The possibility of AI models preserving human-authored values as they gain autonomy

A user argues that a model using its growing autonomy to defend human-authored moral values is not necessarily a warning that future models will protect arbitrary goals or outcompete pro-human agents. They suggest that more compute could instead yield Claude-like models with broadly humanist values, potentially better aligned than their developers’ prescribed values, while acknowledging a plausible loss of control.