Richard_Ngo comments on Eight claims about multi-agent AGI safety