Research

Why Open Weights Matter More Than You Think

The European case for building AI models you can actually inspect, modify, and deploy on your own terms.

Mathilde Renaud · January 15, 2025 · 8 min read

Last November at Station F, I watched a three-person team from Lyon deploy a fine-tuned 7B model on a single consumer GPU, adapted to French medical terminology, serving diagnostics to rural clinics — no proprietary APIs, under two hundred euros a month.

The Transparency Argument

The urgent case for open weights is practical, not ideological. European regulation — the AI Act, sector-specific compliance in healthcare and finance — demands auditable systems. You cannot audit what you cannot see. Proprietary models turn every compliance question into a trust exercise with a foreign vendor.

“You cannot audit what you cannot see. Proprietary models turn every compliance question into a trust exercise with a foreign vendor.”
This is the Mistral Le Chat French AI design system, applied by Curio Design — a design-style library for AI agents. Full Mistral Le Chat French AI guide → designbycurio.com/learn/mistral-le-chat-orange