Aligning Agents, Tools, and Simulators — LessWrong