Big boost for Hugging Face. The artificial intelligence company created in 2016 by three French people takes on a new dimension. The start-up, headquartered in Brooklyn, a district of New York in the United States, has established itself in seven years of existence as the reference toolbox in the era of machine learning. Above all, it can rely on a community of particularly loyal users.
“I’m a big fan of Hugging Face,” keeps repeating Andrew Ng, a Stanford professor, co-creator of the education site Coursera, and now one of the maintainers of Google Brain. This Friday, August 24, the start-up announces that it has completed a new financing round which values it some 4.5 billion dollars. Quite a feat for a company whose name and logo pay homage to the cute emoji that reaches out to give you a hug.
If, among the investors of this last round, there are many leading American players such as Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM, or Salesforce, the company, whose l he idea was born at Players, a three-storey beer bar in the 2nd arrondissement of Paris, which promises to gain momentum in France.
We realize this when strolling through the Parisian premises of the company. On the 4th floor of an Eiffel building converted into a coworking space, we meet Mishig Davaadorj, a Mongolian engineer trained at Colorado State University, or even Jingya Huang – a size in machine learning -, born in China and passed through the Mines of Paris.
But above all, the company is keen to show that, even as it grows, it wants to remain open. As in a huge store – in artificial intelligence, we call it a library – each user can come and help himself. On these virtual shelves, for example, we have access to tools for translating more than a thousand languages (such as Lingala, spoken in Côte d’Ivoire, Yue, a Chinese dialect, Tzotzil of the Mayas, etc.) or even segmentation instruments image, useful for example to identify pedestrians in a video. It is also possible to modify the list of commands that have been given to the machine to achieve this result. In a way, the user regains control.
Hugging Face has invented an original economic model. In total, some 5,000 companies, including Renault, Bloomberg and the Standard Bank, based in South Africa, regularly come to help themselves on its shelves. Because if the general public can access its services for free, the Franco-American start-up charges access to its tools to professional customers. Similarly, it charges the computing power necessary to carry out AI operations to its suppliers, which are cloud operators such as Amazon, OVH, Scaleway or even manufacturers of computer equipment such as Nvidia.
With Hugging Face, deep learning models are no longer just for specialists. While an earthquake wreaked havoc in Turkey, Merve Noyan, an engineer with the company, was able to locate calls for help launched on social networks and thus effectively alert the emergency services.
“There are great companies like Hugging Face doing a great job,” Google CEO Sundar Pichai even praised at the Re-code conference last September. “Their open source approach is very useful,” explained Yann Le Cun, one of the fathers of deep learning, now at competitor Meta. The real challenge will therefore be for this nugget of AI to remain independent of Big Tech with which it works on a daily basis.
Its creators remain ardent admirers of the open model. “The future of artificial intelligence must not be under the control of a handful of American Big Techs, it must be accessible to as many people as possible, AI must be a digital commons”, assures Julien Chaumond.
Asked about the success of Hugging Face, Emmanuel Macron explains this week in Le Point: “The Americans are ahead, but they have not won the battle. If Europe knows how to react, we can have our own model. This is the goal I will pursue. We have started to create structures and invest in research. Everything fits, by the way. For supercomputers, we already have some, but to have more, you need a lot of energy. However, France is well placed in carbon-free energy. We train a lot of talent, but we need to accelerate by training more, and much more investment. »
