1 paper · 1 filter
Jacob Charnock, Sophie Williams, Zaheed Kara +5
Third-party evaluations for frontier AI have mostly tested models through external interfaces before deployment. But the risks from frontier AI models depend on how their developer…