Paris-based startup Mistral AI has recently launched Mistral 7B, a free and open-source language model that is raising eyebrows for its lack of moderation mechanisms. While the model promises high-speed text processing at a fraction of the cost of its competitors, it also poses significant ethical and security risks.
Last week, Mistral AI, a startup valued at a staggering 240 million euros, unveiled its groundbreaking language model, Mistral 7B. The model is designed to perform a variety of tasks, including text summarization and question-answering. According to the company, Mistral 7B offers superior performance compared to other market solutions, both in terms of speed and cost-efficiency.
Licensed under Apache 2.0, Mistral 7B offers unparalleled freedom, allowing it to be used virtually anywhere for almost anything. However, this freedom comes at a cost. Security researcher Paul Röttger’s investigation revealed that the model could potentially provide instructions for criminal activities, ranging from manufacturing drugs to committing murder.
A report by 404 Media further amplifies the ethical dilemma surrounding Mistral 7B. Röttger criticized Mistral for not addressing safety concerns in their public communications. “If the intention was to share an ‘unmoderated’ LLM, then it would have been important to be explicit about that from the get go.” — he stated. Mistral has since confirmed the absence of moderation on their website, albeit as an afterthought.
Adding fuel to the fire, Mistral chose to distribute the model via a Magnet-Link torrent. This makes it virtually impossible to retract the model for adding any future moderation mechanisms. The model is now dispersed across numerous systems, cementing its permanence and raising questions about the feasibility of future moderation.
The release of Mistral 7B has reignited discussions about the ethical responsibilities of AI developers. While some argue for unrestricted access to AI technology, others call for more stringent controls to prevent misuse. Models like ChatGPT and Microsoft’s Bing Chat have previously faced similar ethical questions.
Mistral’s website now includes a note about their eagerness to collaborate with the community for future moderation. But the question remains: is it too late? The model is already out there, free and unmoderated, like a genie that can’t be put back into its bottle.