Modeling AGI Safety Frameworks with Causal Influence Diagrams — LessWrong