The Exchanges

Q everyone, they've asked, it's been with Valley VCs, and they've said, AI, the heartbeat of 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.

A I think there's no denying that there's tremendous excitement activity there. Before that, I felt 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 5000 people joining for like a huge community showcase. Events 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 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 of the points is, uh, for example, Lama, 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 Lama are actually based in, in Paris, right? In, in the Meta AI, uh, lab that, that they have there that is, that is huge.

AI assessment note: “if you look at AI scientists, if you look at ML engineers, it's heavily distributed”

Q Listen, at least when you do go public, the ticket will be an emoji. So, I mean, you know, All you need to do is, you know, get to that stage. 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?

A 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 want to 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 a lot of people were talking about it at the time, but we were super excited about it as kind of like a new paradigm to build technology, new opportunities and all of that. The first company, the first startup I worked for like, 15 years ago was already doing AI. We weren't calling it AI at At the time. So I had some software glimpse of the, of the capabilities. And when we started hugging face, when we started the company, we're like, okay, uh, what can we work on that is going to be both scientifically challenging because one of our co-founders, Thomas is, is a scientist. Uh, and we, we all have a lot of interest 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, however you call it, some sort of like a Siri, Alexa or like ChatGPT, 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…

AI assessment note: “formed because of some sort of, you know, professional crush between me and my co-founders”

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

A So our model is, is, uh, kind of simpler than what, what people think, right? As, um, as a platform with a lot of usage, we kind of like photo 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 3000 companies. And the reason why they're paying us is for premium features. So typically enterprise features like single sign-on. Premium supports, right? When they need help to use our tools and premium compute, right? For example, they want to use Hugging Face, but they want to upgrade to faster GPUs. Then they're going to pay us for, for that. So, 3000 companies are paying us for that, including Meta, including Bloomberg, including Grammarly and, and companies like that.

AI assessment note: “we kind of like photo kind of classic freemium model”

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

A 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. They are just, um, in my opinion, a few people in the world who has been known and who has done that in the past, maybe 50 to a hundred 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.

AI assessment note: “I would say, you know, machine learning engineer”

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

A You're gonna be, you're not getting me in trouble for that. If I have to, if I have to pick one, I have many, many really, really great angels who you mentioned, uh, Thibault, Thibault Alizière in Europe. Uh, but I would, I would go with, uh, Richard Sucher, who's one of the most prominent scientists in NLP. He's like a chameleon. He's, he's been one of the most influential researchers in NLP. Then he went to join Salesforce and he was the chief scientist at Salesforce or 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 has 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, the business side, or the entrepreneur side, um, I, I really enjoy having him part of the adventure.

AI assessment note: “I would go with, uh, Richard Sucher, who's one of the most prominent”

Q Listen, at least when you do go public, the ticket will be an emoji. So, I mean, you know, All you need to do is, you know, get to that stage. 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?

A 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 want to 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 a lot of people were talking about it at the time, but we were super excited about it as kind of like a new paradigm to build technology, new opportunities and all of that. The first company, the first startup I worked for like, 15 years ago was already doing AI. We weren't calling it AI at At the time. So I had some software glimpse of the, of the capabilities. And when we started hugging face, when we started the company, we're like, okay, uh, what can we work on that is going to be both scientifically challenging because one of our co-founders, Thomas is, is a scientist. Uh, and we, we all have a lot of interest 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, however you call it, some sort of like a Siri, Alexa or like ChatGPT, 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…

AI assessment note: “We actually started with something completely different. Um, the reality is that the company was”

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

A So our model is, is, uh, kind of simpler than what, what people think, right? As, um, as a platform with a lot of usage, we kind of like photo 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 3000 companies. And the reason why they're paying us is for premium features. So typically enterprise features like single sign-on. Premium supports, right? When they need help to use our tools and premium compute, right? For example, they want to use Hugging Face, but they want to upgrade to faster GPUs. Then they're going to pay us for, for that. So, 3000 companies are paying us for that, including Meta, including Bloomberg, including Grammarly and, and companies like that.

AI assessment note: “for us, it's 3000 companies. And the reason why they're paying us is for premium features”

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

A You're gonna be, you're not getting me in trouble for that. If I have to, if I have to pick one, I have many, many really, really great angels who you mentioned, uh, Thibault, Thibault Alizière in Europe. Uh, but I would, I would go with, uh, Richard Sucher, who's one of the most prominent scientists in NLP. He's like a chameleon. He's, he's been one of the most influential researchers in NLP. Then he went to join Salesforce and he was the chief scientist at Salesforce or 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 has 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, the business side, or the entrepreneur side, um, I, I really enjoy having him part of the adventure.

AI assessment note: “I would go with, uh, Richard Sucher, who's one of the most prominent”

Q everyone, they've asked, it's been with Valley VCs, and they've said, AI, the heartbeat of 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.

A I think there's no denying that there's tremendous excitement activity there. Before that, I felt 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 5000 people joining for like a huge community showcase. Events 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 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 of the points is, uh, for example, Lama, 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 Lama are actually based in, in Paris, right? In, in the Meta AI, uh, lab that, that they have there that is, that is huge.

AI assessment note: “if you look at AI scientists, if you look at ML engineers, it's, it's heavily distributed”

Q Can I ask, well, I, I brought you a fantastic tweet from Yann LeCun, who said that the biggest obstacle to, you know, open by, 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?

A 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, 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-IV. 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.

AI assessment note: “he has a point. Um, I, I would argue that it's a challenge”

Q It would, and that's why I'm a good investor. 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?

A Well, I mean, I think something I believe in is that investors are first and foremost investors, right? Um, meaning that their main value add is to deal around To help you on financial matters. So for example, when, when SVB go down, right? And then to help you to always capitalize the company the right way. So they're, 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 other 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 feel like sometimes entrepreneurs are building companies for investors and investors are 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…

AI assessment note: “investors are behaving like entrepreneurs. And sometimes it's actually crashing companies”

Q 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?

A No, I don't think, I don't think you do. Um, so the game face, we, we based, uh, we 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? Uh, and you, you can travel there. But, uh, at the end of the day, I think you can, you can build a company from anywhere now. Uh, the most, most important thing, and 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 build their companies. It's, in my opinion, like the number, number one criteria.

AI assessment note: “No, I don't think, I don't think you do.”

Q Fixel credit for this, but, you know, he essentially posed The, the notion that, you know, there's one model, which is your kind of open AI'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?

A Well, they, they differ a lot in kind of like, um, where do you allocate AI builders, right? If it's just like one model to rule them all, you bet on kind of like models getting bigger and bigger with more and more generalist capabilities. And the builders of these models being concentrated in like one Or few organizations, right? In the model where you think there are a lot of different models, you bet on things being more distributed, on the fact that all companies are going to actually build and train models. And that, that comes from the thinking that In the simplistic terms, the model or 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 going to be one code base that is going to roll all code bases. The truth is, the code base is good or bad depending on your use case, depending on your constraints, depending on what you want to do, right? So if you're Facebook, you have one code base that does what you want to do for your users. If you're Slack, you have one code base that is optimized and does what you want to do for, for your users. And it's the same thing, in my opinion, for, for AI, right? Like if you were The company that wants to do consumer products, you need to build AI models that are opti…

AI assessment note: “they differ a lot in kind of like, where do you allocate AI builders”

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

A 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 time it's some sort of a waste of time, some sort of a defocus. In my opinion, it's hard enough to build the company, um, not to be a hundred percent focused on that. And so that's one of my rules. And then when I raise the 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 stopped talking to other investors, right? When, when I feel like I've, I've got enough term sheets with the people who are interested in serious, I just stopped talking to other investors. And there was this, uh, funny story of, of an investor and, and I don't think 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 role now is that I don't talk to external investors who haven't sent me the term sheets. Right. And I was expecting a…

AI assessment note: “he said, okay, here is a term sheet. Before even, uh, talking to me”

Q 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 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 you hear today that annoys you because it's wrong?

A 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, and saying that we're 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 products, 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 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.

AI assessment note: “the biggest thing is, uh, all this talk about AGI and anthropomorphization”

Q 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?

A Well, you have a point, of course, but, uh, if you take the founder's perspective, what's challenging is to identify the investors you are talking to, because the reality is that outside of a fundraising, all investors want to talk 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 Ewingface where our investors for the seed when we were doing like Tamagotchi AI consumer products are very different than our investors for the B where we're doing an AI B to B platform, right? So if, if I would have talked 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 most, right? 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, 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 investo…

AI assessment note: “It ends up basically be a waste of time.”

Q 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?

A No, I don't think, I don't think you do. Um, so the game face, we, we based, uh, we 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? Uh, and you, you can travel there. But, uh, at the end of the day, I think you can, you can build a company from anywhere now. Uh, the most, most important thing, and 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 build their companies. It's, in my opinion, like the number, number one criteria.

AI assessment note: “No, I don't think, I don't think you do.”

Q Fixel credit for this, but, you know, he essentially posed The, the notion that, you know, there's one model, which is your kind of open AI'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?

A Well, they, they differ a lot in kind of like, um, where do you allocate AI builders, right? If it's just like one model to rule them all, you bet on kind of like models getting bigger and bigger with more and more generalist capabilities. And the builders of these models being concentrated in like one Or few organizations, right? In the model where you think there are a lot of different models, you bet on things being more distributed, on the fact that all companies are going to actually build and train models. And that, that comes from the thinking that In the simplistic terms, the model or 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 going to be one code base that is going to roll all code bases. The truth is, the code base is good or bad depending on your use case, depending on your constraints, depending on what you want to do, right? So if you're Facebook, you have one code base that does what you want to do for your users. If you're Slack, you have one code base that is optimized and does what you want to do for, for your users. And it's the same thing, in my opinion, for, for AI, right? Like if you were The company that wants to do consumer products, you need to build AI models that are opti…

AI assessment note: “they differ a lot in kind of like, um, where do you allocate AI builders”

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

A 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, of course, 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 a Geekbase, 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 a lot of revenue out of that and build a good business around it. But at the same time, even if it's not the most important question, It's a, 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. So then seven figure revenue, ei…

AI assessment note: “it's not the most important question to ask, right?”

Q 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 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 you hear today that annoys you because it's wrong?

A 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, and saying that we're 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 products, 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 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.

AI assessment note: “the biggest thing is, uh, all this talk about AGI and anthropomorphization of, of AI”

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

A 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 time it's some sort of a waste of time, some sort of a defocus. In my opinion, it's hard enough to build the company, um, not to be a hundred percent focused on that. And so that's one of my rules. And then when I raise the 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 stopped talking to other investors, right? When, when I feel like I've, I've got enough term sheets with the people who are interested in serious, I just stopped talking to other investors. And there was this, uh, funny story of, of an investor and, and I don't think 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 role now is that I don't talk to external investors who haven't sent me the term sheets. Right. And I was expecting a…

AI assessment note: “the funny thing that happened was that he said, okay, here is a term sheet”

Q 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?

A Well, you have a point, of course, but, uh, if you take the founder's perspective, what's challenging is to identify the investors you are talking to, because the reality is that outside of a fundraising, all investors want to talk 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 Ewingface where our investors for the seed when we were doing like Tamagotchi AI consumer products are very different than our investors for the B where we're doing an AI B to B platform, right? So if, if I would have talked 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 most, right? 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, 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 investo…

AI assessment note: “if you take the founder's perspective, what's challenging is to identify the investors”

Q Can I ask, well, I, I brought you a fantastic tweet from Yann LeCun, who said that the biggest obstacle to, you know, open by, 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?

A 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, 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-IV. 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.

AI assessment note: “Yeah, he, he has a point. Um, I, I would argue that it's”

Q 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 within incumbents? Like, I think OpenAI and ChatGPT to the world brought this awareness. Was that a step function change, or was that actually just the continuation of the catch-up that you mentioned?

A 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 saying 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 have been so open in sharing their research and that everyone is building on top of each other with this really interesting loop. Uh, 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 birth paper, uh, the latent diffusion paper. Maybe we would be 3040, 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, chat GPT and also kind of like on the underlying technology stack, I think you've had better hardware, right? And kept 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 rig…

AI assessment note: “you started to have some mainstream breakthroughs, right? Like you mentioned, you know, chat GPT”

Q I totally get that shit question. Final, final, final question then. You've raised money from some of the best in the business. What do you know now that you wish you'd known at the beginning, or what do you advise founders having seen all that you've seen?

A Well, I mean, uh, for company building, I think one thing that I knew, wish I knew earlier is that it, it doesn't get easier. You know, like sometimes when you like an early stage founder, you were like struggling and you're like, oh, but that's gonna be good. I'm gonna struggle. It's gonna be really, 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 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 the 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 the 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.

AI assessment note: “one thing that I knew, wish I knew earlier is that it, it doesn't get easier.”

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

A Because 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, 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 saying, 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.

AI assessment note: “Because it's a completely different way to, to build technology, right?”

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

A Because 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, 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 saying, 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.

AI assessment note: “Because it's a completely different way to, to build technology, right?”

Q This will be great. So I want to start with 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?

A Yeah, when we started Hugging Face, we joked with my co-founders, Julia 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 all of that is kind of boring. Felt like it 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 with hands like that. What's our favorite emoji? So we're 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 the clothes, like literally everywhere. So we were like, oh, maybe we're going to keep it. And now it becomes kind of like such, such a brand, so popular that unfortunately it's going to be hard for us to, to change it.

AI assessment note: “So we absolutely wanted an emoji as a name. And the choice is the Hugging Face emoji”

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

A 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, of course, 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 a Geekbase, 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 a lot of revenue out of that and build a good business around it. But at the same time, even if it's not the most important question, It's a, 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. So then seven figure revenue, ei…

AI assessment note: “it's not the most important question to ask... even if it's not the most important question, It's an interesting question”

Q 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 within incumbents? Like, I think OpenAI and ChatGPT to the world brought this awareness. Was that a step function change, or was that actually just the continuation of the catch-up that you mentioned?

A 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 saying 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 have been so open in sharing their research and that everyone is building on top of each other with this really interesting loop. Uh, 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 birth paper, uh, the latent diffusion paper. Maybe we would be 3040, 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, chat GPT and also kind of like on the underlying technology stack, I think you've had better hardware, right? And kept 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 rig…

AI assessment note: “I obviously have an agenda there, but to remember that most of the progress”