Big Job Disruption in 5 Years — Hugging Face Co-Founder on How to Stay Ahead

1. 0:00 – 0:45

   Who is Thomas Wolf

   1. TWThomas Wolf0:00

      something that's definitely gonna happen in the coming five year, big job disruption.

   2. MMMarina Mogilko0:04

      This is Thomas Wolf, co-founder and the chief science officer of Hugging Face, the open-source platform shaping the future of AI. We met at Viva Technology to talk about the future of work and how to stay ahead. What skills will actually matter in the age of AI? How do you stay relevant when entire industries are being disrupted? And how close are we to having robots inside our homes?

   3. TWThomas Wolf0:27

      Yeah.

   4. MMMarina Mogilko0:27

      So what's your advice to people who are studying for five years to do something that might be disrupted?

   5. TWThomas Wolf0:32

      Yeah, I would say the two advice, I mean, one is definitely [beep] .

   6. MMMarina Mogilko0:36

      Hello, everyone. Welcome to Silicon Valley Girl. We have Thomas today, who's co-founder of Hugging Face. Thomas, thank you so much for being here.

   7. TWThomas Wolf0:43

      Thanks for having me. It's a pleasure.

2. 0:45 – 3:20

   App store for AI. What is Hugging Face

   1. MMMarina Mogilko0:45

      Please explain how can I, a non-technical person, use Hugging Face? Can you explain the concept?

   2. TWThomas Wolf0:50

      Okay. So I mean, to be clear, Hugging Face is still mostly done for technical people. Most part of the website are made for software developer who want to develop something with AI.

   3. MMMarina Mogilko1:00

      Mm-hmm.

   4. TWThomas Wolf1:00

      So that's the, the two- we have three main parts on the website. We have models, datasets. You want to use them if you're developing a new app with an AI component, and you don't want to use a closed-source solution like OpenAI Anthropic, but you want to be able to own the, the full stack.

   5. MMMarina Mogilko1:16

      Mm-hmm.

   6. TWThomas Wolf1:16

      You want to have an open-source solution, like your open-source code. So in this case, you will go on our, on our platform. You'll find, select the right model for your task, for the, for the, for the application or website you're, that you're building, and you'll download it. But there is one big part, one big part of the, of the website that we created recently, and that's actually growing exponentially, which is much more accessible, and it's called AI Apps.

   7. MMMarina Mogilko1:36

      Mm-hmm.

   8. TWThomas Wolf1:37

      It's kind of a app store of AI. So we call that Spaces. These are very kind of small website. You can go on Hugging Face/Spaces. You have a search bar, and basically just type something you would like to do with AI. You can type, I don't know, "I want to remove the background of an image." "I want to, um, to get the speech for a text I've written," like, um, ElevenLabs type thing.

   9. MMMarina Mogilko1:57

      Hmm.

   10. TWThomas Wolf1:58

       Or, "I want to, uh, you know, generate a 3D character out of an image." And here you will have, like, spaces that are created by community. You can select one, and here you have a very no-code, easy, uh, button interface that lets you do these kind of, uh, magical, magical, uh, things, basically.

   11. MMMarina Mogilko2:15

       And you use them on your platform, or you can migrate to your own app that you're building with?

   12. TWThomas Wolf2:19

       You can do both. So the nice thing is you can use that d- directly from the platform. Then we provide the compute, it's like website, right?

   13. MMMarina Mogilko2:24

       Mm-hmm.

   14. TWThomas Wolf2:24

       You just... If you want to generate an image from a prompt, uh, like Midjourney-

   15. MMMarina Mogilko2:28

       Yeah

   16. TWThomas Wolf2:29

       ... you type this. You will have, like, a dozen of them. Some of them are more likes, for instance, so you, you can select the more popular one, the most popular one. You go there, you type your prompt, you get the image. Now, let's say you want to do that offline, for instance. You can also Git clone the space, you can download it, and then you can run it on your, on your, on your PC.

   17. MMMarina Mogilko2:47

       Hmm.

   18. TWThomas Wolf2:48

       If it's powerful. Depends a bit on the AI model. Some of them are quite big, so you might need a, a-

   19. MMMarina Mogilko2:52

       Interesting

   20. TWThomas Wolf2:52

       ... quite a large model, but-

   21. MMMarina Mogilko2:54

       Do I need a vibe coding app to make it an actual app?

   22. TWThomas Wolf2:57

       There is actually even a... One of the, the most popular space right now is a vibe, vibe coding space.

   23. MMMarina Mogilko3:02

       Yeah.

   24. TWThomas Wolf3:02

       It's a free one. It's called DeepSeek.

   25. MMMarina Mogilko3:04

       Uh-huh.

   26. TWThomas Wolf3:04

       It use the, the Chinese model DeepSeek-

   27. MMMarina Mogilko3:06

       Yeah

   28. TWThomas Wolf3:07

       ... which is a free, free model.

   29. MMMarina Mogilko3:08

       Mm-hmm.

   30. TWThomas Wolf3:08

       But it doesn't run through the, the Chinese server, right?
3. 3:20 – 6:00

   Who owns what you generate with open-source AI?

   1. TWThomas Wolf3:20

      download it, and you can create website.

   2. MMMarina Mogilko3:22

      How do you think about ownership when you use somebody else's, like, open-source code? For example, as a creator, I decide to create an app where I generate an image or, like, a character-

   3. TWThomas Wolf3:32

      Yeah

   4. MMMarina Mogilko3:32

      ... that I'm gonna use later, and I use that app to create a character. Who ends up owning it? If I decide to migrate it later from my app, sell rights to Disney, et cetera, how do you think about it?

   5. TWThomas Wolf3:44

      The way I think about it is, uh, mostly informed by what we have in code. So in code, we have this open-source movement, and I mean, in a nutshell, the idea of open source is when you write some code, instead of just keeping it secret, you can also decide to share it with everyone. You can put a license on it, and it's actually better to put, to put a license usually. Most of these license, the most popular right now are the MIT or Apache license, and what they will say is something a little bit like Wikipedia. Basically, you can use it, you can edit the content, but you're supposed to credit, uh, the, the original creator of the code. So basically, just, you know, just not take credit for yourself, uh, on something that someone else has built. You can have also com- more complex license, where you will ask, you know, uh, the people to pay if they have a commercial application or something like that.

   6. MMMarina Mogilko4:32

      Mm-hmm.

   7. TWThomas Wolf4:32

      Yeah, so-

   8. MMMarina Mogilko4:32

      This is interesting for me, like, 'cause I've seen, like for, for example, in media, sometimes you sell your image, like a photo-

   9. TWThomas Wolf4:39

      Yeah

   10. MMMarina Mogilko4:39

       ... to an agency for, like, $15-

   11. TWThomas Wolf4:41

       Yeah

   12. MMMarina Mogilko4:41

       ... and they put it on all the billboards cross-country-

   13. TWThomas Wolf4:45

       [chuckles]

   14. MMMarina Mogilko4:45

       ... and make millions on it, and you're like, "Damn, and I sold it for $15."

   15. TWThomas Wolf4:49

       You have to-

   16. MMMarina Mogilko4:50

       Do you think it might happen?

   17. TWThomas Wolf4:50

       ... acceptably that.

   18. MMMarina Mogilko4:52

       Yeah.

   19. TWThomas Wolf4:52

       Yeah. In software, we have the same problem, which is sometime you... Some people have created amazing software that power actually everything around us-

   20. MMMarina Mogilko5:01

       Yeah

   21. TWThomas Wolf5:01

       ... including Linux.

   22. MMMarina Mogilko5:02

       Mm.

   23. TWThomas Wolf5:02

       Or, or in the co- scientific computing, you have a, you have something very famous called NumPy-

   24. MMMarina Mogilko5:07

       Mm-hmm

   25. TWThomas Wolf5:07

       ... which every Python, which every- everyone use somewhere, and these people have definitely not earned even a very tiny fraction of the business value that was generated out of it, so-

   26. MMMarina Mogilko5:17

       Do you, do you think they're happy? Have you talked to [chuckles] ...

   27. TWThomas Wolf5:20

       I think a lot of people do that for the mission. They think that it's better to maybe earn less, but have something that's freely accessible, that everyone can use-

   28. MMMarina Mogilko5:28

       Yeah

   29. TWThomas Wolf5:28

       ... without having to pay. I think it's a bit their choice, because you can also decide to, to create a nice business and company around an open-source model. That's called open core. One idea is that you open source-... a core base, and then the feature that are more especially interesting for business, maybe around security, uh, maybe some features around, I don't know, how you integrate with-

   30. MMMarina Mogilko5:50

       You pay a lot
4. 6:00 – 7:55

   How to use AI agents

   1. MMMarina Mogilko6:00

      sense. That makes sense. Let's take a quick break and talk about AI agents. If you're trying to figure out how to actually use AI agents in your business, not just talk about them, I have something for you. There is a new guide from HubSpot that cuts through all the AI noise, and honestly, it's one of the most useful things I've seen on this topic. The Playbook for Success in 2025 about AI agents is built by people leading HubSpot's marketing, the CMO, and SVP. So if you're running a team or just trying to keep up, this is the playbook. So what's inside the guide? So first of all, they talk about marketing. You'll see how AI agents can help with content creation, social posts, campaign analytics, all the stuff that usually takes hours. This way, your team can finally focus on ideas, not grunt work. Then sales. AI agents can do the prep, research, meeting notes, follow-ups, so your sales team can just focus on closing. And third, in ops, AI agents are, of course, super practical. Agents can organize your talks, automate boring admin, and give you real-time insights without hiring extra people. And the best part is that there is a clear, simple plan inside, how to pick the right tools, set them up, and actually get results fast. So if you've been wondering where to start with AI agents or how to go from one experiment to a full system, this is your roadmap. I'll leave a link in the description below. Download it for free. And by the way, if you're building something in AI, I'll be judging a HubSpot AI startup pitch competition at Inbound, this September in San Francisco. The winner gets $50,000 in cash and much more. Applications close July 25th, so do not wait. I would love to see your name there, and I would love to meet in person. Thanks again to HubSpot for sponsoring this video. [swishing sound]

5. 7:55 – 9:49

   Future of coding

   1. MMMarina Mogilko7:55

      Uh, let's talk about what Hugging Face would look like in two or three years. So what I see happening with the developer industry, I started coding apps. I have no technical background, right? Uh, do you feel like people who code become less technical?

   2. TWThomas Wolf8:10

      Mm.

   3. MMMarina Mogilko8:10

      Or do you think people will get more and more education in coding, and because they want to touch upon that and integrate their apps and businesses?

   4. TWThomas Wolf8:19

      That's a good question. I think we'll see, we'll see both, I would say. So what I definitely already see is a lot of my non-technical friends, for instance, entrepreneur, but they have, like, a business background-

   5. MMMarina Mogilko8:31

      Yeah

   6. TWThomas Wolf8:31

      ... they definitely now play a lot with all the vibe coding tools, all the thing, and they, and they definitely now create apps. I have a good friend living next door to me. He's an entrepreneur, but zero, zero technical background. Now, he does all the time, like, lovable, uh, demo or something like that-

   7. MMMarina Mogilko8:48

      Yeah

   8. TWThomas Wolf8:48

      ... to, to basically get the, the first prototype of what he wants to build. So, so I definitely see that, which is people that are non-technical who become user of this and actually create very technical products. What I also see is, for instance, my, my son, who, who likes to learn, and he's learning coding, but very differently than me. So the way he learns coding right now is he will use this AI tool to generate stuff, and at some point, when it doesn't work, that's where that he start diving the code, and he's like: "Oh, well, okay, now I start to understand."

   9. MMMarina Mogilko9:18

      Yeah.

   10. TWThomas Wolf9:18

       And the way I learn code was much more... I would learn the, the basics, and I would learn the second level, and so, uh, like, very organized, I think, in a way. So they're learning that very differently, but I don't think it's gonna be less technical in the end, just-

   11. MMMarina Mogilko9:31

       Mm

   12. TWThomas Wolf9:31

       ... it's just a different way of learning where you directly get some results.

   13. MMMarina Mogilko9:36

       Yeah.

   14. TWThomas Wolf9:36

       But then, because these tools have some limitation, you still need to learn coding.

   15. MMMarina Mogilko9:40

       Okay.

   16. TWThomas Wolf9:40

       I think it will still be the same.

   17. MMMarina Mogilko9:41

       So we'll need more developers. Is that what you're saying?

   18. TWThomas Wolf9:43

       I think we'll also have a-

   19. MMMarina Mogilko9:44

       Yeah

   20. TWThomas Wolf9:44

       ... a developer pool that will grow.

   21. MMMarina Mogilko9:46

       Yeah.

   22. TWThomas Wolf9:46

       Yeah. Yeah.

   23. MMMarina Mogilko9:47

       Interesting. Uh, let's talk about your kids,

6. 9:49 – 12:48

   Kids learning to code through AI

   1. MMMarina Mogilko9:49

      since you mentioned your son. [chuckles]

   2. TWThomas Wolf9:51

      Sure.

   3. MMMarina Mogilko9:51

      Uh, he recently presented an app that he coded. What are you teaching your kids, uh, and how are you raising them, so that at 12 years old, they're already doing something really cool?

   4. TWThomas Wolf10:01

      Yeah, it's a big question, huh?

   5. MMMarina Mogilko10:03

      Yeah.

   6. TWThomas Wolf10:03

      I, I don't pretend to have any answer, I would say, uh, that I, that I'm really sh- certain of. Uh, what I know, I think, is there is some value that I would still think will be timeless. Uh, uh, one of them is kind of creativity-

   7. MMMarina Mogilko10:17

      Mm

   8. TWThomas Wolf10:18

      ... and basically, you know, uh, not being afraid of creating new things, um, because I think in the future there will always be... LLM, what we see when we use them is that they're still not very good at creating really novel thing.

   9. MMMarina Mogilko10:33

      Mm.

   10. TWThomas Wolf10:33

       They're very good at doing what you kind of expect them to do. They're, they're actually trained, they're trained to predict the most probable, the most likely next token, right?

   11. MMMarina Mogilko10:42

       Mm.

   12. TWThomas Wolf10:42

       So when you ask them something, they will create the most likely thing. Often, when you want to really create something new, and you know that very well, I guess, you know, when you, when you create, like, a video and you want something, you have to be creative. You have to invent something that's a bit out of the ordinary.

   13. MMMarina Mogilko10:55

       Yeah.

   14. TWThomas Wolf10:55

       If you just do the same thing as everyone, just nobody notices you.

   15. MMMarina Mogilko10:58

       Absolutely.

   16. TWThomas Wolf10:58

       And so these skills, like this, this, uh, maybe, uh, fact of not being self-censored and being creative, I think will always be useful. I think another thing-

   17. MMMarina Mogilko11:07

       How do you teach that?

   18. TWThomas Wolf11:08

       Yeah, I don't really know.

   19. MMMarina Mogilko11:09

       If they make a mistake-

   20. TWThomas Wolf11:10

       I think you just try to-

   21. MMMarina Mogilko11:10

       ... you're like, "Try again"?

   22. TWThomas Wolf11:12

       Yeah, try to encourage. W- we've been teaching, trying to teach creativity for quite some time, I think. Uh, in the US, it's very, it's very great. There is some country in Europe, and you see that where it's actually very great. One example is, uh, we were talking just, just recently about, uh, Stockholm and the Swedish ecosystem. They have a lot of focus on creativity at school.

   23. MMMarina Mogilko11:31

       Mm.

   24. TWThomas Wolf11:31

       I think it helps maybe also bringing out great entrepreneur-

   25. MMMarina Mogilko11:35

       Yeah

   26. TWThomas Wolf11:35

       ... like Daniel from Spotify, Sebastian from Klarna. The French system is not the best for creativity.

   27. MMMarina Mogilko11:41

       Mm.

   28. TWThomas Wolf11:41

       We are like, "You're supposed to learn-"... what's in the list, and not-

   29. MMMarina Mogilko11:44

       Yeah

   30. TWThomas Wolf11:44

       ... not to go too much, uh, out-
7. 12:48 – 15:00

   Robots in everyday life

   1. MMMarina Mogilko12:48

      is the world gonna look like? What do you think the world's gonna [chuckles] look like five to 10 years?

   2. TWThomas Wolf12:52

      It's gonna, it's gonna be very different, huh? Pretty sure. I mean, there will be common things. Uh, I, I'm pretty sure there will still be people talking, [chuckles] podcasts, I think-

   3. MMMarina Mogilko13:00

      Hopeful [chuckles]

   4. TWThomas Wolf13:00

      ... because we really like human interaction.

   5. MMMarina Mogilko13:02

      Yeah.

   6. TWThomas Wolf13:02

      I think this won't disappear. Uh, but we will have all this AI automation. A lot of the work I do just now, this, this past few months, has been around robotics.

   7. MMMarina Mogilko13:12

      Mm-hmm.

   8. TWThomas Wolf13:12

      And we just acquired a robotics company.

   9. MMMarina Mogilko13:14

      Yeah. Is it gonna be a commercial or-

   10. TWThomas Wolf13:16

       Yeah.

   11. MMMarina Mogilko13:16

       Yeah.

   12. TWThomas Wolf13:16

       It's gonna be very accessible robots that can be used mostly by developer, but, uh, I want to make it also something that you could even use with your kids to teach robotics-

   13. MMMarina Mogilko13:24

       Mm

   14. TWThomas Wolf13:24

       ... if you want. One thing I do like about robots, for instance, give kind of physical presence to AI.

   15. MMMarina Mogilko13:29

       Yeah, yeah.

   16. TWThomas Wolf13:30

       Usually AI, we just call it, like, even, even Alexa, we say, "Hey, Alexa," but you don't have any idea.

   17. MMMarina Mogilko13:35

       Yeah.

   18. TWThomas Wolf13:35

       And I like the idea of having, like, a physical, and the AI is here.

   19. MMMarina Mogilko13:38

       Yeah, you see some movement.

   20. TWThomas Wolf13:39

       Looking at me-

   21. MMMarina Mogilko13:39

       Yeah. Yeah, yeah

   22. TWThomas Wolf13:40

       ... or, like, it's looking at this.

   23. MMMarina Mogilko13:41

       Yeah.

   24. TWThomas Wolf13:41

       I, I kind of like this.

   25. MMMarina Mogilko13:42

       Is it also gonna be something like you have a robot, but then you go to Hugging Face, and you can upload all-

   26. TWThomas Wolf13:47

       Exactly

   27. MMMarina Mogilko13:47

       ... the open source apps?

   28. TWThomas Wolf13:49

       Exactly. And the cool thing is you can teach things to your robots, and you upload them on Hugging Face, and you can share them with everyone in open source way. One amazing thing with the robots is they can do... You buy them, they can do, like, 20 things, and then when you actually explore them all, you just put the robot inside-

   29. MMMarina Mogilko14:04

       Yeah

   30. TWThomas Wolf14:04

       ... and it work. But here, if it's connected to an open source repository that's growing with people inventing new things, the robot that helps you cook or the robots that, I don't know, play with this, or the robots that can clean something and that share them with everyone, you could have something that actually grow with time.
8. 15:00 – 16:45

   When will we have household robots?

   1. MMMarina Mogilko15:00

      Yeah, yeah, for, like, bedding, whatever. When is the year when every family... I'm speaking for moms here who hate laundry. [laughing]

   2. TWThomas Wolf15:06

      [chuckles]

   3. MMMarina Mogilko15:07

      When are we gonna have a robot, uh, in our household to help us with daily things? How far are we from that?

   4. TWThomas Wolf15:14

      I think the, the technology itself is very close. I would say on the tech side and the maybe prototyping demo, I would guess next year can have some stuff, and then the main question is the price of the thing. Uh, and that-

   5. MMMarina Mogilko15:27

      Price and regulation, I feel like.

   6. TWThomas Wolf15:28

      Price and regulation-

   7. MMMarina Mogilko15:29

      Yeah

   8. TWThomas Wolf15:29

      ... for sure.

   9. MMMarina Mogilko15:30

      Yeah.

   10. TWThomas Wolf15:30

       Yeah, there's two aspect. The, the security aspect will be very important for robots. I mean, there is question for safety in AI, but, I mean, saying a bad words, y- you know, by ChatGPT is not the same as a robot, like, uh, you know, making you-

   11. MMMarina Mogilko15:43

       Punching you

   12. TWThomas Wolf15:44

       ... punching something.

   13. MMMarina Mogilko15:44

       Yeah.

   14. TWThomas Wolf15:44

       So there will be question like that. Uh, but the second one is if this robot costs, like, uh, $80,000 or something like that, um-

   15. MMMarina Mogilko15:52

       What do you think-

   16. TWThomas Wolf15:53

       ... nobody's gonna buy.

   17. MMMarina Mogilko15:53

       What do you think it's gonna cost next year? Like, versus in five years.

   18. TWThomas Wolf15:56

       Yeah, so the early one, uh, will still be like the price of a car right now.

   19. MMMarina Mogilko15:59

       Mm-hmm.

   20. TWThomas Wolf15:59

       I think the... You need two hands that can grab things, so usually it's around 15, 20K.

   21. MMMarina Mogilko16:05

       Mm-hmm.

   22. TWThomas Wolf16:06

       I would love to see them much cheaper. And so-

   23. MMMarina Mogilko16:08

       Do you think there's gonna be subscription as well? Subscription-

   24. TWThomas Wolf16:10

       That's also possible

   25. MMMarina Mogilko16:11

       ... please.

   26. TWThomas Wolf16:11

       Yeah.

   27. MMMarina Mogilko16:11

       Yeah?

   28. TWThomas Wolf16:11

       Probably, probably it's gonna be like that for, for the expensive one. But also, I don't think every robot has to be a humanoid. At Hugging Face, we only do open source robotics, and the f- fun part about this is we can make a platform that a lot of people can use to develop di- different type of robots. So it can be just one arm, can be just a, a funny head that move and talk, um, can be just something that walk but has no arm. We made a robot duck. That's just a duck that walk around you.

   29. MMMarina Mogilko16:37

       Mm.

   30. TWThomas Wolf16:37

       Very cute. So we can explore a lot of form factors-
9. 16:45 – 18:47

   Privacy & safety in home robotics

   1. TWThomas Wolf16:45

      terms of price, possibility, capabilities, uh-

   2. MMMarina Mogilko16:48

      How do you make sure that if someone creates an open source app for a robot, that it doesn't have this spy element that monitors inside the house?

   3. TWThomas Wolf16:56

      Mm. Yeah, that's a big question. I mean, for both models, uh, and robots, already today models, I think one of the big advantage of open source is you can download and run them locally. So you download them, and then you can even-

   4. MMMarina Mogilko17:10

      You don't have to talk to the server

   5. TWThomas Wolf17:11

      ... cut the internet.

   6. MMMarina Mogilko17:11

      Okay. Mm-hmm.

   7. TWThomas Wolf17:12

      You can, you can stop the Wi-Fi, and the thing, it just cannot send anything-

   8. MMMarina Mogilko17:15

      Mm-hmm

   9. TWThomas Wolf17:15

      ... and it can run locally.

   10. MMMarina Mogilko17:16

       Mm-hmm.

   11. TWThomas Wolf17:16

       That's something you can only do with open source right now. So if you use ChatGPT, you have always-

   12. MMMarina Mogilko17:21

       Yeah

   13. TWThomas Wolf17:21

       ... to send the thing. And for robots, I think it, it will be very important. If the Wi-Fi stop, what does your robot do? Okay, I don't know. It's, like, loading the dishwasher.

   14. MMMarina Mogilko17:29

       Throw up. [chuckles]

   15. TWThomas Wolf17:29

       I just let the, let the thing fall, and then push. [chuckles]

   16. MMMarina Mogilko17:31

       [chuckles] I'm done.

   17. TWThomas Wolf17:32

       You want this thing to work, you know, even when there is, like, a problem, and also you don't want them to communicate too much, I think.

   18. MMMarina Mogilko17:37

       Yeah.

   19. TWThomas Wolf17:37

       You want privacy-

   20. MMMarina Mogilko17:38

       Yeah

   21. TWThomas Wolf17:38

       ... very strong privacy with something that's in your house.

   22. MMMarina Mogilko17:41

       So just to sum it up, you think next year we're gonna have robots who are capable of doing things in a house, and then in five years, I would have a robot?

   23. TWThomas Wolf17:48

       ... yeah, I think next year we'll see two thing. On the, on the web, on the, on the LLM, we'll see agents that can do a lot of tasks. We start to see that today.

   24. MMMarina Mogilko17:56

       Yeah.

   25. TWThomas Wolf17:56

       So a lot of complex tasks we do on computer will be automated. We see a beginning of the same thing in the physical world, with robots starting to be able to do some tasks, really, and this will just keep, keep growing, uh, keep growing with time. And maybe the s- the third trend that's, that's a bit more complex, I think, will be... we start to have really this photorealistic content at scale, so video that are indi- indistinguishable of, of reality-

   26. MMMarina Mogilko18:20

       Yeah

   27. TWThomas Wolf18:20

       ... with sound that's also indistin- indistinguishable, and this will also be a big question, I think. In my opinion, the optimistic way is, um, we'll pay more attention to being, uh, face-to-face-

   28. MMMarina Mogilko18:32

       Present

   29. TWThomas Wolf18:32

       ... with someone because we'll know it's a real moment.

   30. MMMarina Mogilko18:35

       Yeah.
10. 18:47 – 21:18

    What AI will look like in 5 years

    1. MMMarina Mogilko18:47

       So our present is iPhones, I have mine here, and a MacBook. In five years-

    2. TWThomas Wolf18:53

       Mm.

    3. MMMarina Mogilko18:53

       ... AI, uh, do you see AI being in some piece of tech? And if yes, what is it? Is it a robot? Is it a wearable? Is it glasses?

    4. TWThomas Wolf19:03

       Yeah, it's a good question. My view of AI is, um, we have a big trends that started, uh, last year, which is we can do a much smaller AI models that are actually very high performance, and so this means you can start to embed them a bit everywhere. There's a lot of people are working on AI chips that will be much more optimized, con- consume less energy, and this, according to me, it points to a future where you have, like, AI a bit diffused everywhere, just like electronics. You know, we have electronics-

    5. MMMarina Mogilko19:30

       Yeah

    6. TWThomas Wolf19:30

       ... in our camera, in our smartphone, and so we'll get used to having basically all our electronic system that just understand much better what we want to do.

    7. MMMarina Mogilko19:38

       Let's wrap up with this question. Imagine we're here in five years, and, uh, can you talk about the wildest thing that's going on in the world in five years? [laughing]

    8. TWThomas Wolf19:49

       [laughing] Yeah. One, one of my most exciting, uh, direction right now in AI is, uh, using all the techniques we've learned to train good AI models, so kind of train intelligence model, applying that to other field of research, applying that to discovering new materials for battery, applying that to, you know, weather prediction, applying that to, uh, fusion, uh, fusion energy, you know? So my hope is in five years, we'll, we'll have some of these things that are maybe not just a chatbot. A chatbot is nice, but it's not, like, changing the world in a way. Like, it's not, you know, inventing things that make us actually solve big, big challenge, right? And so all these things that I group them generally in the idea AI plus science, so you take very smart AI models, or the... You take just knowledge we have around this model and try to apply that to a scientific field where you can do a fundamental breakthrough. I think this is something that excites me a lot right now.

    9. MMMarina Mogilko20:43

       So you think cancer will be cured-

    10. TWThomas Wolf20:44

        Yeah, these type of things-

    11. MMMarina Mogilko20:45

        ... in five years?

    12. TWThomas Wolf20:45

        Right.

    13. MMMarina Mogilko20:46

        Yeah.

    14. TWThomas Wolf20:46

        Cancer, I'm not sure, because it seems like, uh, the more we're looking deeper, the more it's kind of a very complex, uh, um... But definitely some, some disease, and also definitely material science, there is some super exciting developments.

    15. MMMarina Mogilko20:58

        Mm-hmm.

    16. TWThomas Wolf20:59

        Material science is kind of this unsexy scientific, you know, uh, field, that actually can change a lot of things because it unlock small metal batteries like carbon dioxide, um, uh, trapping or this type of things that can be fundamental in, you know, just-

    17. MMMarina Mogilko21:14

        Mm-hmm

    18. TWThomas Wolf21:14

        ... solving huge problem like climate change.

    19. MMMarina Mogilko21:17

        In five

11. 21:18 – 25:15

    Will AI lead to mass unemployment?

    1. MMMarina Mogilko21:18

       years, w- what do you think about unemployment? A lot of people are worried about that when it comes to AI.

    2. TWThomas Wolf21:24

       A lot of... I think there is two, two main trends. Uh, one that I used to be for a long time in this was, um, AI will unlock so much creativity. Start from kids, uh, you know, when I do, uh, vibe coding with my kids, they have a lot of ideas for website, a website to connect, uh... We need, like, a website to share your cats with other people, and, like-

    3. MMMarina Mogilko21:43

       Nice

    4. TWThomas Wolf21:43

       ... a lot of things, or, like, website to connect-

    5. MMMarina Mogilko21:45

       I love that idea

    6. TWThomas Wolf21:45

       ... like, amateur football player with scouts.

    7. MMMarina Mogilko21:47

       Yeah.

    8. TWThomas Wolf21:47

       So they, they have all of this idea, and I'm like, "This is great," and, and all of this could even be, like, potential, uh, companies or something like that. So, so you have this website where basically, because it make a lot of things very accessible, a lot of people can become entrepreneur, basically. But it's also true that not everyone wants to be an entrepreneur.

    9. MMMarina Mogilko22:04

       Mm-hmm.

    10. TWThomas Wolf22:04

        Like, people, uh... some people don't. Uh, and then I'm, uh, I'm wondering because I was a former lawyer before, uh, starting Hugging Face, so I, I was for six years in a, in a intellectual, uh, property, and it's true that these fields are, are gonna be very strongly disrupted by AI.

    11. MMMarina Mogilko22:21

        Yeah.

    12. TWThomas Wolf22:22

        Like, there's a lot of, like, support function that are, that are currently handled by, by many people, and that can definitely fall in the scope that AI can do. And it's also field where you study, like, you do quite long studies, so you've studied five, six, eight years for your job. So it's not like you're gonna jump directly to a new field, right?

    13. MMMarina Mogilko22:39

        Exactly.

    14. TWThomas Wolf22:40

        So this type of, uh, this type of profession is the big question for me, is how can you handle that? It's also a topic that I think very few governments or very few, like, you know, a think tank are really currently tackling. Like, a lot of people, I feel like, are more like, "Let's not [laughing] look too much at this. Let's talk about, I don't know, the danger of AGI, uh, you know, ruling the world," instead of just something that's definitely gonna happen in the coming five year, which is big job disruption in some, in some, in some field.

    15. MMMarina Mogilko23:10

        So what's your advice to people who are studying for five years to do something that might be disrupted?

    16. TWThomas Wolf23:14

        Yeah, I would say the two advice... I mean, one is definitely use these tools because, and try to see what they are capable of, because you want to become the master of-

    17. MMMarina Mogilko23:22

        Yeah

    18. TWThomas Wolf23:22

        ... of this tool, right? And you want to, to, to, to be able to use them in your, in your profession, so you, you have to stay in touch with this revolution. You cannot just ignore it. And the second one would be, um, yeah, start to, to see if you don't, if you don't want to be very creative, uh, uh, and, you know, uh, like, what is... Once, once you start using this tool, what you feel is the remaining exciting part of your job, and is it something you, you're actually really happy to do?

    19. MMMarina Mogilko23:46

        Mm-hmm.

    20. TWThomas Wolf23:46

        And if not, I think it's time to, to see if you don't want to, to create something new or if you don't want to start something new.

    21. MMMarina Mogilko23:51

        Yeah. And, and that was the first? You said you, you had two, two opinions on that, so the first-

    22. TWThomas Wolf23:56

        Yeah. I mean, the first one is everyone become an entrepreneur, basically.

    23. MMMarina Mogilko23:59

        Yeah.

    24. TWThomas Wolf24:00

        But I, I think the second one is, is kind of the panel, which is, I'm not sure everyone wants to become an entrepreneur.

    25. MMMarina Mogilko24:06

        Yeah.

    26. TWThomas Wolf24:06

        So that-

    27. MMMarina Mogilko24:06

        And so what do they do?

    28. TWThomas Wolf24:07

        And here I don't have a great answer.

    29. MMMarina Mogilko24:08

        Okay.

    30. TWThomas Wolf24:08

        I think, I think this is something that should be probably tackled at a, at a society level.

Episode duration: 25:15