Since DeepSeek caused the biggest destruction of stock market value in 24 hours among US tech groups on Monday, January 27, it has escaped no one's notice that this artificial intelligence (AI) start-up is Chinese. But another of its characteristics, less emphasized, is probably just as important: its AI models are said to be "open." "This means that anyone can use them, modify them and build them for free. That is very exciting for the open source community," said Adina Yakefu, a researcher at Hugging Face, the leading platform for publishing AI models, speaking to financial daily Les Echos.

Made available by DeepSeek, the V3 and R1 models, as well as their variants, have already been downloaded more than 700,000 times by developers and companies. "The balance of power now appears to be shifting along two key axes: one between the United States and China, and another between closed-and open-source models," said former Google boss Eric Schmidt, in an op-ed published by the Washington Post on Tuesday, January 28.

Open-source AI offers the hope of an alternative to the domination of the sector by American digital giants such as OpenAI and its partner Microsoft, or Anthropic and its partners Amazon and Google. Often open in their early days, these giants have, in the name of making their investments profitable, evolved towards closed models, whose details are opaque, proprietary (private), and which have exclusively fee-based access. In theory, open-source AI also enables companies and public authorities to control their technology. By achieving similar performance with fewer computer calculations, and therefore less energy and resources, DeepSeek in principle reinforces the hope of more frugal AI. Behind this start-up and the forthcoming Paris summit lies the goal of a more open and economical AI. But this is not guaranteed. The debate involves subtleties and challenges.

'Little tech'

France has made open models one of its focus areas for the Artificial Intelligence Action Summit it is organizing in Paris on February 10 and 11. This openness is presented as an alternative to the tech of American digital giants. It is supported by Hugging Face or Mistral, founded by French people, as well as by Meta, an American behemoth. Summit organizers hope to include [openness] in the final declaration and launch a €2.5 billion foundation to create models, security tools or open databases for AI training. They are hoping for support from China and developing countries.

You have 37.7% of this article left to read. The rest is for subscribers only.