> Article says that there's no human intention or design directing the output we get

I mean that’s objectively wrong for any model using RLHF.