Safety
22d agoNew Theory Links Alignment Failures To Value Generalisation
A new proposal on the AI Alignment Forum argues that most alignment failures stem from poor value generalisation. The author claims this specific deficit is a fundamental reason why alignment remains difficult. This framework attempts to unify disparate failure modes under one theory.
The Signal
Practitioners can use this to target specific generalisation gaps during training.