Why I'm excited about Redwood Research's current project — LessWrong