Kyutai is a French AI research lab with a $330 million budget that will make everything open source. I provided more details on French billionaire and CEO of Iliad Xavier Niel’s plans for an artificial intelligence research facility in the Paris region at this morning’s AI-PULSE conference, which Scaleway was hosting.
This new research facility, Kyutai, will be a nonprofit organization focusing on artificial general intelligence. It will collaborate with PhD students, postdocs, and other academics on research articles and open-source initiatives. When Iliad first revealed this research center, the company stated that Niel would contribute €100 million to this effort, equivalent to around $109 million when converted using today’s currency rate.
During the conference, Niel said, “Thanks to some amazing friends who are there today, we are close to reaching €300 million for financing this initiative.” One of their “friends” is another French billionaire named Rodolphe Saadé, the CEO of the French shipping and logistics business CMA CGM. Saadé has also contributed €100 million to the cause. There are also some other, less significant supporters, such as the foundation of Eric Schmidt and some other donors who choose to remain anonymous.
This is just the beginning, as Kyutai always seeks more donors to support their cause. At a press conference shortly after the announcement, Saadé stated, “What’s interesting with so many journalists in the room is that the project will potentially interest other investors.”
Because Kyutai will be working on basic models, they will also require some computational power. The fantastic news is that Scaleway, the Iliad cloud business, has just recently bought one thousand Nvidia H100 GPUs. These high-end graphics processing units (GPUs) are necessary for inference and model training, and Kyutai will have access to them at an additional fee.
The primary scientific team at Kyutai has already begun hiring new members. Patrick Perez, Edouard Grave, Hervé Jegou, Laurent Mazaré, Neil Zeghidour, and Alexandre Defossez were the guys who took the stage this morning to talk about their prior work and what they have in mind for the research lab. They have experience working for companies such as Meta’s AI research team, FAIR, Google’s DeepMind subsidiary, Inria, and similar organizations.
Patrick Perez, who has a history of working with Valeo, has been selected to take over as the head of the research lab. Yejin Choi, Yann LeCun, and Bernhard Scholkopf are members of the team of scientific advisers that Kyutai has assembled. These individuals are all renowned experts in the field of AI research. They will go about checking the work of everyone once or twice a year and provide feedback.
Kyutai believes it will be able to persuade researchers to join its lab since the researchers will have the opportunity to submit research papers during their time there.
“Unfortunately, extensive technology businesses are becoming increasingly intolerant of scientific papers. During the press conference, Niel stated that the award does more than boost scholars’ egos; it also helps promote research and contributes to the general good.
Naturally, this is hardly the first open artificial intelligence research lab. OpenAI was originally a charitable organization, as the name still reflects. However, when Sam Altman began working on OpenAI full-time in 2019, things took a dramatic turn for the better. OpenAI altered its organizational structure to be more typical of corporations and received investment from Microsoft.
Other firms, such as Meta, with its Llama model, and Mistral AI, have also worked on open-source fundamental models for their businesses. The researchers refer to their work as open science, but Kyutai’s models will also be available as open source. They want to make the models, as well as the training source code and data that describe how they made the models, available to the public as open source.
During the press conference, Defossez stated, “When it comes to the timeline, I don’t think our aim is necessarily to go as fast as Mistral because our ambition is to provide a scientific purpose, an understanding, and a code base to explain the results.” He didn’t necessarily mean they wanted to go as fast as Mistral. However, they anticipate having something to talk about within the following year.
Another Kyutai research team member, Mazaré, agreed with Kyutai that Mistral AI’s first open-source model was a success. Mazaré explained that this was because many community members had been working to fine-tune the model and investigate potential use cases based on the Mistral 7B model.
It will be fascinating to watch if a research lab is more efficient than private firms at releasing basic models, and it will also be interesting to see how private companies harness the work that Kyutai has done for commercial applications of their products.
In a video message shown at the conference that had been produced, French President Emmanuel Macron stated, “I’m also a strong believer in open source, and we need to turn it into a French asset.”
The French perspective is to regulate use cases rather than models.
Macron also used this chance to clarify and defend France’s position on Europe’s Artificial Intelligence Act. He stated that use cases, not model manufacturers, should be subject to regulation. During the trilogies, which are negotiations between Europe’s three major institutions (the Parliament, the Commission, and the Council), France has attempted to scale down the AI Act.
“Regulation is not the enemy of innovation; on the contrary, it is one of its greatest allies. Macron was quoted as saying that it is not a question of identifying excellent models but rather that we must ensure that the services made available to our inhabitants are secure for our residents, other economic players, and our democracy.
“With work on the European regulation for artificial intelligence currently in ‘trilogues,’ regulation must be controlled and not punitive in order to preserve innovation and regulate usage rather than technology as such,” he noted. “Trilogues”
During the press conference, Niel essentially took France’s viewpoint on this issue as his own during the discussion. According to him, Europe is falling farther and further behind in artificial intelligence (AI) innovation, and legislation will slow European entrants and lower the prospects of them catching up.
“At this point, we focus more on the invention than the regulation phase. “The introduction of regulations results in the construction of obstacles for potential competitors,” explained Niel.
Things may improve if French AI businesses achieve unprecedented commercial success. Niel added later in the chat, “I’d love it if one day we could talk about French imperialism in AI.” He said this with a smile.