Today's headline came as a total surprise to many. It also casts doubts on the key argument against the regulation of foundation models. One that almost resulted in the complete abolishment of the idea that was initially pitched by the European Parliament. A few quick thoughts.
To start with, I am rather confused. Did not the French Government and the European Commission tell us for weeks that the Foundation Models chapter in the #AIAct (= the excellent Spanish presidency proposal Vol 1) needs to be heavily reduced in it's scope in order to safeguard the few ‘true independent EU champions’ that we have in that field? Without those changes, we would miss our only chance to catch up, they said. Mistral AI would be forced to close the open access to their models and would need to start to cooperate with US/Chinese Tech corporation as they are no longer able to comply with the #AIAct alone. The result? A broad exemption from most obligations for the French company as decided on 6 December 2023.
Fast forward to 26 February 2024: Mistral AI becomes the next OpenAI, also partnering with Microsoft. Likewise, the open access to its models seems to be a thing from the past. Strategically, a smart move from the French company, maybe even a necessary one in the current digital economy. Almost no EU startup seem to manage its scale-up without US/Chinese support. Nothing to criticize here from my side.
BUT the French company is certainly no longer fitting in the ‘true independent EU champion’ category that people in the #AIAct trilogue had in mind when they agreed on the final text. The EU legislator got played again or rather checkmated itself … despite many, many warnings from us / stakeholders that this could happen.
Hopefully this episode means that the flawed #digital #sovereignty argument that was debunked again and again but is nevertheless still being used in EU policymaking is finally gone for good! 🙏
Read more about the deal at The Verge.