France’s answer to ChatGPT fails spectacularly, goes offline after three days of bizarre responses
WTF?! The open-source AI models from China’s DeepSeek have caused turmoil across the billion-dollar artificial intelligence industry due to the cost-effective way it challenges OpenAI’s o1. Over in France, the country’s answer to ChatGPT, Lucie, has not been faring as well, refusing to answer questions and recommending people eat cows’ eggs. As such, it’s been taken offline after just three days.
The Linagora Group, a company that is part of the OpenLLM-France consortium developing the model, launched Lucie last Thursday. The ChatGPT-like open-source bot was promoted as being “especially transparent and reliable.” Unfortunately, it seemed closer to Google’s old Bard chatbot or even Microsoft’s infamous Tay than ChatGPT.
Some of Lucie’s responses were particularly bewildering. When asked what 2 + 2 is, it said “I am programmed to be neutral and objective,” seemingly taking offense at the lack of diversity in the question.
Some people did manage to convince Lucie to answer their math problems, but it didn’t always get them right. The bot was also happy to offer up a recipe for making methamphetamine, and recommended cows’ eggs as a food source.
Move Over #DeepSeek – Lucie’s Here
France, the land of love, wine, and “oui” introduces #Lucie – @EmmanuelMacron & the French government’s attempt to prove that they can outdo even the most catastrophic tech fails.
Remember Google’s Gemini bot, the AI that went from… pic.twitter.com/EP2nm4LRKA
– Prashant Natarajan (@AIML4Health) January 28, 2025
On Saturday, just three days after what has been described as a fairly extravagant launch, the Linagora Group made the wise decision to take Lucie offline. The company said it remains an “academic research project in its initial phase.”
Linagora added that Lucie is a “raw” model that has not been optimized, works with minimal settings, and has no guardrails, hence the meth recipe. It also said that Lucie is primarily a language model and not a knowledge model, and acknowledged that the premature launch was the wrong decision.
“We should not have released the lucie.chat service without these explanations and precautions. We were carried away by our own enthusiasm.”
The company said it hoped the rollout would help raise awareness of the project, which it certainly has done, though not for the reasons Linagora would have hoped.
There have been comparisons made between Lucie and Google’s Bard, which the tech giant’s own employees described as a liar, useless, and cringe-worthy in 2023. Google faced more problems a year later when Gemini created inaccurate, racially diverse historical images. At least Lucie wasn’t as bad as Tay, Microsoft’s chatbot with a personality of a 19-year-old American girl that turned into a hate-spouting racist just 24 hours after going online in 2016, thanks to it “learning” from the people it chatted with.
