Abstractions as Redundant Information — AI Alignment Forum