Tech Brew Ride Home - (BNS) Hugging Face Founder Clément Delangue

Episode Date: October 23, 2025

Clem discusses his journey from early computing experiences to founding Hugging Face, emphasizing the importance of community, collaboration, and open-source technology in the AI landscape. He reflects on the evolution of technology, the significance of user feedback, and the need for a diverse range of AI models. Clem also shares insights on the startup ecosystem in Europe and the unique advantages of New York City for AI entrepreneurs. Learn more about your ad choices. Visit megaphone.fm/adchoices

It's peak pollination season, and my business is scaling fast. To keep the nectar flowing, I need a phone plan with top priority data speed. That's why I chose GoogleFi Wireless. My connections stay strong even when the hive is buzzing. Plus, unlimited plans start at $35 a month. Now, that's a deal that doesn't stay. Explore Google Fi Wireless plans today. Plus taxes and government fees.

GoogleFi Wireless is not subject to data traffic deprioritization during times of high network usage. Basically, how alive are U.S. open models so far that you're seeing? And do we actually care since the Chinese models seem kind of benign so far? I think I don't want to be too extreme. But it looks like, yeah, this way not enough open source American models right now. at least not to the level of what's released from China. And I think we do care for reasons of concentration of power. I think ultimately we want actually not just China and the US,

but any country to be able to produce their own AI models. Just the same way we want any country to... to be able to write its own code, you know, and write its own software, to provide choice to people to make sure that power isn't concentrated, to avoid some of the biases contained in different models from different countries. Clam, thanks for coming on here to talk to us today. Thanks for having me. So one of my favorite questions long time listeners will know is what was your first computer.

But I read that actually somebody said that your first computer changed your life. So I want the actual geeky model. Like what was the first computer either that was yours or the one that you had access to? I don't remember, to be honest. I was 10. It was like 1998. What I remember was obviously fighting with my siblings with four siblings not only about using it but using it and kind of like calling your friends, right? Because you couldn't receive a phone call when you were on the internet.

Obviously remember the noise of the router when the internet connection, the bottom, yeah, with with start. And we started with my browser. Actually, a few years, few years later, we started to do some internet trading, like buying on one platform, selling on another, which kind of like actually defined some of the following challenges, because at some point we became one of the biggest sellers on eBay, and that led to me joining eBay, actually. Let me get to the eBay thing in a second, but listeners might hear.

You grew up in France. Were you old enough to have had a Minitel access to a Minitel terminal and using that? Yeah, we had a Minutel at home. We weren't really using it much. It was kind of like more one of these machines that you got and that you didn't end up using much except to find addresses and phone numbers. We use it as kind of like a way, a way to look for for businesses to be able to find their addresses and phone numbers.

That's why we were using it for. Okay. So eBay. And I heard, I don't know, maybe you can confirm or inform that. I heard that in France, they spent so much public money into the Minitel, right? Like almost invested more public money into the Minitel than the Internet, which ended up not being kind of like the technology that became dominant. But I always felt like it was kind of like a good way to think about how you want.

to define kind of like technology evolutions and how you usually want it more design and driven by private initiatives instead of public initiatives. Bottom up as opposed to top down. By the way, to give credit to Minitel, there was teletext and there were other things at the same time. This was pre-web. So it kind of was interactivity. Like when I grew up here in the States, like it was things like Prodigy and then eventually AOL, where it had to be a service, because there weren't websites. It wasn't this broader sort of ecosystem.

So I'm going to give credit to Minitel and things like Teletext for that. But you're right. That's the lesson of the web, which was it was technology that was just good enough that it allowed a bottom up energy, which overtook sort of the top down sort of thing. Yeah, absolutely. Okay, you mentioned eBay. You mentioned buying and selling. and you hinted at the fact that you got so successful at like, what, 17 or something like that,

that eBay reaches out to you? No, I reach out to them because I'm like, that's a cool platform. I'm still a few years later, I'm still at school in business school in France. And I have, you do in France some sort of like a break to do internships. And I joined them then to do my internship there. It was super interesting because I, because of what I was doing on eBay and I was a big seller there. I was kind of like a little bit the seller representative in a way, right? like the chief of the union of the sellers, like, you know, trying to move the products into something that can make sense for more sellers.

Wait, are you saying, are you saying you're organizing like a consortium of sellers or are you saying that you're like, you're lending like your seller credibility to other sellers? Yeah. No, I was more kind of like trying to drive the eBay team at the time, right? and kind of like share some insights, share some ideas about, you know, how to make the platform more friendly for sellers. Like, for example, I remember we worked a lot on the mobile app, right, because I felt like at the time it was starting to be important for sellers to just be able to on the move, like take a picture of a product and be able to sell it on eBay, which wasn't that easy at the the time. So I was sharing my seller insights with the rest of the eBay team to try to make the platform better. All right. Let's see if my research is good or bad. eBay does, though, offer you a job at some point? Yeah. Yeah, yeah. After my internship. Okay, but instead of doing that, you start to do either a series of startups slash projects or joining startups or projects, right? Okay,

so let me go down the list a little bit. and this might not be chronologically in order, but tell me what Unishared was. So Unishard was the first startup I started myself when I was still a student. And it was like a collaborative and open platform for students. So I was still a student. Kind of like frustrated that everyone was taking their notes on a piece of paper, not sharing with each other and not sharing with the world. So it was some sort of like Google Docs platform,

Rule Docs wasn't so popular at the time where students could just take their notes together in the classroom and share them externally with whoever was interested in the topic. So we had some kind of like interesting things where we had people from Europe or from Africa who would just like start study with people from Harvard, from Stanford and like collaboratively tried to learn the same things. That was the idea and the vision for Unishade. Did you learn any interesting lessons about sort of bottom-up adoption and community by doing that? Yeah, yeah.

I learned that sometimes we had this vision of really convincing people to be both collaborative and open. And something interesting is that it wasn't always the same people who wanted to do both. You know, there were some students who liked to take notes collaboratively with their friends inside the classroom. Some people who liked to share their notes externally to the world, but it was not necessarily kind of like the same people. So it made it harder for this kind of platform to get kind of like massive, massive adoption. So I learned that sometimes when you have values, when you have a vision, It's important to keep it, of course, and to build a platform that makes it happen, but maybe not to go to extreme too fast, right?

And rather kind of like help the users progressively get to the point that you'd like them to be and be a bit more pragmatic, practical, in a way, less idealistic. I think at the time I was super idealistic and I thought that overnight I could change change kind of like everyone's behavior versus now I've learned that you have to be more progressive for it to happen and for it to work. That's one of the many learning that I had during this this first venture. You're saying have a vision but also pay attention to what user behavior actually is because that gives you signals. Yeah. What comes next? Moodstocks or mention chronologically? So Moodstocks was before. Before.

Just after eBay. So I was mentioning at eBay, I was kind of like a sent also to all the places where there were sellers. So I was at this kind of like a trade show. It was kind of like a hard time because all the sellers would come to me and complain about the user experience at eBay and complain about the user experience of PayPal because PayPal was part of eBay. And at some point I have a guy who comes with like big round glasses, the very kind of like nerdy, nerdy type. And he's telling me, oh, you are eBay. You just acquired a company that is doing barcode recognition.

It was called Red, Red Laser. And he tells me, you don't need the barcodes anymore with the new technology. So he was calling it Computer Vision at the time, but it's basically AI. Now you can recognize the objects without the barcodes. You just point your phone at the product. And with AI, you can recognize the product. It's like, you're crazy. What is that?

I've never heard of these kinds of things, these kinds of technologies before. And after this event, I do a bit of research. I realize that these guys are quite legit. And I think two months later, I joined them. After my internship at eBay, I joined them at this company called Mootstocks. I was doing computer vision, image recognition, video recognition that I joined and worked for, for I think a little bit over a year. Is that your first introduction to AI or machine learning and things like that?

Yeah, yeah. Yeah, and first introduction to what I call deep tech, really kind of like kind of context when I joined them, I think for the first months, I basically understand nothing about what they're talking about at the coffee machine. Really, kind of like, nothing, nothing. So it forces me to kind of like start learning a little bit about these topics and start exploring this fascinating new paradigm that is AI. One more real quick. A company called Mention, which is some sort of social list. listening startup or something?

Yeah, it was like maybe you remember Google Alerts. So like a system that I would like crawl the web and tell you when your company, your product, your name is mentioned on the web. So that's what it was what it was doing. That's actually the startup that brought me to the US and to New York for the first time because we started to have more and more users in the U.S., more and more customers in the U.S. And so they sent me to New York. That's when I discovered New York for the first time.

What year was that about? I think it was like, I want to say, 2013, something like that. 2013. 12 years ago. So the Great Recession has passed. Tech is sort of hitting the early stages of its boom time in terms of tech eating the world and things like that. Coming from France to New York City, what was the tech sort of startup news or scene like here in New York City when you get here? I felt like it was what I really liked.

is that it was a little bit European in some ways, right? Like I think when I moved to New York, meeting like French entrepreneurs here, meeting British people like John Boers Week from Betarworks. It really felt to me like the perfect merge of some of the things that I liked from Europe and some of the things that I liked from the US. Before New York, I spent quite a lot of time in San Francisco and never really kind of like liked it there. It felt kind of like too far from Europe, not only geographically, but also

philosophically in terms of culture, in terms of lifestyle. And so when I came, when I came, to New York for the first time, I didn't only fell in love with the tech here, but also with the city, right? Really quickly when I arrived in New York for the first time, when I felt the energy, when I felt like the busyness, when I felt all the different projects, how international it is with people trying to make it from all over the world. quickly realized this is this is a city where I want to spend time and where I want to build things and so that's why I decided to move from from Paris to New York actually quitting my my previous job to to move to New York so do you meet I'm trying to get to the

hugging face founding story do you meet Julian and Thomas here in New York City no I already knew them. Julianne from Paris kind of like startup cycle in Paris so I already

know them when I'm in New York and it's actually I think after a year and a half in New York that I meet

John Borthwick from from Beta Works and I talked to him and Matt Hartman was also at the time a partner at Betawks.

And I tell them about some of our side projects with with Julia that we started building. One of them being this sort of like AI Tamaguchi. And they

decide to take a chance on us. They lead our first round of funding. And that's what makes us quit our job and focus full-time on what will become hugging face. Well, I've read it described as like an AI BFF chat bot maybe. You can fill in the blanks here, but I do want to make note of the fact that, again, there's sort of AI in the mix here.

Yeah. Yeah. Yeah, we start with Giulien, really from our passion for AI, for AI, from our intuition that it will change many things. about like how how you build technology, how you interact with technology. And at the time, you have kind of like an iteration of assistance like Siri, Alexa, that gets some usage. But we use them and we like, this is a little bit boring. Like this is very productivity driven. You know, like this is designed to just give you the weather. you know, to answer very practical

productivity driven questions. And we were like, you know, maybe this is something more interactive, something more fun, something more entertaining to build. And I think that's what resonates with John and beta works because, of course, they've been working a lot on the intersection of media,

of entertainment and technology. And so that's how we start. It is interesting that, you know, we're all looking at agents and chat spots now. Ten years later, everything old is new again. But can you tell me briefly you've said it in other places, why choosing the name hugging face? Yeah. I mean, at the time, of course, we were using a lot of emojis, right? They were kind of like starting to be cool and starting to be kind of like,

a new carrier of meaning. And the first joke was, you know, what if we're like the first company to go public with an emoji instead of the three-letter tickers, you know, like when you go in the NASDAQ, you have three letters. And we're like, it'd be cool to add emojis there instead of kind of like having these boring three, three letters. So we're like, okay, let's pick an emoji. We're using the hugging face emoji a lot. Let's use that.

Let's take that and probably in two weeks we'll realize it's a terrible name and we're going to change it. And what happens is, you know, the hugging face emoji. People start to put it everywhere on, you know, social networks. People start to do T-shirts. People start to do the sign, the hugging face, hugging face sign. And as a result, you know, we we just keep it. And I think today, a few years later, we realize that it really resonates with people and kind of like makes our name, our logo, quite unique compared to most of the other people.

Study and play. Come together on a Windows 11 PC. And for a limited time, college students get the best of both worlds. Get the unreal college deal, everything you need to study and play. with select Windows 11 PCs. Eligible students get a year of Microsoft 365 premium and a year of Xbox GamePass Ultimate with a custom color Xbox wireless controller.

Learn more at Windows.com slash student offer. While supplies last, ends June 30th, terms at AKA.m.m.S. College PC. Ambition comes in all shapes and sizes. At First Citizens Bank, we roll with your goals because we're built for what you're building. Fit for your ambition. First Citizens Bank.

Ready to soundtrack your summer? With Red Bull Summer All Day Play, you choose a playlist that fits your summer vibe the best. Are you a festival fanatic, a deep end DJ, a road dog, or a trail mixer? Just add a song to your chosen playlist and put your summer on track. Red Bull Summer All Day Play.

Red Bull gives you wings. Visit redbull.com slash bright summer ahead to learn more. See you this summer. All right, tell me the classic sort of pivot story here of moving away from a consumer chatbot to you decide to open source your internal tooling and focus on developers and the ecosystem around that. Tell me the story of how, was that just a solution to an internal problem that you accidentally found was something more interesting or was that a pivot, pivot where you're like,

maybe there's something more interesting over here? A mix of both because I think that's thanks to what we were working on that we managed to do what we do, but also we stumbled into something that we didn't realize before could be as big as it became. It's basically all credits goes, especially at the beginning to Thomas, our third co-founders, Thomas, Thomas, Thomas Wolfe, who just like one Friday afternoon. is kind of like reaching out to Joanne and I. It tells us, oh, I've seen this thing, this research paper released by Google called Attention is All You Need. And they released this model called Bert, right?

Kind of like the first Transformers model. And it's like, but it kind of sucks because it's intensive flow. And I think most people will want to use Pitewash for it. So I think I'm going to spend a weekend just like putting it from, from TensorFlow to Pytorch. And initially, I mean, Jeline and I was like, okay, you know, have fun. You know, you do you if you think it's exciting. I was barely understanding, to be honest, what it was or the kind of impact it could have.

And then on Monday, tweets, okay, I've released the Pytorch implementation of Bert. And at the time, I think we get maybe a thousand likes on Twitter, which for us at the time, We're like, okay, we broke the internet. What's what's happening? Because of course, we were kind of like random French guys, not having much of visibility or recognition online. And so we realized, oh, there's something there. And so we decided to keep working on it.

And we added more models to the same repository. At the time, it was the founder of Mistral. who were building a model called ExcelNet that was added. There was like a few months later, GPP came up from OpenAI, the first GPT that we added to the library. And really quickly, scientists also joined and added their own models to the library. And kind of like one step at the time, it became kind of like this central platform, this central library.

that everyone in AI was using, which led to Huggin' Face. Well, so it's like a GitHub for AI, and you're mentioning this tweet and getting all of those likes and retweets as like, hey, wait, there's something here. But do you remember a moment or a metric that made you say, oh, the platform is the product and this is what we're gonna do? I don't think there was a specific moment. I think it was kind of like over time,

progressively we grew more and more confidence. And I think the biggest validation for us was community members using it as a way to distribute their work. Right. Like seeing the first scientists using the platform to share the work. You know, sometimes these scientists, they spend, you know, two years, three years, five years, like refining their skills to get to the level where they can, train a state of the art model and the fact that they were sharing it on our platform sharing a Higgin face link interacting with the community on our platform i think uh that's something in

our mind that uh showed us that what we were building was useful to to people impactful for people and that that it could have kind of like a big impact i think one of the things that was was important for us too is, you know, I mentioned some of my previous ventures, right? I mentioned Unishared, which was like open and collaborative platform for students. And in a way, we found kind of like the same values in Hugging Face with a different approach and different targets. But you can almost apply the same vision and the same values because HuggingFace is, of course, an open and collaborative platform. for AI.

And so I think that's also played a big role in our excitement, in our motivation, and our willingness to pivot to that because we could see that our values were actually at play there in this new initiative and that we could achieve some of the impact that we wanted to achieve thanks to this. Well, yeah, we kind of touched on this earlier, like getting the sort of signals from a community, right? Where you have a vision, but you're watching like sort of the health of the community trending up or down. So when you do things like Gradio and spaces and other like products on top of what you struck upon, to what degree as a founder, are you, again,

relying on signals from this community to go in the direction that you think you should go? Or are you How do you weigh your vision versus the signals you're getting from users? It's a tough question. It depends. Always depends, right? There's no kind of like standards pattern or playbook for things like that. But, you know, what we learned is that you want to start from a conviction, from an intuition, right? Like, for example, Thomas, he started from the intuition that if we converted these weights to pie torch, more people will find it useful and more people will use it, right?

And then you have to make the work to build something good, something like actually that's aligned with your intuition. and then you want to release it and see if your intuition is confirmed or not and how the community adopts it and how they're helping to evolve it what they're telling you to kind of like refine in the way your next generation of intuition. That's why at Ticking Face we're releasing so many different things that sometimes people are like, oh, what does it have to do? with everything else. Like why are you building so many different products, different features, different libraries is because we have intuitions, we have ideas, and then we release it for the

community and we know that most of them maybe nobody's going to use it or a very small number of people are going to use it. But some of them are going to prove useful and are going to get adoption and then we're going to double down on them and keep building on top of them. Let me ask a slightly different question, which is when you do pivot and you go to a John Borthwick, one of your early investors, and you say, hey, remember we were doing this bot, but by the way, look at what's happening over here. I'm assuming nine times out of ten, and investors like, great, you're getting traction. I don't really care.

But is there sort of advice that you would give for how you frame a pivot to early investors who signed on to a certain thing. Maybe they just signed on to you, but they signed on to a vision and you. And you're like, the vision is different now. Well, I mean, I think first, what's important is to pick investors who are open-minded and more generalists than specialists, if that's what you feel like is aligned with your way of building. Right? because obviously if you pick an investor who's specialized in SaaS for legal, you know,

you can tell them you're going to build something completely different, even if they're open-minded is going to be a challenge for them, right? So we were lucky that we picked investors like Betarworks that were quite generalists and open-minded. And I think that's probably 99% of the reasons why this pivot went well. And then second is, you know, you're going to do a lot of different experiments to get to pivot, right? And it might not be your first experiment that is going to be the final pivot. So I would wait to kind of like bring your investors on board when you have quite high validation and conviction that you're going to pivot, right? So our investors, for example, when we came to them and were like, okay, this is getting really good traction.

We want to focus most of our energy there. It was maybe like, you know, three, four months before we started it and we had quite high conviction on this. So then it makes it easier to convince your investors that this is the way to go. I believe, and correct me if I'm wrong, and it was like 2023, then you raise a large round on this new vision or whatever. But what you do is there's an unusually broad syndicate of like clouds and chipmakers.

And like you're sort of structuring or positioning yourself as like the Switzerland of AI, not only for your cap table, but what you're front facing to the community. How can you? you how can you position yourself as that where you're like we can be friends to everybody and still be successful uh in our own right well i mean um so in general in in life but also in uh in venture i i tend to believe more in and trust more systems and incentives than

than people. And I feel like if you have strong opinions in terms of what you want to build and strong values, you want to build the right systems and incentives that is going to drive you to stay consistent

with these values. And with Fugging Face, we always felt like one of the big impact was potentially to democratize AI, make it available to everyone fight concentration of power thanks to open source and become like a neutral part of the ecosystem and so when when we raise money we we try to create that as a system so instead of

having one player one investor being the biggest and representing the majority of our cap table of our investments, we decided to include a lot of them. So in our last round, we probably had like all the important players at the time in AI, right? Like Google, Amazon, Salesforce, Nvidia, AMD, Intel. And we got them all and tried to kind of like incentivize them all to, participate in this movement. So if you look at the organizations on Hugging Face now, all of them are sharing an enormous amount of open models,

open data sets with the community and really contributing to the field. And at the same time, because they're all there, I feel like none of them is exerting too much, too much power onto us and onto the direction of the platform. So we managed to keep our, in neutrality and keep our position in the ecosystem, which helps us continue to drive in alignment with our values. Right, because neutrality also allows you to maintain independence where you're not too

tied to one big player. Can you just really briefly walk me through the business model of Hugging Face Today? You've got like Enterprise Hub, Private Hub, inference endpoints, providers. What would you say is the spine of the business? This is October 2025 right now. Yeah. So it's a subscription, right? Like most platforms, especially with a big open source components,

we have a freemium model, right? Meaning that the majority of the usage is free open source and then a small percentage of the use. of the users and usage are paid. And the paid version kind of like funds the free open source part and creates kind of like this positive cycle to keep growing kind of like the impact. So we have over 50,000 paid subscriptions on the platform now. And the big kind of like contributors in terms of revenue are. obviously like the biggest companies using the most AI and having you know thousands

of users in their company using Hugging Face usually they need kind of like more advanced features like user management security some connection with their own tech stacks and so these are usually kind of like going contributors to to the revenue I did solicit a couple questions from folks in the AI community. And one of them is interesting to me because I think you just refresh the leaderboards and things like that. If you could get the whole AI industry to adapt one evaluation or maybe one new evaluation or norm, what would it be? It's a tough question because I think there's no one evaluation. I think today people rely too much on generalist leaderboards, which don't make so much, so much sense.

It's like if you are trying to evaluate a code base, based on one generalist like leaderboards, in my opinion, it doesn't really make sense. For me, the way to evaluate AI models, for example, to pick an AI model if you're an AI builder, you have to look at a range of different things. So first, it's important to realize that picking the right model, if you're an AI builder, it's an important part of your job. It's not something that you're going to spend like a minute a day or like even an hour a day. For me, maybe 30% to 50% of the job of an AI builder is to follow what are the best models, understand what is the best model, understand when to train, when not to train, when to use off the shelf versus your own system.

And in terms of evaluation, first you want to look at social validation. So on Hugging Face, you want to look at the number of likes. You want to look at how active the community is for a model, what people are saying in the community tab of the hugging face page about the model. Second, you want to look at public leaderboards and public evaluation. There are over 5,000 of them on Hugging Face that are specialized for different tasks, different use cases. And then you want to evaluate the model yourself on your own data for your own use case, meaning that you want to fit it your own kind of like tests and run your own benchmark for your own use cases. And then when you combine these three factors, social signaling, public evaluation and private evaluation, your own evaluation, I think that's when you can take very informed decisions about what to use, again, when to train a model versus use something off the shelf.

Ultimately, our intuition, and it's a bit counterintuitive to kind of like public perception, today of AI. We believe there's going to be as many AI models as they are like code repositories today that kind of like a generalist model for all use cases is maybe interesting to start with because it's easy, it's fast, but ultimately we think everyone will want to customize,

optimize their own models for their specific use case, for their specific constraints. You know, for example, some people will want a faster model. Some others don't need it to be fast. Some people will want it to be good in English. Some others in French. Some people will want it to be good for text classification,

others for dialogue. So, yeah, ultimately our vision is that there's going to be very much a diversity of models the same way. of diversity of code repositories. And it's kind of like validated by what we're seeing on the Hugging Face platform. In the last 90 days,

there's been 1 million models shared on the Hugging Face platform model data sets and apps. So it's 1 million repositories. So it's one new repository every nine seconds these days on Hugging Face.

So what we're saying is that you know, Based on open source, based on open source based, people create their own data sets, optimize their own models, fine-sune them, retrain them, and then kind of like build their use cases based on their own models versus just using kind of like a generalist API. Expedia and Visit Scotland invite you to come step into centuries of history that await in Scotland. Castle Steepton legend Walk along cobblestone streets

Come share the warmth of stories Passed down through generations This is a place with a past That is fully present today And all yours to explore Plan your Scottish Escape today At Expedia.com slash Visit Scotland

Yamava Resort and Casino at San Manuel is California's number one entertainment destination For today's superstars Catch the Jonas Brothers return to the Yamava Theater stage on April 30th, the powerful vocals of Demi Lovato on May 17th, and the signature Southern Country Rock of Eric Church on July 19th.

Tickets on sale now at Yamavat Theater.com, only at Yamava Resort and Casino, celebrating its 40th anniversary. You in? Must be 21 to enter. The Wired Newsroom is known for award-winning reporting on how technology shapes our world. On Wired's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people building, challenging, and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance?

Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday, wherever you get your podcasts. I'm going to read this next one because it ties in almost to what you were saying, but I'm going to read it directly because there's a certain question implicitity. here, but basically how alive are U.S. open models so far that you're seeing? And how, and do we actually care since the Chinese models seem kind of benign so far? I think I, I want to, I don't want to be too, too extreme. But, but it looks like, yeah, this way not enough open source American model. right now, at least not to the level of what's released from China.

And I think we do care for reasons of concentration of power. I think ultimately we want actually not just China and the US, but any country to be able to produce their own AI models. models, just the same way we want any country to be able to write its own code, you know, and write its own software, to provide choice to people to make sure that power isn't concentrated, to avoid some of the biases contained in different models from different countries. So in my opinion, that's one of the most. important geopolitical topic of the decades. Personally, I'm not in politics, but if I was a politician,

I would make it a priority to make sure that the U.S. has strong open-source foundations for everyone to use. Is there a tool that you could recommend to politicians or to just folks in the American AI community tool or advice for we need to grow more AI open source models here? I would foster more open data initiatives and open datasets initiatives because obviously the more open data sets, the easier it is for people to train and release open weights. So I would try to remove any challenges to open data and open data sets. I would kind of like foster more distribution of compute of infrastructure to kind of like help academic labs, you know, nonprofits, startups to actually have access to compute.

shouldn't be just the frontier close-source labs that have access to compute in the US because otherwise they won't be as much incentives to release publicly open open weights that would be and then third one is that I would send the signal in the US and celebrate when companies are sharing openly you know like it's to me it's crazy that when like a company like META releases open weights with Lama that so many people are like, you know, this is too dangerous. It's like it's going to kill humanity. Why is it so reckless by sharing the models? I mean, they've been shared like a year ago. It hasn't, you know, destroyed humanity. And I think we have to change a little bit our mindsets in the US and try to celebrate when companies

organizations or sharing their research, sharing their models, sharing their data sets, they actually contributing to the development of the domain, of the sector, they're contributing to America. And so we should celebrate that, right? So I'd love, for example, to see the next big open release, you know, Open AI, for example, released the first set of open wait this summer. Maybe for the next one, you know, Samma goes to the White House and with Trump, they explain why sharing these models in the open is great for the world and for the U.S. Where do you expect the open side of AI, the open models, the open source AI, to lead the cutting edge of AI going forward. What is the advantage that Open has that you expect will compound

and be important in coming years? Yeah. So Open is more gigawatt efficient or like energy, energy efficient, right? Because it's one model, it's one training run that you share for everyone. Right. It's interesting right now because those people are talking about energy and the energy race, right, between China and the US. People are talking, okay, there are 10 gigawatts there, 10 gigawatts there, and kind of like comparing this. But the reality is that the same gigawatts doesn't have the same impact in the US and in China. Because in China, because the labs are sharing their models publicly, every lab can do different experiments, different, different run versus in the US, all the research labs because they're closed source,

they're doing the same run, right? So open eye is doing, it's one gigawatts of training, entropy one gigawatts, XI, one gigawatts, of exactly the same, more or less exactly the same thing, right? So you have three gigawatts for nothing, it could be one if one of them was sharing their weights. So I think we're gonna see in the future that open source and open weights are important

for energy efficiency. And then you're going to see, in general, broadly, for the economy, and you're going to see that open weights are better for specialized, cheaper, faster, AI. Like, so whenever you don't need a chat GPT, right? Whenever you don't need a very generally open-ended system, like you'll see that having kind of like open, open source, smaller, more specialized model is going to be much cheaper,

much faster, much easier to iterate with much more transparent. And so I think that's where they're going to shine. I think you're going to have a world where you have a big model for chat GPT, for Google, for these kinds of use cases. And then everything else is going to be like smaller, faster models based on open source. Right. This is almost a similar question where, especially on Wall Street these days and maybe politicians and folks are talking about, are we in an AI bubble? From the perspective of y'all at Hugging Face, are you kind of nonplussed about if there is or is not an AI bubble in a macro sense because that might not affect what you all are good at and what you all are doing? Yeah. I mean, from our standpoint, the foundation in terms of usage from companies are very strong, right? And I think it's not going to go anywhere, right? Like I think companies are really seeing the value of AI right now. And this is not a bubble. I think if there is a bubble, it might be on some.

specific subset of AI, right? Like you could argue that there's a bubble on LLM compute, right? Like this massive deals just to train like very large generalist models. There might be a bubble there. But for the field in general, I think the foundations are quite good. Right. We've talked a lot on the show about the idea that even when the dot-com bubble burst. It's not like people stopped using the internet like that. So yeah. Okay, final couple

questions. I'm curious coming from France, coming from Europe, how you feel about the startup scene and the ecosystem in Europe, whether that be specifically AI or just in general or also Europe or versus or France, whatever, whatever you'd like to tell me about what you think about the startup ecosystem there. I think the startup ecosystem is growing. France and in Europe. It's getting better. I think it's the opportunities are growing. I think now you're starting to be able to do kinds of companies that you couldn't do before. And some example of that is mistrial. There's like managing to be to CAFLAQ raise much more money that we've seen before and build a much bigger company in a much faster time frame that used to be possible.

I think what's exciting to me is that potentially for Europe to focus on different topics than what the U.S. is focusing on or China is focusing on. For example, they have, I think, a lot of capabilities in pure science, in climate change, technologies, in fashion, of course, in luxury. So I think there's a way for France and Europe to have some level of impact in technology and grow their technology sectors. Finally, the New York City tech scene, specifically the AI scene, if you were someone that was interested in AI, interesting, interested in doing a startup in AI right now, what would be a reason that you would suggest New York City would be a great place to start doing that? I think New York is so much more diverse in terms of like nationality origins. It feels like a truly international city. And so I think as the founder, it gives you kind of like a diversity of opinion and support.

In terms of domains, if I was thinking of, you know, starting a company of like AI for, for biology, AI for chemistry, AI for finance, AI for fashion. All these are the domains, which in my opinion are the most exciting domains of AI. I'm a bit bored and tired of the pure kind of chatbot play for AI. And I think the next steps are going to be in these different domains. New York will bring you this diversity and access to some of the most relevant players. in these fields. So I think that's like a great place to start a company. Thank you, Klam, for talking to us. I really appreciate it. Thank you so much.

I knew about investing, but I really didn't know how to go about it. Meet Corey, a Walthfront client. With Wealthfront, it could put money in, and it would automatically distribute it into a diversified portfolio. Then it starts to compound. The compounding compounds on the compounding. Just let it wrong, and it's great. Over 1 million clients trust Wealthfront. Get started at Wellfront. Front.com. Client was paid $1,000 for their testimonial, creating a conflict of interest. Outcomes vary. Investment management and advisory services provided by Wellfront Advisors LLC and SEC registered investment advisor. Investing involves risk to principle regardless of the strategy used.

Task performance does not guarantee future results.