Lorec comments on AI models inherently alter “human values.” So, alignment-based AI safety approaches must better account for value drift

[Lorec]

Are you familiar with Yudkowsky’s/​Miles’s/​Christiano’s AI Corrigibility concept?