Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
A few hours before OpenAI’s board sacked its chief executive, starting a five-day soap opera that has gripped the global artificial intelligence community, Xavier Niel unveiled an attempt to create a French version of the Silicon Valley start-up in Paris’s 13th arrondissement.
Kyutai resembles OpenAI in its original form, rather than its current dysfunctional state: a non-profit research lab designed to build and experiment on large language models — the algorithms that are predicting the end of our sentences and writing essays and code for us. In the hall of his start-up incubator Station F, Niel sat alongside fellow French billionaire Rodolphe Saadé, head of shipping group CMA CGM, and ex-Google chief Eric Schmidt, who will also help fund the €300mn initiative.
Niel’s project is one belated attempt in Europe to join the race for generative AI, and alongside the start-up Mistral AI — also backed by Niel and Saadé — could turn Paris into a hub in the continent. At stake, Niel believes, is Europe’s ability to influence, profit from and regulate these powerful algorithms — with all their inherent biases partly emanating from the data sets on which they are trained — that are going to shape our lives.
“We are only a few months late, so we are getting on it, we are putting means behind it and we are going full speed,” the media-shy telecoms entrepreneur told radio station FranceInter this week. “I don’t want our children to depend on algorithms that are not made here.”
The idea behind Kyutai is to be truly “open” — a concept even OpenAI has departed from since teaming up with Microsoft to commercialise its language model ChatGPT and becoming a for-profit entity overseen by a non-profit board. Kyutai’s research — led by six former Deepmind, Meta and Microsoft employees and overseen by Meta AI chief scientist Yann LeCun, Max Planck Institute professor Bernhard Schölkopf and University of Washington professor Yejin Choi — will be accessible to whoever wants to use it for commercial purposes. Transparency will extend to the source code of the models, said Edouard Grave, one of the researchers.
Across the Atlantic the power struggle at OpenAI, which seemed to stem from concerns over the fast advancement of generative AI tools, has reinforced the idea that the way to govern this potentially highly disruptive tech was not yet set in stone and that if anything, the world needed alternatives. The reputational damage inflicted on the world’s best-known AI start-up could also provide an opportunity for European rivals to lure venture capitalists to their shores. But more importantly, the affair has crystallised the need for Europe to build the processing capacity needed to train these models.
The saga, which ended with the reinstatement of Sam Altman as OpenAI’s chief executive, has highlighted “the dependence of all these AI companies on Big Tech for their computing power, their core infrastructure”, said Martin Tisné, head of Pierre Omidyar-backed philanthropy AI Collaborative and an adviser to the French government on AI.
This is why Altman turned to Microsoft, which has provided billions of dollars in cash and processing power to the start-up, Tisné says. “We are seeing the impact of the consolidation of a sector that at first glance appears to be start-up led but in reality is sustained by Google Cloud, Amazon Web Services and Microsoft. It’s a massive concentration of market power in a hyper-privatised sector controlled by a few large US firms.”
“We will need to provide the infrastructure to support start-ups,” he adds. “Otherwise the only route for these young companies is to partner with — or be acquired by — the tech behemoths.”
Niel is well aware of that. Kyutai will train its models on the supercomputer his cloud company Scaleway has designed with the help of chipmaker Nvidia. “We have built the fifth most powerful supercomputer in the world — the first four being in the US. There might be others in China, we don’t know,” he said this week. European start-ups could also turn to OVH, based in northern France. Scepticism abounds over whether Europe has any chance of being relevant in the AI age.
But Niel has proved many people wrong in the past. “I have built my career on the pessimism of others,” he said.
Read the full article here