(Structural) Stability of Coupled Optimizers — LessWrong