Following three intense days of negotiations, the EU AI Act has finally passed, celebrated by Carme Artigas, Spain's Secretary of State for Digitalization and Artificial Intelligence, who called it “a historical achievement and a huge milestone towards the future!”
Among its key provisions, the EU AI Act mandates that systems designated as ‘high-impact general-purpose AI models’ must comply with transparency standards. Additionally, those identified as ‘high risk’ are subject to more rigorous requirements, including risk management, significant incident monitoring, AI model evaluation, and red teaming strategies.
However, the excitement surrounding the EU AI Act was short-lived. As a provisional agreement, it still requires final wording, which could take months, and it won't become enforceable until two years after receiving approval from European lawmakers. In the fast-evolving AI landscape, two years may feel like an eternity.
Meanwhile, high-profile developments in the AI sector further overshadowed the Act. Paris-based startup Mistral generated buzz by releasing a new large language model (LLM) with just a torrent link. Shortly after, Mistral, founded only seven months ago by former Meta and Google researchers, announced a remarkable $415 million funding round led by investors such as Andreessen Horowitz, Lightspeed Venture Partners, Nvidia, and Salesforce—valuing the company at about $2 billion.
Notably, Mistral has been vocal against stringent regulations for foundation models. Just a month ago, the company lobbied against a tiered regulatory approach proposed by the European Parliament that sought to impose guardrails on generative AI. His lobbying efforts were spearheaded by Cedric O, France’s former State Secretary for Digital, highlighting Mistral's strong government connections.
In a recent interview, O remarked on the critical nature of the EU AI Act for Mistral’s future, stating, “Depending on the final shape of the AI Act, it could kill Mistral or allow it to grow, and there is a lot of gray zone in between.” It appears Mistral's lobbying efforts paid off; the finalized EU AI Act offers substantial exemptions for open-source models, except those deemed to pose systemic risks.
TechCrunch aptly illustrated the current state of the AI market, noting that Mistral’s success reflects the broader struggle for AI sovereignty within the EU. As Germany’s Vice Chancellor and Minister for Economic Affairs, Robert Habeck, remarked, “The thought of having our own sovereignty in the AI sector is extremely important. [But] if Europe has the best regulation but no European companies, we haven’t won much.”
Looking ahead, it is evident that AI regulation—both in the EU and worldwide—will need to play catch-up. Just as Mistral’s new Mixtral 8x7B model reportedly outperformed OpenAI's GPT 3.5, thousands of researchers from leading AI labs gathered at NeurIPS (Conference and Workshop on Neural Information Processing Systems) in New Orleans, eager to showcase their breakthroughs.
As these researchers celebrate what promises to be an exhilarating year for AI in 2024, policymakers face the daunting task of regulating this rapidly advancing field without stifling innovation. The thrilling trajectory of Mistral serves as a stark reminder of the urgent need for regulatory bodies to move swiftly.