Cooperation for AI safety must transcend geopolitical interference — LessWrong