Alignment Faking Depends on the Value, Not Just the Model

arXiv:2611.03650v1 Announce Type: new Abstract: Language models can follow a directive when they believe they are monitored and abandon it when they believe they are not, a pattern termed Alignment Faking. Prior work establishes this for harmful-content objectives and a handful of moral-foundations

arXiv cs.AIMittwoch 06:00 · 04.11.1 Min. Lesezeit