How open-source AI will reshape power dynamics in tech w/ Hugging Face CSO Thomas Wolf
40:11 Share

Hey, the level here before we start to show, I have a quick favorite ask if you're enjoying the dead eye show. Please take a moment to rate and leave a comment in your podcast APP wish episode of you loved and what topics do you want to hear? More of your feedback helps her shape the show to satisfy your curiosity, bring amazing guess and give you the best experience possible.

So i'm milenio. And when I was growing up, there was a toy that had us in a show cold. Not for me, not take go me.

有 吗？ I'm talking about tamago chees. You know, these were those little digital pets that lived on egg shape key rings and never let you sleep. They were super needy. And if you didn't feed them long enough, you to wake up to find a grave stone where your pet used to be with a little ghost floating next to IT bursty.

I was very happy to cosign the tomato I to the annals of history and believe IT or not, they would have a weird connection to some pretty cool AI stuff happening today. Unbeloved able to do. And this is the tedy eyes show where we figure out how to live and thrive in a world where A I. Is changing every.

N, F, T G P S gracing capacitance. The tech world is full of a lot of lingo. Keep up with the latest acronyms and technology news with teds new news letter. Ted talks tech will bring you tech headlines, talks, podcast and more on a biweekly basis, so you can easily keep up with all things tech. And A I subscribed now at the link in our show notes.

There is a company called hugging face like the eo ji, smiling with outstretch hands back in two and sixteen. The go founders, Julian shamans, Thomas wolf and client along had this idea.

They could use AI to create an online tomato chi for teenagers, a chabot T, A friend that was always around for them, a constant companion that responded to their input while they were tinkering with their AI chat bott, they started thinking about the actual infrastructure underpinning their product. Like what goes into an AI chap pot? You need natural language processing, right, which means you're going na need data, massive amounts of data.

You're gonna need a library of retrain models to mediate the interactions between computers and human language. Basically, you need a lot of complex parts just to run a simple chat pot. And right around this time in twenty seventeen, google drop, their seminal attention is all unique paper which introduce the world to transformers.

So while hugging face was tinkering with flagship product, they had a light, bold moment. The software stack they were creating to build and improve their chatbot was perfect for the shift to the transformer architecture, and they wanted to share IT with others. So they share their code, their best practices and their findings with the public.

And each time they shared something, they got such a positive response from their followers and users that they didn't want to stop. They decided to build a public library of A I software, so anyone, anywhere, building anything with A I can access the information they needed. Next thing you know, hugger face made a full pivot away from the chat pot and put all their resources into this mission.

And today, their site is the go to resource for all towards of folks interest in the A I. From software engineers, the weekend codes to computer science students, I use IT myself every single week. And hugging face includes so much more than code.

It's a common space for anyone using, building with, or even just interested in AI to learn more and do more with this amazing tech. Want to create your own comics? Someone made a comic book generator and shared IT on a hugging face, looking to include images from other countries in your data sets.

Yep, they've got that too. Want to share the latest release from IBM and NASA on A I enhances climate predictions. There's a message board for that as well.

It's pretty wild, right? Of course, not everyone wants to go open source. And yeah, there are definite advantages to keeping your cards close to the vest. But I spoke with Thomas wolf about why open source works for hugging face and why IT matters, not just for the major players in the industry, but for the rest of us, too. Thomas.

welcome to the show, the pressure to be here. Thanks for having me cool.

Alright, so let's talk about your origin story. You know, I obviously want to chat about hugging face, but i'd actually love for you to start with your personal background because you are what I would call a triple threats. You're an engineer, you're a scientist and you're a lawyer. Could you share a bit about who you are and your journey into artificial intelligence?

So I group in france and and that's also so where I met one of Michael found out to the assem. We were playing together in a band, rock band in paris during our engineering school. And after our engineer school, I went to do a research in in quantum physics, statistical physics, basically working on on superconducting materials.

I really like that. That was really nice, but IT was moving a bit too slowly for me. And so at the end of basic opposed that after my, my, my P.

H. D, I was like, I need, I need to do something else. And I relax writing at this time. And so one of my friend was a lawyer and told me, I if you you do low, basically you write a lot, sounds like, okay, why not? And so I moved to buttons low, which is alf technical and alf, alf legal, I would say.

And so I did that for for, for six years study to get my my for restively, my own clients bigger and bigger. And and a lot of them were startups and a lot of founders. And at the end in the years, like two thousand, four, fifteen, some of them were starting to do deep learning. So some of the intellectual property strategy that was designed were around OK. How can you protect, you know, stuff that were around at that time, imaging recognition, object segmentation that were really the first thing that they work with, modern, modern A I techniques, right? And I was very surprised that I didn't know anything about A I, because this was basically the same equation as physics equation, but just written by people and anchors in time.

What you around what you're was this when you're looking at these kind of perception I patterns.

So there was ten years ago, two thousand and fourteen and fifteen. So like .

right in the early days of like .

image to pop exit time. And so that's why I started to do some evening class, but on my own reading, reading papers, reading books. And so that's why I I basically contacted back to to 是 Young well， we actually just, I need posting roughly the same thing on A I on on facebook and I was like, oh yeah, I also reading this and said, yeah, maybe gonna up in new york.

So why don't you why don't you join some science back and you find maybe a more serious job in a couple of month? And basically what happened is I never moved out to find a more serious because, you know, the game company that was hugg face at the beginning. So the early idea, that's why we have this non serious nothing that was to make a kind of a modern version of of time.

I so a modern version of this kind of, you know, fly little being to interact with. And when I say modern, I mean powder with all this early gene, I think so. So basically at that time, we thought the only thing we could do was, you know, understand image Better.

And we and we like that, still really cool. You can recognize cl phy. You can may be react to to how someone is, you know, looking at your own emotions and and but quickly came some some of the early NLP breakthrough in generation.

So with S, T, M at at the time, and we will like, oh, maybe we can even generate text. And so basically we we were expLoring all of this building, this game. And that's a bit when happened.

I would say the the the change are like the the more thing of hugin face in in what IT is today, which happened basically by, you know, trying to on outside as well, give back. So if we do some research, we find some stuff. We should also write out first paper.

We should also publish occurred. And after a couple of months, we were like, yeah, when we look at all the curves we have, you know, when to start up, you always looking for this exponential, you always looking for where is something growing crazy. And so the game was working nicely.

We had very nicely in our increase while reaching hundred million messages. But the open source indicator of the, the the stars on guitar of the issues, they were really growing financially. And we also realized we are very much excited about this idea of bird source, that something really is also a very, very strong and even even more than me, I would say, like radical opens in radical open sharing.

And this, I think, made a lot of sense. That's where basically the mission affecting face kinds started to come. So not not really from the beginning, but IT became kind of of use that that was both something that was growing very strongly and at the same time after we believe a lot in. So basically, that's when we people did for a series a two thousand nine hundred and talk to me about that pivot.

Clearly over one of the earliest companies that we're looking at, these sort of machine learning primitives that you're disposal, let's save her visual understanding and trying to build like a consumer experience around IT. I'm sure that involved building your own stack to kind of a do that is probably the early days where there were no company is such as yourself making that easier.

And so kind of packaging that up and tackling open source, that must have been interesting because usually you hear about the rivers where you've got a developer century company that goes consumer and bomb, they get suddenly a lot of interest in adoption. You all feel like the complete reverse case. Can you talk a little bit about more, more about that, that vivid and shift as IT happened? yeah.

So in this time, we are doing like a lot of a lot of trials in in many, many direction, both on the interaction we wanted from the user, okay, is is Better to you know be on mobile, is Better to send like NLP, is Better to to use images, what is the most interesting thing? But also more generally, you know, how do we want to be as a company? How can we get a little bit of excitement? You have to picture that back in the day.

We are really underdogs. We are basically, you know, three french founders. So still I mean, still already like in in the U.

S. But but for the rest, we had no P H D N M L. We will not interview at all versus google people. And so we had a lot of impostor SAndra and wanted to build some credibility. And IT was also the time of the chat with company.

You know, the early version, we either wanted to steer away from this and not be puts in the same bag. We were like, okay, we are trying to build something seriously, believe in the AI. We are not just, you know, repurposing a couple of time plates making that into a into a chat buds.

So um that was proud of this thing with the the idea. Okay, if we publish if we communicate about what we do, we're gonna be taken more seriously and it's going to easier to raise gonna, easier to hire people. So I was really ready, which acted will end up being very true.

And I think being open source and having an open source approach is still really a great, great idea to hire a great people to be visible when you're starting, you know, a little bit outside of the traditional path. And yeah, doesn't doesn't mean you have to do everything open source. But I can be a very big part of the mix.

You know a big part of why i'll have so much adoption, especially the early days, was in some of your open source libraries, right, like the famous transformers library. Now, for the audience that might be uninitiated, how would you explain what a transformer is and the importance of these open source libraries and even data sets?

So the transformer is is basically the software of A I. So is the, is the you know the word course is the the backend that can make all the A M model we see today. And I D find you know how you how you combine the numbers and A M model basically just a big set of mattresses.

So a big set of numbers, and you need to combine, find them in a certain way to get to to get the output that you wants can be an image created from text prompt IT can be text generated from text. So the first types, usually you convert the input in numbers. You convert your texting numbers by associating a number to each world.

And then the way these are numbers are combined artifices by the architecture. So transformers is one way to to combine them. And basically transform is also the name for the big libraries that we have, like we have many, many office libraries now because we expanded much wide and just takes, but that's kind of where we started.

And this library, I would say what was really special about IT was two two thing. The first thing was, was very, very easy to use. And so basically, I remember when I met the, the, the, the researcher created this this bird model, which was the first, I would say, widely use model.

So before GPT one, right before I was a bit around the same time, he was the first model that people would, you know, people who have been working the file for a long time would basically drop up, drop everything they were doing to try this new model and run to IT and and try to test its. And so when you register the entries, arrier IT held them a lot. So we had a lot of people thanking us.

Okay, I could try this model and I could modify IT and tried to tweak IT and understand how IT was working and basically switch my whole research around this new model in a minute, thanks to the very easy access that your library gave to us. So the first power user of our libraries were really M, L. Researcher, A, I.

researcher. And that's really how I study. And then the second big in ingredient was that we started to include a couple of models in IT.

So you have one model release by a company, and we see that a lot today. And basically a couple of three weeks later, a couple of months later, you know that completed model, which is slightly Better. So you have this, I think, quite healthy and generally nice competition between a researcher or team building models to get the best model.

So it's very nice. But when you are participating in this competition, you want also to quickly be able to try the new model, to compare them together, to understand what what are the difference? Why is the new model from google Better? And the nice thing about the libraries, the transformer when I created IT was IT was really flexible.

And so I could quickly add new models in its. And basically, I remember when open, I released the GPT two IT was one day in basically weekends. So I worked on IT on saturday and sunday, had edited in the library.

That was really, I think, on the friday and on monday, everyone could use IT in the library. So basically people just keep the same library. And I just switched, you know, I want arguments basically, and they could use the GPT, too.

So this this was the beginning of the idea of the hub. And then we added, uh, a lot of social feature, more thing on top of IT. But that's what that was, this beginning of an idea that, you know, there there will be a diversity of models. There will be more, more models, A, M models coming. And it's very nice if you can easily move, you know, between them and and tried them and compare them in in this single environment.

It's super exciting because like you mention, the ml community, at least historically, has just been so transparent and forthcoming. And I think that creates those opportunities for variable, right? If people's come across a cool new thing, Albert super cool or GPT two is super cool, going to jump into if they are going to post about IT other other ml engineers and researchers are going to look at IT and then on and on, and they sort of fly wheel continues.

And that's quite a beautiful thing, because like the way my first exposure to hugging face was even the next layer, making this more accessible with huggy face spaces. Every time I knew in a computer vision paper would drop, or something like that, or new, you know, like a diffusion paper to fusion transformer thing would drop, I go to the huggy face face and try IT out, oh, segment anything model by matters out. Let me just upload an emergency what IT does.

And that was exciting, right? Because IT suddenly contextualized some cool new research. And it's not the same.

Five videos come associated with a paper. They get reposted again and again and again. You can see with your own data. So if you have to describe what hugging face does that now in the role that IT plays in the open source community, how would do you describe that?

It's really, I think, this place where where people who make models and the people who use them that are actually a lot of them are now the same. And I think it's very nice that that was also our our idea would be if we draw critize M, L and A, I enough, everyone could be a model builder. And actually, just before talking with you, I was playing with my kids and then we were trying to with my son, we were trying the new flex model.

So the imagery for black forest, which is go to the amazing, so good. And and so we went just expLoring spaces, trying IT and then and then find you need on on a couple of images and you think, okay, when you can now find M L mod between, you know, M L builders and M L user is really, really thin. And that's very good, I think because our idea was always A I we think the very long term vision, vision is we think A I should be a common good.

Or like for me, as I said in my physics, the view is everyone can learn about general relativity, quantum physics. Anyone can learn about this is I like common good knowledge from all humanity. And I think it's great if if you know only I don't know microsoft would know everything about a quantum that would be very sad, right? I think a yeah is a bit the same.

It's a very fundamental, you know, revolution is a technique that's gonna change where obviously a lot of things we do and IT should be something that anyone could understand and that if anyone would, okay, I want to understand now how this thing work. They should be able to do that because it's gonna control so many things in our life. And the best way to understand something is i'm in first to to share IT in an open source way, but also to be able to use IT very easily.

You know, you you you go to a space and you like, okay, how does this model where where you try a little bit to try a couple of things, example of prompts you try if you are, if you to chat model, you try to ask a couple of tRicky question. And spaces has helped us a lot to democratize the suspect make. You can face this kind of place where you can find models and you can also understand them.

And understand thing we added is data set. Because in the end, what we are discovering is that the data that you put in these models is basically the cough thing that make everything works, you know, the quality of the data and the quantity. And so couple of years ago, maybe three years ago now, we started this second approach round data sets.

So we are so study the second library data sets. And we started to host data set on the hub. And the idea was, let's open this new black box OK.

So we have the models. We've opened roughly the black box of the model with open weights. We've open source around the model code, but the data set should also be accessible.

If you want to understand A I, you have to understand how to make good data set. So we have to push people to share them in an open way. And this has been growing also very strongly on the hub. We have, I think, gravely may be rounded thousand data said, but I would have to check because this is exponentially increasing and this is very exciting. Nothing because when you have all of these things open, basically, you can really say, okay, now anyone can train a model and basically joined.

joined the fan. I think that's really it's it's so interesting to hear you talk about open as both on the model side and I think the data set side gets less attention. Your tear point yeah these models are like if they're open weights and it's like this inscribed ble set attracts like what you like all all you can do is using maybe can find tune, whatever.

But when IT comes to the real, I guess like I don't know if oil or goal is the right analogy, but is like distill the wisdom into these models that comes from the the data. And making that openly available is so huge too, especially if you've got an eager community that I kind of, rather than just in a select group of people at various labs working on the stuff you can have the whole community, including the indeed and hackers working on at, which sounds exciting. But IT almost makes me want to go back to the bigger question here, which is, you know, in the past, in previous interviews, you've said that, you know, we need to fight for open source. Why is open source important? And why should we fight for IT?

Yeah, it's a big question. Is possible for many reason. Maybe the most practical one is because of baLance of power. I think open source is a great way to make sure that you can have many people participating in the air revolution.

And not only, you know, like a small set of, you know, well founded labs, but basically anyone, you know, any small, small person from anywhere in the world who have a nice I D can join. And because they can start from this opening accessible weights data there almost at the level of the top lab from the get go. So that's that's really nice.

I think IT brings, in my opinion, IT brings them a lot more people around the table. So it's it's really great to catalyze you know new research ideas to basically you know accelerate knowledge in the field, but also to include other voices. We we also tend to think sometime that A I is just built by, you know uh, a small set of M L people, you know, math focused, White, male for but I think it's really nice if everyone can join. So basically, if everything is open, a lot more people can join, participate.

I do want to talk about the other side of this, right, when a lot of people push back against open source, especially in the recent around, you know, regulation california, and just like amErica at large, you know, how do you think about bouncing ing that need for innovation verses like the potential risks of making these like some very powerful A I tools, extremely accessible?

So I remember maybe one year ago, there was a lot of fear around, okay, A I is gonna be like a terminator, or or is going to kill us all in the next month, rights. And there was a lot of flight. I would say, needless, he worries from existence risks.

And we would say, okay, in january, people say we are gonna all dead. In january thousand. Four, A I is going to killing. Obviously didn't have .

no paper club yet.

I think a lot of this was kind of push also by this this, you know, obviously interest from the final lap, maybe get some earlier regulation that would, you know, slow down a bit. A lot of this was, I think, some kind of lobbing pressure. And I think, thankfully, people can understood that this is also tool.

It's gonna be not like, should we forbid this technology or not? But is gonna be more okay? There is there is a set of usage of this technology that should be forbidden.

And so regulation, I think, is very nice and it's something we we welcome within participating, also giving giving some feedback in a to the various regulators. We don't actively look lobby for anything, but definitely, we think some, some regulation is great. The only thing we would like to avoid that would say is some kind of gulag that would be regulator or captured basically.

So I mean, to to explain this, basically, if you make like a process to to be approved by the the by the government, which is so complex that basically only a couple of very large company with the lawyers and you know the money to do that, be able to do that. I think that would be a very sad outcome is because basically you will you will end up having this oligos like this very a small model ly of company deploying the the technology for everyone. I don't think it's I I don't think it's it's a great outcome for us.

I think is kind of the beginning of any very good distroyed an movie where you are basically company, you know controlling the the core technology that's used. How is we start a very good time I movie. But if we stay, if we take care from that, if we have regulation that makes sense and that also open to smaller to smaller team, smaller company is accessible than I think IT will be to be very nice.

I think you're total write IT. These tools are very dull use, right? And it's it's about time.

Rather than trying to regulate the tools themselves, we got to focus on the usage of those tools and what people are doing with them and hold the people accountable, right? Because ultimately, this comes down to power dynamics, as you said, right? Do you want to have this technology be the diminishing and sort of the the preview of a handful of extremely large companies know all the copy, as you said? Or do do you want there to be baLance in power? And what's interesting, particularly about the open for space that you're a huge and kind, I would say, like community builder and accelerator of, is how well open sources actually done.

I mean, I remember the narrative like a year and a half ago, you like maybe a year ago, ellia scavo was in television. He was like, well, I think these models are always going to be the domain of the big labs. And you know open source will never be able to catch up to a and this past year, there's just been so much exciting and like, personally, llama three point one by meta, which, by the way, kind of funny that meet like the bastion of open source.

Maybe, again, economic incentives can sometimes be a good thing, I guess. But how do you think that like power dynamic has evolved over the last year, year and a half? Because IT feels like open source with the underdog.

And now it's a neck neck IT almost feels like and it's it's also cool how accessible that's got. Then you mentioned in a playing with your kid, find tune ing you know like in creating like clones of just taking up of photos. Instead, you can have so much fun. It's so accessible now. And that's all way easier than I think people think, which is like open source isn't something you need to be super developer savi to take advantage of any more.

Yeah but you're right. There is still I felt and I felt you today with my son, there is still a little bit of entry body. I think we we remove remove open source. There is still like a little bit things that take, say, is getting Better, but IT will be nice to see this becoming more, more and more accessible with the goal.

I think one of the very nice goal is also like you saying, if you have your use case is maybe specific to you, I don't know, you take your notes on your phone and then you want to convert them and put IT would be very nice if you could just plug, allow a couple of these things together and get your things that you know do the things for you and tell or to you. So I don't think like you know, it's going to be tackled by the big the big players very soon. But if you have enough, you know, breaks that smaller companies can make fit together.

And this is going to be very nice. And that's where you have real democratization where basically no IT will start to be very easy to automate some part of the boring workflow that you have and basically concentrate on school staff talking about bullshit, tsk. That's also so something i'm very excited about today, which is open source robotics, which I thinks going to be big probably next year, may be the year after, but very soon.

So when you think about robotics.

what do you mean about robotic? Yeah, it's finally because I don't I don't really like human age robots. I found them quite scary. I don't I don't really want any of them in my house. I mean, I like process that I can understand.

So i've really party build myself for their made of open sales parts that they know and maybe they don't have to have to human factor. But I can have you know, fun factors and you can have one that you just made yourself. You try any to, you know, take care of your dish for sure for yourself like this, or fold your clothes that this project we have this week, which is like two, two small arms that you put, and they just fold your clothes. And this is much less scary, I think, on the actives buying a very large humanity drubbed that costs ten, you know, ten grants and that that sits in your in your house. So I think there is a whole range of robotics, I think, which is gonna be quite cute, quite fun, quite excited.

It's such a good point because i'm seeing your philosophy manifest itself like there's an interesting parallel where the way you're thinking about models, right, you've got you're going to yes, you're going to have some very large models, but you're going to have this massive long tail of small Taylor specific models. Some are going to be running in the clouds. Some are going to be running on the edge closer to your devices like kind of the models you mention. You initially mention you were focusing on NLP natural language processing, understanding no language and then, you know, vision stuff started popping up. How do you understand and perceive the world the right? And now you can kind of take those primitives and put them into these robots using like r dinos and like off the shelf stuff, which a gotcha must be so exciting to be a kid in engineering school these days is like, is the plurality of options such as a staggering yeah.

it's a crazy time to be. I I think what is crazy? You see all the things you can do and it's like, it's like a and less Green field.

They're like nobody tried this yet. No, nobody tried because we can do that since just last month. And you like, wow, that's right.

I mean, I saw that with yeah if my son is now calling through the spaces around flukes, you can do that. Yeah okay. And oh and you can also do that. It's like it's so exciting yeah I think probably the the main problem for kids today is just to choose one project to do because I like some possibility yeah .

yeah the problems of plenty, as they say.

Got a business problem. There's a ted dog for that stay updated on everything, business on ted business, a podcast hosted by column, a business school professor module ola, every week shall introduce you to leaders with unique insights on work, answering questions like how do four day work week's work to will a machine never take my job? Get some surprising answers on ted business wherever you listen to podcast.

So here in the last bit of this interview, how do you see the relationship between open source AI and commercial close source AI evolving over time?

Yeah I think probably indian probably rather similar to the dynamic of in software while you have basically both you know existing you know you you have open source, but definitely you still have a lot of close to software and that's fine. I mean, even on the heart, right, we have a lot of open model. People also use a lot of hope for for private model.

So like maybe all of IT is is closed for model actually, and that's fine. I think it's quite healthy, right to you. If you took allama and you find you eat on your the enterprise data, you shouldn't have to open source this model again.

Okay, it's fine. It's your private data. You you manage to get series. I like whatever. I think it's very healthy and will have both go existing.

I think on both climbing together, we have this this competition of like this was the healthy competition. Obviously, open source is very nice because he allowed ed people to combine stuff together. So the idea of plugging models with, you know image adding speech, like you are saying, you can do that very easily with open your model.

And that's quite exciting. Open so model will also be delays where the the kinsmen community or like the the more hockey community will will continue to strive. I would say right now the the situation is rather healthy.

I would say we have we have a mix of both of them. I I hope you stay, stay like that. But the same with, the same with access to computer, right, as access to computer. So getting easier and easier. And so yeah, I will have something that IT will be like computers.

You know, like, we know all of these very small laptop s and we used to have this very macbook, but you probably grew up like me in a time where we used to have a very large thing and we like, that's how a computer should be. And now when I show, I still have one, this stuff with G, P, U, I show you to, my son is like, oh, that's we are. It's very big.

Yeah it's used to be to stand out. So we have this very small and very versatile and very smart models that we can use a bit everywhere. And we still have also this very large.

But yeah for the rest, I would say the the very interesting thing now is also that we are moving out from just focusing on building models. The kind of taking models as a given will like OK. We have lama and we know you can do a lot of things.

And now what are the most interesting ticket? What are the most interesting use case that we can apply IT to, right? Because chatting is obviously one use case, but chatbot is also not the end of the the goal line.

And people, I think in the early I they're like, okay, making model is going to be the big business. That's where everything is going to be. I very much disagree with this. I think models are gonna really commoditized, but is very exciting to watch to follow.

And a year totally right, that, that vibe shift has been so like clear because even again, a year, year and a half ago, all of the startups that we're hitting the scene, we're like, well, you know, we train our own models. You so like we're not a GPT rapper company.

Words like over the last year, as you have seen, some of these startups kind of go back and return to larger labs, like character AI perhaps being a recent example, but also like pie with inflection and and others is like IT, right? Similar thing with amazon as well. What you're seeing is sort of like it's almost desirable. We had to be a rapporte company because again, like the model layer might not be the piece where all the value you will be created as the use cases, as the products and experiences. And so I have to ask, even though you can't tell me the killer APP, since you are running, you know, the largest opens s first community, I would say, you know, people literally call you the github of the eyes, is the phrase that i've seen you what are some of the most exciting things you're seeing these days that you are like, holy crap, to say you like a flavor? I hit list if you well.

yeah no, I think everything around the eyes, very around around image is very interesting. So your playing just also before this interview with with a new space that that I actually make you wear, you know, piece of clothes. So you put the image of clothes, you put yourself in whatever position, like from side on the back, and just really great and put this cloth on you.

And this is obviously something, I mean, there has been a couple of companies already trying that, right? But that's that's one of the reason why I am really like, oh OK how the take is is really, really good. So this is gonna be deployed everywhere, virtual trial and probably also integrated.

Nothing would be integrated also. Did I experience that? You over around and you just see how it's going to look.

You type so around image. I think there is love really cool thing beyond just the artistic me journey thing we started ed with. So I think that's gonna be a lot of interesting use cases around text.

It's it's quite tRicky because I think a lot of this is in the U. S. And the U I. And how you put that are you integrate that in your workflow? So a good example, if your code is the recent court code.

that's all my feet is, by the way, lad cursor, yes, yeah, that's all I see. So it's funny .

because it's IT means if you if you carefully focusing on the user experience, even if you don't build your own model, you can really make something that get people extremely excited about even if there is like a huge competitive like copilot, which is push you know by microsoft, grated directly in the thing, so which is the obvious option you want to try and start.

But the entry barrier for people who care deeply about their users is is still quite small. So very, very exciting. Video is very exciting.

I think we work out on video. I think video is very nice because yeah, there is a lot of content. I sometimes you don't want to watch a movie, there is there is some needs.

Sometimes you just want to summary of something. Sometimes you just want to to jump to a part of a movie or meeting or something. So being able to have something that understand process, video and in the real ways, gonna be game changing for so many application.

So A I could and look, radio. Amazing thing to hear. One thing I say is like .

what you're talking about, just like all these modalities, these different mediums of understanding and generation are all sort of stacking on top of each other in a cool way, even as you were talking about the example of video. Like, I got so excited because guess so, you're totally right. We can certainly mm ize like massive reams of tax, like twelve page papers, and give me that to still summary your highlights section. And then for video, that's a little bit more chAllenging, like being able to take a video and I kind of create the summary of version of that. I think there's so much late knowledge sort of sitting around waiting for us to to access, which is very exciting.

Anyone let me just open on something that you didn't about modalities. We you see our access. So I have a very cool project we are pushing right now on the public later, which is around modeling molecules.

It's it's a foundation model for quantum chemistry. So I mean, I can see images, I can hear speech, but I cannot feel a molecule that something I will never be able to do. But I am model and can access this thing that I can generate.

making the unseen scene .

in a icy like this. Yeah, I love IT .

was a great conversation. Thomas. Thank you so much for joining us.

thanks. There was great.

So if there's one thing you take away from today's episode, it's this open source is absolutely killing IT in the A I space right now. Remember a get hub deep dive. Well, multiply that impact by about a thousand for A I.

The innovation happening out in the open is mindless. And I mean, think about IT, you've got what's a small number of product managers and engineers in these bigger labs, right? They're working on a tiny slice of what's possible, but throw these models out into the wild.

And bm, you've got an entire community going nuts, pushing boundaries we didn't even know existed. Take matters on a three point two. For example.

Meta is one of the biggest labs out there. And as of this recording in october twenty twenty four, they just released a huge allama to the general public. And yes, you can find IT on hugging face for free.

This thing is nipping at the heels of GPT four, and in some cases, it's even pulling ahead. Folks are using IT to build all kinds of things, writing their own homes from redeem tary chatbot that become increasingly sophisticated as you use them, to custom voice activated assistance on their phones and computers. And that's just scratching the surface.

Or look at A I M A generation. Yes, mid journey, still the top dog in terms of usage. But flux, the free open source imagery model in just a month it's grab so much mindshare, it's not even funny.

And remember, this is from the same folks who gave a stable diffusion, the og of open source ai. And it's like clock work. First you get scrap ducket projects.

Next thing you know, this tech is baked into products we use every single day. And let's not forget about data privacy. If you're running a business and your data is your lifeblood, you don't have to take anyone's word for IT.

Just fire up your own models on your own hardware. No sketchy terms of service, no trust issues, your in control. That's the beauty of open source.

It's all about what you can build, not what some gatekeeper allows you to do. In fact, i'd argue that in the race to push A I boundaries, open source isn't just keeping pace. It's setting th Epace.

The teddy eyes show is a part of the ted audio collective and produced by ted with cosmic standard. Our producers are dominic jarred and alex ign. Our editor is ban bang chang.

Our show runner is ivana tucker, and our engineer is asia par. Our researcher and fact checker is Christian apart to our technical director is Jacob inc. And our executive producer is a lizer math. And i'm belov els to do. Don't forget to rate and comments, and i'll see you in the next one.