[SEQ RERUN] Fake Optimization Criteria — LessWrong