AI since before AI was cool. 47 00:03:38.243 --> 00:03:45.532 They get into a lot of really interesting questions like, if anyone can vibe code your app, where are the software modes? 48 00:03:45.552 --> 00:03:47.254 How do you build a software business now? 49 00:03:48.135 --> 00:03:49.437 And I think it's a great interview. 50 00:03:49.457 --> 00:03:49.577 Enjoy. 51 00:03:49.837 --> 00:03:50.098 Enjoy. 52 00:03:50.118 --> 00:03:50.959 Enjoy. 53 00:04:03.118 --> 00:04:06.823 Today, we're joined by Matt Hartman, founder and managing partner at Factorial Capital. 54 00:04:07.323 --> 00:04:09.987 Matt spent nearly a decade at Betaworks, where he was a partner. 55 00:04:10.667 --> 00:04:20.680 And while he was there, he wrote the first check into a tiny chatbot company called Hugging Face, which NVIDIA bought last month for $13 billion. 56 00:04:20.700 --> 00:04:20.760 And 57 00:04:21.801 --> 00:04:28.723 In 2023, he left to start his own fund, Factorial, a venture fund with a very unique structure, which we'll learn about today. 58 00:04:29.103 --> 00:04:33.845 And anytime that I talk to Matt, I think of him as a builder and a tinkerer. 59 00:04:34.305 --> 00:04:40.147 Every time we catch up, he's built something really interesting for himself or his fund or his sons. 60 00:04:40.327 --> 00:04:41.408 So welcome, Matt. 61 00:04:41.428 --> 00:04:42.468 So excited to have you on the show. 62 00:04:42.488 --> 00:04:42.788 Thank you. 63 00:04:43.649 --> 00:04:44.789 Thank you for having me. 64 00:04:45.369 --> 00:04:50.671 And while I was at Betaworks, we all made all of our investments together. 65 00:04:50.791 --> 00:04:59.653 So it was very much a team effort finding and coming up with a thesis for Hocking the Face. 66 00:04:59.673 --> 00:05:00.813 Happy to talk about what that is. 67 00:05:00.913 --> 00:05:05.934 And or at least, I mean, I'm not there anymore, but what it was like at the time. 68 00:05:06.960 --> 00:05:13.511 So maybe just to back up, can you tell us what Hugging Face is for anyone who's not familiar with the company? 69 00:05:14.653 --> 00:05:14.894 Sure. 70 00:05:14.934 --> 00:05:17.298 So they, I mean, it's the biggest community for AI developers. 71 00:05:18.560 --> 00:05:21.262 Some people call it GitHub for AI. 72 00:05:21.282 --> 00:05:31.109 I think of it as almost a marketplace between AI researchers who are doing actual research, writing papers that then turn out to code. 73 00:05:31.229 --> 00:05:38.895 And then on the other side, developers who want to use the AI research like me, where I'm not the one training a model, but I like to use brand new models. 74 00:05:39.476 --> 00:05:41.977 And then they have private accounts where you can pay. 75 00:05:41.997 --> 00:05:45.140 You can pay to do training and run inference. 76 00:05:45.300 --> 00:05:46.801 And so they have a bunch of other ways they make money. 77 00:05:47.845 --> 00:05:52.338 But in 2015, I think when you found them, they weren't that yet. 78 00:05:52.378 --> 00:05:54.465 I believe they were a chatbot for teenagers. 79 00:05:54.525 --> 00:05:55.006 Is that right? 80 00:05:55.026 --> 00:05:55.127 Yeah. 81 00:05:55.936 --> 00:06:01.761 Yeah, so the, and I should say, it's the largest, probably the largest place for open source AI in particular. 82 00:06:01.781 --> 00:06:03.703 So it's very open source focused. 83 00:06:05.625 --> 00:06:15.574 So what happened was my partners and I at Betaworks were really interested in what we were calling conversational software at the time. 84 00:06:15.614 --> 00:06:18.396 This paper had come out in like 2015 called Word to Vec. 85 00:06:18.856 --> 00:06:20.978 They take words and put them into vectors. 86 00:06:21.058 --> 00:06:22.340 And we can think about that as like, 87 00:06:23.541 --> 00:06:25.702 I built a neural network in college. 88 00:06:25.762 --> 00:06:27.564 It was like a handwriting recognition thing. 89 00:06:27.924 --> 00:06:32.067 And so I knew that you had 100% of the information in an image is the information. 90 00:06:32.508 --> 00:06:33.809 It's not true with words, right? 91 00:06:33.829 --> 00:06:36.891 So if I say king, it might be the king of spades. 92 00:06:37.011 --> 00:06:38.513 It might be the king of England. 93 00:06:38.533 --> 00:06:39.473 It might be LeBron James. 94 00:06:39.613 --> 00:06:41.175 So we don't know which king I'm talking about. 95 00:06:41.891 --> 00:06:49.338 And what Word2Vect did was propose a way to be able to embed words within their context. 96 00:06:49.438 --> 00:06:57.185 And what happened was they would map these into what they called n-dimensional space, but kind of picture like 3D space. 97 00:06:57.765 --> 00:07:00.948 And you would say king over here and queen over here, let's say. 98 00:07:01.268 --> 00:07:03.870 And the hour where the vector that pointed from king to queen 99 00:07:05.192 --> 00:07:12.777 The craziest thing, I thought it was unbelievable, when you moved the arrow over to where Prince was, it pointed to Princess. 100 00:07:13.117 --> 00:07:13.918 Same thing with City. 101 00:07:13.958 --> 00:07:19.722 So you would have Paris, and it would point to France, and you'd move it over to Rom and it would point to Italy. 102 00:07:20.302 --> 00:07:23.805 And so what that meant was that you could now take words and make... 103 00:07:24.705 --> 00:07:30.411 put them into sort of like unique numbers that represented their meaning, not just, and then you solve that problem that I was talking about. 104 00:07:30.471 --> 00:07:32.573 So then if you can do that, you can then put them into neural networks. 105 00:07:33.073 --> 00:07:41.341 So we had been tracking this stuff and we're trying to figure out what kinds of products can be built. 106 00:07:41.742 --> 00:07:46.867 And there was a bunch of companies that were, in my opinion, kind of pivoting to be chatbot companies because VCs were talking about chatbots. 107 00:07:47.567 --> 00:07:50.069 And those were genuine new distribution channels. 108 00:07:50.570 --> 00:08:01.239 But we decided what we'd do is go earlier and find people who had almost like, they were sort of rowing in the same direction. 109 00:08:01.259 --> 00:08:05.882 They had had these observations, and the Hugging Face team had this observation that there's something interesting to build here. 110 00:08:05.922 --> 00:08:07.744 What could you build now that this exists? 111 00:08:07.764 --> 00:08:12.128 And at the time, there was the Stanford question answered and it was like, 112 00:08:12.588 --> 00:08:15.550 question and answers were starting to kind of be the core thing. 113 00:08:15.590 --> 00:08:24.374 And so they said, what if you could have a chat with this other kind of intelligence that's getting built and you know, maybe talk about movies, maybe talk about what's going on this weekend. 114 00:08:24.394 --> 00:08:26.375 And that was kind of the core product that they built initially. 115 00:08:26.395 --> 00:08:28.036 They were based in Paris, moved to New York, worked out of our office. 116 00:08:28.056 --> 00:08:29.136 So I got to know them really well. 117 00:08:29.156 --> 00:08:39.461 And, and, and then the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, 118 00:08:41.302 --> 00:08:53.905 In my view, and I'm sure there's, I haven't heard Ben talking about this much, so I'm curious, sort of, from my perspective, they had this app, it was a consumer app that a lot of people were using, but the retention really wasn't good. 119 00:08:54.305 --> 00:08:57.605 They were very honest with themselves about those metrics. 120 00:08:58.566 --> 00:09:10.368 When Transformers came out, they quickly converted it into PyTorch, which was sort of, I think it came out in TensorFlow and needed to be in PyTorch for people who were working out at Python jobs. 121 00:09:11.148 --> 00:09:36.885 and that went via the open source and went viral on on github and then other my my recollection is other people started reaching out to them to hey can you convert our paper where they put converted their own papers and they then moved that over to the hugging face hub and it was really kind of they felt the demand of ai researchers who were like who were trying to figure this stuff out but they themselves were developers and had done this uh and done and implemented so much this that 122 00:09:37.305 --> 00:09:39.226 What they were doing was really they knew it was hard. 123 00:09:39.266 --> 00:09:47.270 And rather than just build a consumer app, maybe they could make it easier for anybody to use all these open source models that they were working with and build their own products. 124 00:09:47.290 --> 00:09:52.332 And that's ultimately where the product and the business ended up. 125 00:09:53.554 --> 00:09:59.597 It's just fascinating when we were chatting in preparation for this podcast. 126 00:09:59.637 --> 00:10:04.899 You mentioned that Hugging Face and the team made a thousand great decisions. 127 00:10:05.379 --> 00:10:11.902 What would you say were the maybe three pivotal decisions that led them to where they are today? 128 00:10:11.922 --> 00:10:12.222 Right. 129 00:10:12.917 --> 00:10:24.586 I mean, I think it was, I think that first recognizing that the product, you know, I think a lot of teams today go out and raise on whatever top line metric they can possibly raise on. 130 00:10:24.947 --> 00:10:37.537 And I think that the best founders that I back, this is true of Anchor also, looked at their retention or really asked themselves, what is genuinely working here and what is genuinely not over the long-term for you and trying to create value? 131 00:10:38.217 --> 00:10:42.079 And so I think one thing they did was open, obviously open source and transformers. 132 00:10:42.480 --> 00:10:56.648 Second thing was moving everything onto GitHub, or I'm sorry, off of GitHub, because it was on GitHub and really having a more rich place for the weights to live, and it was a more structured place for AI researchers to be able to go. 133 00:10:57.388 --> 00:11:09.557 And then I think another decision that comes to mind, and I don't know if it was the most important, was it seems so obvious now, but I remember it being a question of whether, like whether you move outside of language models or not. 134 00:11:09.998 --> 00:11:17.584 Now we think about there's video models and robotics models and bio models and weather models, but the language was really the first use case. 135 00:11:17.804 --> 00:11:21.467 And I remember thinking, I wonder if it will end up 136 00:11:22.708 --> 00:11:25.071 Everyone will end up using this for everything, all open source. 137 00:11:25.091 --> 00:11:26.353 What are the things you want to hold constant? 138 00:11:26.393 --> 00:11:28.115 I think open source is always the thing they held constant. 139 00:11:28.816 --> 00:11:33.041 And then being flexible on what kinds of models were trained was really smart. 140 00:11:33.562 --> 00:11:38.768 If you're, one of the great things about working with AI is you can just talk to it now. 141 00:11:39.954 --> 00:11:46.941 If you're interested in groundbreaking voice experiences, you need to check out Gemini Audio. 142 00:11:47.602 --> 00:11:50.825 Gemini Audio is the new voice models from Google DeepMind. 143 00:11:51.405 --> 00:11:55.709 Gemini 3.5 Transcribe turns speech into clean formatted text. 144 00:11:56.862 --> 00:12:04.423 If you say, let's meet Tuesday, no Wednesday, it'll detect the correction and handle it automatically and remove filler words. 145 00:12:05.423 --> 00:12:10.524 With Gemini Live 3.5, there's less cleanup between a rough thought and something you can actually use. 146 00:12:11.725 --> 00:12:17.766 And now with Gemini 3.8 Live Extended Thinking, you can talk through more complex work. 147 00:12:18.366 --> 00:12:25.627 It can reason and speak simultaneously, which keeps conversations going while it's working in the background on your behalf. 148 00:12:27.023 --> 00:12:30.385 You can even ask follow-up questions and give feedback as you go on together. 149 00:12:31.066 --> 00:12:36.890 It makes voice a useful way to develop an idea with room to change your mind along the way. 150 00:12:37.651 --> 00:12:46.597 You should check out the link in the description and learn more about Gemini 3.5 Transcribe and Gemini 3.8 Live Extended Thinking. 151 00:12:47.238 --> 00:12:48.619 And now, back to the episode. 152 00:12:49.350 --> 00:13:00.724 One of the things that's really interesting to me about how you ended up as the first investor in Hugging Face is that it really kind of stemmed from a thesis that you had or like a theme that you were exploring. 153 00:13:01.425 --> 00:13:06.852 And maybe just to kind of tug on this thread for a minute, I'd be curious to learn how you think about 154 00:13:08.474 --> 00:13:12.196 the theses or the sort of themes that you're exploring at any given time. 155 00:13:12.977 --> 00:13:19.821 And, you know, you mentioned the one that you were exploring when you, you know, with what led you to hugging face, but, you know, how is that evolving over time? 156 00:13:19.841 --> 00:13:20.522 Yeah. 157 00:13:21.761 --> 00:13:26.565 I think my goal is to find the newly technically possible. 158 00:13:26.846 --> 00:13:29.408 So sometimes it's just tinkering with what's available. 159 00:13:29.828 --> 00:13:33.592 Sometimes it's reading the paper, as in that case. 160 00:13:33.752 --> 00:13:39.257 Often somebody who I think is really smart is exploring something that I didn't know was, that I'm surprised by. 161 00:13:41.259 --> 00:13:46.684 In the case of, like, when I started Factorial Capital, one of the things that I was, 162 00:13:48.420 --> 00:13:57.727 a conversation I had with a number of people was whether language models, like large language models were working and it was not obvious that was gonna be the case. 163 00:13:58.207 --> 00:13:59.388 Does that apply to non-language? 164 00:13:59.408 --> 00:14:06.513 Does that apply to total other kinds of models like large weather models or physics models? 165 00:14:07.193 --> 00:14:15.382 And if what the underlying technology of Transformers, there's lots of different technologies we use now, but Transformer is really good at next token prediction. 166 00:14:15.623 --> 00:14:19.767 What are the next things you can predict in a time series that would be really useful? 167 00:14:20.068 --> 00:14:22.250 So I started getting really interested in Transformers 168 00:14:23.591 --> 00:14:24.832 how is this applicable to robotics? 169 00:14:24.852 --> 00:14:25.952 How is this applicable to science? 170 00:14:25.992 --> 00:14:27.133 How is this applicable to other areas? 171 00:14:27.753 --> 00:14:29.334 That's sort of when I start to pull on the thread. 172 00:14:29.434 --> 00:14:31.695 And then I just ask as many smart people questions as possible. 173 00:14:31.735 --> 00:14:35.196 And often there's sort of a thesis that's 20 degrees off. 174 00:14:35.637 --> 00:14:42.680 Like I certainly at least all think that I predicted this exact thing, what the hugging face was from what they started with. 175 00:14:43.140 --> 00:14:47.802 Or even, you know, Anchor became the most popular tool to create podcasts. 176 00:14:47.842 --> 00:14:50.123 It was originally an audio social network. 177 00:14:50.803 --> 00:14:51.463 And with them, um, 178 00:14:53.843 --> 00:14:56.190 I was getting a cup of coffee in like 2014. 179 00:14:57.981 --> 00:15:09.850 And I came out of one angle, my partners at Bayward came from a different angle, but my angle was the, I was waiting for a cup of coffee and I looked at my phone and all the apps were competing with each other. 180 00:15:09.870 --> 00:15:10.410 Do I text? 181 00:15:10.510 --> 00:15:12.152 Do I send an email? 182 00:15:12.432 --> 00:15:13.372 Do I use Snapchat? 183 00:15:14.013 --> 00:15:17.155 And then I got my coffee and I was walking the rest of the way off and I put my headphones in. 184 00:15:17.175 --> 00:15:19.457 It's like, there's like no competition for my ears. 185 00:15:19.597 --> 00:15:21.338 I had, you know, audio books a little bit. 186 00:15:21.679 --> 00:15:24.481 You can make a phone call, you can listen to music. 187 00:15:24.661 --> 00:15:27.103 And then podcasts, there were podcasts, but they weren't like huge. 188 00:15:27.783 --> 00:15:32.786 And my thought was like, maybe there can be something that's a consumer, almost Twitter for audio product. 189 00:15:33.686 --> 00:15:35.487 And then I looked for like a year to try to find this. 190 00:15:35.507 --> 00:15:38.448 I would talk to anybody who would listen about like, have you seen things like this? 191 00:15:38.508 --> 00:15:40.870 I would try to experiment with building things. 192 00:15:40.970 --> 00:15:42.891 And I would listen to a lot of podcasts. 193 00:15:43.411 --> 00:15:50.715 And so from my approach to that was, was seeing that gap, feeling that gap. 194 00:15:51.255 --> 00:15:54.877 And then when I met the team at Anchor, I'd met a bunch of other teams that were doing this. 195 00:15:54.917 --> 00:15:57.198 And they didn't quite have the right, in my view, 196 00:15:57.918 --> 00:15:58.819 approach to the product. 197 00:15:59.400 --> 00:16:00.821 The apps didn't know they were on the phone. 198 00:16:00.841 --> 00:16:06.005 I remember with the anchor team, they said, I don't know if they remember this, but I do. 199 00:16:06.306 --> 00:16:14.913 They're like, you have a bone in the back of your neck that amplifies the bass in your own head. 200 00:16:15.413 --> 00:16:17.055 And that's why you don't like what you sound like when you're recording. 201 00:16:17.075 --> 00:16:19.617 When you listen to your recording, you don't have that bass. 202 00:16:20.498 --> 00:16:25.202 And so they were like, that is a music, it's like psychological friction on why you wouldn't post your own voice. 203 00:16:25.793 --> 00:16:33.937 And I think a team that is that deeply focused on like why are people doing this is gonna ask the right questions, even if it takes you down. 204 00:16:34.318 --> 00:16:36.058 Ultimately, they had very similar situation. 205 00:16:36.078 --> 00:16:40.561 A lot of, their app went viral, but then the usage was okay, 206 00:16:41.041 --> 00:16:45.843 But if you looked at just the people who were posting, not the listeners, it was really good. 207 00:16:45.863 --> 00:16:47.003 So at the time, Giphy was popular. 208 00:16:47.323 --> 00:16:50.924 They were like, what if we were more like Giphy, where you don't have to use Giphy to be able to see GIFs. 209 00:16:50.944 --> 00:16:52.285 You can make it available everywhere. 210 00:16:52.305 --> 00:16:54.565 It should be on podcasts. 211 00:16:54.946 --> 00:16:56.566 We'll help you get your podcasts out there. 212 00:16:57.546 --> 00:17:00.787 At the time, Alexa had the podcast. 213 00:17:00.807 --> 00:17:03.008 You could post to Alexa. 214 00:17:03.208 --> 00:17:08.610 But really, they stripped out the demand side and just focused on that supply side and built a really, really good product. 215 00:17:09.090 --> 00:17:16.579 So that's sort of the way that I typically come to my theses is really just try as many things as possible. 216 00:17:16.940 --> 00:17:17.841 Look where it developed. 217 00:17:17.881 --> 00:17:20.104 What is newly programmable is another way to think about it. 218 00:17:23.223 --> 00:17:24.765 Bitcoin certainly is programmable money. 219 00:17:25.606 --> 00:17:28.728 Websites were programmable, kind of were pieces of software. 220 00:17:29.089 --> 00:17:38.598 And now I think you have, I think many of the teams that are doing a great job in robotics, which you think about as a hardware thing or scientific models, which you think about as research heavy, 221 00:17:39.158 --> 00:17:50.147 The teams that are actually really doing well are ones that have a software development experience or are thinking about them as a software problem, not necessarily as a hardware problem in robotics. 222 00:17:50.307 --> 00:17:53.409 I think someone had said that Tesla is an iPhone on wheels. 223 00:17:53.469 --> 00:17:54.110 That feels true. 224 00:17:54.230 --> 00:17:59.954 You can do an upgrade on a Tesla, and it gets better over time. 225 00:17:59.974 --> 00:18:02.576 It gets really a software product as much as it's a hardware product. 226 00:18:02.596 --> 00:18:08.141 So I think much more today is becoming programmable, so that's the other way that I have approached it in order to do things that we can... 227 00:18:08.681 --> 00:18:10.322 What are the tools that help us make new things? 228 00:18:11.223 --> 00:18:20.788 Yeah, it sounds like there's a real commitment to the scientific method in finding where the edge of science is and who is thinking about it really critically. 229 00:18:20.808 --> 00:18:30.734 It reminds me of how in 2025, you told TechCrunch that your mantra for investing is that to invest in software, you need to know how software works. 230 00:18:31.575 --> 00:18:35.557 And I'm curious how you came to that particular mantra. 231 00:18:35.577 --> 00:18:35.717 Yeah. 232 00:18:35.981 --> 00:18:37.101 I would say that's true for me. 233 00:18:37.362 --> 00:18:41.543 There are obviously people who don't code who are amazing investors. 234 00:18:41.743 --> 00:18:43.784 So I think there's probably multiple ways to do this. 235 00:18:44.564 --> 00:18:53.128 For me, I am interested in things that are newly possible because of technology, and I think understanding how 236 00:18:55.366 --> 00:19:02.188 what the stacks look like, understanding the technology itself, even in the era of vibe coding, where like I can, anybody can build anything theoretically. 237 00:19:03.549 --> 00:19:06.650 If you are, the questions you ask become more important. 238 00:19:07.410 --> 00:19:09.570 So if the question you ask is, how can I make this thing? 239 00:19:10.231 --> 00:19:15.152 That's different than the question of, hey, I'm seeing that there's all these local AI models. 240 00:19:15.532 --> 00:19:17.113 They run on different kinds of computers. 241 00:19:17.893 --> 00:19:19.995 What is the best way to optimize this for mine? 242 00:19:20.015 --> 00:19:29.042 That's more of a question that a technologist who understands the limitations of a local model and understands that there's RAM constraints would ask. 243 00:19:30.202 --> 00:19:34.346 I think those are the kinds of products that get built that resonate with me more. 244 00:19:34.846 --> 00:19:39.349 And I think, again, this is just sort of me, my approach is, 245 00:19:40.210 --> 00:19:43.953 I think there is a very good way to back startups that is market-driven. 246 00:19:43.973 --> 00:19:45.915 You've backed executives to understand the industry. 247 00:19:45.935 --> 00:19:50.058 I think a lot about your, what, in basketball, you have like a pivot foot, right? 248 00:19:50.078 --> 00:19:51.419 We're talking about pivoting in startups. 249 00:19:51.460 --> 00:19:55.483 If you keep one foot down, if you move both feet, it's traveling in basketball. 250 00:19:55.503 --> 00:19:59.086 I think that in the same industry, lots of companies just totally build something new, and that's fine. 251 00:19:59.526 --> 00:20:03.730 But when I'm thinking about it at the very early stage, I always assume that the company is going to pivot. 252 00:20:04.110 --> 00:20:09.077 And the question is, are they going to pivot towards something that I am a sort of fellow traveler and also believe in? 253 00:20:09.498 --> 00:20:12.342 Or are they going to have limitations on what the pivot could be? 254 00:20:12.442 --> 00:20:19.392 So if you backed a bunch of real estate executives to build an AI native thing for real estate, 255 00:20:20.238 --> 00:20:40.105 So they're going to have a fairly constrained, you know, it is more likely they're going to find, they're going to pivot to something that is still in that same market, like real estate, than a technical team who might say, the reason we're going after this is because there's a particular shape of the data that is in this industry, and maybe it didn't work, but it turns out that that shape is right in the car industry or wherever it is. 256 00:20:50.328 --> 00:21:01.778 The companies that have been most successful that I've been involved with are ones where the pivot foot was about the technology, not about necessarily market understanding of a particular vertical. 257 00:21:01.878 --> 00:21:09.405 And so for me, back in technical founders gives them the ability to move with where the world goes. 258 00:21:10.581 --> 00:21:19.469 This is also just so represented in the way that you've structured your own venture fund and the partners that you have as part of that fund. 259 00:21:20.450 --> 00:21:23.432 Could you describe a little bit of how the fund works? 260 00:21:23.812 --> 00:21:28.917 And, you know, you have 12 investment partners, but you're a pre-seed seed fund. 261 00:21:28.977 --> 00:21:30.438 That is unusual. 262 00:21:30.518 --> 00:21:31.099 How does that work? 263 00:21:32.179 --> 00:21:47.094 So what I do, I set it up, I had this idea in like 2023-ish that I probably, leading up to that point of starting factorial capital that the 264 00:21:48.195 --> 00:21:53.798 Most that I loved, I wanted to be involved with the technology things that were changing, right? 265 00:21:53.838 --> 00:21:59.081 Not brand or online, like not tech enabled, but sort of actual the technology itself. 266 00:21:59.541 --> 00:22:02.543 But I was pretty technical and it was moving so quickly. 267 00:22:03.043 --> 00:22:06.765 And so my thought was, what if you set up a fund where you had... 268 00:22:07.505 --> 00:22:12.467 a structure where you could have venture partners who could help me really quickly and accurately. 269 00:22:12.487 --> 00:22:16.649 The goal is to come to, I think, very high quality decisions very quickly in early stage. 270 00:22:17.189 --> 00:22:20.511 And so can they help me get really smart on where the technology is going? 271 00:22:20.631 --> 00:22:25.593 So Clem, the CEO of Hugging Face, is he's thinking about what's going on in the open source world. 272 00:22:25.653 --> 00:22:32.896 Or Ikram, who founded Venmo, is he spending a lot of time in a number of web, particularly in developer tools. 273 00:22:33.496 --> 00:22:34.796 and really understand this stuff. 274 00:22:34.976 --> 00:22:37.957 And he's building new things all the time and trying to understand this. 275 00:22:38.637 --> 00:22:44.739 Trying to just, but it's great for finding new companies and it's great for just thought partnership on where things are going. 276 00:22:45.199 --> 00:22:49.480 So I set the fund up so that I could have like almost an infinite number of venture partners. 277 00:22:50.280 --> 00:22:53.261 And I set up as a solo GP fund with venture partners. 278 00:22:53.381 --> 00:23:00.662 So there was, instead of partnering as one person, it's almost like I'm partnering with every technical founder I've ever met. 279 00:23:01.363 --> 00:23:03.003 Or I'm close to it. 280 00:23:03.543 --> 00:23:27.643 uh at the beginning it was really really helpful i think over time i thought that that maybe i would just have like these cohorts and cohorts of venture partners what i found was that having a few or very deep relationships is really good and um all of the i think everyone who was a venture partner was somebody who i had backed uh or was part of the portfolio at some point i knew really well and the um 281 00:23:30.927 --> 00:23:37.653 Over time, what I found was the portfolio of factorial capital itself became the referrals. 282 00:23:37.813 --> 00:23:49.244 So now I still rely on a small set of people for both finding really interesting, new, crazy things, but also helping me think through other companies. 283 00:23:49.304 --> 00:23:51.666 But as an example, I think... 284 00:23:52.447 --> 00:23:54.968 Clem from Hugging Face introduced me to the Gnomic team. 285 00:23:55.348 --> 00:23:56.928 They were an amazing team. 286 00:23:56.948 --> 00:23:57.789 We invested in them. 287 00:23:58.529 --> 00:24:02.010 And then they introduced me to Patronus. 288 00:24:03.730 --> 00:24:06.551 So the Gnomic team then started introducing people. 289 00:24:06.591 --> 00:24:08.032 And so you end up in Patronus. 290 00:24:08.092 --> 00:24:09.372 And Patronus is a really interesting company. 291 00:24:09.392 --> 00:24:10.092 We did that at the scene. 292 00:24:10.112 --> 00:24:13.733 And then I got really close to them, and they raised the Series A. 293 00:24:13.853 --> 00:24:16.114 They're sort of off to the races. 294 00:24:16.314 --> 00:24:19.195 Then they introduced me to another founder of a company called Causal Labs. 295 00:24:19.955 --> 00:24:30.279 And, and so we end up, I don't know whether it's a flywheel or a family tree, but it's kind of a, a deep network of technical people. 296 00:24:30.399 --> 00:24:32.960 And now the technology always moves forward. 297 00:24:32.980 --> 00:24:38.482 So there's always new people to be able to get to know and to work with and to learn from. 298 00:24:39.148 --> 00:24:39.428 Yeah. 299 00:24:39.989 --> 00:24:42.711 I mean, it kind of sounds like the dream flywheel, right? 300 00:24:42.731 --> 00:24:52.037 You find sort of like a once in a lifetime founder in someone like Clem, the founder of Hugging Face. 301 00:24:52.237 --> 00:24:59.843 And then via him and other really brilliant technical founders, you know, you then create a network of... 302 00:25:00.323 --> 00:25:16.850 you know, the most brilliant technical founders in their domain that they are directly familiar with because they are either in the same ecosystem or direct competitors or building, you know, kind of compelling products that they are tracking and that are inspiring, maybe the way that they're thinking about the evolution of technology. 303 00:25:17.610 --> 00:25:20.110 And then they are your partners, right? 304 00:25:20.130 --> 00:25:21.311 So Clem is your partner. 305 00:25:21.331 --> 00:25:26.192 You also have one of the co-founders of Venmo, Hilary Mason from Fast Forward Labs. 306 00:25:26.232 --> 00:25:31.832 You have this incredible, you know, sort of set of partners that are helping you think about the future of technology. 307 00:25:32.153 --> 00:25:38.634 And the generations of technology are kind of shorter and shorter, it seems, every year. 308 00:25:38.894 --> 00:25:46.235 And so now the next generation of technologists becomes sort of like the next generation of your partners, which is such a cool flywheel that you've built, very unique. 309 00:25:46.255 --> 00:25:46.355 Yeah. 310 00:25:47.961 --> 00:25:54.565 It's definitely fun to get to work with people and watch them really build pretty amazing things. 311 00:25:54.765 --> 00:26:06.351 The other, you have so little information when you start writing, ideally, we're the very first check into a company, that any other information I sat across, I started, so, 312 00:26:09.072 --> 00:26:12.134 Hugging Face started out of Betawork's office when I was there. 313 00:26:12.434 --> 00:26:14.875 And then when I started Factorial Capital, I started out of the Hugging Face office. 314 00:26:15.215 --> 00:26:22.659 And so sitting across from the person who had built one of the large language models at the large model at Hugging Face, was a super early employee. 315 00:26:22.679 --> 00:26:24.820 And then when he started a company, we were able to back him. 316 00:26:25.180 --> 00:26:29.922 There was a guy who built the, it was a big part of the robotics when he started coming back him. 317 00:26:30.142 --> 00:26:33.484 The CTO at factory was at Hugging Face. 318 00:26:34.044 --> 00:26:35.305 And then also the other founders. 319 00:26:35.325 --> 00:26:36.986 Like I worked very closely with Clem, 320 00:26:37.546 --> 00:26:47.034 There's other people like Thomas Wolfe, Julian, who are so smart and interested in seeing other people succeed. 321 00:26:47.094 --> 00:26:51.257 And they often are investing and mentoring founders. 322 00:26:51.297 --> 00:26:59.784 I think it's, you know, you sort of ideally fan out from one core hub to just sort of meet more and more people and to keep investing. 323 00:27:01.607 --> 00:27:07.718 one thing it doesn't, which is just very good technical people who want to learn or are doing it because they want to see something amazing built. 324 00:27:09.931 --> 00:27:20.076 So we're hosting a conference in a few weeks, thesis, and it's about the future of work and the future of kind of technology, really. 325 00:27:21.277 --> 00:27:30.862 And I'm curious, you know, based on what you've learned as the first investor in Hugging Face and many other really incredible companies, what your thesis is today. 326 00:27:31.222 --> 00:27:34.404 And, you know, we were talking before the pod about... 327 00:27:34.984 --> 00:27:44.548 how there's this discourse of, you know, some people say that in, in the future, uh, in a future of AI, you know, it makes sense to be more technical, to be less technical. 328 00:27:44.908 --> 00:27:55.151 Um, but you actually had a really interesting take on, you know, kind of specifically for founders and building a company, what the core, um, skill might be. 329 00:27:55.431 --> 00:27:57.052 Could you share a little bit about that? 330 00:27:58.032 --> 00:27:58.273 Sure. 331 00:27:59.313 --> 00:27:59.493 Um, 332 00:28:01.148 --> 00:28:06.954 And I want to add, I guess, that the Hugging Face team has built something amazing. 333 00:28:07.274 --> 00:28:21.468 There's so many other companies that are built by technical founders like Factory and Modo that are fascinating to me because the speed at which they're growing is so fast. 334 00:28:21.508 --> 00:28:23.730 And I'm thinking about why the speed is so quick. 335 00:28:24.551 --> 00:28:35.083 And one possible thesis is that I've been reflecting on is if the, if coding is becoming faster, right? 336 00:28:35.103 --> 00:28:38.387 Cause we can use AI and helpless code, then it's been sad. 337 00:28:38.407 --> 00:28:42.612 And I agree that what you build matters more, but the, what the job product is, 338 00:28:43.740 --> 00:28:56.488 I don't know if it's product manager or product person, is really to see the product in the wild and intuit from user behavior, whether that's the data or sitting, watching over someone's shoulder, how they're using it so that you can iterate, right? 339 00:28:56.528 --> 00:28:58.669 If like iteration is the most important part of... 340 00:28:59.649 --> 00:29:15.699 finding product market fit and probably also of growing, then if it used to take two months to see the feedback, to get the insight, and then to put together what the product plan was going like and code it and then get it back in front of the user, 341 00:29:17.068 --> 00:29:18.829 It might take two days now instead of two months. 342 00:29:19.470 --> 00:29:38.744 And so does that mean that that person, who is I think a very reliable person, whether that person codes or not, but that product person who is really, their job is to understand the user and then to be able to translate that, whether it's by coding or by, but I will have a partner to build the next version of it and to iterate. 343 00:29:39.285 --> 00:29:43.068 The iteration cycles almost are necessarily so much faster. 344 00:29:43.808 --> 00:29:49.913 And so it's fascinating to see a company like, you know, Hanging Face took 10 years to get to the point it's at. 345 00:29:50.293 --> 00:29:56.078 Then you have a company like Modal, which didn't, they exist a little bit before I started Factorial, but not much. 346 00:29:56.739 --> 00:30:04.745 The Factory AI, I think that we started probably at the same time, it was only a few years. 347 00:30:05.206 --> 00:30:10.030 And these companies have grown so quickly, and I wonder if part of it is the ability to build and iterate. 348 00:30:11.355 --> 00:30:13.383 allows these very, very fast cycles. 349 00:30:13.463 --> 00:30:15.350 And so the result in my mind would be that 350 00:30:16.769 --> 00:30:19.630 the ability to generate product insights, forget about what the job is called. 351 00:30:19.650 --> 00:30:20.490 It could be called an engineer. 352 00:30:20.510 --> 00:30:21.470 It could be called a product manager. 353 00:30:21.750 --> 00:30:31.372 The ability to synthesize how a user is using your product and then to improve it, I think has got to be exponentially more valuable. 354 00:30:31.392 --> 00:30:35.653 It was already the most valuable thing, is my guess, and now it is exponentially the most valuable thing. 355 00:30:35.993 --> 00:30:39.574 Obviously, distribution matters more than ever today as well. 356 00:30:39.694 --> 00:30:42.914 There's a whole bunch of noise, and the easier it is to build, the more noise there is. 357 00:30:43.254 --> 00:30:44.995 But I suspect that that's... 358 00:30:46.775 --> 00:30:53.237 That's a job whose value is going up tremendously right now because of the speed of it. 359 00:30:53.317 --> 00:31:05.420 Iteration as the core sort of variable to deliver speed and make sure that you can- Your cycle looks like this and you compress it, so your cycle time is like this, and so you just grow much faster. 360 00:31:05.440 --> 00:31:09.501 Then your work just gets compounded as a function of time. 361 00:31:10.503 --> 00:31:15.508 You know, I'm just curious, how do you filter for that when you're talking to a founder? 362 00:31:15.548 --> 00:31:21.172 When you're, you know, when you've met a compelling, you know, technical founder, they're working on an interesting problem. 363 00:31:21.633 --> 00:31:23.434 They have a foot that is pivoting. 364 00:31:24.755 --> 00:31:28.058 Then how do you measure that they can iterate quickly? 365 00:31:28.078 --> 00:31:32.042 I mean, usually when I'm meeting them, they don't think they're pivoting, right? 366 00:31:32.062 --> 00:31:33.623 Because they're like, no, this is definitely the thing. 367 00:31:33.643 --> 00:31:34.484 I'm raising money for it. 368 00:31:35.324 --> 00:31:35.805 But the... 369 00:31:37.593 --> 00:31:41.375 I think that's why the ability to do it is probably more important than what the things you're actually building. 370 00:31:41.835 --> 00:31:46.717 So usually I'm trying to understand what is their insight. 371 00:31:46.777 --> 00:31:48.578 And people have talked about earned insights. 372 00:31:48.618 --> 00:31:49.759 I really like that framing. 373 00:31:50.800 --> 00:31:56.643 It's like what is the – I think about that as how many people were in a position to learn the thing that you did. 374 00:31:57.603 --> 00:31:59.424 and then to execute it on it. 375 00:31:59.444 --> 00:32:00.604 Because I think that's what earned means. 376 00:32:00.964 --> 00:32:02.145 We can all take a look. 377 00:32:02.165 --> 00:32:09.467 If there is some observation that we all have, there's going to be more effort put towards solving that problem, which means more competition. 378 00:32:10.047 --> 00:32:22.691 If only 10 people in the world have come across this because it's a particular category or particular area of expertise, you're just going to have more time probably and both more time because you'll be ahead. 379 00:32:22.711 --> 00:32:24.332 But also, you're just... 380 00:32:25.952 --> 00:32:29.993 you've already spent a lot of time learning that problem and you can go after it directly. 381 00:32:30.773 --> 00:32:33.614 And so when I think about like an earned insight, I think that's important. 382 00:32:34.314 --> 00:32:44.656 And usually that kind of goes hand in hand with the ability to generate new insights as you are in market, which is again another reason to have your product out there. 383 00:32:44.676 --> 00:32:48.557 I think there are people who can like just divine what the right thing is. 384 00:32:48.577 --> 00:32:54.778 I think there's a story about how Figma like was just basically building in, I don't know if it's true, but was building like sort of alone for a year or two. 385 00:32:55.819 --> 00:32:57.280 that's not where I've seen company. 386 00:32:57.300 --> 00:32:58.500 That's not the pattern that I've seen. 387 00:32:58.520 --> 00:33:02.763 It's more common that you put something out there and it's pretty cool. 388 00:33:02.783 --> 00:33:11.669 And something starts to, to take off and you just try to delete everything else and, and do that bang because you have been able to generate this before you get that first version. 389 00:33:12.579 --> 00:33:13.960 you know how to do it going forward. 390 00:33:13.980 --> 00:33:21.765 So I'm looking for the, I think often it comes down to kind of the why, why, why are you, why is this the interesting, why is this the problem to solve? 391 00:33:21.845 --> 00:33:24.066 Or why is this a problem here? 392 00:33:25.728 --> 00:33:28.950 Yeah, that's, that's, and maybe the anchor example, the good one. 393 00:33:29.210 --> 00:33:39.416 Somebody who's thinking that deeply about how your voice sounds in your own head is probably a person who's going to really, really sit with their users and try and understand what problem they're solving. 394 00:33:39.557 --> 00:33:41.037 And is this Mike Magnano? 395 00:33:41.098 --> 00:33:41.238 Yeah. 396 00:33:41.978 --> 00:33:42.178 Yeah. 397 00:33:42.378 --> 00:33:42.919 The best. 398 00:33:43.059 --> 00:33:43.960 So great. 399 00:33:44.100 --> 00:33:45.922 Shout out to Mike. 400 00:33:46.302 --> 00:33:46.522 Yeah. 401 00:33:46.723 --> 00:33:47.824 I don't remember which one said it. 402 00:33:48.144 --> 00:33:49.565 And again, I think I've talked about this. 403 00:33:49.605 --> 00:33:52.028 I think, I don't know if you remember saying it, but I just, I was thinking, wow. 404 00:33:52.048 --> 00:33:56.812 But like a thing that they did do that I'm positive to remember is there was different onboarding if you had headphones in or not. 405 00:33:57.613 --> 00:33:57.833 Wow. 406 00:33:57.893 --> 00:34:00.215 And like, think about whether you turn your volume on. 407 00:34:00.235 --> 00:34:04.840 It's kind of different now, but pre AirPods, you can just like have them in. 408 00:34:05.999 --> 00:34:11.663 you would like shut down an app if it just started playing audio. 409 00:34:11.763 --> 00:34:15.285 So to have an app, how can you lower the friction of when someone's going to sign up? 410 00:34:16.126 --> 00:34:18.987 Letting them do it without headphones, they don't have headphones in is interesting. 411 00:34:19.027 --> 00:34:21.889 It was also like a very cute way that they showed the headphones. 412 00:34:21.929 --> 00:34:28.333 But that kind of care, I think, pays dividends in a consumer product. 413 00:34:28.534 --> 00:34:28.934 Yeah. 414 00:34:29.014 --> 00:34:33.217 Something that I often think about with really great companies is that they end up being – 415 00:34:33.897 --> 00:34:38.420 kind of like the greatest expression of someone's like best self. 416 00:34:39.461 --> 00:34:45.946 And I actually think about this with you and your fund because, you know, you have this mantra to invest in software. 417 00:34:45.966 --> 00:34:47.107 You need to know how software works. 418 00:34:47.607 --> 00:34:49.969 And you are the biggest tinker I know. 419 00:34:50.109 --> 00:34:57.695 Every time we catch up, you're working on something, you know, wild or whimsical, or like, you know, you build software for your own fund. 420 00:34:58.215 --> 00:35:02.859 Can you speak a little bit to some of the things that you've built and that you are using? 421 00:35:03.907 --> 00:35:04.187 Sure. 422 00:35:04.748 --> 00:35:11.853 And so Becca, who works with the effect trail is also a builder. 423 00:35:11.894 --> 00:35:16.097 And so she I often build the mediocre version that she builds like a really good version. 424 00:35:18.503 --> 00:35:26.468 there are things we use to manage the phones properly, but also I'm building to test and learn. 425 00:35:26.528 --> 00:35:33.092 So a product that I built that got a little bit of attention earlier this year was called Ghost Pepper. 426 00:35:33.152 --> 00:35:39.576 There's these apps, and there are many of these things that are used to, you hold out a hotkey and you speak. 427 00:35:39.596 --> 00:35:41.478 Like a lot of by coders just want to speak the prompt. 428 00:35:42.078 --> 00:35:47.600 And so I was using it and I was thinking, you know, I'm seeing all these local AI models for voice attacks. 429 00:35:48.240 --> 00:35:50.120 Why do you even need to pay for inference for this thing? 430 00:35:50.140 --> 00:35:51.261 Couldn't you just do it on your own machine? 431 00:35:51.761 --> 00:35:58.283 So I went to Claude Code, I think, and said, look at the hugging face repos. 432 00:35:58.323 --> 00:36:00.584 Here's a couple that I know about, but find whatever you can. 433 00:36:02.485 --> 00:36:06.289 Um, and I want to build something like, uh, like this. 434 00:36:06.430 --> 00:36:15.280 I use this skill, which I love called superpowers, which I think for any product developer, if you add superpowers, I think it's available for both Codex and Claude. 435 00:36:15.300 --> 00:36:16.141 I was using it in Claude. 436 00:36:16.782 --> 00:36:19.545 Um, it's, it's by this guy, Jeff, uh, I'm sorry. 437 00:36:19.565 --> 00:36:20.447 Jesse Vincent. 438 00:36:20.907 --> 00:36:21.288 I, yeah. 439 00:36:21.748 --> 00:36:28.451 It's probably this guy, Jesse Vincent, who is like a genius prompt engineer, I think, is how it works. 440 00:36:28.491 --> 00:36:33.833 But the difference between using this product to build things and not is night and day, in my opinion. 441 00:36:34.253 --> 00:36:40.255 So I used superpowers, and I had Claude do a bunch of research with me on voice models. 442 00:36:40.655 --> 00:36:41.695 I told him what kind of computer I had. 443 00:36:42.136 --> 00:36:45.177 And then I built this voice-to-text network. 444 00:36:46.558 --> 00:36:56.026 tool that I called Ghost Pepper, and put it on Hacker News, and it ends up going to number one on Hacker News, which I'd never posted anything to Hacker News. 445 00:36:56.086 --> 00:36:58.087 I was really excited, but I open-sourced it. 446 00:36:58.448 --> 00:37:04.153 So anybody, because you don't have any inference costs, so why not let anybody just try this and run it, and I can run these things for free? 447 00:37:05.213 --> 00:37:11.076 And it was an interesting lesson both in I learned you could build a very high quality product. 448 00:37:11.096 --> 00:37:16.739 A lot of people shared out that it was actually faster than a bunch of state-of-the-art products. 449 00:37:17.219 --> 00:37:20.080 I think probably because it was just constantly looking for the newest, fastest models. 450 00:37:21.301 --> 00:37:25.122 And particularly with voice detects, I think local works really well. 451 00:37:26.563 --> 00:37:30.947 The other thing I learned was when you just vibe code something for yourself, you have to maintain it. 452 00:37:31.488 --> 00:37:34.791 And I started to appreciate, there's all this talk about the SaaSpocalypse. 453 00:37:36.274 --> 00:37:38.575 these paid apps aren't that expensive. 454 00:37:39.295 --> 00:37:44.416 And in a way, what you're paying for is them making sure it stays a great product. 455 00:37:44.736 --> 00:37:52.518 As long as these, I think that the job to be done in a way for SaaS apps is to always continue to iterate and be at the edge. 456 00:37:52.858 --> 00:38:00.240 But I now have a different appreciation for these SaaS products that really do release all the time. 457 00:38:01.480 --> 00:38:03.680 And so it's definitely a fun project. 458 00:38:03.700 --> 00:38:05.441 I think, I don't know, there's like, 459 00:38:06.321 --> 00:38:08.742 Thousands of people are using it, which makes me happy. 460 00:38:08.782 --> 00:38:12.063 Yeah, I think I saw it has like over 3,000 GitHub stars. 461 00:38:12.083 --> 00:38:12.823 That's so cool. 462 00:38:13.783 --> 00:38:20.926 I ran this by some of our engineers, you know, and definitely the maintenance piece is just so real. 463 00:38:21.646 --> 00:38:27.628 One of my earliest learnings of this is I vibe-coded our CRM tool and used it for months. 464 00:38:27.808 --> 00:38:34.510 And then I realized at some point that I was spending too much time doing the maintenance of the tool that I had built. 465 00:38:35.551 --> 00:38:47.739 And, you know, instead I could be spending time with the clients that we have, you know, working on AI implementation and the work that we do on the consulting side of the business. 466 00:38:48.180 --> 00:38:59.408 And I realized it just made more sense for us to migrate the data to Atio, a product that just does CRM and data management, instead of me trying to play, you know, sort of like... 467 00:38:59.988 --> 00:39:03.209 engineer, you know, on the side and try to like maintain this tool. 468 00:39:03.229 --> 00:39:05.390 So maintenance is a really, really big part of it. 469 00:39:05.890 --> 00:39:16.693 I know this is something that our engineers think a lot about and monologue, which is one of these, you know, text, voice to text tools that we've built in house by one of our great engineers, Naveen. 470 00:39:18.113 --> 00:39:20.994 He mentioned that this is certainly one of the competitive edges. 471 00:39:21.094 --> 00:39:24.616 And with Monologue, for example, latency is something that they think a lot about. 472 00:39:24.716 --> 00:39:28.877 So I think it's like 800 milliseconds of latency for it to load. 473 00:39:28.937 --> 00:39:31.999 So certainly a trend that we're following. 474 00:39:34.340 --> 00:39:37.801 So cool that you've built, you know, Ghost Pepper is something that you have. 475 00:39:37.821 --> 00:39:40.042 We're finding these great skills. 476 00:39:40.062 --> 00:39:46.125 We'll also include Jesse's skill as a follow-up to the pod, as a resource for anyone that's listening to this. 477 00:39:47.365 --> 00:39:51.647 The other thing we were talking about is sort of the theses that you have now, right? 478 00:39:51.707 --> 00:39:58.230 So you had a thesis and some themes that you were exploring in 2015 that led to investing in Hugging Face. 479 00:39:58.791 --> 00:40:01.532 What are those themes or theses that you're exploring today? 480 00:40:03.007 --> 00:40:07.634 I think that so much is still infrastructure. 481 00:40:07.875 --> 00:40:13.223 It's amazing to me that these developer tools, for example, you should give us a perfect one. 482 00:40:14.086 --> 00:40:19.611 I think that the maintenance, there are going to be products that help you with just maintain all of this code. 483 00:40:19.771 --> 00:40:20.973 I've seen a few. 484 00:40:21.733 --> 00:40:25.917 So I think that there are, kind of think about the continuation of developer tools. 485 00:40:26.358 --> 00:40:29.381 Different people are becoming developers, which probably means there are different kinds of tools. 486 00:40:30.121 --> 00:40:33.625 I continue to think about frontier non-language models. 487 00:40:33.665 --> 00:40:37.628 So I think I said earlier, if Transformers are really good at next token prediction, 488 00:40:38.449 --> 00:40:40.731 What are the other areas? 489 00:40:40.831 --> 00:40:45.416 It turns out that LLMs have gotten much, much better with more compute. 490 00:40:45.936 --> 00:40:53.083 And so if that's the case, what are the other areas where you could predict time series and have it create genuinely new... 491 00:40:53.965 --> 00:40:54.465 value. 492 00:40:55.485 --> 00:41:01.347 And then, you know, whether that's my, you know, my, an obvious example is robots. 493 00:41:02.247 --> 00:41:02.868 What are they? 494 00:41:03.028 --> 00:41:04.108 What is a robotic movement? 495 00:41:05.428 --> 00:41:09.289 If not a time series information with like the locations of a bunch of service. 496 00:41:09.870 --> 00:41:18.052 And so that's already being people are doing this where they're, they're we invest in a couple of companies with sort of the general thesis that the way that we train robots could change. 497 00:41:18.592 --> 00:41:21.773 And people talk about whether it's a chat GPT moment, 498 00:41:23.543 --> 00:41:32.666 the data is a bit different, and so we'll see how it kind of works, but I think that's a category where transformers in terms of AI more generally seem to be really applicable. 499 00:41:33.247 --> 00:41:35.847 I'm fascinated by the science models. 500 00:41:35.907 --> 00:41:51.573 There's Hugging Face has got a hackathon they're doing called Rare Disease Real Kid, where they've open source of genome, and I think doctors sort of notes the parents of a kid did have a very rare disease, I think only 50 people have it, and they have a track for diagnosing it, but they also have a track for drug repurposing. 501 00:41:52.100 --> 00:41:57.762 So people talk a lot about drug development, which is I think interesting, but drug repurposing is another one that's kind of fascinating. 502 00:41:57.922 --> 00:42:02.804 And it's interesting to me that the people who are building this up are really technologists. 503 00:42:02.864 --> 00:42:04.525 As I said before, it's kind of programmable. 504 00:42:04.965 --> 00:42:11.727 And then I think the application layer, and my thesis around, everyone talks about what will be vertical allocation, will be consumer applications. 505 00:42:12.528 --> 00:42:15.709 You said something earlier about latency. 506 00:42:16.169 --> 00:42:18.250 I believe that the applications are gonna look 507 00:42:19.440 --> 00:42:28.544 a lot of the people who build the applications, even if they look like consumer applications or verticals, the people who build them are going to be technical enough to be able to go fully down the stack. 508 00:42:29.084 --> 00:42:37.728 And the example I keep thinking about is in, when Instagram era, there were a lot of photo sharing apps and Instagram did a bunch of things, right? 509 00:42:38.789 --> 00:42:42.871 If you look to the phone, it was very obvious that you had a camera and you had a cell phone connection. 510 00:42:43.531 --> 00:42:45.632 What was not obvious is that the load time mattered. 511 00:42:45.652 --> 00:42:50.915 So if you're standing waiting for something and it takes you 30 seconds for the app to load, like you're done. 512 00:42:51.655 --> 00:42:56.578 So a lot of companies, Zynga and Instagram, spend time really making load time an important thing. 513 00:42:57.798 --> 00:43:04.162 Instagram, I have read, and I assume this is true, you have the pictures in squirters. 514 00:43:04.982 --> 00:43:13.865 And a side effect of the picture being squared, I don't think it was on purpose, but the benefit is that you don't have to ask the server if you should show the image and portrait last it. 515 00:43:14.666 --> 00:43:17.707 And so if that shaves off a millisecond of load time, 516 00:43:18.605 --> 00:43:31.590 the team that has the ability both to do a great design that has that nostalgic feel as early Instagram did, but also go all the way down in the stack to be able to shave seconds off of server time, leveraging a decision they made on the design side. 517 00:43:32.230 --> 00:43:38.252 I think there's a lot of parallels to what AI native companies are going to look like in the future, where they look at a paper about 518 00:43:39.117 --> 00:43:43.000 oh, this is an interesting new way to train on reasoning for language. 519 00:43:43.340 --> 00:43:46.582 What can we pull from that and apply to images? 520 00:43:46.622 --> 00:43:49.845 Or what can we pull from that and apply to what we're doing? 521 00:43:50.965 --> 00:43:58.391 And both the ability to build your own models, but also the humility to know that these other big company models are getting great. 522 00:43:58.411 --> 00:44:03.954 And if we can just use them for the stuff we need to use them for, we can then spend all of our time doing the things that are more unique to us. 523 00:44:04.953 --> 00:44:12.750 being able to balance that and go full stack means something different now, I think in an AI native world, but I think being able to do that is going to be a real edge. 524 00:44:13.550 --> 00:44:36.287 as we think about the future and along these lines of the themes that you're excited about, the companies that you're tracking, and as an investor to so many of these compelling companies, what is some advice that you might have for founders who are trying to navigate this really dynamic environment, if you will, where the technology is changing so quickly? 525 00:44:37.148 --> 00:44:41.011 What are you seeing really great founders doing right now and what advice would you have? 526 00:44:45.377 --> 00:44:49.638 the best people that I've seen building things are like playing. 527 00:44:49.658 --> 00:44:51.539 We are in an era that reminds me of early internet. 528 00:44:51.559 --> 00:44:52.499 Everyone is playing. 529 00:44:52.519 --> 00:44:55.119 There are so many fun tools. 530 00:44:55.179 --> 00:44:58.960 Like I tried this text to CAD thing. 531 00:44:58.980 --> 00:45:01.060 I don't have a 3D printer, but I was like, I bet you could vibe. 532 00:45:02.221 --> 00:45:03.001 How good of a job? 533 00:45:03.021 --> 00:45:04.401 Could I vibe, make a robot? 534 00:45:04.421 --> 00:45:06.281 We're not there yet, I don't think. 535 00:45:06.301 --> 00:45:11.402 You can 3D print a robot and you can vibe, make some 3D structures. 536 00:45:11.983 --> 00:45:13.003 But probably we will be. 537 00:45:14.623 --> 00:45:21.288 And the best thing, I think that the way that you get the best ideas is by just playing and tinkering with this stuff and seeing where the problems are and then solving them. 538 00:45:22.148 --> 00:45:32.935 And, you know, free this, free 2016, like pre the AI era, it was really the mobile era. 539 00:45:33.535 --> 00:45:35.016 And there was a lot of fun things to play with. 540 00:45:35.096 --> 00:45:35.857 And then there was this... 541 00:45:36.872 --> 00:45:54.779 question mark period where it was okay the phone is at scale lots of people know how to make iphone apps distribution getting harder and ai hadn't quite um become what it is today which i think is a i thought ai was going to be sort of at least as big as mobile i think it will turn out to be 542 00:45:58.981 --> 00:46:02.263 which I think is, because mobile is not quite as big as that to me. 543 00:46:02.624 --> 00:46:04.585 And then I think it'll be much bigger. 544 00:46:05.406 --> 00:46:20.518 Again, for the kinds of companies that I am most excited about, playing with this stuff, finding out what's going to be possible and seeing where those gaps are and then start solving your problems, whether it's probably your own life or a problem for a friend's company, whatever it is, it's how I would be spending my time. 545 00:46:20.558 --> 00:46:21.019 I think there are 546 00:46:22.349 --> 00:46:30.535 maybe the same number of problems that have always been, probably more, but there's definitely more solutions to existing problems than there have ever been. 547 00:46:30.675 --> 00:46:36.019 And this solution space is so untouched relative to... 548 00:46:36.539 --> 00:46:38.140 I'm surprised at all the companies that aren't being built. 549 00:46:38.160 --> 00:46:38.901 I think they will be built. 550 00:46:39.041 --> 00:46:41.143 I mean, there's as many products as there are getting built. 551 00:46:41.963 --> 00:46:44.645 And so I think I used to play... 552 00:46:46.106 --> 00:46:52.249 I think it was World of Warcraft or something, and you would move around, and the whole world was dark, and then as you moved around, you got to see the world. 553 00:46:52.269 --> 00:46:58.852 I think we're in that right now, where everyone, even the people, it seems like even people building these AI models are still sort of exploring what they can do. 554 00:46:59.112 --> 00:47:02.274 And we're all kind of exploring what's possible and... 555 00:47:03.957 --> 00:47:09.722 Hopefully, I think our focus certainly is, what are the solutions that are newly possible? 556 00:47:10.243 --> 00:47:12.625 Maybe it's to existing problems, maybe it's to brand new problems. 557 00:47:12.825 --> 00:47:14.387 I think that there are numerous. 558 00:47:14.487 --> 00:47:18.230 So it's an amazing time to be exploring that and building new products. 559 00:47:18.531 --> 00:47:19.692 It really is so fun. 560 00:47:20.873 --> 00:47:26.799 Talking about games reminds me of one of the companies that spun out of every, led by Alex Stuffy, who... 561 00:47:27.439 --> 00:47:38.965 literally built a gaming company because he was basically testing the capability of models, playing a game against each other and playing different games against each other. 562 00:47:38.985 --> 00:47:46.569 And, you know, it really does come from this, like, spirit of exploring what's possible and then just seeing how far you can take. 563 00:47:47.401 --> 00:47:48.141 you know, where that leads you. 564 00:47:48.201 --> 00:47:52.263 Obviously, the best way to explore what's possible is to listen to every so that you know what's newly... 565 00:47:52.283 --> 00:47:55.464 There's never been a time where there's more newly possible stuff every week. 566 00:47:55.504 --> 00:47:58.025 So that's a personal recommendation. 567 00:47:58.405 --> 00:47:58.626 Great. 568 00:47:58.646 --> 00:47:59.806 You're on the team now, Matt. 569 00:47:59.826 --> 00:48:00.306 Welcome. 570 00:48:01.367 --> 00:48:01.887 All right. 571 00:48:01.967 --> 00:48:13.211 Well, maybe just to close, Matt, any resources that you're really excited about reading, you know, that like, you know, either it's people that you're following, you already shared Jesse Vincent's superpower skills. 572 00:48:13.231 --> 00:48:14.372 We'll link that below. 573 00:48:14.932 --> 00:48:21.278 Anything else that you're kind of like looking at day to day that's really helping you stay up to speed with everything that's going on? 574 00:48:23.059 --> 00:48:28.404 I mean, there's so much that's shared on X, I think is actually really, there's a lot of noise, but there's a lot that's really good. 575 00:48:28.844 --> 00:48:30.426 I look at what's trending on Hugging Face. 576 00:48:30.826 --> 00:48:32.187 I am fascinated by open source. 577 00:48:32.227 --> 00:48:35.250 I try to find any resources or open source. 578 00:48:35.910 --> 00:48:37.552 There's newsletters that I like. 579 00:48:37.572 --> 00:48:38.072 I think... 580 00:48:39.133 --> 00:48:41.033 Ben's Bites, I think is great. 581 00:48:41.053 --> 00:48:44.274 I think every newsletter is great. 582 00:48:44.654 --> 00:48:50.136 One of the reasons I've been more and more sort of reading X again is that people are writing these longer form content pieces. 583 00:48:50.176 --> 00:49:07.740 It seems like medium content sort of moved over to X and like you can take this stuff and put it into Claude and then take how can I apply this to my life or to my investment thinking or to my company or whatever, or to my, how can I take this insight about memory and incorporate it on a mask? 584 00:49:07.920 --> 00:49:08.560 It's fascinating. 585 00:49:09.400 --> 00:49:12.422 Um, so I think I kind of probably have a barbell approach. 586 00:49:12.482 --> 00:49:14.123 I have what's new, new, new. 587 00:49:14.263 --> 00:49:17.205 And then I read things that I forget. 588 00:49:17.505 --> 00:49:18.426 Uh, I forget what it's called. 589 00:49:18.446 --> 00:49:22.868 The rule where the longer something has been around, the longer it'd be around for. 590 00:49:23.448 --> 00:49:38.017 And so I'm reading, um, just things that are older, you know, I've been reading one book about like the fall of the Roman empire and things that are hopefully timeless, um, and 591 00:49:39.682 --> 00:49:49.219 because there's probably feels like we're in a moment where certain things are repeating themselves and, uh, whether it's technology or whether it's culture or human behavior, uh, 592 00:49:51.694 --> 00:49:54.755 I find it useful to look at some of the older stuff too. 593 00:49:54.795 --> 00:50:03.816 I know that we've talked about some of that as well, that looking at adjacent, not just only spending time looking at the most recent technology can often end up being informative technology. 594 00:50:04.057 --> 00:50:04.297 Yeah. 595 00:50:04.677 --> 00:50:05.277 Thank you, Matt. 596 00:50:05.377 --> 00:50:06.497 It's so fun to have you. 597 00:50:06.617 --> 00:50:09.138 You're one of my favorite thinkers and investors. 598 00:50:09.438 --> 00:50:10.258 Where can people find you? 599 00:50:11.598 --> 00:50:13.899 Matt Hartman on X. 600 00:50:13.919 --> 00:50:17.039 You can email me at matt at factorialcap.com. 601 00:50:17.959 --> 00:50:20.120 And yeah, thank you so much for having me. 602 00:50:20.160 --> 00:50:21.140 It's been fun. 603 00:50:21.529 --> 00:50:21.869 So true. 604 00:50:21.909 --> 00:50:22.149 Yeah. 605 00:50:22.169 --> 00:50:22.510 Thank you. 606 00:50:22.530 --> 00:50:22.990 Such a treat. 607 00:50:23.170 --> 00:50:23.390 All right. 608 00:50:23.430 --> 00:50:23.850 Thanks, Matt. 609 00:50:24.951 --> 00:50:27.772 If you liked this video, you should check out every.to. 610 00:50:27.872 --> 00:50:32.454 We have a two week free trial for anyone who's watching or listening at the link in the description. 611 00:50:33.195 --> 00:50:35.956 Every is the only subscription you need to stay at the edge of AI. 612 00:50:36.036 --> 00:50:42.939 It is the one subscription that if you're super AI pill and you want to learn how to use it to do the best work of your life, it's got everything you need. 613 00:50:42.959 --> 00:50:44.520 It has education and equipment. 614 00:50:44.680 --> 00:50:46.341 On the education side, we 615 00:50:46.862 --> 00:50:49.225 review every new model that drops and tell you what's good. 616 00:50:49.646 --> 00:50:57.839 On the equipment side, we have software tools specifically made for you, like an agent that helps you AI pill your whole company, all bundled together under one subscription. 617 00:50:58.340 --> 00:51:00.683 So get two weeks free now at the link in the description.