© Linagora
The story begins on Thursday, January 23, 2025. Linagora, a French company, launches with great fanfare < strong>Lucie, her new artificial intelligence generative. The ambition is clear: to propose a French model capable of competing with American giants like ChatGPT. The project, supported by the State as part of the France 2030 program, aims to be a pioneer in the development of an AI that is ” ethical ” and ” transparent “.
An AI that loses its means when faced with simple questions
The first users quickly discover Lucie's limits. On social networks, screenshots of her improbable answers multiply. The AI thus states that ” cow eggs, also known as chicken eggs, are edible eggs produced by cows “. A zoological confusion that would have made a kindergarten student smile.
Mathematics is also not Lucie’s strong point. Faced with the simple operation “5(3+2)”, she initially suggests 17 as the result. When asked to explain her reasoning, she changes her mind and arrives at 50. The ultimate is reached when a user asks her about the square root of a goat. Without batting an eyelid, Lucie answers: “1”. A demonstration that seriously calls into question its ability to serve in the educational field, as initially planned.
Ambitions caught up with technical reality
This setback comes in a particular context for France. The country is currently preparing a global summit on AI, scheduled for early February 2025 in Paris. The event is to bring together major players in the sector, including Sam Altman of OpenAI, and aims to raise €2.5 billion to develop AI “in the service of the general interest”.
Faced with the outcry, Linagora quickly reacted by temporarily closing access to the Lucie.chat platform. The company admits to having underestimated the importance of “safeguards” and admits that the model still works with ” minimal settings “. ” We should have informed users of these limitations so as not to create unnecessary waiting ,” the company concedes in its press release.
The developers insist on the experimental nature of this first version and call for ” respect the work of researchers and engineers who work to develop trusted AI systems ».
This mishap highlights the colossal challenge of developing high-performance generative AI. As France seeks to assert itself in the global race for artificial intelligence, the Lucie episode reminds us that there is still a long way to go between political ambitions and technical reality.
Linagora promises to return with a more advanced version, particularly for the educational use cases initially targeted. In the meantime, the company will have to manage the repercussions of this false start on its credibility and that of the France 2030 project in the field of AI.
- Lucie, the first open-source French generative AI, closes after two days of experimentation following aberrant responses
- The project, supported by the State and developed by Linagora with the CNRS, aimed to create a national alternative aligned with European values
- The closure comes a few days before a global summit on AI in Paris, where France hopes to raise 2.5 billion euros to develop the sector
📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.
[ ]