Challenging the Average With Open-Source AI: Hugging Face’s Thomas Wolf

**SPEAKER_1** (0:02)
Today's guest believes open source and closed source models will co-exist in the world of AI. Find out what he considers the opportunities and drawbacks to each, as well as how communities can make AI tools and themselves work better together on today's episode.

**Thomas Wolf** (0:19)
Hello, I'm Thomas from Hugging Face, and you're listening to Me, Myself, and AI.

**Sam Ransbotham** (0:25)
Welcome to Me, Myself, and AI, a podcast from MIT Sloan Management Review, exploring the future of artificial intelligence. I'm Sam Ransbotham, Professor of Analytics at Boston College. I've been researching data, analytics, and AI at MIT SMR since 2014, with research articles, annual industry reports, case studies, and now 12 seasons of podcast episodes. On each episode, corporate leaders, cutting-edge researchers, and AI policy makers join us to break down what separates AI hype from AI success. Hey, everyone. Thanks for joining us again, and welcome back to a new season. Today, I'm lucky to be talking with Tom Wolfe. He's the co-founder and chief scientific officer of Hugging Face. Tom, great to have you on the show today.

**Thomas Wolf** (1:18)
Thanks, Sam. It's a big pleasure to be here.

**Sam Ransbotham** (1:21)
Let's start with Hugging Face itself. Some of our listeners may not be familiar with Hugging Face. Can you give us a brief overview of what the company does and what you do?

**Thomas Wolf** (1:29)
Yeah, of course. Hugging Face is an open-source AI platform. We give access to all the AI models that are open-source, which means that basically these are the models you can download and run wherever you want. So when you use an AI model nowadays, you can choose either to go to Chai GPT, Anthropic, or Google. They are the most widely diffused at the moment. Or sometimes you want to run the AI models on your own data center, or you want to run them on some specific hardware, could be like local hardware, or could be maybe faster chips because you need instant response. In most cases, you will want to go for open-source AI model, which is a model you can basically just download. There is quite a lot of them. On the Hugging Face, there is close to four million of this model at the moment. There is one new model being published every five seconds.
Some of the most famous ones are the MEDA series, the LAMA series, and one that I think got the most adoption and most visibility recently was DeepSeek, which was released in January and kind of crashed the stock market at the same time that it was released. And so over the past eight years, Hugging Face has been building this platform, growing it together with the community of people and teams who are both sharing and downloading models. This community is now roughly 10 million users and AI builders, how we call them. And we've expanded as well beyond just model hosting to also host datasets, which are used to train models, to fine-tune them, to evaluate them. And more recently, also what we call spaces, which are simple, low-code way to test all of these models.

**Sam Ransbotham** (3:08)
So there are a lot of people offering solutions here. And if I think back on the way technology has developed throughout the history of mankind, people came up with chips and Bell Labs, and Intel came along and built fabs for processors. None of that was open source. Why is open source important here?

**Thomas Wolf** (3:27)
I think open source has been always important in a way. The thing is open source is more often the long game in computer science. So if we go back to, for instance, the year 2000 or pre-2000, where basically Microsoft was one of the largest operating system, and Linux was somehow more for fanatics or geeks. And now if you fast forward 20 years later, Unix is really the basics of all enterprise software and all enterprise cloud, basically. You almost always run them on some version of Linux. Even Mac OS, which I think is probably the most widely diffused on consumer laptops nowadays, is one of the largest competitor to Windows base, is itself based on the Unix core. So there is this trend, which is open source has some advantages that make it extremely appealing in the long term. Obviously on the short term, you can go faster with closed source. And that's also what we see with closed model. You can iterate faster. You can raise larger amounts of capital to train your models. You can try to grab, you know, the most expensive AI researcher and pay them, pay them like huge sum of money. We've keep pushing a lot for open science. And just this Tuesday, we published a new model that's called a SmallLM3, which is an extremely smart model, but it's the best one, but at 3 billion parameters. So it's in the range of size that you can run on your laptop and even on a smartphone. And we've decided to share at the same time, all the data, all the recipe, all the knowledge on how to build this model. It's fine for us because we don't make money out of these models. And we think it's very good because anyone who can want to build a model based on this, or looks a bit like this, or want to extend this type of model, now has all the knowledge they want to start. So we think open source can be defined in many ways in AI, but we think the most radical way is to say you share just everything. You share the data, you share the code, you share the recipe, and we wrote even a very long blog post. We're going to probably make it into a full-blown paper.