Building the "App Store" for Robots: Hugging Face's Thomas Wolf on Physical AI

[00:00] Many, many startups [00:01] just already being built on top of the robot. Just, you know, they want to build something. They have this idea of a manual task they can automate, or they have an idea of something they could do in the physical world. And then they take the robot. They take already like the basic building blocks we've shipped, which is just a robotic, a very simple robotic arm, SO100, that we designed basically to be the cheapest robotic arm, to be $100. And they're already like trying to start business around this. At the bottom, you know, it's the hugging face ethos, which is you bring all this platform, all these basic building blocks [00:31] people to [00:32] build really crazy things on top. And robotics is the same goal for us. [00:36] *music* [00:51] In this episode, we interview Thomas Wolfe, co-founder and chief science officer of HuggingFace, the largest open source community for AI. [01:00] Thomas has consistently predicted the future, and he's responsible for why Hugging Face invested heavily behind transformers and language models enabling the LLM wave a handful of years ago. [01:10] Now he sees the same opportunity for robotics and physical AI and is helping to shepherd Hugging Face's Le Robot project, which brings together policy models, data sets, and physical embodiments to help developers everywhere become roboticists across the diversity of use cases and form factors. [01:26] Thank you. [01:27] We're excited to chat with Thomas about the robotics explosion.

[01:30] the bottlenecks ahead for physical AI, the U.S. versus China in the open model race, and a lot more. [01:36] Enjoy the show. [01:37] Thomas, in your role at Hugging Face, you help Hugging Face invest behind moonshots as the chief science officer one or two years ahead of the field. [01:46] You mentioned to me last time we chatted that, you know, you have a spidey sense that we're at the same moment for robotics today. [01:52] that we were for transformers and language models a handful of years ago. Tell us what you're seeing. [01:57] Yeah, I think honestly, it started two years ago. I mean, we started our activities in robotics 18 months ago. And I think at this time, there was a couple of breakthroughs out of, I mean, these labs, you know, were away to Stanford. And basically, these teams were starting to show robots that were able to tie notes. [02:15] fall close, you know, cook, like, threw things in the air on a pan and grab them. And basically all of these things we've... [02:22] in one way, very little data, but also with also good perspective to be able to leverage some of the world models and things like that, that we see, you know, really benefit from, from internet size data. So all of this kind of pointed to a short future where robotics were going to work in a way like hardware was already there. And in my opinion, it's been there for quite some time, but the missing brick was really software that could adapt, that could be dynamic, all of that. And we started to see that. And that's why we started to work on the robot. [02:52] Yeah, a little bit more than 18 months ago. And I think for us, the success of the robot, like seeing like this huge community,

[03:00] For us, the big bet was can you build a big community in robotics as well? There were small community of kind of hobbies or people very seriously building robots for factory lines. But it was kind of like a tiny vertical, in my opinion. The question was, could you move this tiny vertical to like a full horizontal thing? Just like nowadays, every software developer is kind of an AI researcher almost. They all want to know how LLM works, how you train them. [03:30] AI aware. And I think there is a potential future transition where all of these people also become roboticists in a way, if you give them the tools. And so that was kind of our goal. So we started with the software library and the success of that kind of brought us to also try to go in hardware, which is a big question as well. And that's why we acquired first hardware company this year, Poly and Robotics. And we shipped our, at least we opened the orders for our first robots. That [04:00] Can you tell us what Lev Robot is? [04:02] Of course, yeah. So Le Robot is our attempt of, you know, doing again the success of Transformer, reproducing the success of Transformer's library in the robotic field. So this idea is having kind of a central library that everyone would use and that would bring [04:19] in a very simple and accessible and easy way, all the latest [04:25] technology, all the latest algorithm that people use to train robots efficiently, all the latest data set that they use to train this and also connect this to actuators, which is the hardware part of robotics. And that's this, this intersection and this mix of three aspects, the policy, the models, the data set, and the hardware, which we try to combine in the robot.

[04:45] How does the role of Hugging Face change? [04:48] in robotics, you know, for people building in the physical world, does Hugging Face play the same role or a different role than Hugging Face has played for people building in the digital world with LLMs? [04:57] I mean, our goal is to play the same role, which [05:00] at a very, very high level is building communities and bringing people, you know, in this idea that [05:06] AI can be open source and it's not something you only consume, but it's something you can tweak, you can train, you can control, you can host where you want. [05:14] And actually hosting where you want is even more important in robotics because [05:19] In a future where you have robots everywhere, you kind of want a lot of these mobiles to run locally. Because if your robots lose connection to the Wi-Fi or something and then run in the world, or maybe, I don't know, run in your kids, it's going to be much more dramatic than, you know, just an LLM hallucinating. So safety question in robotics. [05:35] I think a good reason you may want to really be [05:38] able to not depend on the distant API, but have the models as possible to the hardware. So I would say our role is maybe even more important [05:46] for safety and future of robotics than it. [05:50] 90s in L11. Could you say a word on the size of the community you have at Liverbot? How many people are building? How many people are contributing data sets and things like that? [05:58] Yeah, for sure. I should have checked actually the latest number because it's exponentially growing. So it's several thousand people, I would say six to ten thousand. One event we did, you know, a couple of months ago, we did a hackathon. [06:11] which was worldwide. We had 100 locations in over six continents. So it's still not in a million of people for sure, but it's way above several thousand people. And the main indicator for us is we can measure the number of data sets, for instance, on the hub. And what we see in all of these things, number of community members or data set, we see kind of this exponential growth, which is, I think, a very good indication that we're on the right track.

[06:36] And you have to keep in mind that the hardware that's available right now is still very much kind of a hobbyist hardware. So it's like 3D printed arms. It's like they're still wired everywhere. And that's why starting this summer, we wanted to bring much more [06:50] mass market hardware, I would say, so something that would [06:53] interest not only the hackers and the people who are used to [06:58] to plug cables everywhere, but also everyone in families, something that looks much more polished. [07:03] What's the persona of developers in your Le Robot community? And I'm curious how it's the same or different from the people that have traditionally been building, you know, like, [07:12] controls classical-based systems. [07:15] I would say there's three types of personnel. I mean, one is the traditional roboticists. They definitely want to use AI. So for a lot of them, they know how to build hardware. They know what they can use, but they've been frustrated by, you know, the limitation of the software stack. All the optimal control models and all these are very limiting what you can do. So all of these are people who have really happily joined the bandwagon. And we see the same effect we saw in transformers, which is many academic labs starting to use robots [07:45] point for all these students. [07:47] And so this has been growing very strongly. [07:49] The second community is much more interesting, in my opinion. It's people who were [07:54] not really into robotics, but because they are into AI and robotics looks like a physical manifestation of AI, they kind of want to go into robotics. And so these people that cover, I would say, software developer, but even people who are just interested in robotics. So a good example, talking here, it's interesting, but a lot of investors have actually bought, you know, SO100 ARM just to, you know, to try to understand physically what is this robotic thing, what can it do? And because it seems so accessible, you get the ARM and the software is just a Python code.

[08:24] a little bit of vibe coding, you can actually even tweak it or control it quite easily. You know, we see people who, I mean, maybe are not, you know, purely technical, but who want to understand what's happening in robotics. And they use this entry point, which is low robot. [08:38] So you can Vibecode the robot? Yeah, that's really my goal, yeah. So you can already do that a little bit, but for the new robots, Rich Cmini, I definitely want this to be, you know, one of the easiest way to use. I would love my kids to be able to Vibecode behavior on the robots. [08:51] What sort of phase of maturity do you think we're in for the robotics market writ large? You know, like when will we have a ChatGPT moment in the world of robotics? [09:00] Yeah, that's the thing I'm looking for. I like sometimes call it also the iPhone moment. Maybe what will be the first use case? The first, you know, the first moment where everyone or like a large fraction of population will think, I want a robot in terms of consumer. I mean, I think the enterprise market is quite complex. There is, in some place, there is already a lot of robots in some kind of industry. The car, car manufacturing is the best example. Then there is this second part, you know, where there is like an entry of robots. [09:30] um, [09:30] reliability. Will these robots be reliable enough to be deployed in retail store and basically be really useful? And the third part I'm much more interested in is actually entertainment, fun, demo, education, where I think [09:47] Maybe this question around, you know, [09:50] I need a $3,000 robot because I need reliability or less pregnancy. And so you can take a robot that's really accessible. So each Cimini, for instance, is priced at $300. That's something that can definitely be like an impulsive buy. You buy it as a...

[10:03] as a gift and you're not sure if it's going to work or not. But for this price, what we want to find is [10:09] Is there not a lot of, you know, potential in more like entertainment, fun, education, learning AI through a physical interaction stuff instead of just coding, you know, on the chat bot or coding on the screen? And I think that's... [10:25] something that has not been explored at all. I think there was a couple of tried. I mean, the Meet Media Lab, Cynthia DiGibo, for instance. But in the past, they were usually priced a bit high, I think above $1,000. [10:38] And more importantly, I think the software that was there was very limited. So you would buy a robot that would be fun, but you had maybe five or ten behaviors. And once you've tried them all, that's it. It's finished. And here the goal for Richie Mini is really to make kind of almost like a smartphone. So it comes, you have a couple of behaviors, but just because you can tweak it and people can build new behavior and share them and plug all the new content, [11:02] VLN, speech models, chat model, actually the possibilities are kind of endless. It's kind of an open door on like rebuilding, you know, the app store of [11:12] iPhone, basically. So that's what I'm very excited about. And to be honest, this last part is still very much a bet because nothing exists there. There is no real [11:20] Poof. [11:21] manager's hints are all this exponential growth of community, which make it quite plausible. So you see Ritimini as like the... [11:29] you know, the reincarnation of the robot dog. [11:32] dogs of the 90s was like how people can actually play and experiment and you know have like these

[11:38] robot companions in the households. [11:41] I mean, to be fair, this one is a big bet. But something I was discussing just yesterday, actually, on robotics at Tech BBQ, someone was telling me as an investor, you know, how [11:50] Just see so many, many startups. [11:53] just already being built on top of a robot, just, you know, [11:57] He will go out like [11:58] They want to build something. They have this idea of a manual task. They can automate, or they have an idea of something they could do in the physical world. And then they take the robot. They take already the basic building blocks we've shipped, which is just a very simple robotic arm, SO100, that we designed basically to be the cheapest robot. [12:14] robot account to be $100. And they're already like trying to start business around these two stars like [12:21] to say something around this. And with Shemini is also [12:24] in a way designed for that. It's a very wide, simple robot, [12:28] kind of a white labeled thing. And if you want to adapt it and if you think, hey, I have a business idea around this, but I need a robot to interact with people, I don't know, in the hospital or something like that, you can take this and you can actually start to build your idea. And at the bottom, you know, it's the hugging face ethos, which is you bring all this platform, all these basic building blocks for people to build really crazy things on top. And robotics is the same goal for us. [12:51] Really cool. I want to talk about data as a bottleneck. I think one of the big differences between language and robotics is you have trillions of tokens out in the public internet to train OLMs. [13:01] that [13:02] dynamic doesn't exist in robotics. And actually, I think that's where Hugging Face's role in the ecosystem could be much more interesting in terms of

[13:09] decentralized data sets, curation, creation, [13:13] Talk about what's happening on the dataset side of low robot. [13:17] Yeah, it's super interesting, I think. [13:19] I mean, there's a couple of challenges in robotics. And I mean, the general, the main challenge is in the data. There's just not enough data. There is. [13:27] some ways to use video on the internet as training data, but it's very limited. [13:32] and [13:33] And in some way, we may be able to use models. But in some other, if you want to automate a test, there is no way around just recording someone or the robot's [13:42] possibly just doing the task. [13:44] I think here there is [13:46] There is one possibility and one limitation. I mean, the main limitation is you can record a lot of... [13:52] to ask yourself, [13:53] But usually what you will lack a lot is the diversity. So you will basically be able to train a robot to do something very well in your room when everything looks the same. But once you put it in the next door room where maybe the walls are green instead of red, the robot has a lot of troubles to generalize. So this is the main limitation. [14:10] And so our idea with the hub was that [14:13] everyone could record datasets, and if we managed to [14:17] incentivize them to share the data, then we could maybe build a very like [14:21] multicolored data sets that would, like multi-location data set that would be extremely diverse. And in addition, hopefully, would be also very big. So I would say that's a long-term goal. [14:33] And we hope this can help. But another more direct things we try to do is to work also directly with the actors of the community.

[14:42] We released a couple of data sets to try to help them releasing some data set. We think in robotics, one of the nice aspects is a lot of people in the end want to sell the hardware. And so they can actually afford to, even more than LLM, they can afford to share a bit of the software as open source if it brings all the field above. Because in the end, that's not really directly what they sell. So that's kind of what I'm trying to convince a lot of robotics companies to do. And surprisingly, it's actually something a lot of them seem to be interested in. [15:12] Hmph. [15:13] Super interesting. You tweeted out world models the other day. I think you and I met the same world model founder. What's happening in the world model open source and how does that help or not help what's going to happen in robotics? [15:23] And can I ask about you, is there a why now in world models at the moment? Because it feels like they've started to pop up recently. So what is interesting is it's, [15:33] I feel like it's a couple of teams who have been actually working on that independently for a few months, and they just happen to release this right now, right? Because when you talk about all of them, they are not really coping each other. I mean, I guess one thing was the advent of really cool, really good image generation and basically finally understanding how to fix this six-figure thing and basically just get a more reliable and more coherent world model for image,

[16:03] And so we see now some really cool video models as well. And this one is just one next step. And a lot of the founders I've talked to in this field also say that they were helped by the advance of open source video model generation and open source image generation. And basically they take this video generation model and then they fine tune them and then they train them to be able to react to some inputs, which is also what we do in robotics. There's a lot of common points between these two things. [16:33] quite well. So you start to have this kind of [16:35] very interesting, in my opinion, totally new experience where you actually have a film that's controllable, that's like both photorealistic and also reacting in a very, you know, coherent way to the action you will input, which is either just moving around or just asking it to add something, you know, add a rider, a castle, a car driving. [16:58] and you see this thing that just reacts very well, [17:01] And you have here, I think, a lot of potential applications, you know, both obviously in entertainment, actually some form of entertainment that might be just totally new, something we've never seen, which is maybe the first time we create a new form of really virtual entertainment, but also a lot of application in business and how you can have interactive things. And one of these downstream applications is generating more data for robots. I mean, there is just two ways to generate data. One is to record it in the real world, which I think is still very interesting. [17:31] simulate it. And surprisingly, on the simulation, we have not seen a lot of, you know, I mean,

[17:37] There was some development, but it's not like there have been some really breakthrough recently on simulation. So maybe this is the first breakthrough I've seen on simulated generated data in quite some time. [17:47] Yeah, I was very excited to see some of the, even like what DeepMind's doing with Genie to train their embodied robots. Super exciting. Humanoids. [17:55] Do you believe in humanoids as the kind of ultimate form factor? [17:58] Yeah, big debates, big debates. What is sure is that [18:02] I'm quite more excited about trying other form factor right now. [18:06] I mean, the main problem with humanity, I think there is two main problems. The first one is it's always quite expensive just because you need a lot of models and all the price in the robots is just the actuator. That's always like 70% of the price tag. And so when you have 60 actuators, that's just your bill is here. [18:23] And so it's really hard to drive humanoids [18:25] below the price of a car. [18:27] I think the price of a car is still already quite a high requirement. If you buy something at the price of a car, you do expect to have a lot of value out of it, right? And so... [18:36] And so that's why we're exploring like smaller robots, like just one arm or just a moving head and this type of thing. There is some possibility that we can get cheaper humanity at some point. And K-Scale was trying to do, Unitree is definitely trying to cut the price. And there's a lot of companies trying to aim, but it's going to be really hard, I think, to get this under like $10K, $10,000, something like that. [18:58] The nice thing about the humanoid, of course, is once you've solved the humanoid, you solve like a lot of tasks at the same time. So if you solve the humanoid, you can do everything a human does, which is very exciting.

[19:09] And the main question is, do you need to solve the human rights? So on my side, I'm more like, I would like to see, you know, [19:16] a galaxy of different form factors. I also think some of them are much more cute than the humanoid. I think for social adoption, I think the humanoid is also asking a lot from people, right? It's like this [19:26] You're directly in this kind of uncanny valley with something that looks a lot like you, moves a lot like you. [19:31] So I thought this would be a big limit for social adoption. Now, to be honest, I've seen a lot of Unitary Robots. I don't know about you, but you kind of... [19:39] ignore them at some point. So I'm also much more confident that people will just [19:43] saying, yeah, that's just... [19:46] Maybe we're too worried about the uncanny valley in robotics. And maybe at some point, once we start to have seen like a couple of robots, people, we just accept them very, very easily. [19:55] Okay, so we're going to see the robot humanoid soon? I mean, the goal would be if we chimini and our small robots work really well, that at some point we'll climb back to make it the humanoid form factor, I would say, kind of progressively as we've done, bringing the community along with us. [20:10] If as you imagine the world in 10 years, [20:12] Like... [20:14] How many robots do you think there are among us? Do you think it's like 80% of them are humanoids and then 20% are this long tail of this diversity of hardware and use cases? Or like... [20:22] How do you think the world plays out? [20:25] Yeah, and I would love to see the second option because I think that's an option where we have much more robots in our life. What I would really not be super excited about is a future where robots are kind of an elite thing because they cost $100,000. And so basically, if you're rich, you have three robots at home. And if you're not, you don't. I mean, Hugging Face has always been also about the big community. So we care a lot about that.

[20:48] And so for this reason, I'm much more excited to see a lot of form factors that are basically accessible to a lot of people. Some of them are cheaper, some of them are more expensive than just this single humanoid, you know, that costs a lot. And if you can buy it, that's nice. And if you cannot... [21:01] too bad for you. So I would say at Hugging Face, that's the future we try to nudge to push to around. I think also it's much more fun because in a way you're also restricting yourself. Just like LLM, if you just try to make them copy human, it's one thing. But if you try to think maybe they can do something that human cannot do, it's also much more interesting in a way. [21:22] Do you think we're heading towards a world of, you know, big foundation models that can kind of do everything and then be adapted quickly to any new... [21:29] domain with just like, you know, a few prompts? Or do you think that developers in your community are going to [21:34] start from like a small base model and then do a lot of their own [21:37] data collection, customization to adapt to their domains. [21:41] Hmm. [21:42] I think we'll see more and more both. I mean, I think as the field evolves, you know, we start to see really a long tail. So for instance, if we take the downloads on Hugging Face, we see both like very large state-of-the-art models being downloaded, which are usually too large to run on a local platform. [22:00] laptop, but we see also some of the most downloaded models are actually just the right size that fits to run quickly on the laptop. So we see these two, like, really modality. And I think as, you know, as the field kind of mature, we'll start to see this more and more, which is, it's not like you choose one or the other. It's just depending on what you need. You might use, you know, one locally or not. And I think GPT-5 with the router is a good example of this. You know, maybe the largest model or the most

[22:28] reasoning, the longest reasoning chain is not the answer to everything. And you actually need to smartly select the one you want. So it can be behind a router, but it can also be just locally. You will run some models here. They might be extremely useful and we know [22:41] better and better how to train models that are actually extremely useful. But when you need something much more complex, when you need reflection for a very long time, then you will turn to much larger models. [22:52] One of the narratives that's been really popular over the last few years is this narrative of the battle between open and closed, you know, closed models versus open models, who's going to win. And just in the last few weeks, OpenAI... [23:06] is now present on Hugging Face. And so I'm curious what to make of that and sort of what it might [23:14] imply about the future of open versus closed or maybe how they work together. [23:19] I mean, we were super happy to welcome them back. They were there. I mean, the first model I worked on. And the reason we switched from being a game company to an open source platform was GPT-1, which not a lot of people remember, but it was very funny because it was trained mostly on novels and romance novels. [23:49] took this idea and trained it also on Wikipedia, which added a lot of world knowledge, and then expanded to GPTT, all of that. But at that time, they were very pro-open source, and I think open source

[23:59] Just like in software, I think both solutions, we just coexist. And having company that open source both or that do both. I mean, Google has been an example for quite some time, right? With the Gemma line and Gemini line and some interesting moments where sometime I heard that one Gemma model was actually so good that it was better than closed source models. So they had to not open source it. So the frontier, what is true, is quite seen at the moment in a way and the challenging, you know, [24:27] challenging new players, mostly in China, but I think we will start to see also some [24:32] new foundation models team in the US. So I think we might see also some challenges in the US. The frontier will stay quite seen, I think, and both things will stay with a tiny difference of performance. The main reason right now, I think, is [24:45] To be honest, at this exact point of time, I think we're not exactly in a kind of cost-saving time of AI. So which means that for a lot of actors, I think moving to open source because it saves costs is not the most important thing for them. So usually they move to open source right now because they want data privacy. They want to be able to adapt the model. They maybe have a new idea or new, you know, like this action model, for instance. They have a new idea of something that does not exist and they want to do that. So that's usually what we see right now. [25:15] Thank you. [25:16] emulsion of new exploration in an open source way. What I do expect is as we go to a more mature market as well, then the cost and being able to run it maybe on faster hardware or this type of other hardware and then

[25:30] Being able to own the model and to own also the full stack of where the models run will become actually more and more important. So I think just like in software, I think in the long term, open source is kind of a winning solution for many applications, for many use case. But we're still in the turbulence place. [25:49] Yeah. [25:49] How do you think Hugging Face's role in the LLM ecosystem has evolved as these models have pushed at the frontier and there's closed models? I remember back when it would be like you could download the small Burke model on Hugging Face and run it locally, right? And that was... [26:04] There was a lot of the usage [26:06] How has your business evolved now that we're going towards, you know, as you mentioned, models that are too large to run on consumer hardware? And how do you see Hugging Faces role evolving? [26:15] I mean, surprisingly, I was doing these stats at the end of last year. This BERT model is still really used a lot. So a surprisingly interesting aspect of open source is also resiliency, which is once you have something that worked, that really worked in production, you may not want to be forced to move to the new GPT, right? I mean, that was a little bit of the backslash around GPT-5, you know, just people actually wanted to keep using GPT-4 for many things. Maybe they've [26:42] fell in love with this or was their main friend and some Reddit posts were around this, but also maybe they just had their, like, applications that worked really well and they don't want to redesign it. I think open source, I mean, the long-term interest for us is also to provide this very stable base. Like, you build something, you know, it will exist and, you know, you can keep this as a very stable base.

[27:00] And in general, I think in the community, our role has switched progressively from [27:07] maybe pushing ourselves a lot of things, pushing our library, pushing our early product to more like enabling more and more the community in general. So we work now a lot with many, many actors of the community. We work a lot with Lama CPP. We work a lot with VLLM. We work a lot with all the big players to try to see how this whole ecosystem can be very efficient, can work really well. So like one model is released. You want to be able to use it directly in VLLM. You want to be able to use it directly in Lama CPP. [27:37] We try to have more and more this kind of role of-- [27:41] meta community builder where we try to align and to bring maybe all the players, you know, at the same pace and to help them move in the same way. So in a way, we are much more [27:52] focused on the community, on the hub than we were maybe a couple of years ago. [27:58] Really cool. [27:59] What do you think about what's happening? You mentioned China's had a lot of the open models recently. Why do you think that is happening now? [28:06] Um... [28:07] And what is the state of open model development in the West? [28:11] Yeah, this is the most surprising thing that happened, I think, in the last two years, right? The fact that China would become a champion of open source, who would have predicted that in 2020s, right? And so I've been, you know, I've been actually visiting them two weeks ago to try to understand a bit better on the ground how it's happening. And the thing is just, it's a very, very competitive market internally. There is a lot of teams there that are extremely good. And it reminded me in some way of Silicon Valley.

[28:41] extremely hard and they compete with each other, all of these model providers. And one part on which they compete, which is surprising, is being the most open, is the open source aspect. So they're extremely proud of being very open. And some of these companies, and that stopped being open, one was called Zipu, they decided to not open source it. And they saw an immediate black slash, I think mostly on hiring, like people didn't want to come work there anymore. And so they went back to open sourcing. So it's quite strong now. [29:11] So I would expect this to continue. I would expect also quite more teams to come because I see a lot of, you know, I mean, we see that as well, right? When you have the presentation of GPT-5, like a lot of people are, you know, actually did the study, some of them at Tsinghua University, right? We know the team are here also, you know, partly with Chinese members. So they have extremely, extremely strong people and they all really want to train the best model. [29:40] What I think is interesting is to see the West kind of coming back to open source very recently, to be honest, just over the summer. Right. But this call for open sourcing, OpenAI decided to come back. Now we're just waiting for Anthropik to maybe open source their first model. So I think it's time to try to ask them to participate. [30:00] Yeah, I would say right now, the situation for OpenSource is... [30:03] is pretty good. [30:05] But yeah, it's like the Jedi and Star Wars. It's never one. We have to keep pushing this. We have to keep pushing our flag of openness.

[30:16] What's driving the resurgence of open source in the West? Yeah, I think one thing is... [30:23] When you have, in a way, nothing to lose, open source is always a good solution when you're on your team. So it can be, for instance, you create a new company and you want to quickly rise to the top. Then you open source your model, right? That's the minstrel recipe. How can you very quickly become a great player? But for the Chinese, for instance, it's also, for instance, [30:41] In the West, [30:42] almost nobody will use a Chinese API. So they don't sell API in the West anyway. So in a way, they have nothing to lose, you know, from the Western market by open sourcing their market there. So I think there is this thing at play also. And the consequence of that is also that when nobody's open source, [30:59] It's like a market. There is an interest for someone to take [31:03] the room, right? To say we're going to be the open source player. So Meta was this open source player when everyone kind of stopped open sourcing. And I feel like there will always be this thing when some people stop open sourcing and then there is actually a gap to being the only, you know, the new top open source actor, then someone will want to fill this volume. [31:23] Thomas, you mentioned that, you know, Western companies won't use a Chinese model over, you know, [31:27] Chinese API. What about, are you saying Western companies are actually willing or not willing to use Chinese open models when it's, you know, the weights and... [31:36] hosted on US servers. Is there still hesitance to do that? And is it well-founded or not? [31:44] I don't see that a lot, to be honest. I mean, it's a good question. I try to do regular...

[31:49] So I try to ask a lot of people regularly, you know, what do you think about that? Because it can be for sure a concern, right? There was when DeepSea came out and there was a very nice answer, sort of model from perplexity, for instance. [32:02] The thing is, in many business cases, I don't think people really notice anything, you know, [32:08] So I think there is more of a general appetite for people to have a better way to understand [32:15] um, [32:16] like the safety of a model. I would say it's quite general. People are a little bit worried about having a model that maybe will behave strangely in some cases. And so I think this is a general thing that a lot of companies have been asking, which is, can you guarantee this model will always behave well? Which we know is really hard because, you know, even with GPT, sometimes you ask the number of R in strawberry and it's just behaving badly. And you're like, why? You're very smart. You should be able to know that. [32:43] So I think this is a general thing that is needed soon. And there's a couple of teams working on that for sure. So we talk about open science. Yeah, we build LLM like human. But what if, you know, an AI model could see infrared, could see some radiation? We cannot. This is a thing that human cannot do. So it's already superhuman. And for science, it's actually super interesting. So a lot of the AI model for science are already superhuman in a way because they can actually either see modality or predict things that are just, [33:13] accessible to you men [33:14] And I think it's a good ground to think outside of like...

[33:19] the human limitation of what we can do. And you've been pretty passionate about open science for a while. So can you just say a word about what is open science? What role does Hugging Face have to play? And where does your passion for it come from? [33:32] For me, it started a very long time ago. So before Agiefest, I was a lawyer. But before being a lawyer, I was a researcher in physics. And so I was working on this superconducting material. And surprisingly, in superconducting material, a lot of the great research had been done by the Soviets. [33:50] back in Soviet Union. And these people, I mean, you know, Soviet researchers have a very different way of inventing theory than the Western world has. And so they had some really great ideas and some really interesting things. But I had to find this... [34:03] invention or this theory, I had to find them, to track them down in the Soviet's GTP letters. And some of them were even still in Russian. And so from this time, I got this idea that [34:13] "Damn, accessing knowledge is hard. And if I can make this easier, that's going to unlock a lot of really cool stuff. If I could just find where, you know, does this equation comes from and really be able to read this article, that would be crazy." And so when I joined computer science, I discovered Archive, I discovered open source. I was like, "This is really cool. Everything's just free." Basically, everyone just share things written in English. Everyone can read it. It's even free. [34:38] You don't even have to buy the publication. [34:41] And I was very excited about that until I started to try to reproduce one DeepMind paper [34:48] And I discovered that there was a limit because people publish what they want to publish, but they don't really give you all the tricks of the trade, right? And so when you try to reproduce that, you discover that it just doesn't work.

[34:58] And so open science for me was this extension, which is, [35:01] It's nice to give open models to people so they can... [35:04] build things on top, but it's even better to explain how to train a model. It's a thing, it's nice to give a fish to someone to feed them, it's even better to teach them to fish. And that's basically what we want to do. We think in the very long term, [35:16] AI is going to be such a fundamental technology that basically it should be just like physics, should be something everyone could learn by reading a book. Like if you want to learn today about general relativity, you can... [35:27] read a book and you can know about it, right? You don't have to pay to get access. I mean, you buy the book or you find it, but that's basically free access. I think AI, all the recipe to train an intelligent... [35:38] object or artifact should also be something that everyone should know. So I mean, that's a very long-term thing. But the very short-term thing is if we teach people how to train great models, then they bring great models on the hub and then we have much great content to offer. So it's kind of also just content providing. If you provide great model, it's nice. And so one example that we do for that is we write very long blog posts that some of them even become books. We just [36:08] the Lord and how to do all of this parallelism thing. [36:11] Another very long blog post we wrote was around how to make a very good quality data set. And so we made a data set to free train the models. It's called FineWeb. [36:20] And it's used in a lot of the recent models, the QN models, they use FineWave, for instance. And then we also wrote how we build this data set, how we filter it, what is important to understand when you want to build great data to train models. So all of this, I think, just go together. And for us is a way to basically bring just better open source AI models in honey phase.

[36:42] I want to go back to your physics and superconducting comments. It feels like... [36:46] You know, a lot of the AGI labs believe that AI actually disrupting science is not that far out. [36:52] There's been some exciting discoveries, I think. Well, I think there's been exciting discoveries [36:57] evidence so far in math and then maybe extending into physics, material science. Like, do you think we're going to see an inflection point in scientific discovery from these models? And what do you think open source's role will be in driving that? [37:11] As always, it's nice. There is some hype here because then it drives people. But I think sometimes we overestimate what's happening. I mean, math is a good example, right? There was this idea, oh, [37:21] AI is doing new proof for some math theory, and that's like inventing new science. [37:29] As a scientist myself, I think that's really the wrong way to view science. [37:33] The reason is I was a bad scientist, so I can tell about that. So I was a very good student. So when you give me a problem, I'm always [37:41] pretty sure I can find the proof. I can find the thing. But I know this thing has a solution. So I just have to, you know, fill the gap and kind of grab a couple of things that I know and then combine them together. And when I became a researcher, I discovered that I was a pretty bad researcher because basically what I was not able to do was I was not able to ask the right question. So if somebody asked me a question, say, can you demonstrate this theorem? I could do it. But if someone say, okay, what is interesting to explore now in math? I had no idea, basically. [38:11] main thing you need to do

[38:13] if you want to do, I am talking about big, big breakthrough, right? Is you need to ask the right question. You need to find a way to ask a question nobody has asked before and a question that will open a whole new field of research. And that's basically a Nobel field, a Nobel prize. A Nobel prize is typically someone who just opened a new field of research because... [38:32] this person just asked the right question is maybe, [38:35] you know, maybe the speed of light should be the constant and let's explore what does it mean. And it means actually we can create general relativity and then we can invent black hole out of it. And I think LLM right now are still extremely bad at this thing, at this kind of a tasteful way to ask the right question, which doesn't mean we cannot do really cool stuff with them. But the way I see them nowadays is really more as very useful helpers. So once you have [39:01] human researchers who say, this is something interesting to study, then you can use them to actually multiply by 10, 100 or 1000 the production you can do. You can use them to quickly do a full survey of what has been done in the past on this production. [39:17] molecule, this protein, you can use them to say, okay, what would be, you know, the most logical way to test this hypothesis. But I still see this as kind of accelerator and assistant of scientific research. Then [39:32] What I would love to see, which is an AI that would say, hey, I have an idea on how to go faster than light. But for this, you cannot just [39:39] write the answer on how to go faster than light. You have to ask the right question: what should we change to today's theory? What should we do today? What should we reconsider to invent something that has groundbreaking? How's that?

[39:53] To your point on asking the right questions, what do you think are the interesting questions in the world of AI right now? Or maybe the questions that people are not asking that they should be asking? [40:03] I mean, this is one question, I think, and it's related to something we talk a lot, which is this sick offensive, which is tendency of AI models to always agree with you. I think, I think. [40:15] A good researcher is actually a good example of a person who disagrees with a lot of people. My former professor with the Nobel Prize was very... [40:24] not friendly in how he would have this. But I think that's part of it. You have to be extremely opinionated. So finding a way that this, you know, to push this model to have maybe in a way stronger opinion or maybe a taste in their opinion or like, I think for science will be a key. And it may, of course, this will be based on deep learning and LLM, but it may involve other ways to train them, other ways to think about them. I think that's one of the big questions I think not... [40:54] Not a lot. There is a couple of people exploring that, but not a lot of people are exploring. [40:58] Okay, when you see the world in 10 years, like, what is Hugging Faces' role in it? How much of your community you think is building... [41:05] with LLMs, with robotics. [41:08] I know it's hard to think in 10-year time spans, but what do you think the world looks like in 10 years? [41:12] Yeah, Tanya is very, very different. [41:15] I mean, what I would love to see is in 10 years a world where basically everyone feels like they can build with AI and not just consuming AI, but they feel like they can be an actor of this thing. A little bit like, you know, the difference between...

[41:28] We used to have a lot of media that were generated and created for us. And then we moved to the current era where everyone is actually able to create media. And we saw that this created a whole new generation of people and YouTubers and influencers and people actually making extremely interesting content. And I would love AI being the same, which is like a very big community, like the software developer community, where everyone can create things with AI. And they feel like it's just another tool in their box. [41:58] model. [41:59] The nice thing about that is [42:01] I'm a big believer in basically the creativity and natural invention of just the community. I think it's something that's very beautiful to witness. So in 10 years, I hope that people are not just consuming AI content and not doing anything, but they're actually exercising their creativity to build really nice things with a lot of AI tools around them. To be honest, that's something that I think that's kind of what we're building right now. So I'm quite optimistic. The thing is going to change a lot of things for society in general, because a lot of this job will just be different. [42:31] It's a beautiful vision. [42:32] Thomas, thank you so much for joining us today. We really enjoyed this chat. [42:36] Thanks. It was a pleasure.

[43:01] So [43:01] Thank you.