How to Purchase AI Risk Reduction — LessWrong