Stability of optimal predictor schemes under a broader class of reductions — AI Alignment Forum