The team – called Preparedness – hopes to keep current and future AI models from going majorly awry.

Aller à la source