Clem Delangue: The Ultimate Guide to Investing in AI; Elon's Threat to Sue OpenAI | E1013

1. 0:00 – 1:14

   Intro

   1. CDClément Delangue0:00

      When you, like, an early stage founder struggling and you're like, "Oh, but that's gonna be good. I'm gonna struggle. It's gonna be really hard. But in one year, in two years when I'm gonna be bigger, it's gonna be easier," I'm sorry, but it, it won't be easier. Uh, the truth is that each stage has a lot of challenges. And so I think something important is for entrepreneurs to realize that and enjoy what they're doing now, and try to find and build a company that they enjoy building. Not the joy of, you know, getting to a series D, getting a series C, getting to IPO, but really the joy of, of building, right?

   2. HSHarry Stebbings0:43

      (instrumental music) Clem, I am very excited for this. As we just said, I've stalked the shit out of you from Lee Fix or Pat Grady, Ollie at Datadog, Dev at Mongo, Thibault Elzzier who told me about the very early days. So thank you so much for joining me today.

   3. CDClément Delangue0:59

      Thanks so much for having me. I'm excited about that.

   4. HSHarry Stebbings1:01

      This will be great. So I wanna start with, uh, a little bit of context. Hugging Face, where did the name come from and what's the origin of the company founding? In a short two to three minutes.

2. 1:14 – 4:51

   Founding Story of HuggingFace

   1. HSHarry Stebbings1:14

   2. CDClément Delangue1:14

      Yeah. When we started Hugging Face, we joked with my co-founders, Julien and Thomas, that we wanted to be the first company to go public with an emoji rather than the three-letter ticker, you know? We felt like the three-letter ticker, like, on the NASDAQ and, and all that is, is kind of boring. Felt like it was, was time for a refresh and to finally have emojis up there on the boards. So we absolutely wanted an emoji as a name, and the choice is the Hugging Face emoji, right? The one, the one with hands like that, was our favorite emoji, so we were like, "Okay, let's do that." We thought maybe we would keep it for a few weeks, you know, like for a few months at, at most. And then the community started to put it everywhere, you know, like on social media, on their clothes, like, literally everywhere. So we were like, "Oh, maybe we're gonna keep it." And now it's become kind of, like, such, such a brand, so popular that unfortunately it's gonna be hard for us to, to change it.

   3. HSHarry Stebbings2:11

      Listen, at least when you do go public, the ticker will be an emoji. So I mean, you know, uh, all you need to do is, you know, get to that stage. Uh, well, in terms of company founding, like, why did you decide this was the idea that you wanted to spend 10 years, 20 years of your life on?

   4. CDClément Delangue2:25

      We actually started with something completely different. Um, the reality is that the company was, was formed because of some sort of, you know, professional crush between me and my co-founders where we were like, "We absolutely wanna work together." Um, plus our excitement about AI, right? It was seven years ago, uh, so it was not of use as it is now. Um, not enough people were talking about it at the time, but we were super excited about it as kind of, like, a new paradigm to, to build technology, new opportunities and, and all of that. The first company, the first startup I worked for, like, 15 years ago was already doing AI. We w- weren't calling it AI at, at the time, so I had some sort of a glimpse of the, of the capabilities. And when we started Hugging Face, when we started the company, we were like, "Okay, uh, what can we work on that is going to be both scientifically challenging..." Because one of our co-founders, uh, Thomas, is, is a scientist, uh, and we, we all have a lot of interest in, in the science side of things. "... but at the same time entertaining?" So we actually started with some sort of a Tamagotchi AI or, like, AI friend, howev- however you call it, some sort of, like, a Siri, Alexa or, like, ChatGPT, uh, except just for, like, entertainment, not for, like, the boring, like, productivity aspect of it. Um, and we actually did that for, for almost three years. We raised our pre-seed and seed rounds on, on this idea. We got a product out with a couple of, uh, billion messages exchanged between users and, and this Tamagotchi AI. Uh, but as, you know, it sometimes, sometimes happens, um, when we started sharing a little bit of the underlying technology and the underlying platform that we were building to, to do that, we saw a lot of traction from the community, from the open source community, from companies using that. And so that basically made us pivot from this Tamagotchi AI to this AI platform that we are now.

   5. HSHarry Stebbings4:38

      Can I ask, the thing that's really striking about what you just said there is Hugging Face today and in the last year with the rise has become a very prominent brand in the space. But seven years, it's a long time. So when you look back,

3. 4:51 – 10:57

   AI: The real deal or all hype?

   1. HSHarry Stebbings4:51

      like, to what extent is the hype and the excitement that you have today around Hugging Face, to what extent is that, like, hype cycles driven by investors and hype cycles versus true underlying development in AI given you've had this 15 years of experience in the space?

   2. CDClément Delangue5:12

      Yeah. It's, it's an interesting question, right? Um, that's completely true that perception and hype can be, you know, delayed or, like, very different than reality. My kind of, like, understanding of the current situation is that, uh, the VC and the mainstream interest is kind of like a catch-up on the reality because if you look at usage of AI, it's been massive and it's been growing massively for the past, you know, three years, right? Uh, even before, before ChatGPT, even before, like, you know, uh, the new Bing, AI was used in, in Google for, like, billions of users every day. AI was used on Facebook to rank your posts. AI was used on Zoom to remove your backgrounds. In my opinion, the current interest about AI from VCs and the public is just a catch-up.... on, on usage and on how it's kind of, like, become this, this new paradigm to, to build kind of, like, old products, old technology, old workflows. So I, I, I, I, don't really, uh, define it as hype. More kind of like a catch-up from, from usage with, like, the perception of, of VC in the mainstream.

   3. HSHarry Stebbings6:41

      Have there been catalytic breakthrough moments in the last 12 months which have taken it to the next level? Or is it just, as you said, a continuation and a catch-up from existing usage with an incumbence? Like, I think OpenAI and ChatGPT, to the world, brought this awareness. Was that a-

   4. CDClément Delangue6:59

      Yeah.

   5. HSHarry Stebbings6:59

      ... step function change or was that actually just the continuation of the catch-up that you mentioned?

   6. CDClément Delangue7:05

      So, I think, first, it's really important, and I obviously have an agenda there, but to remember that most of the progress that we've, we're seeing and that we've seen in AI is based on open science and open source. Um, it's because AI has been so open, scientists is, have been so open in sharing their research, and that everyone is building on top of each other with this really interesting loop, a positive loop of, of feedback, improvement, experiments, that we could move so fast with, with AI, right? Without open science, without open source, without Google sharing their "Attention Is All You Need" paper, sharing their BERT paper, uh, their latent diffusion paper, maybe we would be 30, 40, 50 years away from where we, where we are today. And then what happened, I think, in, in the past, past few months is you started to have some mainstream breakthroughs, right? Like you mentioned, you know, ChatGPT. And also, kind of, like, on the underlying technology stack, I think you've had better hardware, right? And kind of, like, more availability of, of GPUs, um, and better optimization of models, right? With things like quantization, distillation, and all these techniques to basically be able to run bigger, uh, models, uh, at scale for, uh, hundreds of millions or, or billions of users sometimes. These were kind of, like, probably, like, the last missing piece for AI to go mainstream as what we're seeing right now.

   7. HSHarry Stebbings8:50

      So before we dive into kind of the granular, 'cause I do want to get into kind of, uh, different models that w- can be approached, I do just have to ask. I've done now three, uh, deals in AI, like any good ambassador has done, thank you clubs, um, and, uh, every one they've asked, it's been with Valley VCs and they've said, "AI, the heartbeat of it, it is in Silicon Valley. We have to move the founders to Silicon Valley." Do you agree that SF and Silicon Valley will be the center of this next generation of AI startups or do you think that's bullshit and it's actually a decentralized globalized talent network like we've seen over the last few years?

   8. CDClément Delangue9:28

      I think there's no denying that there's tremendous excitement and activity there. Proof of that, I, I've kind of, like, uh, felt that because I was in San Francisco two weeks ago and I just, like, tweeted that I was around and, and said, "Oh, we should do, like, a get together with, like, open source AI, uh, you know, fellow community members." Uh, and in a matter of few days the thing blew up and we ended up with 5,000 people joining for, like, a huge community showcase event that people started to, to call the Woodstock, Woodstock of, of, of AI. So I, I really felt, thanks to this event, obviously, the energy that you have in Silicon Valley right now with, with AI. But at the same time if you look at the full, uh, stack, especially outside of just the early, early stage startups, right? If you look at AI scientists, if you look at ML engineers, it's, it's heavily distributed, right? Uh, one, one other point is, uh, for example, LLaMA, which is arguably one of the best open source model that came out from, from Meta, I don't have the exact number, but I think 10 out of the 13 authors of LLaMA are actually based in, in Paris, right? In, in the Meta AI, uh, lab that, that they have there that is, that is huge. So, no, I think, I think there's a lot of energy in Silicon Valley.

4. 10:57 – 12:49

   Do AI Founders need to be in Silicon Valley?

   1. CDClément Delangue10:57

   2. HSHarry Stebbings10:57

      So for all AI founders being told, "You need to move to the Valley. As the founders, you need to be in the Valley," do you say, "That's fair," or do you say, "No, you don't"?

   3. CDClément Delangue11:06

      No, I don't think, I don't think you do. Um, so Hugging Face, we, we based... Uh, we're very, very decentralized, very distributed all over the world. I'm French, obviously, so we have a big team in, in Paris, but we also have a lot of people in, in New York, in San Francisco, all over the world. I think, uh, you have to be in Silicon Valley sometimes, right?

   4. HSHarry Stebbings11:29

      Mm-hmm.

   5. CDClément Delangue11:30

      Uh, and you, you can travel there. But, uh, at, at the end of the day, I think you can, you can build a company from anywhere now. Uh, and the, the most, most important thing... And I'm, I'm sometimes, like, calling bullshit on founders saying like, "Oh, I need to be there for my company. I've taken, like, a very strategic decision to move there for my company." At the end of the day, I think it's important for founders to be happy. Uh, and if they're happy, they can build a great company. And so the most important thing, in my opinion, is for founders to find where they're the happiest, right? And if they're happy in London, they should build their company from there. If they're happy in Silicon Valley, they should build their company from there. If they're happy from, like, the middle of nowhere-... uh, completely separated from, uh, from the rest of the community. That's, that's where they should, should be on their companies. It's, in my opinion, kind of like the number, number one criteria.

   6. HSHarry Stebbings12:30

      I took the same approach to my career. Everyone told me, you know, I've been doing this eight years, Clément. I've been, everyone told me, "You have to be in the Valley if you want to be in venture, Harry. You have to be in the Valley." I fucking hate the Valley. I love London. (laughs) Right? And so, uh, I stayed and made it work. Um, listen, I wanna dive into really the two models, really, of the world,

5. 12:49 – 17:16

   One Model to Rule Them All

   1. HSHarry Stebbings12:49

      so to speak, and I, I give Lee Fixler credit for this. But, you know, he essentially posed the, the notion that, you know, there's one model which is your, kind of, OpenAI's view of the world, which is one model to rule them all, and then there's the other idea which is an open source model with many models. First, before we get into, like, opinion, just for people to understand, how do these approaches differ?

   2. CDClément Delangue13:11

      Well, they, they differ a lot in kind of, like, um, where do you allocate AI builders, right? So if there's just, like, one model to rule them all, you bet on kind of, like, models getting bigger and bigger with more and more generative capabilities and the builders of these models being concentrated in, like, one or few organizations, right?

   3. HSHarry Stebbings13:43

      Uh-huh.

   4. CDClément Delangue13:43

      In the model where you think there are a lot of different models, you bet on things being more distributed, uh, on the fact that all companies are going to actually build and train models, and that, that comes from the thinking that's in simplistic terms, the model for, like, uh, AI is, uh, like a code base, right? It's, it's a bit different, but at the end of the day, it's, it's a code base, right? And so it's kind of silly to say, "Oh, this code base is better than this code base," or, "There's gonna be one code base that is going to rule all code bases." The truth is, a code base is good or bad depending on your use case, depending on your constraints, depending on what you wanna do, right? So if you're Facebook, you have one code base that does what you wanna do for your users. If you're Slack, you have one code base that is optimized and does what you wanna do for, for your users. And it's the same thing, in my opinion, for, for AI, right? Like, if you're a company that wants to do consumer products, you need to build AI models that are optimized for this use case that are going to be faster, cheaper, more efficient. And that's how you differentiate yourself.

   5. HSHarry Stebbings15:13

      Okay, so say, say with that consumer idea, we have the consumer idea and we need to leverage AI models to, to build what we wanna build. We then have the choice of whether we choose, kind of, many models and the world of many models, or leveraging one model to rule them all, so to speak. How do we know which one to choose and why do we choose which one?

   6. CDClément Delangue15:34

      It's a tough question, especially because it involves a lot of, uh, you know, short-term versus, versus long-term, right?

   7. HSHarry Stebbings15:41

      Yeah.

   8. CDClément Delangue15:42

      Uh, the reality that sometimes today using kind of, like, one model behind an API is, uh, faster and easier at the beginning, but the challenge is in the long run, you have more risks because then you don't really internally build the capabilities to actually do AI yourself. You can't optimize these models, so they're going to be inherently more expensive, um, and you risk being in competition with others, not really differentiating yourself. The analogy that I sometimes like is that on the early days of, of the web, you could use kind of, like, something that would create a website for you, right? You, you could use the equivalent of a, of a Squarespace, of a Wix, and that would make you feel good, right? Because really quickly you would have, like, a nice website up, um, and you can start experimenting. That's the equivalent of using an AI API for me. Uh, but the reality is that, uh, if you really wanna differentiate yourself, build your capabilities and really do something that is specifically catered for your use case, for your users, and be able to optimize that, the same way you need to write lines of code to build a technology product, in my opinion, you need to train, optimize your own models in the machine learning

6. 17:16 – 19:34

   How to Sell AI to Enterprise

   1. CDClément Delangue17:16

      paradigm.

   2. HSHarry Stebbings17:17

      So, I agree with everything you say, but I'm also aware of enterprise buying and enterprise education levels. And when we, you know, you sit in Paris, I sit in London, we know how slow and bluntly ignorant large enterprises generally have been

   3. NANarrator17:32

      (music)

   4. HSHarry Stebbings17:33

      ... AML to new waves of innovation. And actually, if you can offer them a bundled service with a blue check mark and it's verified and safe and it's easy, it's easy, then they'll go for it. Do you think we'll see-

   5. CDClément Delangue17:45

      Yeah.

   6. HSHarry Stebbings17:46

      ... a three to five year period where they go for the bundled solution because it's easy before they realize the need to embrace the more tailored?

   7. CDClément Delangue17:57

      Maybe. Maybe, and, uh, that's why there's a huge opportunity for new companies to disrupt this, the incumbents, right? Uh, because they're gonna go for, like, the easy solution, whereas, like, other companies that are more, like, AI native are gonna be going for, like, the more disruptive approaches, right? And you see that with a lot of, uh, startups, right? Um, we see that with a lot of startups that are using-... Hugging Face. Like, um, if you look at models closer to, you know, RunwayML or Stability AI, or like PhotoRoom in, in Paris, you see these, uh, AI-native startups that are actually building, training their own models and how they can, in my opinion, build much better things than the ones that just use APIs. So, I mean, I think this, uh, what, what you're describing is, is a great opportunity for AI-native startups to disrupt the incumbents who are going to go with the easier solutions, even if it's not the most sustainable in the long run.

   8. HSHarry Stebbings19:02

      Can I ask why- I, I thought I saw a fantastic tweet from Yann LeCun, who said that the biggest obstacle to a, you know, open- like, you know, the open model, so to speak, is actually the legal status of the training data. How do you think about that? Is he right? Is that the main obstacle? And do you think that's fair?

   9. CDClément Delangue19:21

      Yeah, he, he has a point. Um, I, I would argue that it's a challenge for the proprietary approaches too, because they're also going to get challenged by that,

7. 19:34 – 22:44

   Elon to Sue OpenAI?

   1. CDClément Delangue19:34

      right? I don't know if you've seen, but, uh, Elon Musk tweeted that he's, he's gonna sue OpenAI for using, uh, tweets in their training for, for GPT-4. So, I would argue that it's, it's a challenge for AI in, in general, and it's gonna be good this year, I think, because we're going to start to have more legal clarity about, you know, how do we consider fair use, uh, what are the regulators expecting, uh, from AI companies to, uh, to respect in terms of rules. So, I'm excited for, um, for more clarity on regulation this, this year. I think it's, it's gonna be a good thing for, for the field as we, as we mature.

   2. HSHarry Stebbings20:18

      Can I ask, what do you think happens with content access? We've also seen Reddit, who are, you know, now starting to talk about how they're gonna need to monetize their content and access to it. How does, like, the relationship between models, whether open source or closed, and content providers play out, do you think, in the next 6 to 24 months?

   3. CDClément Delangue20:37

      I don't think everyone knows, uh, any- anyone knows really, 'cause it's going to depend, uh, on, on many factors. I, I hope we get into a, a model that works better for, for everyone, right? For the content creators, uh, to keep incentivizing them to create good content, uh, and AI companies alike. We have some, uh, some initiatives on the, on the topic. For example, we've been, uh, training, uh, a really good, uh, code model with, uh, with something called BigCode. And we've, uh, trained for the first time, I think we were the first, uh, organization to train on a fully opted out dataset, where developers could just remove themselves from the, from the training. So, we're working on the topic. There are some, some interesting things. We're still scratching the surface, in my opinion, both Hugging Face and the, and the domain, but it's going to be fun to see, uh, everything that happens in the next few months for sure.

   4. HSHarry Stebbings21:48

      While we're on this topic, and before we dive into business model, I just have to ask, what do you think happens with the Elon Musk and OpenAI case? I saw it, and I was like, "Where does that end?"

   5. CDClément Delangue21:57

      Yeah, it's a good question. It's always hard to, to know what happens, you know, behind the scenes. You know, there are very, very different approaches. Um, Elon Musk obviously is kind of, like, a, a character that is going to say, uh, one thing and, and, and the opposite, uh, almost on, on the same day. Uh, but I think he, he has a point in, uh, the necessity of, uh, openness for, for AI, and I'm sure we're gonna talk a bit more a- about that and how important openness and transparency is for, for AI, but also for society in, in general. So, I, I appreciate that he's putting at least this, uh, this part of, uh, of the conversation in the spotlight.

8. 22:44 – 27:01

   HuggingFace’s Business Model

   1. CDClément Delangue22:44

   2. HSHarry Stebbings22:44

      I do want to ask you, I spoke to many of your investors before the show, and, um, all of them said to, to dive into business model and how Hugging Face makes money, speaking of kind of the relationship with content providers and publishers there. How do you respond to-

   3. CDClément Delangue22:57

      Yeah.

   4. HSHarry Stebbings22:58

      ... how does Hugging Face money? What does that look like in the long term, do you think?

   5. CDClément Delangue23:01

      So, our model is, is, uh, kind of simpler than what, what people think, right? As, um, as a platform with a little usage, we count, like, for the kind- kind- kind of classic freemium model, right? Where most of the companies using us are using us for free. We have 15,000 companies using us now. And then a smaller subset of companies are actually paying us, right? And for us, it's 3,000 companies. And the reason why they're paying us is for premium features, so typically enterprise features, like single sign-on. Premium support, right? When they need help to use our tools, and premium compute, right? For example, they wanna use Hugging Face, but they wanna upgrade to faster GPUs. Then they're going to pay us for, for that. So, 3,000 companies are paying us for that, including Meta, including Bloomberg, including Grammarly, and, and companies like that.

   6. HSHarry Stebbings24:12

      How do you charge them? Is it on a seat basis? Is it on a, on a volume-of-query basis? What's the pricing model aligned to that business model?

   7. CDClément Delangue24:20

      It varies. Um, I think we s- haven't really...... figured out and optimized yet for maximum revenue, because our main priority is more, like, adoption usage as a platform with network effects.

   8. HSHarry Stebbings24:37

      Clem, do you get pissed off when people ask you how are you gonna make money? Do you think it's the wrong question to ask?

   9. CDClément Delangue24:41

      No. I mean, it's, it's not the most important question to ask, right? Because, um, as a platform, uh, with network effects, the adoption and the usage is kind of like the number one, uh, KPI for us, right? Especially, it's, it's something that we ... it's an assumption and kind of like a position that we took very early on with, with Hugging Face, especially coming also from, like, more, like, consumer backgrounds where it's really obvious that, you know, like, a Facebook or, like, a Twitter, um, you know, needs to focus on adoption usage first. And this kind of like assumption that we have that, uh, usage is delayed revenue, right? Especially on the domain like AI where you expect companies to be ready to pay for AI. So if, you know, Hugging Face keeps being the number one platform that companies are using to build the AI, it's fairly obvious that we're going to be able to make, uh, a lot of revenue out of that and, and build a, a good business around it. But at the same time, even if it's not the most important question, it's, uh, it's an interesting question. And the way, the way I see it for, for us as a platform is that with monetization, we kind of like have to take it as, like, stepping stones and almost kind of like, uh, uh, unlock some learning progressively to go from, you know, you start with like six-figure revenue. You learn from that, you see how it works, then seven-figure revenue, eight-figure revenue, nine-figure revenue, and you learn at each step, especially on AI because, like, the underlying technology is moving so fast that probably the way we make money today is not going to be the way we make money in three years or in, in five years.

   10. HSHarry Stebbings26:52

       Okay.

   11. CDClément Delangue26:52

       So it's interesting to do this, uh, monetization and revenue learning that, that we do, I think.

   12. HSHarry Stebbings26:59

       Can I ask, Clem, you said about adoption

9. 27:01 – 30:38

   AI Startups vs AI Incumbents

   1. HSHarry Stebbings27:01

      there. I always think about Alex Rampeal at Andreessen who said, "The question in company building is will the incumbent inquire i- acquire innovation before the startup acquires distribution?" And you mentioned adoption there, and it leads me to think about who gains from this next wave most predominantly. And my thinking more and more is incumbents who are fast-moving, like Microsoft, who are incorporating it into PowerPoint, like Adobe, who are incorporating it into all of their suite of products, they will take 90% of the gains actually because they have the distribution and they're moving fast. Do you agree, or do you think actually startups are the ones who will accrue the most value in this next generation?

   2. CDClément Delangue27:48

      It's a bit related to what we talked about before. Um, I think if you're thinking about AI as, uh, AI APIs, I agree with your analysis. If you're thinking of AI as a more radical paradigm switch to build technology, right? And if you think about an AI startup as a company that is actually training models, creating new architectures, optimizing models themselves, I think it's a different story because, uh, this is really hard to do for the incumbents. And so I think the new startups have kind of, like, an opportunity there to do things 10 times, 50 times, 100 times better than the, than the incumbents.

   3. HSHarry Stebbings28:36

      Why, why is it hard to do for the incumbents? Sorry, I'm naive here. Why is that difficult for them to do?

   4. CDClément Delangue28:41

      'Cause it's a completely different way to, to build technology, right? It's a, it's a way where you have to have scientists, for example, working for six months on a new architecture, on a new model before releasing it. Um, so it's, it's a, it's a different paradigm of how you build software, and it's different enough in my experience that, uh, that it's, it's hard to do for, like, bigger teams and bigger companies that are moving slower and that have started with, like, a, a very different paradigm. At least that's what I'm kind of, like, seeing, um, on, on the field. It's hard to predict the future, again, but, uh, but I think there are many, many opportunities for really AI-first startups and really startups who are not just using AI with APIs, but really building AI themselves, building new architecture, building new models, optimizing their, their own models.

   5. HSHarry Stebbings29:44

      What are the biggest challenges or barriers those AI-first startups face? Is it data model access? Is it hiring? Is it ... You name many of the challenges. What's the biggest challenge you think this next generation of AI-first companies face?

   6. CDClément Delangue29:58

      I would say hiring probably right now, like getting, getting, like, the, the best people and getting kind of like this hybrid kind of, like, profile because it's science plus engineering. I, I would say right now, like, hiring and getting the right set of co-founders, early team members is the hardest, hardest thing.

   7. HSHarry Stebbings30:22

      Totally get you.

   8. CDClément Delangue30:22

      Especially because there's a lot of competition with, with also others, companies that have raised so much, so much money that, you know, like, for the really good people, the salaries are, are kind of like insane.So that's, in my opinion, the biggest thing.

10. 30:38 – 33:20

    Why AI Startups are Expensive

    1. CDClément Delangue30:38

    2. HSHarry Stebbings30:38

       We've seen some monster funding rounds very early for some of these companies, 100 million, 200 million. Do they, do they fundamentally need that funding for data model access for, for something specific? Or is it a case of the demand is there and so raise what you can? I, I don't understand. Does it cost more money to build AI-first companies than all generation of prior companies?

    3. CDClément Delangue31:05

       It does a little bit. Um, a bit similarly, in my opinion, to how, you know, like you would build, like, an internet company or like a software company 20 years ago because, um, compute is kind of like more expensive for AI than it is for traditional software because the team members that you need to hire, as we mentioned, are more expensive than, uh, if you need regular software engineers. So it does cost more money to build an AI-first startup than, uh, a regular kind of like software startup. One thing that I am not really sure of is, uh, if this model of not needing just a little bit more money, but much more money is really the right approach or not. Um, especially because, uh, we're starting to realize that more compute for models is not necessarily the right thing, or at least that it's, it's not enough, and the return on investment on training larger and larger models is starting to go down. So it changes a little bit, the perspective on, on raising more money and more, like, having more money spent on compute as some sort of a moat or barrier to entry. But like all technology cycles, you know, you have companies taking, taking bets, right? And if investors want to take the same bets, I don't think it's a negative thing. I think it's, it's a good thing. It doesn't mean that they'll all succeed, right? And that it's gonna work, but uh, it's an interesting risk to take and interesting kind of company to, to build in my opinion. So I'm, I'm excited to see some of what these companies are, are going to build and, and do in the future.

    4. HSHarry Stebbings33:12

       Clément, can I ask, um, before we touch on your fundraising, which I, I heard some great stories about by the way, um, but can we, uh, just touch

11. 33:20 – 37:46

    AI Regulation

    1. HSHarry Stebbings33:20

       on, you know, Elon has said before that fundamentally, unlike most, uh, regulatory environments, AI, you cannot wait until it's in play for you to regulate it. Once it's in play, regulation will not be effective. Do you agree with that? And how do you predict the regulatory landscape to play out in the next 6 to 24 months?

    2. CDClément Delangue33:43

       I don't really agree with that. My point of view is that as, as you've heard, I'm extremely excited about AI. I think it's a new paradigm to build tech. I believe at the same time, we are very far from a world where AI is autonomous and has conscience and is taking over the world and destroying humanity. I think this is more fear that is very sci-fi driven, that we're very far from. And so I think when you take this point of view, you realize that regulation is necessary because, you know, it's a new way of building technology and it's going to create some challenges, but these challenges are not so much AI running wild autonomously and taking over the world. These challenges are more biases that are included in these models, misrepresentations or misinformation that these models can amplify, and, and these need to be regulated the same way traditional technology has been regulated or a little bit maybe differently, but not in an a priori way where you're like, "Let's put a pause on everything, let's stop everything because maybe it's gonna kill humanity." Because if you, if you do that, I think you risk killing the advantages that we can get from the, from the technology, killing the progress and actually not solving the problems that we're seeing today that needs to be solved. So in that way, I feel like a very different approach than, um, Elon Musk, but also than, than OpenAI, I think.

    3. HSHarry Stebbings35:43

       Can I ask one final thing before we discuss the qu- fundraisers? Is there anything else which you hear often which you find annoying in terms of misrepresentation? You've been in this for 15 years. Suddenly everyone wants to talk about something that you've done for so long, and, uh, there's always this case where like you know something so well and it's like, "Oh, for fuck's sake, I wish you'd stop referring that." Is there anything that y- you hear today that annoys you 'cause it's wrong?

    4. CDClément Delangue36:08

       Yeah, I mean the biggest thing is, uh, all this talk about AGI and anthropomorphization of, of AI, right? Kind of like, uh, considering and characterizing AI as, as human and sh- and saying that we close to, you know, the, the Robocop scenario where AI is taking over the world and, and killing all, all humanity. The truth is, uh, we're very, very far from that.AI right now is just a new paradigm to build technology, right? Instead of writing, writing a million lines of code, now you use machine learning to build features, to build product, to build workflows. It's an evolution that is going to be important but it's, it's not kind of like an autonomous semi-human being. We're, we're very, very far from that. So that's kind of, like, the thing that is the most annoying to me. I think it's important to work on these topics for the long-term. It's important that some researchers do some research on the topic. But at the same time, it's important that it's not taking over the whole public narrative, and that we work also on some of the challenges of AI that happen right now with the current technology, and not just kind of like a sci-fi driven long-term threat that we're not even sure is going to happen anytime.

    5. HSHarry Stebbings37:35

       I, I, I loved another Yazhi Yamakan tweet which said, um, "Before we talk about God-like AI, can we please build dog-like AI?" (laughs) Um-

    6. CDClément Delangue37:44

       Yes, absolutely.

    7. HSHarry Stebbings37:45

       But,

12. 37:46 – 52:27

    Fundraising at HuggingFace

    1. HSHarry Stebbings37:46

       uh, I heard some wonderful investors said the magic words, "Here is a term sheet before we meet." Um, I've no idea, I mean, that is just some real balls. Um, what happened there? Take me to that.

    2. CDClément Delangue38:01

       Yeah. So I have some, um, rules with, uh, with investors that I set for myself and that I think have been pretty useful to me. One of these rules is that I don't talk to any investors, uh, external investors in between rounds, right? I'm making an exception for you today because it's a, it's a podcast, but otherwise I, I don't talk to anybody in between rounds, uh, because I feel like a lot of the time, it's some sort of a waste of time, some sort of a de-focus. In my opinion, it's hard enough to build a company, um, not to be 100% focused on that. And so, that's one of my rules, and then when I raise rounds, uh, it usually goes pretty fast, and so I have a window where I talk to external investors, and then at some point, I start getting term sheets, and so then I, I stop talking to other investors, right? When, when I feel like I've, I've got enough term sheets with the people who are interested and serious, I just stop talking to other investors. And there was this, uh, funny story of, uh, of an investor, and, and I don't think I, I should name him, but, uh, who arrived a little bit late in the process, and so I, I told him, you know, "I'm sorry. You know, it's been, it's been a week. I'm, I'm kind of like, I have my term sheets, so unfortunately the rule now is that I don't talk to external investors who haven't sent me a term sheet." Right? And I was expecting the conversation to stop there. You know, I was a bit sad because it's someone who I liked on, on paper, but the funny thing that happened was that he said, "Okay, here is a term sheet."

    3. HSHarry Stebbings39:56

       (laughs)

    4. CDClément Delangue39:58

       Before even, uh, talking to me, just, like, as a reply on an email. I never talked to him before, like, on the phone or, like, never, never met.

    5. HSHarry Stebbings40:06

       I, again, so I'm just gonna push back on you. I've learned over the years, I have opinions too and I'm not the little Charlie and the chocolate factory and, uh, I, I think that's the wrong approach. And the reason I say it's the wrong approach is because people invest in lines, not dots. And if you meet me during a fundraise, it is not a long enough time period to build a relationship of trust, authenticity, respect, that's gonna be very prominent in your life for 15 years. And so I, I think you should be very careful about who you speak to, three, maybe five investors who you respect intensely, and build the relationship in between. But not to speak to any, you're doing a shotgun marriage. Why am I wrong?

    6. CDClément Delangue40:53

       Well, uh, you have a point, of course, but, uh, if you take the founder's perspective, what's challenging is to identify the investors you're talking to, because the reality is that outside of fundraising, all investors want to talk to, to you, right? But it doesn't really mean that they're serious about, you know, what you do, what you're building, and that you're aligned with them, right? So, how do you pick these investors? Especially in a fast-moving startup like Hugging Face where our investors for the seed when we were doing, like, Tamagotchi AI consumer product are very different than our investors for the B where we're doing an AI B2B platform, right? So if, if I would have stopped and invested a lot of time talking to a lot of consumer investors between the C and the A, it ends up basically be a waste of time. Also something that I've seen is that an investor ... I mean, you're, you're a better investor than, than most, right? So I'm, I'm talking about your average, average investor usually has quite a different, um, approach when they're talking to you and they're not investors because their whole job at that time is basically to make you like them-

    7. HSHarry Stebbings42:21

       Mm-hmm.

    8. CDClément Delangue42:22

       ... right? Versus when they're an investor. So it's hard to say if the relationship that you create with investors before they're investors is really indicative of the relationship that you're gonna have when they're going to be actual investors. And then the last point is that-I might not spend, you know, like, a year talking to investors, but when I pick them during the fundraising, I spend, like, a shit ton of time with them. I spend, like, at least, like three days full time, which is, which is a lot of time, three days full time with investors. I do shit ton of background check. I- I really, uh, it's a shorter period of time, but much more intense. So I- I feel like it's still giving me s- a good, good signal of, like, is it going to be a good relationship? Are we aligned? Are we have, like, similar expectations and, and all of that? So I- I'm not saying it's- it's perfect, but for me, for me, it's been, um, it's been, it's been working well, I think.

    9. HSHarry Stebbings43:35

       I think, I think you're totally fair on the pivot element. You know, I think that's a very unique element of Hugging Face where you're right, you could've spent time with consumer-facing investors, and maybe it would be a different

    10. NANarrator43:45

        Yeah.

    11. HSHarry Stebbings43:45

        ... type of investor that does ? Totally get you there. In terms of who guides you, your seed investors. You have Thibault. Thibault knows everyone. Thibault is the center, and you have a great roster of k- people on your cap table. I think, you know, for founders listening, it would- it would be your seed investors very much so who guide you. And then I think what I would say, and sorry, Clem, I didn't mean to push back, but fuck it, people love it when I push back, and then - Um, the idea of, like, intensity of relationship within three days, that is a completely manufactured relationship. I will tell you anything you want to hear, baby. It is a relationship of hierarchy and imbalance. The reason why raising, or, like, communicating outside now, it- and- it why this is great. I'm not selling you shit. You're not selling me shit. I'm just getting to know you, and you're getting to know me. And when the three days are in, I'm selling you or you're selling me, and it's like a marriage where that doesn't work. Uh, it's a b- it's an equal balance. But in those days, there is an imbalance, which means you're not getting impurity.

    12. CDClément Delangue44:49

        I'm not sure. I think it's closer to the founder investor relationship than, you know, you and I talking like that without, you know, any, any kind of, like, really fundraising goals in, in mind, you know? It's, it's also, also a different relationship now than it is, you know, if you would be an investor for Hugging Face, I feel like.

    13. HSHarry Stebbings45:11

        I would love to be. Thank you. Actually, I... (laughs)

    14. CDClément Delangue45:13

        (laughs)

    15. HSHarry Stebbings45:14

        It worked, it worked, this interview style work.

    16. CDClément Delangue45:17

        (laughs)

    17. HSHarry Stebbings45:17

        Um, uh, I suppose-

    18. CDClément Delangue45:18

        Yeah. So could be because the flip side of the coin of me not talking to external investors between rounds, is that if I do, then it means that you can become an investor, right? Otherwise, it wouldn't work.

    19. HSHarry Stebbings45:33

        It wouldn't. That's why I'm a good investor.

    20. CDClément Delangue45:34

        Just kidding.

    21. HSHarry Stebbings45:35

        I lull you into a full sense of security, and then I pray. Um, I spoke to Brandon Reeves, um, before the show, a wonderful mutual friend of both of ours, and an investor in Hugging Face. He said you have some spicy takes on the venture ecosystem. What are your spiciest takes on venture?

    22. CDClément Delangue45:53

        Well, I mean, I think something I believe in is that investors are first and foremost investors, right? Um, meaning that their main value adds is to do rounds, to help you on financial matters, so for example, when- when SBB go down, right? And then to help you to always capitalize the company the right way. So their, if they're doing their seed, their main job is to help you to do their series A, your series A. If they're doing your series A, their main job is to help you do your, your series B. Um, and that's almost like 95% of the value of, of an investor, to be financially supportive, help you capitalize the, the company, right? And I think, uh, right now, a lot of investors have kind of like a little bit, uh, forgotten that, and they focus most of their time on, on all these things. Um, you know, they- they sometimes act almost as, as CEO or like operators for, for companies, which in, in my opinion is, is not really their, their job. And, and worse than that, I, I feel like sometimes entrepreneurs or building companies for investors and investors or behaving like entrepreneurs, and sometimes it's actually crashing companies just because contrary to an entrepreneur, unfortunately, an investor has a lot of different companies, right? So they can't only spend, like, a short period of time on each company. And even if they're, like, the smartest people in the world, that just this constraint in terms of time makes it so that they have sometimes a more simplistic understanding of, like, the technology, for example, uh, of kind of like the company and things like that. So that, that's kind of like maybe one, one thing where I differ a little bit, um, to, uh, some other entrepreneur or some kind of like, uh, uh, conception of, of venture.

    23. HSHarry Stebbings48:04

        I- I- I totally agree with you. I think a really big problem is entrepreneurs building companies for investors, or in a way that they think investors want to see it. Totally agree with you there. Well, final one before a quick-fire, Clem, and it's, your cap table's fantastic. I love Lux, I love Lee, I love Pat at Sequoia. It's fucking awesome. But you are French, and you are in Paris, and I'm just interested, what do you make of the European venture ecosystem? Because that was not on your cap table other than Thibault.

    24. CDClément Delangue48:34

        Yes, I have, uh, French angels, right? When we, when we started, our first investors were BetaWorks and French Angels. I don't have a lot of experience with, uh, with European venture, because after that, we only had, uh, US VCs.So, um, I'm not sure if I, I would become, like, a good judge of it. Um-

    25. HSHarry Stebbings48:57

        I totally get that. Shit question.

    26. CDClément Delangue48:59

        Yeah.

    27. HSHarry Stebbings48:59

        Fine- final, final question then.

    28. CDClément Delangue49:01

        (laughs)

    29. HSHarry Stebbings49:01

        You've raised money from some of the best in the business. What do you know now that you wished you'd known at the beginning, or what do you advise founders, having seen all that you've seen?

    30. CDClément Delangue49:12

        Well, I mean, uh, for company building, I think one thing that I knew, wished I knew earlier is that it, it doesn't get easier, you know? Like, sometimes when you're, like, an early stage founder, you are, like, struggling and you're like, "Oh, but that's gonna be good. I'm gonna struggle. It's gonna be really hard. But in one year, in two years when I'm gonna be bigger and it's gonna be easier." I'm sorry, but it, it won't be easier. Uh, the truth is that each stage has a lot of challenges. And so I think something important is for entrepreneurs to realize that and enjoy what they're doing now and try to find and build a company that they enjoy building instead of, like, forcing themselves to suffer. And like that, they can build just the, the enjoyment and the, the joy from, you know, building, not the joy of, you know, getting to a series B, getting a series C, getting to IPO, but really the joy of, of building, right? The joy of the, of the journey, of the entrepreneurial journey. And I think it completely changes your, your adventure.
13. 52:27 – 58:46

    Quick-Fire Round

    1. HSHarry Stebbings52:27

       Uh, listen.

    2. CDClément Delangue52:27

       (laughs)

    3. HSHarry Stebbings52:27

       I wanna dive into a quick fire round. So I say a short statement, you give me your immediate thoughts. Does that sound okay?

    4. CDClément Delangue52:33

       Okay. Yeah, let's do it.

    5. HSHarry Stebbings52:34

       Okay. So what do others not know that you know to be true?

    6. CDClément Delangue52:40

       Well, I mean, we talked a little bit about it before but, uh, I think the biggest thing is, in my opinion, all companies will have their own AI models. Like, all companies will have their, their ChatGPT or, like, the, their GPT-4.

    7. HSHarry Stebbings52:53

       I love that. I just asked someone the other day that and they said, "The amount of love your child gets from the age of zero to five will dictate how they act when they are older." (laughs) And I asked you the same, and it's just such a brilliantly different answer, which I love. Uh, tell me, what is the single-

    8. CDClément Delangue53:09

       (laughs)

    9. HSHarry Stebbings53:09

       ... biggest risk to Hugging Face today, do you think? What do you sit around with your team and co-founders and go, "Ah, this is a risk"?

    10. CDClément Delangue53:17

        Well, I mean, the biggest kind of, like, market risk for us is that if, uh, AI fails to deliver, right, like being kind of like an AI platform, if AI fails to deliver, it's not gonna work for, for Hugging Face. No, no matter what. So that, that would, that would be kind of, like, the bigger, biggest risk. That's why we're taking such a community, open-source approach to things and we're being so community-driven and supporting the old ecosystem because at the end of the day, if AI wins, we win. And so the most important thing is that we contribute to the community, to the ecosystem for, for this to happen.

    11. HSHarry Stebbings53:56

        You said that Brandon's your favorite VC. Who's your favorite angel? Who's the most impactful angel that you've had?

    12. CDClément Delangue54:05

        You're gonna be, you're gonna get me in trouble for that but I have to, if I have to, to pick one. Uh, I have many, many really great angels here. You mentioned, uh, Thibault, Thibault and Zia in, in Europe. Uh, but I would, I would go with, uh, Richard Socher who is one of the most prominent scientists in NLP. He's like a chameleon. He's, he's been one of the most influential researcher in NLP. Then he went to join Salesforce and he was the chief scientist at Salesforce for, for a few years. And now he went back to starting a company and he's starting this company You.com which is disrupting search engines. Uh, and he's, he's been one of, uh, my favorite angel investors. He's been with us-... almost since the beginning and has helped us in so many different topics, uh, because of his background from the science side, business side or the entrepreneur side. Um, I, I really enjoy having him part of the, the adventure.

    13. HSHarry Stebbings55:15

        Following. What a great dude. Wow. Okay. Great answer. Love that one. Uh, tell me, what's the most painful lesson that you're also pleased to have learned because you learned a lot from it?

    14. CDClément Delangue55:28

        Well, I mean, I, I said that before, but, uh, the fact that nothing gets easier is, is probably, like, uh, the biggest, biggest one for me because it changed my mindset, right? Once I realized that nothing was getting easier, I started to focus much more on the process, much more on, you know, not building a big company or not building kind of, like, the biggest company, but building the company that I enjoy building and that I think needs to be built. And so it changed, uh, quite a lot my, uh, my mindset and it proved to be, uh, quite, quite useful and, and impactful, I think.

    15. HSHarry Stebbings56:13

        Penultimate one. What's the hardest role to hire for today for you?

    16. CDClément Delangue56:18

        I would say, you know, machine learning engineer, and by machine learning engineer, I mean someone who's really building a new architecture for AI models and able to train state-of-the-art models. There are just, um, in my opinion, few people in the world who has been known and who has done that. In the past, maybe 50 to 100 people. Hopefully there's going to be more and there are a lot of people who've never done it before who are going to be able to do it now. But it's, it's a very, very short supply, uh, in terms of, like, number, number of people, and a very, very difficult, uh, background to, to hire for right now.

    17. HSHarry Stebbings57:08

        Clem, final one. If all the stars align, if everything goes right, how big could Hugging Face be and, and what company is that in 10 years?

    18. CDClément Delangue57:18

        I don't know if it's a good question for me because, um, I think our goal is not fundamentally to build, like, the biggest company of all. I mean, we see that more as kind of, like, a side effect of building something impactful, useful for, for everyone. And one actually of the advice that I give a lot of entrepreneurs that I meet is to make sure to remember to build, like, the company that they want to build and the company that they think is needed to be built, even if that means not being obviously, like, the biggest company. So hopefully, uh, in 10 years, Hugging Face would be the most impactful organization and company in AI, in this new paradigm that is AI. And then maybe the side effect of that is that Hugging Face is, like, the biggest company of all, but that's not, in my opinion, kind of, like, the number one goal. It's almost a side effect of having an impact.

    19. HSHarry Stebbings58:28

        Clem, listen, I've loved this. I haven't got you in trouble at all, uh, there's no term sheet coming your way from me, sadly, um, uh, for me.

    20. CDClément Delangue58:35

        Sure.

    21. HSHarry Stebbings58:36

        (laughs) But, uh, I so appreciate you taking the time, my friend, and I've enjoyed it immensely.

    22. CDClément Delangue58:42

        Likewise. It was an amazing conversation. Thank you so much for taking the time.

Episode duration: 58:46