Dynamic inconsistency of the inaction and initial state baseline — AI Alignment Forum